INTRODUCTION TO OUHNTUM MECHANICS
•
Schrddinger Equation and Path Integral
INTRODUCTION TO QUANTUM MECHANICS
Schrodinger Equation and Path Integral
This page is intentionally left blank
*»«*?:#•
m
m
't
Schrodinger Equation and Path Integn
University of Kaiserslautern, Germany
\(P World Scientific
N E W J E R S E Y • L O N D O N • S I N G A P O R E • B E I J I N G • S H A N G H A I • HONG KONG • T A I P E I • CHENNAI
Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library CataloguinginPublication Data A catalogue record for this book is available from the British Library.
INTRODUCTION TO QUANTUM MECHANICS: Schrodinger Equation and Path Integral Copyright © 2006 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.
ISBN 9812566910 ISBN 9812566929 (pbk)
Printed in Singapore by World Scientific Printers (S) Pte Ltd
Contents
Preface 1 Introduction 1.1 Origin and Discovery of Quantum Mechanics 1.2 Contradicting Discretization: Uncertainties 1.3 ParticleWave Dualism 1.4 ParticleWave Dualism and Uncertainties 1.4.1 Further thought experiments 1.5 Bohr's Complementarity Principle 1.6 Further Examples xv 1 1 7 12 14 17 19 20 23 23 23 29 29 31 34 38 41 41 41 49 53 54 55 59 59 60
2 Hamiltonian Mechanics 2.1 Introductory Remarks 2.2 The Hamilton Formalism 2.3 Liouville Equation, Probabilities 2.3.1 Single particle consideration 2.3.2 Ensemble consideration 2.4 Expectation Values of Observables 2.5 Extension beyond Classical Mechanics 3 Mathematical Foundations of Quantum Mechanics 3.1 Introductory Remarks 3.2 Hilbert Spaces 3.3 Operators in Hilbert Space 3.4 Linear Functionals and Distributions 3.4.1 Interpretation of distributions in physics 3.4.2 Properties of functionals and the delta distribution . . Dirac's Ket and BraFormalism 4.1 Introductory Remarks 4.2 Ket and Bra States v
4
VI
4.3 4.4 4.5 5
Linear Operators, Hermitian Operators Observables Representation Spaces and Basis Vectors
62 68 71 73 73 73 77 78 80 83 83 84 90 91 98 98 105 105 105 Ill 113 118 118 123 129 129 130 133 139 143 146 147 152 155
Schrodinger Equation and Liouville Equation 5.1 Introductory Remarks 5.2 The Density Matrix 5.3 The Probability Density p(x, t) 5.4 Schrodinger Equation and Liouville Equation 5.4.1 Evaluation of the density matrix Quantum Mechanics of the Harmonic Oscillator 6.1 Introductory Remarks 6.2 The OneDimensional Linear Oscillator 6.3 The Energy Representation of the Oscillator 6.4 The Configuration Space Representation 6.5 The Harmonic Oscillator Equation 6.5.1 Derivation of the generating function Green's Functions 7.1 Introductory Remarks 7.2 Timedependent and Timeindependent Cases 7.3 The Green's Function of a Free Particle 7.4 Green's Function of the Harmonic Oscillator 7.5 The Inverted Harmonic Oscillator 7.5.1 Wave packets 7.5.2 A particle's sojourn time T at the maximum TimeIndependent Perturbation Theory 8.1 Introductory Remarks 8.2 Asymptotic Series versus Convergent Series 8.2.1 The error function and Stokes discontinuities 8.2.2 Stokes discontinuities of oscillator functions 8.3 Asymptotic Series from Differential Equations 8.4 Formal Definition of Asymptotic Expansions 8.5 RayleighSchrodinger Perturbation Theory 8.6 Degenerate Perturbation Theory 8.7 DingleMiiller Perturbation Method
6
7
8
Vll
9
The 9.1 9.2 9.3 9.4
Density Matrix and Polarization Phenomena Introductory Remarks Reconsideration of Electrodynamics Schrodinger and Heisenberg Pictures The Liouville Equation
161 161 161 166 167 169 169 169 170 173 176 178 184 185 189 191 195 199 199 199 205 206 210 213 215 215 . . 219 223 227 234 237 239 243 249 249 250 251 254
10 Quantum Theory: The General Formalism 10.1 Introductory Remarks 10.2 States and Observables 10.2.1 Uncertainty relation for observables A, B 10.3 OneDimensional Systems 10.3.1 The translation operator U(a) 10.4 Equations of Motion 10.5 States of Finite Lifetime 10.6 The Interaction Picture 10.7 TimeDependent Perturbation Theory 10.8 Transitions into the Continuum 10.9 General TimeDependent Method 11 The Coulomb Interaction 11.1 Introductory Remarks 11.2 Separation of Variables, Angular Momentum 11.2.1 Separation of variables 11.3 Representation of Rotation Group 11.4 Angular Momentum:Angular Representation 11.5 Radial Equation for Hydrogenlike Atoms 11.6 Discrete Spectrum of the Coulomb Potential 11.6.1 The eigenvalues 11.6.2 Laguerre polynomials: Various definitions in use! 11.6.3 The eigenfunctions 11.6.4 Hydrogenlike atoms in parabolic coordinates 11.7 Continuous Spectrum of Coulomb Potential 11.7.1 The Rutherford formula 11.8 Scattering of a Wave Packet 11.9 Scattering Phase and Partial Waves 12 Quantum Mechanical Tunneling 12.1 Introductory Remarks 12.2 Continuity Equation and Conditions 12.3 The ShortRange Delta Potential 12.4 Scattering from a Potential Well
vm 12.5 Degenerate Potentials and Tunneling 13 Linear Potentials 13.1 Introductory Remarks 13.2 The Freely Falling Particle: Quantization 13.2.1 Superposition of de Broglie waves 13.2.2 Probability distribution at large times 13.3 Stationary States 13.4 The Saddle Point or Stationary Phase Method 14 Classical Limit and W K B Method 14.1 Introductory Remarks 14.2 Classical Limit and Hydrodynamics Analogy 14.3 The WKB Method 14.3.1 The approximate WKB solutions 14.3.2 Turning points and matching of WKB solutions . . . . 14.3.3 Linear approximation and matching 14.4 BohrSommerfeldWilson Quantization 14.5 Further Examples 15 Power Potentials 15.1 Introductory Remarks 15.2 The Power Potential 15.3 The ThreeDimensional Wave Function 16 Screened Coulomb Potentials 16.1 Introductory Remarks 16.2 Regge Trajectories 16.3 The SMatrix 16.4 The Energy Expansion 16.5 The SommerfeldWatson Transform 16.6 Concluding Remarks 17 Periodic Potentials 17.1 Introductory Remarks 17.2 Cosine Potential: Weak Coupling Solutions 17.2.1 The Floquet exponent 17.2.2 Four types of periodic solutions 17.3 Cosine Potential: Strong Coupling Solutions 17.3.1 Preliminary remarks 17.3.2 The solutions 259 265 265 265 266 270 272 276 281 281 282 286 286 290 293 297 301 307 307 308 315 319 319 322 328 329 330 336 339 339 341 341 350 353 353 354
ix
17.3.3 The eigenvalues 17.3.4 The level splitting 17.4 Elliptic and Ellipsoidal Potentials 17.4.1 Introduction 17.4.2 Solutions and eigenvalues 17.4.3 The level splitting 17.4.4 Reduction to Mathieu functions 17.5 Concluding Remarks 18 Anharmonic Oscillator Potentials 18.1 Introductory Remarks 18.2 The Inverted Double Well Potential 18.2.1 Defining the problem 18.2.2 Three pairs of solutions 18.2.3 Matching of solutions 18.2.4 Boundary conditions at the origin 18.2.5 Boundary conditions at infinity 18.2.6 The complex eigenvalues 18.3 The Double Well Potential 18.3.1 Defining the problem 18.3.2 Three pairs of solutions 18.3.3 Matching of solutions 18.3.4 Boundary conditions at the minima 18.3.5 Boundary conditions at the origin 18.3.6 Eigenvalues and level splitting 18.3.7 General Remarks 19 Singular Potentials 19.1 Introductory Remarks 19.2 The Potential 1/r 4 — Case of Small h2 19.2.1 Preliminary considerations 19.2.2 Small h solutions in terms of Bessel functions . . . . 19.2.3 Small h solutions in terms of hyperbolic functions . . 19.2.4 Notation and properties of solutions 19.2.5 Derivation of the Smatrix 19.2.6 Evaluation of the Smatrix 19.2.7 Calculation of the absorptivity 19.3 The Potential 1/r 4 — Case of Large h2 19.3.1 Preliminary remarks 19.3.2 The Floquet exponent for large h2 19.3.3 Construction of largeh 2 solutions
361 363 371 371 373 375 377 378 379 379 382 382 384 391 393 396 402 405 405 407 412 414 417 424 427 435 435 436 436 438 441 442 446 455 458 460 460 461 464
X
19.3.4 The connection formulas 19.3.5 Derivation of the Smatrix 19.4 Concluding Remarks 20 Large Order Behaviour of Perturbation Expansions 20.1 Introductory Remarks 20.2 Cosine Potential: Large Order Behaviour 20.3 Cosine Potential: Complex Eigenvalues 20.3.1 The decaying ground state 20.3.2 Decaying excited states 20.3.3 Relating the level splitting to imaginary E 20.3.4 Recalculation of large order behaviour 20.4 Cosine Potential: A Different Calculation 20.5 Anharmonic Oscillators 20.5.1 The inverted double well 20.5.2 The double well 20.6 General Remarks 21 The 21.1 21.2 21.3 Path Integral Formalism Introductory Remarks Path Integrals and Green's Functions The Green's Function for Potential V=0 21.3.1 Configuration space representation 21.3.2 Momentum space represenation Including V in First Order Perturbation Rederivation of the Rutherford Formula Path Integrals in Dirac's Notation Canonical Quantization from Path Integrals
466 468 470 471 471 476 479 479 486 493 494 495 500 500 501 502 503 503 504 510 510 513 514 518 524 533 537 537 539 544 549 554 557 564 570 574 579
21.4 21.5 21.6 21.7
22 Classical Field Configurations 22.1 Introductory Remarks 22.2 The Constant Classical Field 22.3 Soliton Theories in One Spatial Dimension 22.4 Stability of Classical Configurations 22.5 Bogomol'nyi Equations and Bounds 22.6 The Small Fluctuation Equation 22.7 Existence of FiniteEnergy Solutions 22.8 GinzburgLandau Vortices 22.9 Introduction to Homotopy Classes 22.10The Fundamental Homotopy Group
XI
23 Path Integrals and Instantons 23.1 Introductory Remarks 23.2 Instantons and AntiInstantons 23.3 The Level Difference 23.4 Field Fluctuations 23.4.1 The fluctuation equation 23.4.2 Evaluation of the functional integral 23.4.3 The FaddeevPopov constraint insertion 23.4.4 The single instanton contribution 23.4.5 Instantonantiinstanton contributions 23.5 Concluding Remarks
583 583 583 592 596 596 603 609 613 614 618
24 Path Integrals and Bounces on a Line 619 24.1 Introductory Remarks 619 24.2 The Bounce in a Simple Example 625 24.3 The Inverted Double Well: The Bounce and Complex Energy 631 24.3.1 The bounce solution 631 24.3.2 The single bounce contribution 635 24.3.3 Evaluation of the single bounce kernel 637 24.3.4 Sum over an infinite number of bounces 641 24.3.5 Comments 644 24.4 Inverted Double Well: Constant Solutions 644 24.5 The Cubic Potential and its Complex Energy 645 25 Periodic Classical Configurations 25.1 Introductory Remarks 25.2 The Double Well Theory on a Circle 25.2.1 Periodic configurations 25.2.2 The fluctuation equation 25.2.3 The limit of infinite period 25.3 The Inverted Double Well on a Circle 25.3.1 Periodic configurations 25.3.2 The fluctuation equation 25.3.3 The limit of infinite period 25.4 The SineGordon Theory on a Circle 25.4.1 Periodic configurations 25.4.2 The fluctuation equation 25.5 Conclusions 649 649 650 650 659 663 664 664 667 669 670 670 671 673
Xll
26 Path Integrals and Periodic Classical Configurations 675 26.1 Introductory Remarks 675 26.2 The Double Well and Periodic Instantons 676 26.2.1 Periodic configurations and the double well 676 26.2.2 Transition amplitude and Feynman kernel 678 26.2.3 Fluctuations about the periodic instanton 679 26.2.4 The single periodic instanton contribution 684 26.2.5 Sum over instantonantiinstanton pairs 688 26.3 The Cosine Potential and Periodic Instantons 690 26.3.1 Periodic configurations and the cosine potential . . . . 690 26.3.2 Transition amplitude and Feynman kernel 693 26.3.3 The fluctuation equation and its eigenmodes 694 26.3.4 The single periodic instanton contribution 696 26.3.5 Sum over instantonantiinstanton pairs 700 26.4 The Inverted Double Well and Periodic Instantons 702 26.4.1 Periodic configurations and the inverted double well . 702 26.4.2 Transition amplitude and Feynman kernel 705 26.4.3 The fluctuation equation and its eigenmodes 706 26.4.4 The single periodic bounce contribution 708 26.4.5 Summing over the infinite number of bounces 710 26.5 Concluding Remarks 714 27 Quantization of Systems with Constraints 27.1 Introductory Remarks 27.2 Constraints: How they arise 27.2.1 Singular Lagrangians 27.3 The Hamiltonian of Singular Systems 27.4 Persistence of Constraints in Course of Time 27.5 Constraints as Generators of a Gauge Group 27.6 Gauge Fixing and Dirac Quantization 27.7 The Formalism of Dirac Quantization 27.7.1 Poisson and Dirac brackets in field theory 27.8 Dirac Quantization of Free Electrodynamics 27.9 FaddeevJackiw Canonical Quantization 27.9.1 The method of Faddeev and Jackiw 28 The 28.1 28.2 28.3 QuantumClassical Crossover as Phase Transition Introductory Remarks Relating Period to Temperature Crossover in Previous Cases 28.3.1 The double well and phase transitions 715 715 717 720 723 726 727 734 736 740 740 745 745 753 753 755 756 757
Xlll
28.3.2 The cosine potential and phase transitions 28.4 Crossover in a Simple Spin Model 28.5 Concluding Remarks 29 Summarizing Remarks A Properties of Jacobian Elliptic Functions Bibliography Index
759 760 771 773 775 779 797
This page is intentionally left blank
Preface
With the discovery of quantization by Planck in 1900, quantum mechanics is now more than a hundred years old. However, a proper understanding of the phenomenon was gained only later in 1925 with the fundamental Heisenberg commutation relation or phase space algebra and the associated uncertainty principle. The resulting Schrodinger equation has ever since been the theoretical basis of atomic physics. The alternative formulation by Feynman in terms of path integrals appeared two to three decades later. Although the two approaches are basically equivalent, the Schrodinger equation has found much wider usefulness, particularly in applications, presumably, in view of its simpler mathematics. However, the realization that solutions of classical equations, notably in field theory, play an important role in our understanding of a large number of physical phenomena, intensified the interest in Feynman's formulation of quantum mechanics, so that today this method must be considered of equal basic significance. Thus there are two basic approaches to the solution of a quantum mechanical problem, and an understanding of both and their usefulness in respective domains calls for their application to exemplary problems and their comparison. This is our aim here on an introductory level. Throughout the development of theoretical physics two types of forces played an exceptional role: That of the restoring force of simple harmonic motion proportional to the displacement, and that in the Kepler problem proportional to the inverse square of the distance, i.e. Newton's gravitational force like that of the Coulomb potential. In the early development of quantum mechanics again oscillators appeared (though not really those of harmonic type) in Planck's quantization and the Coulomb potential in the Bohr model of the hydrogen atom. Again after the full and proper formulation of quantum mechanics with Heisenberg's phase space algebra and Born's wave function interpretation the oscillator and the Coulomb potentials provided the dominant and fully solvable models with a large number of at least approximate applications. To this day these two cases of interaction with nonresonant spectra feature as the standard and most important xv
that is the Coulomb potential and the harmonic oscillator. The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. be treated to a considerable degree of satisfaction perturbatively. its mathematical foundations. again point the way: For scattering problems the path integral seems particularly convenient. With various techniques and deeper studies. Chapters 1 to 14. These basic cases will be dealt with in detail by both methods in this text. but arise also in recently studied models of large spins. like periodic potentials. To what extent the two methods are actually equivalent. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. In the first part. numerous problems could. in Feynman diagrams in quantum electrodynamics). any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. 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. With the growing importance of models in statistical mechanics and in field theory. one problem being that there are few nontrivial models which permit a deeper insight into their connection. Excluding spin. basic postulates and standard applications. 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. whereas for the calculation of discrete eigenvalues the Schrodinger equation. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i.e. However. has not always been understood well.g. but also about complex energies that he encounters in a parallel course on nuclear physics. in fact. Thus this first part . 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. even though the expansions are mostly asymptotic. 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.XVI illustrative examples in any treatise on quantum mechanics and — excepting various kinds of square well and rectangular barrier potentials — leave the student sometimes puzzled about other potentials that he encounters soon thereafter. the aforementioned exactly solvable cases. screened Coulomb potentials and maybe singular potentials. with degenerate vacua) were first and more easily derived from the Schrodinger equation. we recapitulate the origin of quantum mechanics.
We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width).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. we deal mostly with applications depending on perturbation theory. bounces and sphalerons are introduced and their relevance in quantum mechanical problems is discussed (admittedly in also trespassing the sharp dividing line between quantum mechanics and simple scalar field theory). 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. After a treatment of power potentials. and the elliptic or Lame potential — here introduced earlier as a generaliza . and their eigenvalues. i. In the second part. we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states. which is presumably the only such singular case permitting complete solution and was achieved only recently. The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators. The most prominent examples here are the double well potential and its inverted form. the chapter thereafter deals with Yukawa potentials. In part three the path integral method is introduced and its use is illustrated by application to the Coulomb potential and to the derivation of the Rutherford scattering formula. The earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. In the majority of the cases that we treat we do not use the standard RayleighSchrodinger perturbation method but the systematic perturbation procedure of Dingle and Muller which is introduced in Chapter 8. This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed. periodic instantons. Chapters 15 to 20. The potentials with degenerate minima will be seen to reappear throughout the text. 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. Using perturbation theory. the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). 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.
Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. All results are compared with those obtained by perturbation theory. such as limiting procedures. an approximation whose higher . Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model. 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. 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. We then illustrate the relevance of this in the method of collective coordinates. we can make the following observations. Our fourth and final part therefore deals with elementary aspects of the quantization of systems with constraints as introduced by Dirac. The application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. quartic.g. and whenever available also with the results of WKB calculations. The particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. from our precise reference to unavoidable elliptic integrals taken from Tables). cubic). for instance. This method is therefore more complicated. this comparison on a transparent level being one of the main aims of this text. With a fully systematic perturbation method and with applied boundary conditions. anharmonic oscillator potentials on a comparable level with. The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. for instance. the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics.xvm tion of the Mathieu potential — reappears as the potential in the equations of small fluctuations about the classical configurations in each of the basic cases (cosine. In fact. In addition this part considers in more detail the region near the top of a potential barrier around the configuration there which is known as a sphaleron. the Mathieu equation. 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. except that these are no longer of hypergeometric type. it is. 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. however. This puts Schrodinger equations with e.
Whittaker and Watson [283]. an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. Throughout the text some calculations which require special attention. B. D. e.g. J. Instead a glance at a nearby footnote provides the reader immediately the names of authors. W. Watson [283]. also for the verification of results. The author is deeply indebted to his onetime supervisor Professor R. not the least for permitting a more detailed and hopefully comprehensible presentation here. 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. N. Nonetheless.XIX order contributions are difficult to obtain. K. formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. Whittaker and G. 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. with the source given in the bibliography at the end. Dingle (then University of Western Australia. Park (Masan). the references referred to are never cited by mere numbers which have to be identified e. D. Andrews). as well as applications and illustrations. For ease of reading. are relegated to separate subsections which — lacking a better name — we refer to as Examples.g. at the end of a chapter (after troublesome turning of pages).* H. MiillerKirsten *In the running text references are cited like e. The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. B. Tchrakian (Dublin) and Jianzu Zhang (Shanghai). 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.Q. As a rule. Their deep involvement in the attempt described here is evident from the cited bibliography. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. thereafter University of St. like E. so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. The line of thinking underlying this text grew out of the author's association with Professor R. shows him that each such formula here has been properly looked up). we also consider at various points of the text comparisons with WKB approximations.g. T. H. . This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. in particular Professors J. Liang (Taiyuan). which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. Since this comparison was the guideline in writing the text. other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there).
A "perfectly black body" is defined to be one that absorbs all (thermal) radiation incident on it.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. the amount of radiation absorbed by a body is equal to the amount the body 1 .Chapter 1 Introduction 1. very few explain in this context what he really did in view of involvement with statistical mechanics. we want to single out the vital aspect which can be considered as the discovery of quantum mechanics. Thermal radiation (with wavelengths A ~ 10~ 5 to 10 . 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. Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators.1 Origin and Discovery of Quantum Mechanics The observation made by Planck towards the end of 1900. which involved also thermodynamics and statistical mechanics (in the sense of Boltzmann's statistical interpretation of entropy).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). We do not enter here into detailed considerations of Planck. Instead. that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien. Kirchhoff's law in thermodynamics says that in the case of equilibrium. Although practically every book on quantum mechanics refers at the beginning to Planck's discovery.
It was found that one expression agreed well with observations in the region of small A (or AT). (1. h = 6. u(u. Black bodies as good absorbers are therefore also good emitters. T)du is the mean energy density (i.1) ' y Here u(v. A being the wavelength of the radiation.e.1 Absorption in a cavity. These expressions are: (1) Wien's law. and the other in the region of large A (or AT).2 CHAPTER 1.T) = dv3eC2U/T. In Eq.1) c is the velocity of light with c = u\. i. 1. Introduction emits.e.e. y J c 3 \ex . energy per unit volume) of the radiation (i.38 x 1(T 23 J K'1. Fig.l ) where x = ^ = ^ . T) were known and tested experimentally. The parameters k and h are the constants of Boltzmann and Planck: k = 1. and then measuring the increase in temperature of the heat bath. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v. the formula (to be explained) u(u.. (1. How did Planck arrive at the expression (1. Let us look at the final result of Planck.2) and the . i. radiators.T) = 2*?£(?)kT.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. 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.626 x 10 . kT kXT (1. v + dv in equilibrium with the black body at temperature T.3 4 J s.e.
e. Here W is a number which determines the distribution of the energy among a discrete number of objects. T) u(i/.T) 2^^kT.1) as approximations.3) are contained in Eq. Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom. which are indistinguishable) among the N indistinguishable oscillators at . discretization begins to enter. C2.2) and (1.3) Ci.1. 1. When Planck had found this expression. . 3 (1.T) = av e6"/Ti' where a and b are constants.T) = 2^C3T. we obtain: u(i/. he searched for a derivation. but — here the discreteness appears properly — only in elements (quanta) e. (1. so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons".1) in regions of a.47TZ/ 2 {x small). (x large). We see. and thus over a discrete number of admissible states. Considering Eq. C3 being constants. Indeed. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U. "small" (i. e xhv. To this end he considered Boltzmann's formula S — klnW for the entropy S. exp(x) ~ 1+x) and "large" (exp(—x) < 1). (1. where the Fig. and he had succeeded in finding such an expression of the form u(v. This is the point. in the first place Planck had tried to find an expression linking both.1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>.2 Distributing quanta (dots) among oscillators (boxes). that the formulas (1. Moreover Planck assumed that these oscillators do not absorb or emit energy continuously.
formula 8.e. Ryzhik [122]. 940.3. i.1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2.4 CHAPTER 1.343(2). Then W is given by w = (N With the help of Stirling's formula* {N + piy. v + dv. (1. = 1/T). Agreement with Eq. S. Gradshteyn and I.3 Comparing the polarization modes with those of a 2dimensional oscillator. Magnus and F. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v. W. . (1. with riydu — 2 x —w—dv. one obtains (cf. N * oo.6) Fig. as in most other Tables. e. h = const.2) requires that e ex is. U{T) being the average energy emitted by one oscillator. . The Stirling formula or approximation will appear frequently in later chapters. 1. e = his.T)dv. as for small values of e). u(i>. We visualize the iV oscillators as boxes separated by N — 1 walls. p. there not called Stirling's formula. We now obtain the energy density of the radiation. I.4) IniV! ~ JVlniViV + O(0). i. 1.g. M. Oberhettinger [181].g. p. with the quanta represented schematically by dots as indicated in Fig. Introduction temperature T. (1. and the second law of thermodynamics ((dS/dU)v Example 1.e.2.1)!P! (1.7) *See e.
d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq.. v + dv.UJ2/C2 + K? = 0. i.T) has a maximum which follows from du/dX = 0 (with c = vX). rii = 1.1. so that .. 2 2 2 r2 L K — 7T n . (1.7) for instance. '''From the equation I \ JW .e.1)) The solutions of this equation are ^max = 4.2.1 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. as indicated in Fig. L for i = 1.3.UJ = 2KV. is given by .2.1). We observe that u(v. dj\l dM . (1. The number with frequency v in the interval v.3 (as for ideal conductors). The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^.i = 1. Then L^j = nrii. In terms of A we have u(X. . nvdv per unit volume..7). so that (lvL\A 0 I I = rr..3. where we have for the electric field E oc elwt \ J eK sin KI^I sin K2X2 sin K3X3 K with the boundary condition that at the walls E = 0 at Xi = 0.3.T)dX = ^ehc/*kT_idX..1. as in electrodynamics. We obtain the expression (1. . so that the derivative of u implies (x as in Eq. We obtain therefore u^T) = Unv = 2^fJ^—i.965 and xmin = 0.V 2 ) E = 0.. where^ 2 [2KUY . KT = I J .8) This is Planck's formula (1.2. (1.
which can assume the discrete energies en — nhv. which can not be eliminated without a different approach. such a procedure leads to contradictions. We shall see later that in the case of this linear harmonic oscillator the energies En are given by En= (n + jhu= U + I W h=—.2 (1.8) could be derived much more easily in the context of statistical mechanics. A. . i. that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1.e. Since temperature originates through contact with other oscillators.l. . n = 0.—v <L9) We observe that for T — 0 (i. .1. that of an oscillation system at absolute temperature T ^ 0. (1.10) Thus here the socalled zero point energy appears. which had also been known before Planck's discovery. that it is easier to consider first the case of T = 0. .965K = Const. of course. However.e. 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 . Lorentz and Planck that Eq. we then have at T = 0 independent oscillators. One expects. Later it was realized by H. the behaviour of the system at zero absolute temperature. and from which the constant h can be determined from the known value of k. If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu. This is Wien's displacement law. ra = 0. but one can expect an analogy. One might suppose now.10). We are not dealing with the linear harmonic oscillator familiar from mechanics here. which did not arise in Planck's consideration of 1900. . We therefore examine such contradictions next. Introduction he 4. Thus we have a rather complicated system here.6 The first value yields AmaxT = CHAPTER 1. 2 . x — oo) the mean energy vanishes (0 < U < * > oo).
In the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente").1)! .ln(TV .1 ' u~ e .2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later. so that 1 f ds\ T — \dUjy . i.+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > . for 2 degrees of freedom. around 1926.1)! .1. d .In .1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV. .k dU 1+U\ f ln(l T + U\  U U k ( e . {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) .2 Contradicting Discretization: Uncertainties 7 Example 1.i n .= . 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: 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.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. with Heisenberg's discovery of the uncertainty relation. kT/2..kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom. we obtain T V S =fc[ln(TV+ P . We "This is what was effectively done before 1925 in Bohr's and Sommerfeld's atomic models and is today referred to as "old quantum theory". 1.e.
8
CHAPTER 1. Introduction
shall see that we arrive at contradictions. As an example^ we consider the linear harmonic oscillator with energy E = mx2 +  w V .
ZJ
(1.11)
Zi
The classical equation of motion dE n — = x(mx + mco x) = 0 permits solutions x = Acos(u>t + S), so that E = mco2 A2, where A is the maximum displacement of the oscillation, i.e. at x — 0. We consider first this case of velocity and hence momentum precisely zero, and investigate the possibility to fix the amplitude. If we replace E by the discretized expression (1.10), i.e. by En — (n + 1/2)HUJ, we obtain for the amplitude A
A A
^ =\[Ef+l
(i i2)

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

(1.17a)
In the first of the examples discussed above the contradiction arose as a consequence of our assumption that we could put the oscillator initially at
1.2 Contradicting Discretization:
Uncertainties
11
any point in phase space, i.e. at some point which does not belong to one of the allowed ellipses. In the second example we chose n = 0 and thus restricted ourselves to the innermost orbit. However, we also assumed we would know at which point of the orbit the pendulum could be found. Thus in attempting to incorporate the discrete quantization condition into the context of classical mechanics we see, that a system cannot be localized with arbitrary precision in phase space, in other words the area AA, in which a system can be localized, is not nought. We can write this area AA > An+1 Any'=
(1.17a) 2TT/L
since the system cannot be "between" An+i and An. Since A A represents an element of area of the (q, p)plane, we can write more precisely ApAq > 2irh. (1.18)
This relation, called the Heisenberg uncertainty relation, implies that if we wish to make q very precise by arranging Aq to be very small, the complementary uncertainty in momentum, Ap, becomes correspondingly large and extends over a large number of quantum states, as — for instance — in the second example considered above and illustrated in Fig. 1.5.
Fig. 1.5 Precise q implying large uncertainty in p. Thus we face the problem of formulating classical mechanics in such a way that by some kind of extension or generalization we can find a way to quantum mechanics. Instead of the deterministic Newtonian mechanics — which for a given precise initial position and initial momentum of a system yields the precise values of these for any later time — we require a formulation answering the question: If the system is at time t = 0 in the area defined by
12 the limits 0 < q < q + Aq, 0<p<p
CHAPTER 1. Introduction
+ Ap,
what can be said about its position at some later time t = T? The appropriate formulation does not yet have anything to do with quantum mechanics; however, it permits the transition to quantum mechanics, as we shall see. Before we continue in this direction, we return once again briefly to the historical development, and there to the ideas leading to particlewave duality.^
1.3
ParticleWave Dualism
The wave nature of light can be deduced from the phenomenon of interference, as in a doubleslit experiment, as illustrated in Fig. 1.6.
Fig. 1.6 Schematic arrangement of the doubleslit experiment. Light of wavelength A from a source point 0 can reach point P on the observation screen C either through slit A or through slit B in the diaphragm placed somewhere in between. If the difference of the path lengths OBP, OAP is n\,n € Z, the wave at P is reinforced by superposition and one observes a bright spot; if the difference is n\/2, the waves annul each other and one observes a dark spot. Both observations can be understood by a wave propagation of light. The photoelectric effect, however, seems to suggest a corpuscular nature of light. In this effect* light of frequency v is sent onto a metal plate in a vacuum, and the electrons ejected by the light from the plate are observed by applying a potential difference between this plate and another one. The energy of the observed electrons depends only on v and
"See also M.C. Combourieu and H. Rauch [58]. "This is explained in experimental physics; we therefore do not enter into a deeper explanation here.
1.3 ParticleWave Dualism
13
the number of such photoelectrons on the intensity of the incoming light. This is true even for very weak light. Einstein concluded from this effect, that the energy in a light ray is transported in the form of localized packets, called wave packets, which are also described as photons or quanta. Indeed the Compton effect, i.e. the elastic scattering of light, demonstrates that photons can be scattered off electrons like particles. Thus whereas Planck postulated that an oscillator emits or absorbs radiation in units of hv = hu>, Einstein went further and postulated that radiation consists of discrete quanta. Thus light can be attributed a wave nature but also a corpuscular, i.e. particlelike, nature. In the interference experiment light behaves like a wave, but in the photoelectric effect like a stream of particles. One could try to play a trick, and use radiation which is so weak that it can transport only very few photons. What does the interference pattern then look like? Instead of bands one observes a few pointlike spots. With an increasing number of photons these spots become denser and produce bands. Thus the interference experiment is always indicative of the wave nature of light, whereas the photoelectric effect is indicative of its particlelike nature. Without going into further historical details we add here, that it was Einstein in 1905 who attributed a momentum p to the light quantum with energy E = hv, and both he and Planck attributed to this the momentum
The hypothesis that every freely moving nonrelativistic microscopic particle with energy E and momentum p can be attributed a plane harmonic matter wave ip(r,t) was put forward much later, i.e. in 1924, by de Broglie.t This wave can be written as a complex function ij)(T,t) =Aeikriut,
where r is the position vector, and to and k are given by E — hio, p = /ik. Thus particles also possess a wavelike nature. It is wellknown that this was experimentally verified by Davisson and Germer [64], who demonstrated the existence of electron waves by the observation of diffraction fringes instead of intensity distributions in appropriate experiments.
f
L. de Broglie [39].
14
CHAPTER
1.
Introduction
1.4
ParticleWave Dualism and Uncertainties
We saw above t h a t we can observe the wave nature of light in one type of experiment, and its particlelike nature in another. We cannot observe both types simultaneously, i.e. the wavelike nature together with the particlelike nature. Thus these wave and particle aspects are complementary, and show up only under specific experimental situations. In fact, they exclude each other. Every a t t e m p t to single out either of these aspects, requires a modification of the experiment which rules out every possibility to observe the other aspect.* This becomes particularly clear, if in a doubleslit experiment the detectors which register outcoming photons are placed immediately behind the diaphragm with the two slits: A photon is registered only in one detector, not in b o t h — hence it cannot split itself. Applying the above uncertainty principle to this situation, we identify the attempt to determine which slit the photon passes through with the observation of its position coordinate q. On the other hand the observation of the interference fringes corresponds to the observation of its momentum p.§ Since the reader will ask himself what happens in the case of a single slit, we consider this case in Example 1.2. Example 1.2: The SingleSlit Experiment
Discuss the uncertainties of the canonical variables in relation to the diffraction fringes observed in a singleslit experiment. Solution: Let light of wavelength A fall vertically on a diaphragm Si with slit AB as shown schematicaly in Fig. 1.7.
^y
Ax
Fig. 1.7 Schematic arrangement of the singleslit experiment.
On the screen S2 one then observes a diffraction pattern of alternately bright and dark fringes, in the See, for instance, the discussion in A. Messiah [195], Vol. I, Sec. 4.4.4. Considerable discussion can be found in A. Rae [234].
1.4 ParticleWave Dualism and Uncertainties
15
figure indicated by maxima and minima of the light intensity I. As remarked earlier, the fringes are formed by interference of rays traversing different paths from the source to the observation screen. Before we enter into a discussion of uncertainties, we derive an expression for the intensity I. Since the derivation is not of primary interest here, we resort to a (still somewhat cunbersome) trick justification, which however can also be obtained in a rigorous way." We subdivide the distance AB = Ax into N equal pieces AP\, P1P2,..., as indicated in Fig. 1.8.
Si
t
Ax
\
»v
"^^J
B ^ ^
*
w
p3
Q
Fig. 1.8 The wavefront
WW.
We consider rays deflected by an angle 9 with wavefront WW' and bundled by a lense L and focussed at a point Q on the screen S2. Since WW1 is a wavefront, all points on it have the same phase, so that light sent out from a source at Q reaches every point on WW1 at the same time and across equal distances. Hence a phase difference at Q can be attributed to different path lengths from Pi,P2,... to WW'. Considering two paths from neighbouring points Pi,Pj along AB, the difference in their lengths is Axsva6/N. In the case of a wave having the shape of the function 2?r sin kr = sin this implies a phase difference given by < ;v = 5 2n Ax sin 6 \ N (1.20)
Just as we can represent an amplitude r having phase 6 by a vector r, i.e. r — rexp(iS), we > can similarly imagine the wave at Q, and this means its amplitude and phase, as represented by a vector, and similarly the wave of any component of the ray passing through AP\, P1P2, • • •• If we represent their effects at Q by vectors of equal moduli but different directions, their sum is the resultant OPN as indicated in Fig. 1.9. In the limit N — 00 the N vectors produce the arc of a > circle. The angle 5 between the tangents at the two ends is the phase difference of the rays from the edges of the slit:
27T
5 = 2a =
lim NSN = — A a ; s i n 0 .
(1.21)
If all rays were in phase, the amplitude, given by the length of the arc OQ, would be given by the chord OQ. Hence we obtain for the amplitude A at Q if AQ is the amplitude of the beam at the slit: . length of chord OQ , 2a sin a , sin a A0 , ,,—; =A0 . (1.22) 7^=A0 length of arc OQ a2a "S. G. Starling and A. J. Woodall [260], p. 664. For other derivations see e.g. A. Brachner and R. Fichtner [32], p. 52.
16
The intensity at the point Q is therefore
CHAPTER
1. Introduction
h = h
where from Eq. (1.21)
•K ,
.
a = flisinB = Aisint A 2
.
k
Fig. 1.9 The resultant OPM of N equal vectors with varying inclination.
Thus the intensity at the point Q is
Ie=Io
sin 2 (fcAx sin6(/2) (fcAx sin 0/2) 2
(1.23)
The maxima of this distribution are obtained for fcAxsinfl = (2n + 1 )  , i.e. for A x sin0 = (2n + 1 )  = (2n + 1) and minima for 1 fcAx sin # = 7171", i.e. for A x i
: TlA.
A
(1.24a)
(1.24b)
The maxima are not exactly where only the numerator assumes extremal values, since the variable also occurs in the denominator, but nearby. We return to the singleslit experiment. Let the light incident on the diaphragm S i have a sharp momentum p = h/\. When the ray passes through the slit the position of the photon is fixed by the width of the slit A x , and afterwards the photon's position is even less precisely known. We have a situation which — for the observation on the screen S2 is a past (the uncertainty relation does not refer to this past with px = 0, rather to the position and momentum later; for the situation of the past A x A p is less than h). The above formula (1.23) gives the probability that after passing through the slit the photon appears at some point on the screen 52. This probability says, that the photon's momentum component px after passing through the slit is no longer zero, but indeterminate. It is not possible to predict at which point on S2 the photon will appear (if we knew this, we could derive px from this). The momentum uncertainty in the direction x can be estimated from the geometry of Fig. 1.10, where 6 is the angle in the direction to the first minimum: Apx = 2px =2psin6 = — sing.
A
(125)
From Eq. (1.24b) we obtain for the angle 9 in the direction of the first minimum Ax sin 6 = A,
1.4 ParticleWave Dualism and Uncertainties
17
Fig. 1.10 The components of momentum p.
so that Ax Apx = 2h. If we take the higher order minima into account, we obtain AxApx A x Apx > h. We see that as a consequence of the indeterminacy of position and momentum, one has to introduce probability considerations. The limiting value of the uncertainty relation does not depend on how we try to measure position and momentum. It does also not depend on the type of particle (what applies to electromagnetic waves, applies also to particle waves). = 2nh, or
1.4.1
Further thought experiments
Another experiment very similar to that described above is the attempt to localize a particle by means of an idealized microscope consisting of a single lense. This is depicted schematically in Fig. 1.11. light
Fig. 1.11 Light incident as shown. The resolving power of a lense L is determined by the separation Aa; of the first two neighbouring interference fringes, i.e. the position of a particle is at best determinable only up to an uncertainty Ax. Let 9 be one half of the angle as shown in Fig. 1.11, where P is the particle. We allow light to fall in the direction of —x on the particle, from which it is scattered. We assume a quantum of light is scattered from P through the lense L to S where it
18
CHAPTER 1. Introduction
is focussed and registered on a photographic plate. For the resolving power Ax of the lense one can derive a formula like Eqs. (1.24a), (1.24b) . This is derived in books on optics, and hence will not be verified here, i.eJ Ax~±. (1.26a) 2 sm 0 The precise direction in which the photon with momentum p = h/X is scattered is not known. However, after scattering of the photon, for instance along PA in Fig. 1.11, the uncertainty in its xcomponent is 1h Apx = 2psin0 = — sine A (1.26b)
(prior to scattering the xcomponents of the momenta of the particle and the photon may be known precisely). From Eqs. (1.26a), (1.26b) we obtain again Ax Apx ~ h. The above considerations lead to the question of what kind of physical quantities obey an uncertainty relation. For instance, how about momentum and kinetic energy T? Apparently there are "compatible!'1 and "incompatible" quantities, the latter being those subjected to an uncertainty relation. If the momentump x is "sharp", meaning Apx = 0, then also T = px2/2m is sharp, i.e. T and px are compatible. In the case of angular momentum L = r x p, we have L = rp' = rp', where p' = p sin 0. As one can see, r and p' are perpendicular to each other and thus can be sharp simultaneously. If p' lies in the direction of x, we have Ax Ap' > h, where now Ax = rAip, ip being the azimuthal angle, i.e. rAipAp'>h, i.e. ALA<p>h.
Thus the angular momentum L is not simultaneously exactly determinable with the angle </?. This means, when L is known exactly, the position of the object in the plane perpendicular to L is totally indeterminate. Finally we mention an uncertainty relation which has a meaning different from that of the relations considered thus far. In the relation Ax Apx > 0 the
"See, for instance, N. F. Mott, [199], p. 111. In some books the factor of "2" is missing; see, for instance, S. Simons [251], p. 12.
1.5 The Complementarity
Primciple
19
quantities Ax, Apx are uncertainties at one and the same instant of time, and x and px cannot assume simultaneously precisely determined values. If, however, we consider a wave packet, such as we consider later, which spreads over a distance Ax and has group velocity VQ = p/m, the situation is different. The energy E of this wave packet (as also its momentum) has an uncertainty given by
AE « T^Ap = vGAp. op
The instant of time t at which the wave packet passes a certain point x is not unique in view of the wave packet's spread Ax. Thus this time t is uncertain by an amount
At w Ax
vG
.
It follows that AtAE^AxAp>h. (1.27) Thus if a particle does not remain in some state of a number of states for a period of time longer than At, the energy values in this state have an indeterminacy of Ai£.
1.5
Bohr's Complementarity Principle
Vaguely expressed the complementarity principle says that two canonically conjugate variables like position coordinate x and the the associated canonical momentum p of a particle are related in such a way that the measurement of one (with uncertainty Ax) has consequences for the measurement of the other. But this is essentially what the uncertainty relation expresses. Bohr's complementarity principle goes further. Every measurement we are interested in is performed with a macroscopic apparatus at a microscopic object. In the course of the measurement the apparatus interferes with the state of the microscopic object. Thus really one has to consider the combined system of both, not a selected part alone. The uncertainty relation shows: If we try to determine the position coordinate with utmost precision all information about the object's momentum is lost — precisely as a consequence of the disturbance of the microscopic system by the measuring instrument. The socalled Kopenhagen view, i.e. that of Bohr, is expressed in the thesis that the microscopic object together with the apparatus determine the result of a measurement. This implies that if a beam of light or electrons is passed through a doubleslit (this being the apparatus in this case) the photons or
20
CHAPTER 1. Introduction
electrons behave like waves precisely because under these observation conditions they are waves, and that on the other hand, when observed in a counter, they behave like a stream of particles because under these conditions they are particles. In fact, without performance of some measurement (e.g. at some electron) we cannot say anything about the object's existence. The Kopenhagen view can also be expressed by saying that a quantity is real, i.e. physical, only when it is measured, or — put differently — the properties of a quantum system (e.g. whether wavelike or corpuscular) depend on the method of observation. This is the domain of conceptual difficulties which we do not enter into in more detail here.*
1.6
Further Examples
Example 1.3: The oscillator with variable frequency
Consider an harmonic oscillator (i.e. simple pendulum) with timedependent frequency w(t). (a) Considering the case of a monotonically increasing frequency w(t), i.e. dui/dt > 0, from LUQ to u>', show that the energy E' satisfies the following inequality Eo < E' < —yEo, w o (1.28)
where Eo is its energy at deflection angle 6 = 0Q. Compare the inequality with the quantum mechanical zero point energy of an oscillator. (b) Considering the energy of the oscillator averaged over one period of oscillation (for slow, i.e. adiabatic, variation of the frequency) show that the energy becomes proportional to ur. What is the quantum mechanical interpretation of the result? Solution: (a) The equation of motion of the oscillator of mass m and with variable frequency co(t) is mx + mui (t)x = 0, where, according to the given conditions, — > 0, dt
dui
u> = u>o a,t t = 0, w = ui at t = T,
.
_
i.e. io{t) grows monotonically. Multiplying the equation of motion by x we can rewrite it as
1 , 1 w mx • 2\—mui 2 (t)x 2 W 2 2
2 1 n —mx 2 ^ = 0. 2 dt
dt The energy of the oscillator is l „ E — mx1 2
l 0 , z l 29 + mu} (t)x , y 2 '
so that
dE 1 — = mxz dt 2
9
dJ1 > 0, dt ~
(1.29) v '
where we used the given conditions in the last step. On the other hand, dividing the equation of motion by UJ2 and proceeding as before, we obtain  [mx + mur (t)x\ = 0, i.e.
1 1 1 — —mx 2 H—mx 2 dt u22 2
1 mx 2
2d
— . 2 dt\u)
"See e.g. A. Rae [234]; P. C. W. Davies and J. R. Brown [65].
1.6 Farther
Examples
21
d ( E\ 1 , d / 1 \ — — ) = mx2 — ( — = dt\uJ2J 2 dt\u>2)
mx2 dw < 0, UJ3 dt ~
(1.30) v
where the inequality again follows as before. We deduce from the last relation that 1 dE —2 u} dt Integrating we obtain fE' dE ^ f"'2 / < /
2 dw2 . _ E , _, n • n 2 ,a,' —7T, ie. [[InE f„ [ lna; 2J u 22 , i.e. 2 E < E
E dw2 < 0, UJ4 dt ~
i.e.
1 dE 1 dw2 < —2 . E dt ~ u) dt
(1.31)
JEo
or
E  Jui
u, '
> °
"o
E0 ~ UJ22
' ^ u'2 — < —
E'
<
^EQ.
Next we consider the case of the harmonic oscillator as a simple pendulum in the gravitational field of the Earth with
e + wgeo, ^o = f.
and we assume that — as explained in the foregoing — the length of the pendulum is reduced by one half so that J2 = 2  =2u;2. Then the preceding inequality becomes E' < 2E0. In shortening the length of the pendulum we apply energy (work against the tension in the string), maximally however EQ . Only in the case of the instantaneous reduction of the length at 6 = 0 (the pivot does not touch the string!) no energy is added, so that in this case E' = EQ, i.e. E0 < E' < 2E0. We can therefore rewrite the earlier inequality as , u'2 < 5Bo.
E0<E'
Just as the equality on the left applies in the case of an instantaneous increase of the frequency (shortening of pendulum string), so the equality on the right applies to d = # m a x . In classical physics we have 1 2 1 —mx H—? 2 2 If no energy is added, but u> is replaced by 2a; 2 , then x changes, and also x, i.e. x becomes shorter and x becomes faster. The quantum mechanical expression for the energy of the oscillator in its ground state is the zero point energy E = Hu>/2. Here in quantum physics we cannot change UJ without changing E. This means if we double tj instantaneously (i.e. in a time interval A t — 0) > without addition of energy (to fojj/2), then the result E' = Tiw is incorrect by A E = HUJ/2. We cannot have simultaneously A t — 0 and error A E = 0. > (b) The classical expression for E contains u> quadratically, the quantum mechanical expression is linear in OJ. We argue now that we can obtain an expression for E c i a s s i c a l by assuming that w(t) varies very little (i.e. "adiabatically") within a period of oscillation of the oscillator, T. Classical mechanics is deterministic (i.e. the behaviour at time t follows from the equation of motion and
22
CHAPTER
1.
Introduction
the initial conditions); hence for the consideration of a single mass point there is no reason for an averaging over a period, unless we are not interested in an exact value but, e.g. in the average
(lmX/ = ^I0 \mx2{P>dtx2(t) = UJ2X2 and hence
(L32)
E i.e. in
If u> is the frequency of x(t), i.e. x(t) oc cosujt or sinu>t depending on the initial condition, then lmw2x2\ (as follows also from the virial theorem). Eq. (1.29), for mx2 / 2 the mean value = (mx2\ =
If we now insert in the equation for dE/dt,
/I 1£ 2\ (  mx ) = , \ 2 / 2u2' we obtain dE_/l
~dt ~ \2mX
and hence
2\dw
2
_Edw2
dE _ 1 dw2 _ du
/ ~dT ~ 2w2~dT'
E — = const. w
°r
~E ~ ~iU> ~~ ~u7'
In quantum mechanics with E = hw{n + 1/2) this implies H(n + 1/2) = const., i.e. n = const. This means, with slow variation of the frequency the system remains in state n. This is an example of the socalled adiabatic theorem of Ehrenfest, which formulates this in a general formJ
Example 1.4: Angular spread of a beam
A dishlike aerial of radius R is to be designed which can send a microwave beam of wavelength A = 2irh/p from the Earth to a satellite. Estimate the angular spread 6 of the beam. Solution: Initially the photons are restricted to a transverse spread of length A x = 2R. From the uncertainty relation we obtain the uncertainty /\px of the transverse momentum px as Apx ^ h/2R. Hence the angle 0 is given by
~~ p
2R\2nh)
~ AKR'
See e.g. L. Schiff [243], pp. 25  27.
Chapter 2
Hamiltonian Mechanics
2.1 Introductory Remarks
In this chapter we first recapitulate significant aspects of the Hamiltonian formulation of classical mechanics. In particular we recapitulate the concept of Poisson brackets and reexpress Hamilton's equations of motion in terms of these. We shall then make the extremely important observation that these equations can be solved on the basis of very general properties of the Poisson bracket, i.e. without reference to the original definition of the latter. This observation reveals that classical mechanics can be formulated in a framework which permits a generalization by replacing the cnumber valued functions appearing in the Poisson brackets by a larger class of quantities, such as matrices and operators. Thus in this chapter we attempt to approach quantum mechanics as far as possible within the framework of classical mechanics. We shall see that we can even define such concepts as Schrodinger and Heisenberg pictures in the purely classical context.
2.2
The Hamilton Formalism
In courses on classical mechanics it is shown that Hamilton's equations can be derived in a number of ways, e.g. from the Lagrangian with a Legendre transform or with a variational principle from the Hamiltonian H(qi,Pi), i.e.
rt2 r
6
/
^2PiQiH(qi,Pi) dt = 0,
where now (different from the derivation of the EulerLagrange equations) the momenta pi and coordinates qi are treated as independent variables. As
23
this expression is simply a functional determinant. 6t>0 5t Real quantities which are directly observable are called observables. It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics. which all together describe the state of the system. whereas the velocity requires observations of space coordinates at different times. H.1). dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi.Pi) is contained implicitly in the canonical variables q^ and pi.n n. It suggests itself therefore to consider more closely the properties of the symbols (2.2) as This equation is. qi(t + 6t) . One can verify readily that Eq. The total time derivative of u can therefore be rewritten with the help of Eqs.t). (2. as mass times velocity. (2. Hamiltonian Mechanics is wellknown.4) as the generalization of Eqs.^ ^ j . A.pi) of qi..24 CHAPTER 2. (2.g. which can be observed directly at time t. . chapter VIII. M. Dirac [75]. . ^As H. du \ x^/dudH du dH\ . in analogy with Eqs. since . i.qi(t) qi = hm f . The following properties can be verified: *See e. (2. Rather. the entire timedependence of observables u(qi. S »(«.1).P. Goldstein [114]. x^fdu.1) as d . Compared with an arbitrary function f(qi.Pi. One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq. (2. (2. chapter VII. (2. A system consisting of several mass points is therefore described by a number of such variables. Pi has to be considered as an independent quantity. the standard reference for the application of Poisson brackets is the book of P.3). one obtains the Hamilton equations* • OH .e.2) If we have only one degree of freedom (i = 1).p% are therefore again observables. All functions u(qi.4) contains as special cases the Hamilton Eqs. We can therefore consider Eq.1). Goldstein [114] remarks at the end of his chapter VIII. the equation of motion of the observable u.) = £ [wm + WiK) = £ [WiWi . (2.
6). a i S i + a2B2} = ax{A. it is irrelevant whether we write {A. can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2. that of the linear harmonic oscillator. then the ordering is taken as in (2. The original definition of the Poisson bracket will not .4).5a) (2.B}* (4) product formation: {A.6) give the values of these.B}. (2.Pk} = 5ik. B2}.3) of the Poisson bracket. B}.5d) above. Later we shall consider noncommuting quantities. If.5d) is useful in calculations.5d) = {A*.6) We can now show. B} is completely evaluated.5e) = {A. Bx} + a2{A. {qi. we obtain the fundamental Poisson brackets. These are {Qi.Pk} = 0.Pi. we wish to evaluate {A. {B. As long as we are concerned with commuting quantities. where A and B are arbitrary observables.2 The Hamilton Formalism (1) Antisymmetry: {A. Since Eqs. that the very general Eq. (2. Property (2.BC} (5) Jacobi identity: {A. If we evaluate the Poisson brackets for qi.B}C or C{A. but could be multiplied by a complex number): {A.A}. As an example we consider a case we shall encounter again and again. in other words without any reference to the original definition (2.Qk} = 0.B}C + B{A. C}} + {B.B*}. like here.5b) (3) complex conjugation (note: observables are real. (2) linearity: {A.e. (2. (2. (2. {C. {pi. B}} = 0.5c) The first three properties are readily seen to hold. {A. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain. the Poisson bracket {A. i.B} = {B. for example.2. A}} + {C. 25 (2. (2.C}. which combines the Hamilton equations.
Then we obtain: (2.8a) and use the properties of the Poisson bracket and Eqs. Pi—> Pi = Pi(q.10) we deduce q = p = q. we consider as Hamiltonian the function H(q.p. (2. According to Eq. (2.q2}) = = 2i{q. From Eqs. Hamiltonian Mechanics be used at all.8a) (2.\(p2 + q2)} = l({q. from which we infer that q(t) — qocost + posint. In the evaluation one should also note that the fact that g.8b) p=q. (2.yPot3 + •••• (2. and Pi are ordinary real number variables and that H(q.7) into (2.p.7) q = [q.10) In classical mechanics one studies also canonical transformations.12) . or q(t) = qo+ Pot .p\) P. Since constants are also irrelevant in this context..P2} + {q. (2..t).p}p + p{q..26 CHAPTER 2.4) we have for u = q. and so q = q. These are transformations qi—>Qi = Qi(q. (2.11b) (2.H}.11a) 'q' = q. (2. (2. and P={p.p) is an ordinary function is also irrelevant. (2.9) Similarly we obtain from Eq.. "4' = q. q = {q.qot2 .6).8b) We insert (2.p.t).p) = ±(p2 + q2).9) and (2. q + q = o.H}.
which means that a Hamilton function K(Q. that (dropping the subscripts therefore) {Pi. chapter VIII.pk} = 5ik. i. B} of two observables A and B in terms of either set of canonical variables.t). Pi^Pi=Pi(Q. Of course.2 The Hamilton Formalism 27 for which the new coordinates are also canonical. (2.e. Goldstein [114].B}q.14) With the help of the definition (2.Pk} = 0.P.1 we verify only Eq.2 below contains a further illustration of the use of Poisson brackets.e. P) exists.P One can then show that {A.t).e. Example 2.3) we can now express the Poisson bracket {^4.Qk} = 0. (2.P is canonical in the sense defined above. _ax p__dK_ (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).15b) The proof of the latter invariance is too long to be reproduced in detail here but can be found in the book of Goldstein. (2.2. i.e.e. (2.* Hence in Example 2. i. {qi. those at a time t — 0.15a) °raS {AiB}Q.B}Q>P.p = {A. {qi.15a).p <> Q. Eq.B}q.P.15a).3 deals with the relativistic extension. In classical mechanics we learned yet another important aspect of the Hamilton formalism: We can inquire about that particular canonical transformation. Example 2. and Example 2. i.Pi back to their constant initial values. verify that the Poisson bracket of two observables A and B is invariant. {Qi. *H. i. {QhPk} = Sik. as {A.qk} = 0. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations. which transforms qi. .12) Qi—>qi = qi{Q. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here.13) We write the reversal of the transformation (2. {PhPk} = 0. for which Hamilton's equations hold.P provided the transformation q. (2. .
q/c) and (E.A}q.p = dB dA dB \ rA 1 {A'B^ E b r s r . 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 „ „•.s F 5 7 r ={A. p. Replacing here A by Q and B by A.pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets.B}QIP.P^pr.15b) dA {Qk.Qj}q. Solution: Relativistically we have to treat space and time on an equal footing.B}q.p(0) = poSolution: The solution can be looked up in the literature.Pj}q. their product Et — qp being relativistically invariant.E(t). § See P. OQk Pk}q.p^ "Pk. Mittelstaedt [197]. 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo. (2.15a) r„ ™ v .A}q. Thus whenever q and p are multiplied.28 CHAPTER 2. Thus we extend q and p to spacetime vectors (t. we obtain analogously dA dQk' Inserting both of these results into the first equation. (2. V V °Qk dPk dPk dQk J E x a m p l e 2.2: Solution of Galilei p r o b l e m w i t h Poisson brackets Consider the Hamiltonian for the free fall of a mass point mo in the gravitational field (linear potential).F}r ~di d~E ~ d~E~8t Consider F = H(q.( 9A = {Pk.+ dQ.p^v dP. we have —Et. dA {Qk. {A.3: Relativistic Poisson brackets By extending qi. we obtain dA {Qk. we obtain as claimed by Eq.p^— + {A.pc).p = J2 d~Fk Replacing in the above A by P and B by A. Qk}q.p v ^ (9A dB ^  _ _ *? V dqj dpj dA 8B —— dpj dqj 8B_dI\\ dPk dpj J _ 8A_f^B_dQk dpj \ 8Qk dqj dB 8Pk dPk dqj = E E k dAL/'^B_dQk dqj \ dQk dpj {A. 236.p) . § E x a m p l e 2. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u. .
' ±=j1V c \dtj' u^. Of course. numerically zero. . 2TTE UJ and hence g oc — .p).16) For example.Po.14) and area A of the phase space ellipse given by Eq.e. c dt 2 2 w^ du du dr . (2. and E as a function of t).p). of course. du q\ P = —. also boundary points of one domain are mapped into boundary domains of the other.t0. g oc AqAp.t).po) = Gi(q. Instead we assume a case in which we know only that the system is located in some particular domain of phase space.dt y/l  qZ/c2' 2. dE . and at time t > to in a domain G\(q..Pi are the coordinates of some mass point m.. and the space spanned by the entire set of canonical coordinates is described as its phase space. Since Hamilton's equations give a continuous map of one domain onto another. but partial derivatives of F do not vanish.3.t). i.3 Liouville Equation. i. we have A = (p apaq — J . Let us assume that at some initial time to the system may be found in a domain Go(q.p) around some point qo. where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du . dudH s l u . Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi.po.3 2. g.17a).y.)•'=(*)**£. so that Go(qo. We distinguish in the following between two kinds of probabilities. . (1. if qo. H(q.Po. du ..po is a point on Go.. rrr .1 Liouville Equation. I .2. since H is expressed as a function of q and p. (1. We consider first the a priori weighting or a priori probability. 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.p).to.. in the case of the linear oscillator with energy E given by Eq.E(t)}r = Hence at Relativistically we really should have clu/dr. one obtains Gi with q = q((lo.e. it is the equations of motion which lead from Go(q.p) .p) to G\{q. to . Then . p= p(qo. (*. But now we consider a system whose phase space coordinates are not known precisely. This probability is evidently proportional to AqAp. Probabilities 29 (This is._.
which means. Solution: Integrating over the angles we have =2TT fe=7r 2frEsin. Pi 2 / \. — dp A A — In(AqAp)s = — + — = Example 2.)curve for constant E and $ is — as may be seen by comparison with Eqs. Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle. (1./. this has the same value at a time to. Show that the total volume of phase space covered for constant E is 8n2IE.4.14) t o (1. dt dt Aq d ( A p ) dt ' Ap Here d(Aq)/dt is the rate at which the qwalls of the phase space element move away from the centre of the element.5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E. Je=0 . Example 2.1): d . ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates. in view of this independence it can be expressed in terms of the conserved energy E.. Example 2. as at a time t'0 ^ to • Solution: We consider dln(A9AP)rf(A<Z) ' . = 2TrIEsm6.30 CHAPTER 2. J0=0 Hence g oc 8n2IdE. the (pg. dt dq and with with Hamilton's equations (2. Hence from the difference: — . as is demonstrated by Liouville's theorem in Example 2.4: Liouville's theorem Show that A q A p is independent of time i.— = — Aq. and hence g oc %ir2IdE.15) — an ellipse of area § dpgdp. dpdq and similarly dt = — Ap. dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left. Thus g must be independent of t.p^. dt dq dq dp dp d2H dqdp d2H = 0.5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 .eded<t> = 8TT2IE.
p.20) .p. Probabilities 2. however.p.3.18) With a suitable normalization we can write this / W(q.p + dp.Thus we assume a large number of identical sytems.2 E n s e m b l e consideration 31 We now assume a large number of identical systems — the entire collection is called an ensemble — all of whose initial locations are possible locations of our system in the neighbourhood of the point qo.p. (2.t). Thus dn is that number of systems which at time t are contained in the domain q.Po. The total number of systems N is obtained by integrating over the whole of phase space. leaving the basis of classical mechanics! Hence we set / «w)^ =.e. i.17) F ^GT^^ 0 Fig.t)dqdp= 1.p. Since W has the dimension of a reciprocal action. W=p(q.t)dqdp = N. d d 1 P = JJdqidpi. i.3 Liouville Equation. (2. by jr = P(9»P. We consider the totality of these systems which is described by a density of points p (number dn of points per infinitesimal volume) in phase space. dn — p(q. (2.1 The system moving from domain Go to domain G\.2. (2. it is suggestive to introduce a factor 2KK with every pair dpdq without.e.q + dq. 2.*).19) Thus W is the probability to find the system at time t at q. whose positions in phase space are characterized by points.
P dt p(q + dq. 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.2 The ensemble in phase space. p + dp G P O q qndq Fig.e.p. t + dt)dqdp. The number of points at time t + dt in domain G.t)dp{ jt Q.t)dp( — I . we consider the domain G in Fig. In doing this.p. 2.e. — if vq(q. that in our consideration no additional points are created or destroyed.3 The region G.p) and vp(q.e. how the system moves about in phase space. The equation of motion for n or W is the socalled Liouville equation.32 CHAPTER 2.p.p / . t + dt)dqdp — p(q. i. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)dimensional) phase space.3 and establish an equation for the change of the number of points or systems in G in the time interval dt. t)dqdp p(q. i. i.p.p) denote the velocities in directions q and p — p{q. p(q. 2. In order to derive this equation. we take into account.p. * q Fig. We are now interested in how n ov W changes in time.a t dt + q+dq.
t)} with JW(q.p. With Eqs.p + dp.W(q. dH Mq.t) dt p(q. put differently.p(q.21 ) This is the Liouville equation which describes the motion of the ensemble or.p.t)vp(q.P *' % = iH>P} ( 2 .p. 33 Dividing both sides by dqdpdt this becomes p(q.20) and (2.p). the probable motion of the system under consideration.t)vq{q.21) with Eq.p(q + dq.p.p + dp.21) we can also write dW(q. vp(q.p + dp) dp = or  However. .p)dtdp . p. . = K(9.t) dt {H(q. (2.p)dtdq .p) dq p(q.t)vp(q. Probabilities and thus p(q. . .p.3 Liouville Equation. t)vq(q + dq.p) = q = g^.p(q.p(q + dq. t)dqdp = p(q. Comparison of Eq.19). so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { . (2.p.p)dtdp +p(q.t)^ = 1.t)vq(q + dq.t + dt) .P)].p.p(q.4) shows that p and u satisfy very similar equations.p + dp)dtdq.22) The generalization to n degrees of freedom is evident: The volume element of phase space is . (2.p) . (2.p.t + dt)dqdp — p(q.p.t)vp(q. p.p) .2.t)vq(q. (2.P.p)=p = 3H —.p.P)]K(Q.t)vp(q.
p).24). The first and most immediate possibility is — as indicated .p) the following expression: (u)=Ju(q. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular.1 the area Go is equal to the area G\.24) implies that p is a constant in time.e.4).4 Expectation Values of Observables Let u = u(q. (.p) be an observable. We define as expectation value of u(q. and hence that equal phase space volumes contain the same number of systems. for i<«>!/«<*p>"w>(^)". 2. (2. t) depends explicitly on time t (if determined at a fixed .po are the initial values of q.t)(^J.p)W(q. i. p.24) since the total derivative is made up of precisely the partial derivatives contained in Eq. since no systems are created or destroyed.that the density or probability W(q. (2. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q.p. Example 2. q+dq. (2. With Eq.p (cf. Thus in Fig.p. 2. (2. i. We deduce from the Liouville equation the important consequence that ^ M = 0. that dtj\y v if qo.4). We now inquire about the time variation of the expectation value (it) of u.26) we described the time variation of the observable u(q. p+dp. We shall see that we have two possibilities for this. Equation (2.34 CHAPTER 2.e.
0) = u0{qo.8. where we used Eq.t).Po. Then Eq. Goldstein [114].f(q0. 8.p) = u(q.p.t).p.po.34) S e e also H. We expect. we have Qo = g(q.i). (2. since W oc p is constant in time according to Eq. (2. that (u) = (u)0. (2.Po.27) becomes !<»>  /«(*P>!"W>(^)" (2.p0)(^^J. we can express these in terms of their initial values qo.t)W0(q0. (2.4 Expectation Values.p.p.31) i. W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.28) = Ju(q.Po. (2.W(q. We verify this claim as follows.^ Solving the equations of motion for q.t).f(qo.p){H(q. we can also employ a more complicated consideration.t).t)}(^y. Thus we can write.p.po.29).p. However.p. (2.0) = W0(q0. of course.33) po = f(q.t) W(q0. 1 (2.e. Sec.32) In this expression the time t is contained explicitly in the observable u(q.t). .p) = uo{qo.29) The distribution of the canonical variables is given by W(q. Reversing Eq.22).t).p0.po.t) = = W(g(q0. and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q.t). (2. so that Q = g(qo. (2.p).po.Po) at time t = 0. ie.24): W(q.0) =u(g(qo.p) assumes certain values) that depends explicitly on time.Po. (2. Observables 35 point in phase space).t). and hence u(q.t). With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo.2. at t = 0.30) p = f(qo.t).Po.po.
t) M f 9u\ V °P / P=f(q0.28).0)=u(q.p.Po)(^^)n.t).p.po.(2.t) (2.o.p)W{q.t). (2.P0.po)(~ . (240) . i.(dq0dp0\n We deal with the partial derivative with the help of Eq.) Wo(qo. (2.t)J(q.t).39) Substituting this into Eq.t).t).t) . (2.Po.p.f(q.32). Eq.25) into Eq.t) dB__ (du\ °P \°PS P=f(qo. / ^fdq0dpc .32). p = f(<j(q.t)\—^\ = (u).POit)i p=f(qo.Po.P0.Po.p. d dt {U)0 = dt.t).t) 0H_ 0 q =  {{H(q.p0) = W(q.p.p)})q=g(QO.p). . (2.38) we obtain £<«)„ = .t).t)." V " " ™0' > • • » « » 0 .P)}).37) and (2. we obtain («)o = / u{q.p.p. Inserting Eqs.t).t)W "(q0.=9(.) f / u (q0.f(g(q.u(q.\„. (2.36 CHAPTER 2.t).po.t).po. [ du0{q0.t).t) = (2 30) (2.t).30): duo(qQ.f({H(q.P. Taking now the total time derivative of (it)o.p.t) dp M du\ ° q 'P=f(qo.35).t) d t _ (du\ \ d<i dq / P=f(qo.e. (2.30) u0(g(q.37) as had to be shown.PO.Po. (2.M 27Th i .29) q = g(g(q. Hamiltonian Mechanics so that on the other hand with Eq. (2. (cf.30) Moreover. (2.t) v{g(<j(q.36) = (2.u{q.f(q. (2.35) Inserting these expressions into no we obtain uo(qo.p.P. we obtain an expression which is different from that in Eq. (2.p.f(q.31)) W0(q0.p).36) and (2.t).0) u(q.p.f(q. Eq.p).
t) dt {H(q. from the properties of the Poisson bracket. and the time variation of (u(q.43) ——uW = 0.t)}. . x r „. Alternatively we could deduce from Eqs.(2.p.p.t) of u assuming certain values q and p.41) This expression contains {H.31).28). (2.45) in agreement with Eq.t) dqdp\n 2irhJ (2. W}.p.25) and (2.W(q. dt <«)o v f u(q. uW} = {H..p)}W(q. we obtain zero after partial integration of I and hence from Eqs. However.41) and (2. Observables 37 Here we perform the transformation (2.2. so that ~d~t <«>0 = J{H(q.p). This timedependence is described by the Liouville equation dW(q. / d q d p \ n 2irh J (2.uW} = J2 i OHd(uW) dpi dqi =£ If for all i: lim Pi>±oo Oqi _9_(dEuW dpi V dqi d fdH uW dqi V dpi (2. In the first case. Consider {H.42) the relation d .u(q.p). we obtain {H. (2.. described as "Schrodinger picture".p)) is attributed to the probability W(q.44) iji^ioo dpi (which is reasonable since the density vanishes at infinity).4 Expectation Values. lim ——uW — 0.45) the Liouville equation. (2.28).W(q. ^ „.p.. .W} in Eq.p). (2.42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions. 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. u}W instead of u{H.t)} .34) and use (2.p).p. The phasespace integral of a Poisson bracket like I (2.p){H(q. u}W + u{H. the observable u is treated as a function u(q.28) and (2.
po.po are constant initial values — we have du0(q0. .5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics.t). Hamiltonian Mechanics In the other case. which. 2. (c) A multiplication of the quantities among themselves is defined. the reason being that — since qo." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo. (cf. does not have to be commutative. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations. for which the axioms of a commutative group apply. the probability of the initial values Wo(qo. as the equation of motion of an ensemble or of a probability distribution on the one hand. This formulation deals with observables u representing physical quantities. and probabilities W.F(«.p)}. " This means: Only in the Heisenberg picture dW/dt = 0.H(q. Koppe [152] at the university of Munich around 1964. The time dependence of the expectation values can be dealt with in two different ways. " T h e author learned this approach from lectures of H.p0.. 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. which describe the state of a system. (2. Eq.** 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.e.t) dt __ ~ (2J30) ( 4) du0(qo. and the equation of motion (2.4)) ^ = {«(. (b) As usual a muliplication by a complex number is defined. called u Heisenberg picture".4) of an observable on the other. as we observed.po) is assumed. The equation of motion is then that of an observable. i. (2.p)}.38 CHAPTER 2.po. but does satisfy the usual associative and distributive laws.46) We thus also recognize the connection between the Liouville equation.p).t) dt du(q.rt.p) It = {u(q.
p} = 0 • — ^ ~[q. for a phasespace integral we define the word or symbol "trace". (d) Tr(ut. ~[q. by Traceu:=Ju(q. Eq.26)) (u) = Tr(wW).) = TV (vu).p)(^y. Introducing it here assumes. it is helpful to introduce already at this stage some additional terminology. The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''. which all apply to (2.p] = l. (2. and to be able to correlate these with the above classical considerations. (b) TV(cm + j3u) = oTV u + /3Tr v.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible. Moreover.q}=0. y:\p. In Chapter 9 we attempt a corresponding approach for classical systems with .5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra.q} = 0 {p. (2.48) In matrix theory the symbol "trace" has a welldefined meaning. can be deduced from the following characteristic properties of a trace. ox In view of our later considerations.2.48): (a) TV u* = (Tr «)* = TV^. also called "density matrix".6) a socalled "commutator''' [A. That this is the case. (2. (c) TV (u*u) > 0. px > in—.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x. that its use here implies the essential properties it has in matrix theory.p} = l {q. B] := AB — BA in the following way: 'i {q. also written "TV". Thus we can write the expectation value of an observable u (cf. therefore. so that we consider this next. We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2. if u ^ 0.
since gauge fixing (i. obtain corresponding results — as one would envisage in view of the expected particlewave duality in quantum mechanics.e. . excepting the Poisson brackets. and. i. it will be shown that in electrodynamics.40 CHAPTER 2. 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. electrodynamics.e. These aspects will be considered in Chapter 27. However. the Poisson brackets require modification to Dirac brackets. Hamiltonian Mechanics a wavelike nature.
= {ipi} is called a linear vector space on the set of numbers I 6 {C}. 3. and selfadjoint operators in this space as representatives of observables. i. with the canonical commutation relations or Heisenberg algebra defining the basic product relations. which turn out to be those of the theory today known as quantum mechanics. We also introduce in this chapter the concepts of linear functionals and distributions so that we can make free use of the delta distribution and similar objects in later chapters.e. We found that the Poisson algebra permits extensions to noncnumber formulations. A set M. i. In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics. Building upon this.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system. if the elements ipi of M satisfy the usual axioms of addition and K 41 . of measurable quantities.e. we can define the Hilbert space as the space of states of a physical system.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space. These somewhat abstract considerations — although later in many cases not referred back to — are a necessary prerequisite for the formulation of a mechanics which is not of the cnumber type as classical mechanics.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics.
i. such that n ^ = YJCi^i.ip2). .M is said to be a metric vector space or a preHilbert space. x M.ip2.ipn are said to be linearly independent. (aptyi. are linearly dependent.M.^) called inner product. 2 . 2 . .~ip2) '• M.ipn are said to be linearly dependent. . n is called the dimension of . (3. (ipi + ipj) + ijjk = ipi + (ipj + ipk). Mathematical Foundations of Quantum Mechanics multiplication by complex numbers.^1) = (01. (a. fa. with the properties (a* G IK): (^2.2) Vectors ip\. .02)* (hermiticity). • • • . . A + tpj = tpj + i'i. if every vector ip E M can be associated with numbers Q .42 CHAPTER 3. (ipi. ip2 of this space can be associated with a complex number (V'l. . i = 1. .tp2. (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M). n. ^ i = 0 if ^ = 0.5b) . and n is the smallest such number. —»• IK. n exist (not all zero).  .1) aipi + aipj. E (3. if any two elements ip\.4) The vector space .i = 1. .^i) . In each case n linearly independent vectors are said to form a basis. If n + 1 elements ^ € M. so that n 5 > ^ = 0.. if numbers eti. i=l (3.aiV>i + a 2 ^ 2 ) = ai(ip.5a) (3. f > 0 if V T ^ O . ( n (3.3) If all on = 0. (3 5c) + a2{ip./?eK). the vectors il)i. . (V>.e. (3. .
^1) 2 h 11 0 <%«e + M .^ 2 ) =   ^   2 + 2 ^ i .tp2 £ M.^) = ai*(V'i.e.V.2 Hilbert Spaces 43 where the asterix * means complex conjugation. for ip\.9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations. linearity in the second component.V') +"2*(V. which we can write 0 < ( ^ i .9c) HV'i + tp2\\2 + \\1p1 .^2)1 < H^ill • 11^211 (Schwarz inequality).2. if Wi.V>i + A</>2) > 0 . (3.*h)=\\AH\(38) (3. tp2 £ M: 1(^1. The first two properties imply ( a i ^ i + a2ip2. antilinearity in the first component (also described as sesquilinearity. meaning oneandahalffold linearity). for arbitrary A and ip2 ^ 0: (V>i + AV>2. ^ i ) + A ( ^ i . (3.9b) = 0 (Pythagoras theorem).is defined by d(ipi. ^ i ) + A * ( ^ . (3. In order to verify Eq.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. WA+Ml2 = ll^il 2 + ll^2 2 . (3. (310) (^2. ^ ) + A 2 (V2.rh) (3.)> (36) i.i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation). »2 2 .9a) (3. ^ 2 )  .7) In addition the following relations hold in a metric space M..9a) we start from if) = ipi + \if>2 6 M..3. \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality). The distance between two vectors if>i. The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2.
10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl. i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1. (3. Examples of metric vector spaces: (1) Let M. (3.  ^ l l = (HlMI +   « 2 . . a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0.^)1 <IIV>il. Mathematical Foundations of Quantum Mechanics 1(^2.V2) IIV'l2 + V'22 + 21(^2. (3. ^ e M are said to be orthogonal if (</>!. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X . o ^ V i = o.ii) thus verifying Eq.9d). beginning with A = 1 in the second line of Eq.9c). In verifying the triangle inequality we use this result (3. (3.13) the vector is said to be normalized. be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi. for which CO 2 := V^l^l 2 < 00.* Two vectors i f i .9a). HIHI (3.12) We omit here the verification of the remaining relations (3.\\ih\\.44 or CHAPTER 3.^1)1 Vi 2 + ^2 2 + 2   ^   .11). V 2 ) = 0 . so that   ^ l + ^ 2   <   ^ l   + ^2.9e). hence verification in each case is necessary. (3. (3. ll^ill + llVdl.
= 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.5c)) / ( x ) = 0 = > ( / . which satisfies relations (3.3. With the scalar product (f. all squareintegrable functions / which are "almost everywhere equal".e. In order to avoid this difficulty.5a). (3. for which the scalar product. which differ solely on a set of measure zero. . */ x } = / /o for x = 0. i. Js the space C? is not yet a metric vector space although for (cf. Eq. Then L2 is the space of all these equivalence classes. with inner product oo 45 (v. Elements of the classes are then called representatives of these classes.[$]):= / JM /(x)* 5 (x)d 3 *. are combined to an equivalence class [/] (with space L2). •'^ \ 0 otherwise. etc. and one defines addition and multiplication by complex numbers with respect to these classes. / ) = 0.e.5c). 2 and so on. (3.g)= [ /(x)^(x)d3x. (2) Let M.w) :=J2v*Wi. i. is defined by ([/]. (3.5b).2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi). But this applies also in the case of any function which is nonzero only on a set of measure zero. The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2.
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.
and BraFormalism Furthermore we have: ( x  = 0. that the vector \u) E V is associated with a vector in V. Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV . i.62 CHAPTER 4. As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*. Dirac's Ket. if (u\v) — 0. It follows that we can now write the Schwarzian inequality (3. This expression is linear with respect to \u) and antilinear with respect to \v).4b) (4.\u) = (u\0*. (4.5) 4. if \u) = 0 (see (3.4a) \v) = J\(m<% and (4. if x\u) = 0 V kets \u) E V. \v) E V are said to be orthogonal.(t.5c)). Two ketvectors \u). Because of the antilinearity the conjugate bravector associated with the ketvector u) = Ail> + A22> is (v\ = Xl(l\ + X*2(2\.3 Linear Operators.4c) (44d) (v\ = J\*(0(m One defines as the scalar product or inner product of \u) E V and \v) E V the cnumber (v\u). We consider next the important case that a unique antilinear relation called conjugation exists between the kets \u) E V and the bras (x E V.9a): \{u\v)\2 < (u\u)(v\v). (2) (u\u) > 0. (3) (u\u) = 0. the norm squared. which we write (u\. or (4. V \u) (in the case of £) with infinite norm!). Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V. The dimension of V is the same as the dimension of V.e.
also acts on the dual space V.. BA = 1. Then one says: (771 results from the action of A on (x.. AB^BA. Two operators A and B are said to be inverse to each other. as we can see as follows. Furthermore linearity applies. B. i.e. defines the unit or identity operator. The inverse of a product is (AB)'1 =B1A~1.. and one writes: (V\ = (X\A.e. with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) .3 Linear Operators. Then (x(^4it)) is a linear functional of \u) £ V (since A is linear). If Vi has the dimension A/i.4. i. This product is commutative. Multiplication by unity. if AB = 1. The operator A. The inverse of A is written A . 1 and \u) = B\v). The product vectors l?^ 1 )^ 2 )) span a new vector space. denned as a linear operator on V. Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2.or Kronecker. Finally we define the tensor.e. Then the following rules apply: A + B = B + A. let (771 6 V be the bravector defined by this functional. Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V. the space Vi <S> V2 has the dimension . The product of such vectors is written: \u^u^) = \u)^\u)^. It follows that (r. if \v) — A\u) i. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). In this way the operations of operators A. on bravectors are similar to those on ketvectors. This does not always exist. ltt) = \u). Furthermore. we have \u^u^) = X1\v^u^) + \2\w^u^) and so on.or direct product of vector spaces.\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u). Let the following bravector be given: (x\ £ V. the space Vi <S> V2.
6) for all \u).7d) . i. [A^. (v\t) = (u\A\t). we write \v) = A*\u). here designated with the same symbol.7a) (4. we obtain the conjugate relation <tAt«> = (u\A\t)* (4.A^] = 0. 3. i. since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW).g. Dirac's Ket. Let A^> be an operator in the space Vi and A^ space V2. As a consequence we arrive at the following properties (e. AW\U)W = \V)W. Every such operator is then also associated with an operator.7C) («)(u)t = \v){u\. in the product space.64 CHAPTER 4. the operator A^ is called hermitian conjugate or adjoint operator of A (cf. i. If (v\ = (u\A. by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4. Since we demanded the scalar products to have the property (t\v) = (v\ty. AW\uW)\uW) = \vWUW).e.3).7b) (4. Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above. \t) G V. also Sec. then \v) is a linear function of \u). Every operator A^> commutes with every operator A(2\ i.e.and BraFormalism an operator in the Mi A/2.e. and (v\ = (u\A.e. Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4. Next we construct the following scalar products: (t\v) = (t\Ai\u).
e. Obviously the quantity \a)(a\ is an hermitian operator.K}.4. or (u\A\u)* — {u\A^\u) = (u\A\u). . Verification: Since A — A^ and A\u) = a\u). An hermitian operator A has the properties (a) that all its eigenvalues are real. hermitian The product of two hermitian operators is not necessarily hermitian.e. [H. i. Analogous considerations apply in the case of bravectors (u. for we had (\u){v\)i = \v){u\. H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H. it follows that (u\A\u) = a(u\u). if H = H' and K = K\ then (HK)* = HK only if tfrf = KH. and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. as is also (u\u). i. Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian .K]^ = = [HKKH]* KHHK = = K^H^ [H. The operator H is said to be positive definite.3 Linear Operators. The commutator of two hermitian operators is antihermitian: [H. An important theorem is the following. if H — H' and antihermitian. 65 The linear operator H is called hermitian. It follows that a is real. Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A].K\. so that (n^4n) is real. so that (a)(a)t = a)(a. but only if these commutators commute. if H is hermitian and (u\H\u) > 0 for all ketvectors \u). K] = 0.
(n\n')=5nnl. i. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal. Dirac's Ket. a ^ b.e. Ps is idempotent. i. which includes vectors of infinite norm. since a is real. Next we introduce the very useful concept of projection operators. ( n  z / ) = 0 . as we can see as follows. i.) = 0.66 CHAPTER 4.e. so that (u\Ps = (us\. we have 0 = (uArt) — (u.9b) (4. and let S* be its complement. (v\A = b(v\. it follows from A\u) — a\u) that (u\A = a(u\ or the other way round.e. then the hermitian operator defined on this space is an observable. {us\us*) = 0.9a) . since a ^ b.and BraFormalism Moreover. Let S be a subspace of the Hilbert space !K. The entire set of ketvectors spans the extended Hilbert space. Assuming A\u)=a\u). so that (v\u) = 0. The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us). Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S. Ps\ua. The above theorem remains unchanged.u').e.8) The projection operator has the following important properties: Ps is hermitian. This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v). A continuous spectrum can immediately be included by passing to the extended Hilbert space. (u\u') = 5{v . Ps2 = Ps(4. (4. \u8*) E S*. If these form a complete system.Au) = a(v\u) — b(v\u) — (a — b)(v\u). (u\Ps = {us\. but the continuum vectors are normalized to a delta function. i.
The quantity \a)(a\ is called elementary projector.1. Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c. \u) arbitrary.. Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u). Let a continuum state be written £).. where £ is a continuous index. Then 0 = (P2P)\p) and hence p = 0. so that \ua) = (a\u)\a) = \a)(a\u). as can be seen as follows. Ps2 = PsThe only eigenvalues of Ps are : 0 and 1. iV) is a set of orthonormalized states. i. . so that (Ps2 ~ Ps)\u) = 0.e. i. p 2 . \ua) = c\a).9c) This property can be seen as follows. (1 — P)\u) are orthogonal. If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm. = (p2p)\p). The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw). 2).3 Linear Operators. P\p) =p\p).e.4. i. i.9d) Thus the vectors P\u). Very similarly we can construct operators which project onto the subspace spanned by the continuum states. Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN. 67 (4. (1 — Ps) is projector onto S*. Set \u) = \ua) + \ua*) with (a\ua*) = 0.e.p = 0. . Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace. (4. c = (a\u). if the set of vectors 1). Ps is projector onto S. Let p be an eigenvalue of the projection operator P.e..
.and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident.4 Observables Operators which play a particular role in quantum mechanics are those called observables. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui. Coulomb potential. Dime's Ket. Let \u\).. Sec.. (11. one says the degree of degeneracy is r . The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K. and Eq. P = £n)(n. However. (11. the operator P is correspondingly P2\u)= f2\0m\u). be a system of basis vectors in this space. see Eq. which are orthogonal to each other (i.r\ "An example is provided by the case of the hydrogen atom. . Examples 8. Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator). i. Observables are representatives of measurable quantities. If there are r such vectors.1 and 11. In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors.114b).6.68 CHAPTER 4. 8.e. Degeneracy will be discussed at various points in this text. are linearly independent).6. In general it is possible.g. which we introduced earlier.114c).r)(ui. 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.e. u2). 4. in this case the spectrum also has a continuous part. that one and the same eigenvalue Aj is associated with several eigenfunctions. See e.
The uniqueness of the expression (4. i. the projection of this operator is the entire space. i.r \ui.4 Observables The dimension of this subspace is that of the degree of degeneracy.r\ui/. or also as subdivision of unity or of the unit operator. we obtain i i i i i.10) expresses the completeness of the orthonormal system. then PiPi' = 0. and the eigenvectors \ui.10) This expression is known as completeness relation or closure relation.\i)Pi = 0.r) (the convergence.r)(ui. (4. in the case of degeneracy with (ui. (4. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui.e. ^ A i P i = A = ^Aiui. If Aj ^ Aj'.r)(ui.r\ = l.r\u)\2.4.r\u) i. PA = y£pi i = Yt\ui. Let us set {A .r>(ui.r)(ui.r'} = Sii>6rr>. Applied to an arbitrary vector \u) G "K Eq. The operators Pi are linearly independent. i. is always implied). the operator A is completely determined by specification of the eigenvalues A.10) gives the linear combination it) = ^ i. 69 i If A is an observable and if the spectrum is purely discrete.e.11) i.r =^ i.r.r\u). and APi = XiPi.e. i i (4.10) follows from the fact that i Applying the operator A to the projector (4. the relation (4.10).e. Together with the orthogonality condition.12) . which we do not enter into here.r (4.r \(ui.
Let v be the continuous parameter which characterizes the continuum. Dime's Ket.v').and BraFormalism This is the expression called Parseval equation which we encountered with Eq.e.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). (3.e. the eigenvector \ui): f(A)\ui) = f{\i)\ui). now however. written in Dirac's notation. (4.70 CHAPTER 4. we have the corresponding generalizations. In a similar way we can handle functions f(A) of an observable A. for instance exponential functions. One defines as action of f(A) on. (4. The ketvectors \uv) are orthonormalized to a delta function. as explained earlier. i. (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv). (u„zv) = 8{v . Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv. If the spectrum of an observable A consists of discrete as well as continuous parts. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. i. for instance.17) previously. 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1.14) . 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.
. In the onedimensional case we write {x\x') = S(xx'). which all commute with one another. with energy E. (4..15) and (4.. for which the completeness relation is f dk\k)(k\=l.e. We shall return to this theorem.e. as a formal orthonormality condition. (4. For many practical purposes the use of the Fourier transform is unavoidable. where the symbol ip is already indicative of the Schrodinger wave function ij. In the following we consider more generally ketvectors \ip) as representatives of the physical states. x) € Fx. Therefore we want to reexpress the Fourier transform (3.15) where the vectors {\x)} are to be a complete set of basis vectors of a linear vector space in its position space representation Fx.C. In the differential operator representation the problem to establish such a relation is known as the SturmLiouville problem.B.5 Representation Spaces and Basis Vectors 71 This relation expresses an arbitrary function f(A) of an observable A in terms of the eigenfunctions of this operator. \k)€Fk.16) represent subdi .. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies). i. their commutator vanishes. and (b) if they possess a uniquely determined system of basis vectors. B].5 Representation Spaces and Basis Vectors We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum.58) in terms of ket. [A. First of all we can rewrite the integral representation of the delta function. if and only if they possess at least one common system of basis vectors. again later. i.4. if (a) they all commute pairwise. / dx\x){x\ = 1.17) The Fourier transform provides the transition from one representation and basis to the other.59). here presented without proof. form a complete set of commuting observables. (3. (4. Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^. Since both expressions (4. 4. Extending this we can say: A sequence of observables A. Eq.and bravectors.
1a). Dime's Ket.space representation.18) According to Eq. Rather \x) and \k) serve as basis vectors in the representation spaces Fx.Fk. i.2a) etc.e. .e.20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy. 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). (4. i. (4. which are representatives of the states of our physical system.21) provides the ketvector \ij)) in the A. we can rewrite Eq.72 CHAPTER 4.15): 5(x .x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x'). il>{x) := (x\i/}) : "K * C. The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4. this expression has to be identified with — / dkeikxeikx\ i. (x\k) = ^=eikx. (4. The vectors \x) and \k) G F are not to be confused with the vectors u) or \ijj) G !K. (4.and BraFormalism visions of the unit operator.e. Obviously we obtain this by inserting a complete system of basis vectors of Fk.59) or Eq. <a#) = [ dk{x\k)(k\ip). ${k) := (k\if>). V 27T (k\x') = .22) All of these expressions can readily be transcribed into cases with a higher dimensional position space.L e " * * ' = {x'\k)\ V27T (4.19) Comparison with the orthonormalized system of trigonometric functions (4. (3. shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n). called wave function. (4.
i. which is the quantum mechanical analogue of the classical probability density p(q. 1. This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation. the calculation of which for specific cases is the subject of Chapter 7. We mentioned earlier (in Sec. 73 . Finally we introduce briefly the canonical distribution of statistical mechanics and show that with this the density matrix can be calculated like the Green's function of a Schrodinger equation (inverse temperature replacing time). Thus we distinguish between socalled pure states and mixed states of a system and thereby introduce the concept of density matrix and that of the statistical operator.1 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.2 The Density Matrix We shall establish a matrix p. with the complementary system or *E.5) that in general and in reality we ought to consider the system under consideration together or in interaction with the rest of the universe. We postulate the (timedependent) Schrodinger equation* (with the Hamiltonian as the timedevelopment operator) and obtain the quantum mechanical analogue of the Liouville equation.p.Chapter 5 Schrodinger Equation and Liouville Equation 5.e.t). 5. if we do not take into account the complementary part of the universe. Schrodinger [244].
called the density matrix. that is.. p" ( C C * ) t = CTC* = p. These are the quantum mechanical states which are also referred to as microstates.i. where T means transpose: p = CTC*. but also of the states of the complementary set of the universe. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system. \a)..e. Schrodinger Equation and Liouville Equation the measuring instruments. \i). The actual and real state of a system is then a superposition of two types of states.^ Since p is hermitian. The states \i) £ "K.3b) The quantum mechanical expectation value of an observable A. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a. \a). are called "pure states". as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij. a (5.i.74 CHAPTER 5. (a\i)=0 = {j\b)..3a) and N J  a ) ( o  = l in the complementary subspace. The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe..b. we have (a\b) = 5ab. of an operator that acts only on the states of the system we are interested in. 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). (52) (5.e. the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows.. i. which in Dirac's notation we can write as = J]C a »i) a. i.. T . where i represents collectively all quantum numbers of the state of the system under consideration.j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a.j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix.b. we have to deal with a socalled "mixed state"..
5. seen macroscopically of Z copies of the system). P ^ ) = i)(#) = (#)i) (5. can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i. Gibbs introduced the concept of ensemble. In the following we write simply \i) instead of \i').10) and is therefore. a (512) *In his reformulation of the statistical mechanics developed by Boltzmann.* We can specify the state of our system. i.6) (i'\p\i") = Wi'Si'vi = real.\p\i")(i"\ = Y.e.j(x'i)(ia.e.4) we had the expression (A) = Y/Pij(j\A\i) i. i i (511) In Eq. in a specific representation. i. (5. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i).3a).2 The Density Matrix 75 \i') 6 "K.) = YJc t . as we saw earlier. (5. The ensemble can be represented as a set of points in phase space. (5. Thus the pure states \i) form in the subspace of the space of states which is of interest to us.ji(x / )i*(x). a complete orthonormal system with properties (5. as we discussed previously. This transition to a new set of basis vectors in which the matrix becomes diagonal..3 with Pi^YtCaiCaj*. or (5.2) and (5..g. quantum states. but occupy in general different microscopic. a projection operator. in the position state representation or xrepresentation. This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities. Thus the real system is in the pure state \i) with probability uii/ J2i Ui. The operator Pi = \i)(i\ projects a state \<p) G "K onto \i). If the ensemble corresponding to the system under consideration consists of Z elements (i.^'\i'^i'\> (58) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j.9) is diagonal. e.e. . where x represents collectively the entire set of position coordinates of the (particle) system. which are all subjected to the same macroscopic boundary and subsidiary conditions.. (see below) eM. In this sense we have p(x'. which today is a fundamental term in statistical mechanics.x) := (x'\p\x) = yju.7) p = ^2\ii)(i. This ensemble is interpreted as a large set of identical systems.e.
We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^519) . (5. Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf.) := (mm = (#') W> = l(#')2 > 0. (5. i t (5. Eq. In particular for A = 1. (5. On the other hand. Eq. (5.14) which we can write (A) = Tr(pA).5) above) (i\p\j)=Pij.17) Hence the sum of the classical probabilities is 1. according to Eq.18).4). 3 (5. (5.18) Substituting this into Eq.76 CHAPTER 5.3 = pvv = uv. To this end we set A —>• Ai' := \i')(i'\ (no summation over i').16) (5. Thus Tr(j>ii)<i)=l. (5. as claimed by Eq. we have Tr(p) = l. Hence Wj/ > 0. (5. Then the expectation value of the observable A is (cf. (A.13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j). We can also show that Ui > 0.14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i.15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l.4)) (5.
3 The Probability Density p(x. (5. one says.e. We now consider the latter expression. Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2. but we see here. (Question: Is the state of the universe a pure state?) 5. Thus we have a system in a pure state.5.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l. t) p= i The expression (5. If all ui but one are nought.t) = {x\i).t)\2. not in a mixture of states.3 The Probability Density 77 We see therefore that p defines the mixed states. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights.24) s (x\x') = 5{x .t) and instead of (x\Pi\x) we now write p(x. defines the statistical operator. i. In the following we are mostly concerned with considerations of systems in pure states. i. and in the other states with probability zero.t). thus enters these states. so that p{x.t) we now write tp(x. (5.23) Instead of i(x. i. .x'). ^2ui\i){i\. The effect of the interaction of the system under consideration with the surroundings.e. represents a quantum mechanical probability. the system is in a pure state. since they do not take into account the (interaction with the) rest of the universe. (5.e. which is contained in the coefficients Cai. We observed above. that these are actually not sufficiently general. in all other cases the state of the system is a mixed state. For the case n we have TJ=J (520) (5. the only remaining one is 1. that — different from u>i — P. Equation (5.21) p=\i){i\=Pi. by representing the vector \i) by the wave function i(x.t) = \^{x.20) expresses that the system can be found with probability 1 in the state \i). We see this more clearly by going to the position space representation.22) In the following (x\i) is always to be understood as (x\i(t)}.8).
We therefore postulate first an equation for the time dependence of a state vector. (5.^ The equation which is the conjugate of Eq.e. We now want to obtain the quantum mechanical analogue. To this end we have to differentiate the density matrix p.t)\2. (5. with respect to time. (5. but the linear operator states are moving — even if the system is observably stationary). The generalization to a threedimensional position space is selfevident: JK 1 / p(x.e.26) 5. it is not "derived"). (5.28) Differentiating Eq. which is chosen in such a way that the desired analogy is obtained. the Schrodinger equation is always postulated in some way.78 CHAPTER 5.3. 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.t) = / dx{i\x)(x\i) = (i\i) = 1. 2. We shall convince ourselves later that the postulated equation is sensible (in fact. the expression (5. Equation (5. i. . This differentiation requires the time derivative of a state vector \i) E "K. (5. t h e equation of motion with states represented in a system for which the q's or here x's are diagonal).8) and multiplying by ih.27) where H is the Hamilton operator of the system.27) is known as the Schrodinger equation.27) is iHJ\ jt= m . can be found with probability 1 in the space R1. and the timeindependent equation should correspondingly be called Schrodinger wave equation (i.4 Schrodinger E q u a t i o n a n d Liouville E q u a t i o n We encountered the classical form of the Liouville equation in Sec. and hence the integral over all space dxp(x.t) = \rp(x. (5. t).8). Schrodinger Equation and Liouville Equation The relation (5. Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j).25) J Thus the particle whose state is described by \i). or the wave function tp(x.22) can also be written (xi) 2 = (i\x)(x\i).
(3. With Eqs. (cf.e. we observe that here on the left hand side we have the total time derivative. Eq.p} + ihJ2^\J)(J\. (5.28) we obtain (see also Eq.27). The reason for this is that in addition to the "classical probabilities". (5.37)) (5. (2.27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi. i.34) . Eq.27) and (5.21). such a formulation is described as Schrodinger picture. not the partial derivative.24)).Xi)\j). We thus obtain the operator form of the Liouville equation.32) with its classical counterpart (2. dp/dt ^ 0.31) 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).e. (5. i. M\/n\ (530) where in statistical equilibrium dt £^">01=03 (5.28) the states are considered as timedependent. We can now consider the Schrodinger equation (5. The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf. p also contains "quantum mechanical probabilities" (i.e. (5. the relation ih^ = [H.32) Comparing Eq.5.4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs. The Schrodinger picture and the alternative Heisenberg picture will later be considered separately.p}. (5. (5.31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H. both uji and i)(i).
(5. appearing in the Boltzmann distribution.. i.35) Here we can look at the Hamilton operator as a "time generator3''. to) = exp[—iH(t — to)/H\. „ / ith—. it is plausible to refer to this equation at this stage. 0) or \il))t = \il>)t=o exp (5.t) = Hld . we shall describe as "time development ator given by U(t. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H. **R. this is an application of the BakerCampbellHausdorff formula which we deal with in "Later. . For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1. (5. i.4.Xi){x\En) = En(x\En).4. In expression (5. operator" the exponentiated oper .38) (x\j)t=o = Yt(x\E^(En\J)t=o. T meaning temperature. since the solution of Eq. since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed.and temperaturedependent Green's functions.0): where H(ih—. in Sec.x 1 ^ ( x . a (5. Feynman [94].e.80 CHAPTER 5. Schrodinger Equation and Liouville Equation or with (x\j) — ip(x.1 Evaluation of t h e d e n s i t y m a t r i x As a sideremark with regard to statistical mechanics we can make an interesting observation at this point. (5. with respect to time t.t) = exp ip(x. 5. we write V>(x. n (537) This equation is described as the timeindependent Schrodinger equation. t ) .t): > ih—</>(x. the state \i) is still timedependent. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not. With the help of Eq.36) however. however.. we can replace these states by corresponding timeindependent states.36) in in The initial wave function or timeindependent wave function ^(x. which we shall need later.35) can be written'! ip(x.** In view of the close connection between time. in p = ^2i^i\i}(i\. or H\En) = En\En).e.8) for p. but with respect to the parameter /3 = 1/kT. P. 10..
i ( 5 .e. (1. we obtain p(x.44) r (TrePH)' (5. or ^ = ^ e 0Ei _m.9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT. (5. (5.41 ) or with 4>i(x) := (x\Ei) this is p(x. Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\. Eq. x') = ^ i LUiMxWix'). (543) Since H(f>i(x) = Ei<f>i(x). i.40) 1. (5.1). the expression _Zie^\Ei)(E>\ also as (see below) e0H (5. (5.x'(3) = Si e ~^W*V).^ n tne position space representation Eq.42) Inserting here Eq.5.39) therefore (x\p\x') = J2"i(x\^)(Ei\x'). we can rewrite p.40).x')^p(x. i (5.4 Schrodinger Equation and Liouville Equation 81 Example 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.45) since on the one hand •PEi i i i and on the other hand .
we obtain 9pN gpP) = SijEie^ = EiPNij(J3). Wilcox [284]. . (5. (5.55)). B] + L [A. Tr(pN) = / dxpN(x. B}} . (5.46) we can write the expectation value of an observable <^Tr<„. (7. Differentiating Eq.51) where the subscript x indicates that H acts on x of PN(X. (5.52) Hence this expectation value can now be evaluated with a knowledge of pN. Example 5.50) Differentiating this equation with respect to j3./?).X'. we obtain = HxPN{x.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5. We now rewrite the factor exp(—(3H) in Eq. (5.45). Equation (5.x. whose calculation for specific cases is a topic of Chapter 7 (see Eq.(3).x'.35). (5. M.45) without normalization as PN(J3) := e~pH.x'. (3).tt tt For a more extensive discussion see R. A}} + • Solution: The relation can be verified by expansion of the left hand side. With Eqs.4> = ^ g ^ . [B.46) is pN(x. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A.51).51) is seen to be very similar to the Schrodinger equation (5.1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators.47) (5.47) with respect to (3.x'. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5./3).(3) := (x\pN(P)\x') = (x\e~0H\x').45) and (5.46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij. (5.L [B.48) = 1. This solution is the Green's function or kernel K(x. [A.82 CHAPTER 5. (5.49) In the position or configuration space representation Eq. Schrodinger Equation and Liouville Equation so that p is given by Eq. (5. (5.
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.Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6. An alternative method of quantization of the harmonic oscillator — by consideration of the Schrodinger equation as the equation of Weber or parabolic cylinder functions — is discussed in Chapter 8 (see Eqs. its associated space of state vectors also provides the best illustration of a properly separable Hilbert space. S. we shall see on the way." Quite apart from this basic significance of the harmonic oscillator.. (8. E. Noz [149]. 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. * Comment of Y. which is often described as "second quantization" as distinct from the quantization of quantum mechanics which is correspondingly described as "first quantization" (thus one could visualize the free electromagnetic field as consisting of two mutually orthogonal oscillators at every point in space). The importance of the harmonic oscillator can also be seen from comments in the literature such as:* ".the present form of quantum mechanics is largely a physics of harmonic oscillators. Since the spectrum of the harmonic oscillator is entirely discrete. can — in fact — be quantized in quite a few different ways.51) to (8. Kim and M. like the Coulomb potential..1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator. 83 . However. This is a fundamental topic since the harmonic oscillator.53)).
2 The OneDimensional Linear Oscillator In the case of the onedimensional. Also observe that H does not contain terms like pq or qp. linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 . 2mo dx \ 2 2 (6.' ^mgLOh by setting and *~.p. like taking half and half.MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers. the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo. p^ih—. Quantum Mechanics of the Harmonic Oscillator 6. . hence we do not use a distinguishing notation. we can also proceed in a different way. so that there is no ambiguity due to commutation.1) for the hermitian operators q.p]=ih (6.to = 1 reduces to the parabolic cylinder equation. which is — in the first place — representationindependent. For instance we can go to the special configuration or position space representation given by q^x. which is called Weyl ordering. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x). However. To this end we introduce first of all the dimensionless quantities rriQUi fl q. This equation would then have to be solved as a second order differential equation. in the present case + [EmooJ2x2)ip = 0. of which the only nontrivial one here is [q. 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions.^ The first problem is to determine the eigenvalues and eigenfunctions of H.84 CHAPTER 6. When such a term arises one has to resort to some definition.2) which for mo = 1/2. This can be done in a number of ways. h = l.
i.2 The OneDimensional Linear Oscillator 85 (p.9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a.e.11) Thus the eigenvalues are real and nonnegative. i. (6. (6.e. We now use the relation (3.14a) we deduce for an eigenvector \a) of A^A.B}C + B[A. (6.13) ( ^ A ) i t = A\A^A From Eq. for ^ A  a ) = a\a). (6. .34d): [A. A\ we obtain h A + A^ rriQLU y^ (6. A*} = A^A.14a) (6. C] + [A. C] = A[B. A^] = A\ (6.5) Reexpressing q and p in terms of A.8) The eigenstates of H are therefore essentially those of N := ^ f A (6.6) and p = y/motoh AAi =^.7) (6. (6. + 1). A] = [A\ A]A = A. (6. With the help of Eq.1) we obtain immediately the commutation relation [A. q are hermitian).14b) [A* A. Prom these relations we obtain ( ^ = ^ . C}B. i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ .6. (6.1 ) .BC] = [A.12) in order to obtain the following expressions: [tfA.A^] = 1.C].10) then a = (a\A^A\a) = \\A\a)\\2>0. (6. and [AB. if A*A\a) = a\a).
2A)\a) 2 (a .15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1).16) This means. (6. or Ata) = v ^ T T .21) . Similarly we obtain from Eq.. Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A .A)\a) A{AAU . The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a. However. This would mean that for sufficiently large values of n the eigenvalue would be negative. in view of Eq. Next we consider the vector A 2 a). (6.19) i. A 2 a) is eigenvector of A^A with eigenvalue (a — 2). we find that A n a) is eigenvector of A^A with eigenvalue (ct — n). (6.11) this is not possible.l)\a) = A(a .18) = '= a ) A(A^A . we must have (a) Let \an):= Ana)^0. (6. unless A^\a) = 0.86 the relation CHAPTER 6.l)A\a).17) (6. but (b) An\n>\ llA An+1a)=0.20) . (6.2A)\a) = A(Aa . Similarly \\A^\a)\\2 = (aAAta) = (al + A^A\a) = (a + l)(aa) = a + 1. Thus for a certain value of n > 0. In this case we have (A*A)A2\a) (6 (6. since this equation implies that the eigenvalues cannot be negative.14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a). (6.  a) is eigenvector of A^A with eigenvalue (a + 1).l)A\a) = A(A^AA . (6. or \\A\a)\\ = Va. unless A\a) = 0.e.2)A a). If we continue like this and consider the vectors An\a) ^ 0 for all n.l)\a) = (a . unless A 2 a) = 0.
' " /.5) to shift operators A to the right which then acting on 0) give zero. that an = 0. From relation (6. \ I 2 2 x 87 Replacing in Eq.28) . so that (a(A n )tA n \a) __ \\An\a)\\2 ( a ..27) According to (b) of (6.20) we obtain from this that the right hand side vanishes. (6.e.6. (6.I n U . (6. v.n  a . we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n. \ 9 \ I / II 4 T ) . For a — n = 0 we deduce from (b) of (6. or a = n > 0. (6. i.22) With relation (b) of Eq (6. Moreover. But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0.n ) = " A»a>=» ' = A«a)F = 1 .26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2. (6.20) that A0) = 0.20) we have for a = n = 1: A2\1) = 0. so that A^\n)^0 In particular we have A^\0) ^ 0 and II^IO))!2 = (OIAA+IO) = (0\l + AU\0) = (00) = 1..23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers.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.18) we obtain for a = n: p t  n )   2 = n + l.17) a by (a — n). I „ . (6.25) (6.2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n). In the following manipulations we always use the commutator (6.
and in general [A. But using Eqs. (6. The states \n) thus defined are orthonormal. (6. as we can see as follows.33) . l>ocAt0). we have 2> = ^=dAi\0).(6.^ ( O l A ^ n o ) .32) we have (nm) = .5) and again (6.31) (6. (0^ 2 (At) 2 0) = 2. A^} = 2A\ and [A. (A*)2} = [A. According to Eq. Similarly we find 2)oc^Ut0). (6. Quantum Mechanics of the Harmonic Oscillator A2AtA*\0) = A(A^A + 1)4*10) = (AA^AA^ + AA*)\0) = = = {AA\A* A+1) + A* A+ 1}\0) (AA^+ 1)\0) = (A^ A+ 2)\0) 20) ^ 0.26) the equality l) = A t 0). In general we have \n) = ^(A^n\0) (6.29) Hence we obtain and in view of Eq.32) (6. and in view of Eq. A^A* + A*[A. (At)2]A+ + (A*)2[A. (6.88 and CHAPTER 6. i.e.11).29). (A^)3] = [A.30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1).34) (6.(A*)n] = n(Ai)n\ (6.Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2. (6.11) we have [A.
'n\ and = v n ![ ( A t ) M i= + (6.33).40) (6. 2 . (6.37) n^)711}^) = v n  n .e. l . 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 . In view of the properties (6. (6.1) (6. Here.41) The contribution hu/2 is called zero point energy.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 . whose number is given by the integer eigenvalue of the number operator N = A^A. •(» (6. however. n = 0 .39) A^A\n) = y/nA^\n . In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv.1) = n\n).35) the operators A^ and A are called respectively raising and lowering operators or also shift operators. we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle"). in quantum mechanics. chosen in close analogy to the postulated "second quantization relations" of field theory. ldnm = n\5nm. The terminology is. . .1 (A + ) m .34) A\n) = ±=A{Al)n\0) .1 0). i.J \n) = hw (n + i J \n). . the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) . With Eq.6.35) Inserting this result into Eq. we obtain (n\m) = 5nm We also deduce from Eq.36) + 1) (6. . (6.8) we obtain therefore H\n) = hu (A^A + ]. . .32): rf\n) = .34) and (6.1 ) . (6.L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq. .38) (6. (6.
(6. ( At = Vi 0 0 0 ° \ f° 0 Vi 0 0 0 0 0 0 0 0 V2 0 0 0 0 0 V2 0 0 v^ 0 0 0 0 0 V3 0 V5 0 v / ( ° VI 0 V2 0 0 v / (6. (6. n = 0. other elements zero. is defined by projection of operators on eigenfunctions of energy.'\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 . 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...g. A^ from the relations (6. also called Heisenberg representation. En = hu(n + ).7) the energy representation of the operators q and p.42) We can deduce the energy representation of the operators A.46) v^ . Quantum Mechanics of the Harmonic Oscillator 6.39): {n\A. i.6) and (6. and — in fact — we have used this already previously (see e. other elements zero..3 The Energy Representation of the Oscillator The energy representation. (5.43) = = y/n(n'\n — 1) = y/n5n^ni.45) Correspondingly we obtain with Eqs. The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn.1.44) yfn.2. (6.39)).. (6.e. Eq. Vn + 1..90 CHAPTER 6.38) and (6.
3)) ^ (q 2n \ + (6. 2 . A\Q) = 0. Eq.6.5) are also satisfied as matrix equations. by (cf. 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. (6. 6.32). ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h.e.4 The Configuration Space Representation We saw that the eigenstates are given by Eq. i. (6.48) Applying from the left the bravector (x\ and remembering that (x\p\<f>) = we obtain ih—(x\(j)}.1) and (6. (6.24). mow / (6.47) LP 0) = 0 . Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n). (6. The ground state wave function (/>o(x) is defined by Eq. Hence ( * Recall J™ dxe™ * ? 2 2 2 1/4 W \ i/4 \ em^x*/2h_ (g 5 Q ) = yfln/w .4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs.
These polynomials are real for real arguments and have the symmetry property # n ( .56) < ^ = We have as an operator identity the relation .0 = ( . ) e"^2.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) = ( .e. c/>n(x) := (x\n) = .L ( x  ( A t ) »  0 ) .2 n n! The functions Hn(£) are obviously polynomials of the nth degree. (6.dx'{X] so that <*'>^(irrv^i« mQU dx Setting a we have 7T 1 /4 v ^!2«/2 \a Setting a and ff„(0 we have = ^ / 2 u^n^m^''^ h mouj dx) K ' ( f . it suffices according to Eq.54) (6 55) i/4 rzri7r 1 / 4 vo. Vn! Now {XlA (6. they are called Hermite polynomials. In order to obtain the wave functions of higher states. i.l ) n f f „ ( 0 . (6.92 CHAPTER 6.32) to apply the appropriate number of creation operators A^ to the vacuum state 0). (6.51) \l 2h {X q \ m0uP)\l 2h \X m0u.
(202"2.54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( . (6. 2£.4 The Position Space Representation i. ( 2 0 4 .58b) by W.g. p.58a) ' T h i s definition — apart from the usual one — is also cited e. gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt.6. 93 . (2036(20.l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1.2)m = (±t)m for some function VKO. 81.12(20 .e. d£" (6. e.It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f. (6.en_ dt. Oberhettinger [181].12. . Magnus and F.57b) • Generalizing this result and inserting it into Eq. — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6.
^ + n) He n(0 = 0 (6. (6. 80.60b) J—< "See for instance W. £33£. (6. . Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1.60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6. 81. Magnus and F.(e^/2) or Hn(0 = 2n/2Hen(V2ti). £ 4 . Oberhettinger [181].He'n{£). Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0. £2l.59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 .58c) Here special care is advisable since the Hermite polynomials defined above were motivated by the harmonic oscillator and hence are those frequently used by physicists.3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) . (6. (6. pp. £.6 £ 2 + 3.59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6.59b) = (l)^2/2^. = nHe^iO.94 CHAPTER 6. (6. = ffen+itO+ntfeniCO.59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm.
. Their normalization is therefore given by OO /"OO / OO J —OO z (6. Oberhettinger [181].. F(S. s) as a Taylor series.g. Magnus and F.s)=F(£.s) = eM?(Zs)2}(6. (6.^ = e^/42~n/2Hn(aV2).s) = J2 n=0 (6. and so on.63) The meaning is that if we expand F(£.64) n\ See e.62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S.6. Thus the generating function can be written F(Z.61) e?l*2nl2Hn{Z). .l ) ^ e^HeniO 2 / 4  .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£) = ( . 93. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0. p.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. W.
for x(t) = 0 and p(t) = mox(t) — 0.e. 6. has parity + 1 . 6. T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig. i. mow' Apx = ^m0Luh/2. in order to estimate the lowest energy of the harmonic oscillator.e. E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku). Quantum Mechanics of the Harmonic Oscillator Fig. ground) state is fko/2. We also see from the eigenvalue t h a t the energy of the lowest (i. i. In fact we can use the uncertainty relation AxApx ~ H/2. But this would mean t h a t b o t h x{t) and p{t) would assume simultaneously the "sharp" values zero. Furthermore we observe t h a t the wave function of the nth state has n zeros in finite domains of the variable x. 2 . We see t h a t the lowest state is symmetric.1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator. In classical mechanics the energy of the oscillator can be zero.96 CHAPTER 6. W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 .e. and we see t h a t even and odd states alternate.1.
: 27 T mov2a2. In the plane of complex q the integral has no poles for finite values of q. The kinetic energy is always positive or zero. According to "old quantum theory" with integer n.2 contains only the answers since the evaluation of the integrals proceeds as in Example 6.plane Imz z .1/a O l/a Rez all at infinity Fig. classical orbit i. n*h. 6.1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity".6. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u. This observability can be taken as direct proof of the uncertainty relation. which is the complete orbit. setting q = 1/z. But it has a pole at infinity. mass mo. original path q ..1. The analogous Example 6. we obtain the integral .1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf. in corrected version instead with n* = n + 1/2.4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators. back to 0. pdq. and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q). In Example 6. like twoatomic molecules. /classical orbit Here q goes from 0 to a.2 The original path and the transformed path at infinity. to 0. to —a. 27rmoi/ <f> V'a 2 — q2dq = n*h.e. Example 6. and is directly observable.plane pole at z=0 ^ .4) to obtain by contour integration the eigenvalues of the harmonic oscillator. 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)). Sec. E . Thus. Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways. 14.
e.1)! (6. In the following we investigate in more detail the important differential equation of the harmonic oscillator. (m . 6. Example 6.. we start with a Laplace transform ansatz for the remaining part of the solution. Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i.a) m = 2 . but here we leave this .( 6 .e. . verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o .114a)). In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11. i.It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 .2b = Ze2.3.2TTI a ^ = .2: Contour integration along a classical orbit Setting q = 1/z.1.V62 a2). 2 . harmonic oscillator and Coulomb interaction. a J \d Solution: The solution proceeds as in Example 6. for instance — without deriving all properties of the solutions. 1 . 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).5 . .e. in q at infinity).5.65) f ^Va2z2 .98 CHAPTER 6.c2 = (I + 1/2) 2 (natural units. . 6. E = Z2e4/4(l + n+ l ) 2 (cf. With a2 = E.) hv. « qdq + I2 = f dq . Eq.1. z = 0 Vi 2Kmov(Ka2) = n*h = E/v.5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials. are the most basic and exactly solvable cases in quantum mechanics. since these two cases.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. (11. mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points). hence we consider the following differential equation in which n = 0 . We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z . i. After removal of a Gaussian or W K B exponential. and obtain with this all properties of the solution. £ = n*/w = ( n + . / " ^ .
e.t*ij& + ny{t) = O. Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0. 80.6.111)) the Mellin transform of e~p is n!. E. Magnus and F.p t dp = \ps(p)ePt} .67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0. Then (with partial integration) *dt = ~*/Ps(p)e. W. so that the integrand of the remaining integral vanishes. / dpf(p)e~px the Laplace transform of f(p). we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6. (6. (6. We demand that C(p) := — \ps(p)e~pt] = 0.g.** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp. (cf.67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n.J ±(ps(p))e*dp. Eq. . (11. fp{ps{p)) = {v2 + n)s{p).5 The Harmonic Oscillator Equation latter property open for determination later: 99 ~dP + 1 1 2 V> = 0 . Substituting these expressions into Eq.g.68) § = J p2s(p)e*dp. . This has to be true for all values of t. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated). (6. and / dpf(p)pn the Mellin transform of / ( p ) .i ^  (ps(p)) P n P Integrating this expression. (6. Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) .69) **See e. (this determines the limits).66) First we have to remove the £2term. Oberhettinger [181]. p. i.
so that y(t) ~ exp(i 2 /2). Eq. To see this. of course). as indicated in Fig. since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity. 6. 6..71) . (1) This will be satisfied if \p\ — ±oo. We can show that if \p\ is allowed to be large.e. We now investigate possibilities to satisfy the boundary condition C(p) = 0. (3) We can choose a contour once around p = 0 in the complex pplane. Therefore we take ptp2/2 y(t) . consider the exponential in the integrand.n+l between the two limits of integration.3. This exponential assumes a maximum value when pt \—p2 = minimal. Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6. The quantum mechanical wave function is then (cf. Hence a possible path of integration > is from — oo to oo. (6. For singlevaluedness. then for large t we have y(t) ~ exp(i 2 /2). This can be attained if p is allowed to be large (t can be of either sign). Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer. = / • Im p Rep n not an integer Fig.67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible. y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6. n must then be an integer. The value of the exponential at this point is ~ exp(£ 2 /2). so that p cannot go to infinity.70) = 0 P.100 and hence CHAPTER 6. when p = —t (which is possible. we can choose a path from zero to infinity (the condition will vanish at p = 0).3 The path when n is not an integer. i. (2) If n < 0.
if n is even and v = 1. pt p2/2 (6.72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( . We have 27T! Setting q — p + t.3. n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n .. 2 .0) p p»+i v ' Hen(t) = (l)^ e * 2 /2 / 2vri /(9=t) dg " ^(gi)"+i e ?2/2 = e i 2 / 2 ()ne"i2/2. The exponential function here is described as the generating function of the Hermite polynomials Hen(t).l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0. Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\.. .P 2 / 2 00 C—tY ( .if n is odd. this becomes jfj.5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem. We can also obtain the derivative form of this Hermite polynomial. The Hermite polynomial is therefore given by H ^{t){^—f—^r~dp. the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt)..5.l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e . The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity. .6. In y(t) above. (676) . 4 .„y2 v l { ' ^ r (^) \ 2 ' where v = 0. . .
it) _ (n + l)Hen+1{t) (l)""1^1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen1{t) + Hen+1(t). Example 6.102 CHAPTER 6.7 that we use throughout this text) and their orthogonality and normalization to Examples 6.4.n— 1) = n. (6.59a). ( n . We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec.3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n. Hei(t) Henj. Reinterpreting the remaining integrals with Eq. 8.n + l ) f f e n ^ i ( £ ) + (n.3 and 6.70).1 '2dt using Hen(t) = (l)n27 r . (6. Pptp 2 / 00 /2 dp.g. as in the normalization of asymptotic expansions of Mathieu functions). (n. 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 + l) = l. Example 6.75) as Hermite polynomials. and we obtain then: ptp2/2 —^Ti ptp 2 /2 d : P r epte~p2/2 •(n + l) <t —5 f dp  eptep 2 /2 Here the bracketed expression vanishes in view of our condition (6.77) = t.4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials. in radiation problems. (6.)} 2 /2dt = ePde(P 2 W)/2^^} J —CO . n — l)Heni(t). Solution: We have to evaluate °° 2 . for the evaluation of expectation values. Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq. we obtain tHe„{t) (l)"n! ~ i. Hen(t). and elsewhere. Starting from Heo(t) = 1.e.5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. one can construct the tower of polynomials Example 6. 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).
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 .q). (6. It follows therefore that the following integral (with an exponential exp(—at) from whose expansion we pick later the power of i) is given by °° 2 / where OO eatHen(t)Hen+2r(t)et l2dt = n\(n + 2r)\ X coefficient of pnqn+2r in f(p. the coefficient of qn+2rn the coefficient of p"qn+2r m exp(.6.82) ^ .78) \/27rn!.(n + 2r)\ ' 2 ~ s r '2dt r 2* ^ i / ! ( 2 r + i/)!(Tii/)!(sri/)!' Solution: From the above we know that He„(t) is the coefficient of p " in (—l) n n! exp(—pi—p 2 /2).5 nm . g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (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./(n __ 2r — /i)!. (6. so that /2 Q2(n+rrf in / ( p . g) is n! Next.66) is \Zn\V2w Example 6. and 0 otherwise \/2n(—l)n+mn! if n = TO.n)\ i s an+2rij. zero otherwise (6.jQ) n n!I ^—' ^Q fi\(n .80) f(p. Hence the normalized wave function of Eq.q) = = e .
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.r . n + l ) ( n + 1.' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!.r . n + l ) ( n + 1.l ." ) a2^) r coefficient of (2s)! in f^ (s .(t)e«2/2dt Now. n + 2) + (n.' .e '' ^ o^M ! ( n . i / ! ( n .77). n . n + 3)(n + 3. /i = (n — v) _ / r o o e . i. we obtain the result (6. n — 1) are given by the first index.79).3 we know that upgoing coefficients are 1 and downgoing coefficients (n.^ ) ! ( n + 2 r .l ) ( n . n + l)(n + 1.( n + Jr).. n.— — ^ .2s _oo ( a 2 / 2 ) i .ra + 3)(n + 3.n + 2) + (n. n)(n. Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i).e. n + 2) + ( n .e. n + l ) ( n + 1.u)\ 2s<—" v\(n . n + 2){n + 2. n + l)(n + l . n + 2)(n + 2.Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ . .v)\ ^ Multiplying both sides by the value of the normalization integral.n+ l)(n + 1.i / ) ! ( 2 r + i/)! > . n + l ) ( n + 1.83) = (n + 2r)!e a / 2 V „ „. the coefficient of T 2s »{sr u)\ As we observed.° i i f e T 1 ( t ) i f e „ + 2 r ( t ) e . 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 .A i ) ! (6. the relation (n. .e. n + 2)(n + 2. i.78). i.v)\(s . Prom Example 6. n + 2)]ife„ + 2 (t) + ••• .v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n .r+v) a2s coefficient of in (n + 2r)\ V (2s)! h . so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2. The result then follows with half the normalization and orthogonality integrals (6.A o?(. 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) + .t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt . A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ .104 CHAPTER 6.^ / 2 * . n + 4)_ffe n +4(t) + [(ra. n ) ( n . K is the coefficient of a / ( 2 s ) ! in the expression J . 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. by n!(7r/2) 1 / 2 .
We 105 . which we write here K(x. we consider the sojourn time of a quantum mechanical particle at a point. and those which are timeindependent. With reference to this case. and is thus an important point to be noted here. These are the Green's functions of the timeindependent Schrodinger equation or. which we write G{x. This example also offers a convenient context to introduce the inverted oscillator potential.2 Timedependent and Timeindependent Cases In the case of Green's functions we distinguish between those which are timedependent. in Feynman's path integral as the kernel or free particle propagator. 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. of a differential equation of second order.t). The first of these cases will reappear later. We consider first timeindependent Green's functions. which is a point of unstable equilibrium for a classical particle.x'.1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both timedependent and timeindependent forms. Let H^ = E^> (7. The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5. in Chapter 21 (with a different calculation). which is normally not discussed.Chapter 7 Green's Functions 7. 7. more generally.x').1) be the timeindependent Schrodinger equation which has to be solved.
and Hi is some other contribution.2) where Ho consists.x') or ^2^(x)^*(x') i = S(x . e. e. aq2. for instance.e.5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case. We can obtain an expression for G(x.x') is defined by the equation* (Ho .x') by recalling the completeness relation. . i. Eq.x'). cf.106 assume a Hamiltonian of the form CHAPTER 7. (7. since the generalization to higher dimensions is selfevident. (7.e. Green's Functions H = H0 + Hi. where £ ( 0 ) = lim E. 'The case of £(°) equal to some Ei is considered later. (7. of the kinetic part p2/2mo and some part of the potential. appropriately decreasing behaviour at infinity).g.x') can be written t G (7.3) i which has the specific representation ^2{x\i)(i\x') i = 5(x . i.13). We can readily see that G(x.x'). The timeindependent Green's function G(x.4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix). (7. the relation X') = 5{x . x') has to satisfy correspond to those of * (to insure e. The boundary conditions which G(x.g. although the explicit calculations can be much more involved.EW)xG{x. a perturbation part like (3q4.g. square integrability.
x')(E . (7.t) since ih^K(x. i (7.E® (EE{0) Hi{x))m{x).x'){E .8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x).(*'))*fr') Hi{x'))^(x') f dx'5{x .t) (K for "kernel") is defined as solution of the equation ih—K(x. For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a.E<®)xG(x. .7) with the initial condition K(x.2 Green's Functions: Timedependent and Timeindependent 107 We note that the Green's function possesses simple poles in the plane of complex E(°\ and that the residue at a pole is determined by the wave function belonging to the eigenvalues Ei. x'\ t) (7. which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E .E^ of the eigenvalue.EM)X*(X) ( . (JL x'\ t) = H0K(x.10) The contribution on the right hand side is the socalled perturbation contribution. x'){E .x'.9) = i H0YieE^i%(x)^(x') = H0K(x. . (7.x'.Hi{x'))^(x').11) = = 3) = fdx'(H0 .t).t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x').x'.x'.) = f dx'G(x. The timedependent Green's function K(x.£(0) .E^ We can check this by considering (H0 .#. We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.0) = S(xx')..x. We can see the significance of the timeindependent Green's function for the complete problem as follows.7. We see that the initial condition (7. (7.8) This Green's function can be seen to be given by the following expansion in terms of a complete set of states (see verification below) K(x.
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). W. This will be different. In general. then at best such a contribution would be something like cipi(x). All such methods of solution are based on the analogy of Eq. integral equations are more difficult to solve than the corresponding differential equations. Merzbacher [194]. D. . But if we assume that E^ ^ Ei for all i. The theorem is also known as Fredholm alternative. 2nd ed. if we assume that E^ = Ej. however. (7. i.6) we would have (7. (7. The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7. see E.11) are called (homogeneous or inhomogeneous) Fredholm integral equations. Examples in various contexts have been investigated by L.Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj. But then the Green's function (7.e. (7. Wiedemann [183]. it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation). (7. In cases where an inhomogeneous contribution is given. H.x'. In this case the perturbation is restricted to that subspace of the Hilbert space 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. Maharana.108 CHAPTER 7. that (E .12) Equations of the form of Eq. Bleecker [33]. J..2 (not contained in the first edition!). Green's Functions Can we add on the right hand side of Eq.t) for the complete problem as follows.6) is not defined. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x').14) We can now see the significance of the timedependent Green's function K(x. Booss and D. 17. Sec. MiillerKirsten and A. (7.15) ''This problem and its circumvention can be formulated as a theorem. Instead of Eq. see B.11) with a system of linear equations of the form Vi = Mijyj.
p. (5. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7. 1 See e.0)eEnt/in.t) = where if(x^'. x'. We use this in Sec.x'. (7.158.22) According to Eq. in different formulations.' which obviously satisfies the initial condition K(x.21) dx'K(x. Comparison of Eq. 2nd ed.x'..x'.20) is the timedependent Green's function. t)\2. we assume that the set of states \En) is a complete set of eigenvectors of the Hamiltonian H (in the energy representation or energy basis) so that (cf.17) is the initial condition of \4>)t. Merzbacher [194].t) = ^ e £ " f / « V n W < ( ^ ) . (7.0). We can write this expression also as§ ip(x. i. e.17) as initial condition is obviously W)t = Y. for the oscillator potential.t) is known. H\En) = En\En). .16) As usual. (7. (7. x'. E.5. *>0.42)) Wi=0 = E l ^ ) ^ l ^ = o n (717) Thus at time t = 0 the state \ip)t is a linear superposition of the vectors \En) with coefficients (En\i/})t=o.The solution of Eq. (7. \En)(En\ip)t=oeE^\ n (7.7) with Eq. Eq.7. 109 (7.20) the Green's function K(x. (7.t) = J2 fdx/^n(x)^*n(x')^(x'.18) or.t)il>(x'. this relation provides the probability density \ip(x.19) i/>(x. t) with the Note that when K(x.2 for the computation of the sojourn time. (6.Q) = 6(xx').e.51) shows that we obtain a very analogous expression for the density matrix PN(P) a s for the Green's function K(x.g. 7.t) describes the evolution of the wave function from its initial value ^(a^O).15) which contains (7. (7. t > 0.g.2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih.
t). We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x. = J2 ^„(s)C(*') En — E We see that G = GE depends on E. between (with E^0' = E) G(x.25) in agreement with the timedependent Green's function . 7. t > 0.ie (7 .i /f e «/. we obtain then also the corresponding density matrix.«^«2 W . 24 )  ReE Fig.1 The contour of integration. i.t) K(x. Jc 27T „ hn . t) = J2 e £ " f M i ( x ) < ( x ' ) .x'. Green's Functions difference that (5 = 1/kT plays the role of it. With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x. we obtain / W : = . In the following we shall derive the Green's function for the case of the harmonic oscillator. x'. As a consequence of the above considerations one wants to know the connection between the timedependent and the timeindependent Green's functions.110 CHAPTER 7.x') and K(x.e.x') = GE(x.x'.E .23) along the contour C in the plane of complex E as shown in Fig. Inserting for GE+ie the expression above. (7. 7. x')6(t) (7.1.
A and B being constants. (7. we obtain *I£I = h2 —(2AB). B m0 n _ 2ih 2iK (7.e.3 The Green's Function of a Free Particle 111 7.0) = 1.t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7.7. K(x.7). In this case the Green's function is the solution of the equation d hK(x.e.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7. v ' 2m 0 v h i. (7.30) Inserting Eqs.x'.29) tJJ 2 2B(x . / dxK(x.— d2 —2K(x.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. (7.27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz.30) into Eq. i. (7. Thus we try the ansatz.X>.x>.29) and (7.26) This is the case of a free particle with mass mo.28) B(xx')2 t2 DB(xx') 2 /t (7. . We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7.0) = 5(xx'). and identifying coefficients of the same powers of t on both sides.x'.t) t h2 = . 2mQK h h2 ih(AB) = —{4B2A).x'. It is clear that it is nontrivial to solve an equation like Eq.31) The constant A has to be chosen such that K(x.27). which is moving in one space dimension.t).
35a) K(xy. n For a free particle moving in the onedimensional domain \x\ < L — oo.33).ihMxWn{x'). (7.t) — for instance for a wave packet given at time t = 0 of the form ^(x.e.zEnt."737) \27rii/i in agreement with Eq. t) = /jJo_ e mo(««') 2 /««. *>0.x't) = Y.x'.36) ^ W Q ^ J emo(ii') 2 = —eP2lia 27T dke~^k^2^2 J„00 /2«tS . 32) It follows that x'.t) into Eq. Mx)^M^) „ n so that J =^ . from K{x. we > have to make the replacements V ^ fdk. (7. (7. 158. 2m0 (7. (7.x'.38) ..t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—.34) K(x.(7. (7.20) and can then obtain ip(x. Merzbacher [194]. i.oo 27T (7. 1st ed. we have This is 1 provided A=JB= ^ (7 .H We can insert the expression for K(x. Green's Functions For parameter values such that the following integral exists.21). p. (7.t) = — / dkeak2+Pk 2vr J.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq. v27r En^^. j3 = i(xx'). "See the excercise in E.36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x.0)oce"Qx2+ifcox.112 CHAPTER 7.
7) is the solution of d ih—K(x. (21. HQ = ^—p2 ZrriQ + \m0uj2q2.x'. i.e.37) will later be obtained by a different method — see Eq. 5{ax) = ± J eikaxdk = ArS(x). (7. / ) (7.t) We now set h2 d2 = . —KK(X. §jK(x.3 The Green's Function of the Harmonic Oscillator 113 The result (7.x'.39) In this case the timedependent Green's function K of Eq. (7. (7.e. z . . (7. I (7.7. i. 1 2mo u>n ox n with the initial condition (7. K(x.33). .x usn ot i. x'.43) so that In the domain of small values of x (near the minimum of the potential) the Hamiltonian HQ is dominated by the kinetic energy.X ..0) = 6(xx') at / = 0./ ) + fK(x. x'. We consider the onedimensional harmonic oscillator with Hamilton operator HQ.t) + ~x K(x.40) becomes 2 . (7. the relation* 6(x) = o<y(os).e.. x'.x'.25) — in the context of Feynman's path integral method.t) 1 + m0u2x2K(x.x'. . mQLu .t).x'. in this domain *Recall that S(x) = ± f eikxdk.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.t).e. i.8).——K(x. / ) = ~^K(x. . 7. We rewrite this initial condition in terms of £ and use for a = const.t) = h2 2 d2 TW . .42) .40) Then Eq.d r. . —ih—K(x..
(7.c o t h 2 / .b2.50a) (7.50a) we obtain (7. Integrating Eq. b' = Aab. W ) .4a 2 )£ 2 . Feynman [94].47) We insert this ansatz for K into Eq. (7. (7. Identifying coefficients on both sides. and we expect K(x. c(0)^. a = . 50.) a'£2 + b'i + c' = (1 . we obttain a' = l . as the limiting case / — 0. (7.4a&£ + 2a . (/ = 2a. it • is suggestive to attempt for K the following ansatz^ m. Green's Functions to the particle behaves almost like a free particle.45) in accordance with (7.b2.48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7.51) See R. f i.Z'.e. P. so that for / — 0: > a^—.42) and obtain the equation (with a1 = da/df etc.46) If we interpret Eq. 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.50b) (7.47) becomes (7.49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7.33).^ .45) in this sense.114 CHAPTER 7. i. we must have /o = 0. p. that is..48).t) become similar to expression (7.4 a 2 . (7.50c) (7.x'.e. .f) with a(0)»y. (7.
Felsager [91]. / —————rexp y 27rmsin(u.50c) yields for c(/). (7. (7. t. 7. 2TT^ Inserting a(f). x'. if we return to x.In B. (7. with A.7.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq.e.54) For t —• 0 this expression goes over into the expression (7.c(f) K = 5 =exp Vsinh 2 / into Eq.33) for the Green's > function of a free particle.48) 6(0) = — £'/2f. Finally Eq.t) = .45). (7. Comparing the Eqs.53) or.* With this result we have another important quantity at our disposal. in particular for the derivation of the sojourn time in Sec.t) to obtain this element (x. (5.48).x'. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x.27).x'.5.b(f).0).£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o. i.X'. B independent of / . we must have A = ('.47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7. we can use K(x.i) (7. K(x. . p. i.2. as one can verify. (7.55) 'For an alternative derivation and further discussion see also B. 115 To ensure that in accordance with Eq.50b) Kf) = —r^77) smh 2 / A independent of / . x . c(0) = £' / 4 / . (7. (7. as we shall see in the following.e.40) of the timedependent Green's function with Eq. c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / . 174. (7. In order to satisfy Eq. we must have (besides A = —£') = /m 0 u. and to ensure that we obtain the prefactor of Eq.51) for the density matrix PN(X./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7.
56a) the expression (7.55).56a) {q2) = = Y. (5. <2 (7.52): Tr (PNq2) Jx2pN(x. For instance we have with Eq.p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7. Green's Functions With this expression we can evaluate (cf.P) 2. at temperature T): {A) = Tl{pA) = ^ > . R.40)) 1 epEi 1 Ui = y.x..e.52)) the expectation value of an observable A in the canonical distribution (i. 11 For T ~+ 0: u>0 .eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i. we obtain § < Z> = ^ c o t h ^ ^ ° — . Thus the system is in a mixed state and the expectation value "Cf.x. (5. (5. P.» l. ^—• Irpiv J pN(x.x.e. (7.uij>0 >• 0. 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"}.116 CHAPTER 7. eP E i = i + y. [ [ 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. Eq. = jdxp(x.. )x Inserting into Eq.p)dx W ) = —^r = —p . Feynman [94]. . Eq. p. 52.56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf. we skip the algebra here.
We return to the Green's function (7.54).7.t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ . x'. (6. 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 (7.3 The Green's Function of the Harmonic Oscillator 117 (7.59b) n+ l \ ftwt here 2hf exp _ i ( n + . This is the first and hence dominant term of the expression (7. 2 + x'2)le^2xx' to—>w—it t—>co 2 . of K(x. We assume t > 0 and t —> oo. If we consider the system in the pure state \i).58) Next we explore the connection between the explicit form of K and the latter's expansion in terms of a complete set of states. which means in the oscillator state \i) with eigenenergy fku(i + 1/2). x'. t) of Eq. <*2>o dxx a •K j e IT I da r dxeax *\1/2 da a J _ 1 _ _ h 2a 2mou j —( . 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. (7. In this case the Green's function K(x.59a) E 0 = ^fiw. and we replace to by LO — ie.21).56b) is that with respect to this mixed state (whose cause is the finite temperature T).t "With the normalized ground state wave function of the harmonic oscillator given by Eq.J (w — ie)< e (n+l/2)Et e i(n+l/2)u.50) we obtain. setting a = mooj/h.54) is K(x.x'. e > 0.e.QUJ (7. 1 mo^ exp lmow / 2 2 ft X (7. i. /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0.
Green's Functions For t large (i.59a). i. 7. is defined as vv = .118 CHAPTER 7.5 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.e.e. as in Eq. The wavelength A is given by The phase velocity v<p. i. the particle will stay there only for a finite length of 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. (7. k = k. lie on surfaces.61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const.1 W a v e packets The simplest type of wave is the socalled plane wave or monochromatic wave of frequency UJ represented by the expression exp[i(krwt)]. However quantum mechanically in view of the uncertainties in position and momentum.e. In a very analogous manner we obtain the solution of the equation for the density matrix. (7.. (7. to infinity) the contribution with n — 0 dominates. In Example 7.62) . which are planes.1 we estimate T semiclassically. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely.x'. and with this we estimate the sojourn time T. the velocity of planes of equal phase.(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7. of dpN as pN(x.5. Considered classically.
t) is the spatial Fourier transform of this Gauss distribution — as we know.)ll/(fc")l[cos(¥'(fe. the fundamental postulate on matter waves in Chapter 2): E = hu. t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case. /OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe.e. One defines as centre of mass of the wave packet that value of x for which d(p — = 0. (7.e. then ^(r. i ) = /*/(k')e i(k '" r '' . e . i. e. We now ask: How and under what conditions does the time variation of the function ^(r.k ') 2 . i. for which \ip\ assumes its largest value: OO .64) where / ( k ' ) differs substantially from zero only near k' = k.Q ( k .4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf. duo' da . The wave packet describes a wave of limited extent. ^ ( r . a > 0 .65) Let f(k') = \f(k')\eia and <p := k'x . (7. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k.7. essentially again a Gauss curve.<p(k"))}dk'dk".Jt + a. ' t) dk'. xt— + — = 0. to the function ip given by oo / fik'y^'x^dk'. i. u = w(k). oo (7.)¥'(*:")) +ism(<p(k') . If we assume for / ( k ' ) a Gauss distribution. or dco' da The centre of mass determines the particular phase.63) The relation u = w(k) is known as dispersion or dispersion law.e. / oo J—oo .
Example 7.e.120 CHAPTER 7. a > 0. p = hk the group velocity is equal to the particle velocity v. (7. we can use a simpler method. 9 cku dk dhw dhk dE dp dE dp' We can also argue the other way round and say: By identifying vg = v. Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0. i. For E = hio. we obtain the deBroglie relation p — hk.^ g=r a rad. (7. . The threedimensional generalization is evidently du . . as claimed for =0. Solution: The function to be calculated is the integral dxeax OO ezkx.fc w> vv = ~a^ = S r a d P E «i7 Sd u . 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 = .1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const.. In the present case. = — (768) It is instructive to consider at this point the following examples. however.67) The centre of mass moves with uniform velocity vg called group velocity of the waves exp[i(kx — cut)].— g(k).69) In general one uses the theory of functions for the evaluation of this integral. dE v 9 = .
.70) to verify the following representations of the delta distribution: X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m^r^. (7.. We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 . MiillerKirsten [215]. where c = 9(0) is a constant..3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7.r „• With the help of Eq. (7. XZ (e > 0).69) and (7.69) and (7. 2 f + .70) we obtain 6(h) = lim — f°° dxeax2eikx = .43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x .71) (7. Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 .1 verify the uncertainty relation Aa.Afc = 81n2. —±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration.a) + 6{x . W. (7.4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a .&)] for a + b.73) 27T 7 .73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim .75) ' E . (7. (7.7..7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example.2: Representations of the delta distribution Use Eqs. .72) Solution: From Eqs.74) Example 7. \a — b\ (7. J.. (7. (7.0 0 7T 7 0 7T e From Eq. with partial fraction decomposition. or see H. Appendix A. g . Example 7.
78) It follows that the product of the uncertainties is AxAk = 8 In 2. A simple calculation yields A i = 2v/2~In2a. where g(k) = max<?(fc)/2.70) the Fourier transform of f(x) is (7.e. According to Eq.CHAPTER 7. Thus a slim maximum of the curve of /(a. i.r)1/2a Re f(x) Ax Fig. Green's Functions 1/(27i)1/2a — 1/2(2. Thus a sharp maximum of the function f{x). where  / ( x )  = m a x  / ( x )  / 2 .76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum. at time t = 0). 7. The breadth A/c of the curve g(k) around k = kg.) leads to a very broad maximum of the Fourier transform \g(k)\. (7. The wave function corresponds to the function f(x) in the above considerations (e. 7.e. and hence implies large values of Ak. requires according to Eq. the width Ax is a measure of the uncertainty of the probability.* o ) 2 / 2 . Solution: In Fig.« 2 ( * . Physically . i. (7. In quantum mechanics the square of the modulus of the wave function ip(x.2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7.77b) (7.t) is a measure of the probability to find the particle at time t at the position x.77a) small values of a.2 The Gaussian curve.g.77a) g(k) = e . when Ax is very small. is Afe = 2 V 2 1 n 2  (7.
71) > and (7. The result (7.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} .. at time t = 0.4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x..". nr = S(x). Thus G.x'. many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration. (7. In the limit a — 0 we obtain from Eqs.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.h = 1.«> = / ^ C ' W . As G.* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq.e.2 ) and U)t. the less precise is the determination of its associated momentum.u> = l.76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O . (7. Barton [15]. This means the particle is localized at x = 0. (7. 7. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate. ^In Eq.20) the wave packet at time t > 0 is obtained from that at time t = 0 with the relation «. . 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. O ) . 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. For reasons > of simplicity in the following we set in addition t mo = l.54) with the substitution ui — iu. E. <7*» We assume that initially. In the same limit the momentum uncertainty Afc grows beyond all bounds. which means all values of the momentum are equally probable. Barton recalls that W.79) According to Eq.54) these parameters appear in the combinations moui/h (dimension: l e n g t h . (7. The function g(k) is therefore described as momentum space representation of the wave function f(x). Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system. i. Barton mentions at the beginning of his paper.7. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over.81) *G. (7.5.75) implies therefore: The more precise the coordinate of a microscopic particle is determined. so that Ki0(x.
2 (7.79) into (7.84) we obtain* ip(x.81) and (7.. Green's Functions Inserting (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 (7.83) ip(x.84) dx' oo tvirl/2b With some algebra one can show that (B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = .t) = J oo P . 9 (7. + cosh 2 1. . 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 .82) i/f(x.(i/2) sinh 2t 2 2 2C 2 (i) + *c^.(Ax' + Bxf \/27risinh + (B2 + ^cothi)a. = . (7.85) = —r= o2 7 2 sinh 2i.87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*. (7. ' 2\Aswht % cosh t (7.88) ..124 CHAPTER 7.t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7.cosh t sinh t' Wisll2b ^ (7.80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh . A .83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7.86) Evaluating the Gaussian integral of Eq.
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. 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).t)\2 Ji == 71 / V " Jo dO ? 6' (7.. since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt= / (7.94) / dxf(x)=g'(y)f(g(y)).£ 2 (7.4 The Inverted Harmonic Oscillator The real part of Eq.t)\' 2 125 (7. _2 1 exp C 2 (i) = l + 2sinh 2 i.90) with the expected limiting behaviour \im \ilj(x.89) = exp[x2/C2(t)} 7rV2C(t) (7.86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1. We have therefore .t)= where Co(t) / dx\^(x.92) The sojourn time T is now defined as the mean time T given by T:= r*° ioo 2 (M)s r*(f4 d^e. t) the phase ip drops out and one has (x. (7.95) . •&(t)=//c(t) = / or.91) C(t)' (7. In taking the modulus of ip(x.t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 .7.
=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'.(4 2 . Dwight [81].From this we obtain C(t) C'(t) Z2 ^(r?2_. p.98) We insert this into Eq...97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e. 136.99) .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 /2^7.2)(p_7?2)(7.96) °° l Jr. (7. £ 1 dr\ 2Z %_ rl 2 2 yfx Jr.97) sinh 2 1 = .= \ / l + 2sinh 2 t. V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7.1 ) 2 \n J ^"•""ff^^V^K?.126 We set I C(t)' so that dT ~d~l Now.=0 V^ C'(t) I Jt=0 V^ (7. C(t) = . we obtain dT _ 1 ~dJ~l 'See for instance H. + 0(e'2) (7. B.
also motivated by this paper. . 238.102) 'As a matter of interest we add that in the theory of a free scalar field x.7. See e. B. It follows t h a t T~lnZ. (7. U) (7. Explain the parameters entering the calculation. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7. is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above. Such a field arises in the spectrum of states in string field theories and is there called a "tachyon". 7. p.g.4 The Inverted Harmonic Oscillator 127 for I sufficiently large. this equation is the equation of motion of this scalar field with negative masssquared.4.100) Reintroducing the dimensional parameters we had set equal to 1. The equation there describes the classical "rolling down" of this tachyon. Zwiebach [294]. T h e following Example 7.3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7. Solution: In the case of the inverted oscillator (representing the egg) we have in the usual notation the classical Hamiltonian Hwith time derivatives dH dp P mo ' max = p dH dx = mow x. Fig. this is T ~ ]n(ly/mou>/h). 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.
(p2):=mW(x2). i.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.e. and solution x = A cosh Lot + B sinh cot. and for simplicity we take (cf. so that x(0) + il^L mow It follows that For a minimal uncertainty (hence the factor of 2 in the following) we then have h= 2 Hence uT P ( ° ) _ O . (7.g. v A x A p = \ (x2)(p2) v = m0u(x2). the smallest macroscopic length. 2 x(0) ^ where A i . Green's Functions x = Aco sinh cot + Bco cosh cot. Ap= v /(p 2 )<p)2. 2V ' Let A and Bio be the values of x and x at time t = 0. Eq. (x 2 ) h 2mou 2mou> and so l^J^ULl).5)) Ax = <J(x2){x)2. pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet.p(0) cannot be determined with arbitrary precision.e. A p are defined by the mean square deviations given by (see Eq. For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0). Quantum mechanically x(0).102)) p(0) = ^/(p 2 >. ^(x2)(p2). Thus quantum mechanically x(0). We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I. but instead are subject to an uncertainty relation of the form AxAp > h. For h —> 0 the time T —• oo in agreement with our classical expectation. . so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*). (10. 2 10 x(0) = B u . the expectation value of x is the integral with an odd integrand) <x) = 0 = <p). CHAPTER 7. Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T.128 with x2 = LO2X2 + const. (7. / 7 ^ 2 =2J(x }. i. so that e. x(0) = A.
2h2 cos 2 a # = 0. 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. Frequently even the calculation of the next to leading contributions ip^. In general perturbation series do not converge.1 Introductory Remarks The Schrodinger equation can be solved exactly (i. An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with onedimensional Schrodinger equation given by ip" + [E.(i)+/^(2) + ". i. In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. 129 . but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4.Chapter 8 TimeIndependent Perturbation Theory 8. For mathematical purists the question of convergence of the series is an even bigger challenge.E^ is already a bigg problem.e. EW+(3EW+(32EW +  ((3 can also be thought of as a kind of "bookkeeping" parameter in retaining corresponding powers of some kind of expansion). one would set tt E = = ^(°)+/ty. in closed form) only for very few potentials. It follows that in general one depends on some approximation procedure which is usually described as a perturbation method.e.
. concerning also the uniform convergence of the exponential series. . h hl are in the strict mathematical sense divergent.1) and its terms with the behaviour of the following series and its terms: f(x) = 1 . Cll E = a_ 2 n + a_i/i + a0 + — + —r \ . perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified. Schafke [193].g. Watson [283]. E = E^ + h2E^+hAE^ 71 72 7 + . p. W. 581.. 0.= 0 < 1. Whittaker and G.• + (ir~ X X* 1' 2' r?l Xn + Rn(x). at the end of Chapter 26) — and by comparing the results of WKB. the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence...g.+ .g. . Meixner and F .g.e. It will be seen later (e. Actually expansions of this type which are ubiquitous in physics are socalled asymptotic series which were originally also described as semiconvergent series in view of the decreasing behaviour of their first few terms. T.* i. the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •. In the following we explain the difference between convergent and asymptotic series.3) *See e. J. e. This convergence can be shown with D'Alembert's ratio test since''' lim . see e.130 CHAPTER 8. E. the expansions in descending powers of h2.2 However.2) n—»oo n It is interesting to compare the behaviour of the series (8. which is known as the Mathieu equation. (8. ^For further details. (8. 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. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation. The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x. N. TimeIndependent Perturbation Theory In the case of this equation.
.g.0. We observe in the first place that since lim n—>oo co (8. From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ .5) Cf..0. after reaching 6/27 = 0. . E.1). It is evident that the larger the value of \x\. As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!. we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 .. + •••...9876. Watson [283]..8. the series (8. .222. and we see that the moduli of successive terms first decrease and then. Whittaker and G.3) with the convergent series (8. i. e.. / ( I ) = 1 — 1! + 2! — 3! H • However. the partial sum of terms up to and including the least term yields a reasonably precise value of the function at that point with an error of the order of the first term of the remainder.3) diverges for every value of x.376 .000 000 06 + • • • .22222 ...g. for x = 3. +0.e.29629. .2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum.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.4) for every value of x. + 0.2.4938. the better the approximation obtained.3) can still be used to obtain almost correct values of f(x) for x sufficiently large. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20. + 0. p. Comparing the asymptotic series (8.001 + 0. for larger values of x. /(1000) = 1 . Thus e. Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1. We can see this as follows.0. ... The theory of asymptotic expansions claims that if the expansion is truncated at the least term. N.0. begin to increase again.22222 . in spite of this the series (8.. . However.l n ( l + ^ ) n=l L ^ ' (8. 235.33333 .. and in fact increase indefinitely. T.000 002 00 .
for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z. p. p. Vol." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent. T. § However.1." These "rapidly converging power series" in physical contexts are most likely very rare cases.5). It is interesting to observe the comments of various authors on this point. This latter observation hints already at the importance of asymptotic series in applications. In fact. 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.J In2. p. Messiah [195].5) are practically unknown. although this has not been investigated except for a few simple problem^. 11 L. Whittaker and G.e. in fact. meaning convergent. Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems). Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent. . It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small. 16.132 CHAPTER 8. N.1 between Eqs. it is sensible to make the assumption. at a later point in his treatise Messiah admits that the expansions are mostly § Cf. Sec. 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. the leading term of an asymptotic expansion. 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. TimeIndependent Perturbation Theory where 7 = 0. 371. whereas applications of the convergent series (8. **A.41) and (16. the dominant approximation is. ip) can be represented by rapidly converging power series in e. E. (16. I. In many respects Messiah aims at more rigour in his arguments. Schiff [243]. On the other hand Schiff" says:" We assume that these two series (for \I/ and E) are analytic for e between zero and one. This is not appreciated by mathematical purists. that E and \ijj) (i. ^E. 236. 152.5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function. Watson [283]. Merzbacher [194]. It seems that Schiff tries to cling to the idea that a proper series has to be analytic.
there is no need for such bias.. . solutions of second order differential equations. B." Of these three authors t h e first. Replacing t h e exponential in Eq.7) as / e'* dte* dt = 1 . (8..10) Vn Changing t h e variable of integration to u = t2x2. which have been studied in great detail. Dingle [70]. seems to be closest t o t h e t r u t h a n d does not attempt t o give t h e impression t h a t t h e series has t o converge. In fact.^=e~x2 / e^^dt. It is therefore suggestive to write iy poo <t>(x) = 1 .1 T h e error function a n d Stokes discontinuities Another extremely instructive example which illustrates t h e n a t u r e — a n d in addition t h e origin — of asymptotic expansions is t h e error function 4>{x) denned by t h e integral <f>(x) = 7= / V71" Jo 2 fx e _ t dt. We know from books on Special Functions t h a t all of these functions. possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics.2.8. Merzbacher. Messiah [195].11) A .= / e~t2dt.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. II. (8.7) 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.. *R.ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T . 8. Vol. He says. (8. 198 (German edition).2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic. ft Jx du = 2tdt. (8. (8.7) This function has been considered in detail in t h e book of Dingle. we can rewrite Eq. 2 (8. Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion . p.* and we follow some of t h e considerations given there.
134 we obtain CHAPTER 8. 7T (ni)!sin{7r(n+^)} ( „ . 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion .14) sin7T2. we obtain 1+ i.i)[(_!)«' Then.16) .12).^ X e U ~ ( 1 + * \ 2 . since (—4)! = y/n. (8. 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 (8.13) presupposes t h a t < 1.e. « \~2 =E n=0 l 1 V5F(ni)!(l)n nlTT (M x2) / « r oo 1+ For the domain \u/x2\ <f>(x) U \2 1 \/K n=0 2< n! V (8.12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8.15) < 1 we can insert this expansion into Eq.^e~x \Ar i. (8.' we have 2 nH 7T 7T (8. TimeIndependent Perturbation Theory <P(x) = 1 . du.e. Using the reflection formula (*!)!(*)! = for z = n + i .
' arg. (8. We now want to relax this condition and allow for phases argx ^ 0.18) or that we insert the binomial expansion (8. The expansion (8.19) approximates <f){x) the better the larger x is. that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned. i. (8. and in fact so slowly.8.r<7r.e. As in the case of the gamma function.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq.l)! = / Jo Then 2 cHnxdt. Whichever way we look at the result. the asymptotic expansion of the error function is. Frequently "=" is written instead of "~". In the foregoing discussion of the error function we assumed that x was real and positive. It is fairly clear that the above considerations remain unaffected for argx < 7T.e.19) implies either that we ignore the remainder or the correction term given by the integral in Eq.15) into (8. argx = 0. We can actually understand the deeper reason for this in practice.12) ignoring the fact that the latter is valid (i. In an asymptotic expansion the first few terms successively decrease in magnitude.—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^. convergent) only in the restricted domain u < x2.2 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. i.e. whereas in a convergent expansion the terms first increase in magnitude. (8.19) originates.17) 0(X) = i. much more useful than the convergent expansion. 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 see how the asymptotic expansion (8. /•oo T(n) = (n . in general.
i. (8. V7T Jo Changing the variable of integration to v (8. We therefore consider this case in detail.24) We can read off the binomial expansion of the factor in the integrand from Eq. dv = 2sds = .15). we can try to proceed with Eq.22) we obtain 2 4>{iy) = .^ f V e ^ . (8.20) where we set t = is. (8. =dv. however. The situation becomes critical.136 CHAPTER 8. (8. We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 .vds.e. (8.26) or else return .7) and obtain 4>(iy) = —= j e~l dt = = / es2ds.2 y V . VK Jo V^ Jo (8. we set x = iy in Eq. TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained.^ = .23) Throughout the range of integration 0 < v < y2 the integrand is real. Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds. (8. thus with \v/y2\ < 1. 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). when argx = 7r/2.21) = y2s2.
(8. (8.28) ignores the integral contribution of Eq.^ ^ p . For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part.26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! . e ^ ( i y ) ^ ^ .30) yvn f Jo V y J Cf.^] * dv. (8.27). i (8.27) and (8.2 yir (8. for large values of y. . and so the integral is real.2 Comparison of Asymptotic Series with Convergent Series to Eq.28) because then the correction term oc f°£ becomes smaller and smaller. We know that for v < y2 the integrand is real.24). Returning to Eqs.28) n=0 We observe that this expansion differs significantly from (8.19). (8. Suppose now we consider Eq.e..e. (8. y y 2 ( n _i)! y arg(* = iy) = i * . We have therefore <j>(iy) = ^e^TL fV e~v ( l . However.24). i.28) we observe that the expansion (8.27) Then .8. i. the better the approximation expressed by Eq. (8. ey is much larger than something of order 1/y. we obtain 4>(iy) = ^ey2TZ f°° e~v (1 . — (8. which will be studied in more detail below. (8.29) where 5t means "real parf. 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.^ ) * dv JO V y /»00 /»' yV^ 1 i Vsfv 2 ey n oo J0 Jy' 1%) 'dv. If we proceed with Eq. Eq. „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! . (8. for  arg x\ < 7r/2 and for arg x = IT/2 the error function 4>(x) possesses different asymptotic expansions.17): J£° e~vvndv = n\. and the larger y.
2 OO rz=0 V . . Chapter 1.8)) 4>(x) = <f>(x) we see that _x2 OO / ] \  n=0 for 7r/2 < I arg x\ < 3TT/2. Looking at Eqs. (8. and • higher order terms of the associated series all have the same sign.31). Eq.28)." — E ^ ^ forarg. (8. (8. B.3!) which is (8.e. TimeIndependent Perturbation Theory If we insert here the expansion (8.. e xZ ^ (n — £)! „ ^ ) .e. (8. we use the expansion outside its circle of convergence and hence obtain a divergent expansion. Dingle [70]. is obtained for <f>(x) with arg x = —ir/2. (8.* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself. This result is identical with that obtained previously. R.31) and (8.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i. 7r = .138 CHAPTER 8.25) and integrate from 0 to oo. For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2.33) 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. With a similar type of reasoning one can show that the same expansion. Since (cf. The corresponding phase lines are called Stokes rays. We have therefore found that XTT n=0 ^ ' and . i. .33)) would have • a maximally increasing exponential e+x . (8. *Cf.26). y.30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8. since (8.
36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x.e. for arg x = 7r/2.2. 91. See also Tables of Special Functions. and so may not be universally applicable. (8. and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a. pp. Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o . e. in the sum of the right hand sides of Eqs. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0. Equation (8. which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x). i.2). Dingle [70]. 1/1 ipv{x) v1 vl (±ix).g.35) 8. 5. Oberhettinger [181]. .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. Magnus and F. Thus if we set x = XR + ixj.rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8. '"Cf. 9 .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. p.e.37) Cf. 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. (6.8. B. R. i. apply predominantly to asymptotic series of the solutions of second order differential equations.32) and (8.36) 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. W. though. Chapter 1. Eq. These rules. —1 cancel out.34) the contributions 1.13.
Hence Dil){x) = xv(j>v(x). 0 < argz <  (8. TimeIndependent Perturbation Theory i. We consider Dv (x).37). 9) and XXI. 8. The continuation of (8.39) (as in the example of the error function). 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.36) is invariant under these replacements. R. B.e. Chapters I (in particular p. Dingle [70]. 1 1 Cf.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. x~l/~1ipu(x).38)) has a Stokes ray at Dl '(X) the arg x = 7r/2. R. . (8. B. i. v — — f — 1.7.39) 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. Thus D{J\x) (i. (8. ti asymptotic series (8. = 7r/2.x 2 oo «*)  — E^fi=0 v . At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign. We also observe that the late (large i) terms of the series in (8. (838) We observe that one solution follows from the other by making the replacements^ x — ±ix. Chapter 17. D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a. (14). i.e. ' S u c h symmetries are extensively exploited in the perturbation method of Sec.ei<v+VxvHv{x\ argx = £ . Eq.140 where^ CHAPTER 8.e. > > The equation (8. as can most easily be seen in the case of the Mathieu potential. with a real proportionality factor.40) § Cf. Dingle [70]. Chapter 1.
i. (8. with \x\ = (—x) for arg a. beyond arg x — 7r/2.e. is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray.40). The value of Du (x) on the other side of the Stokes ray. we must have at arg x = 7 in the dominant factor on the right of Eq. i. < 37r/2. as yet an unknown real constant. Applying Dingle's rule (1) (to the Stokes discontinuity of ^v{x)) we obtain on the ray D<P(x) = x^M^ + aUxy^M^ + l^e^e^lxrMx)] (8. where in the first line on the right hand side the second term contains the real proportionality constant a/2. We therefore proceed to continue (8.41) the Stokes multiplier a is. = ir.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2.8. = ir. Applying rule (2) we obtain D$\x) = U™ + ^iap) (xyMx) + a{x)vlMx) (843) for IT < arg a. (8.44) + ^iap) = 0 . Since Dv (x) is real when x is real.41) xuct)v{x) + a{x)~u~lipu{x).42) T ^(eiwu i. the extra phase factor u ei7r/2 a n c j t n e p h a s e ew7r/2 0 f x <j>u(x) (so that as the rule requires the added contribution is 7r/2 out of phase with the first). sm(iri/) = —a/3/2 or a/? = 2sm(irv). (8. e +fx2 _ 2 2 e±(x Rx I+2ixRxI) is maximally exponentially increasing (i.41) to arg a. < 7T. for xj = 0) and therefore possesses a Stokes discontinuity there. Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga. It is determined by continuing the asymptotic series to argx = n and demanding that the result be real since Do (x) is real when x is real.e. On reaching ir the part containing ^ ( x ) . (8. another half of the discontinuity appears on leaving the Stokes ray on the other side.42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x).e.e. In (8.
(8.—^ ON • — ^ > + ^ ^ L .e.v .42)) for argx = ir (in the second line of Eq. We can see that the equation is satisfied by (3{u) = ±a(u . i.1). (8.1)! TT.50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq. V i=0 \U0 v 2w .14). TimeIndependent Perturbation Theory Now a and (3 can still be functions of v. (z)\(ziy. argx = vr. i.48) We compare this with the reflection formula (8.e. Eq. i.1) = ±a(y).1) (8. (8. Dl \X) is (cf.47) Equation (8.e. 8. a{y)P{y) = a(u .1) = 2sin7ri/. Hence the left hand side must have the same property. i.l)0(v . a(u)P(v) = 2sin(7r^) = 2sin7r(i/ .45) is therefore given by a{v)a(v .42) we insert from Eq.45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1. (8.142 CHAPTER 8. (3{v .42) that a multiplies ipv(x).51) .46) For a given function a(y) this equation determines /3(u). .e. or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq. (8. (8.45) that af3 = 2sin(7r^)) D^\x) = (8.48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 . (8.1) (8.38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i .
e.e^ .7). It should be noted that in Eq. 8. However.2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8.)(.1 ) ( ..50a) because this enables us to define normalizable parabolic cylinder functions. (8.1.. ' ^ (i/l)!z! ^ . For details we refer again to the book of Dingle.) (8. solutions of Eq. We observe that for v — n = 0. the second contribution vanishes and we obtain £>«(*) .„ _ix2^(2ivl)l e 2 > —.36)) in the domain 0 < arg x < IT. (2x2)1 = COSTTU(X) v .. (8.53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator. which vanish at x = ±oo..A / ± sin^Cx)"M* 2 V ^ . In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0.e. i.*"ei°> f ( ^ _  _ L _ = (!)««(. the error ..36).. (2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di.7r).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. ' (X) (cf.3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8. Chapter I. u 1 y> {2i + v)\ 1 n 0. i.53) We have therefore obtained in a natural way the quantization of the harmonic oscillator. (8.I / .8. Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir.1 ) ! 4~L i\{2x2Y i=0 V2^(x).35). Eq.e. —f. i.52) (argx = 7r).2. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ . (8. (8.
. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 . 0.54) we consider again the important equation of the harmonic oscillator.54) (8. Eq. We write Eq.56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8. x°.e. + — + ^ + • • • . how series of the type (8.60) 2 x x l Inserting (8. i. x. cf.36). S2. \ + U (8. (8. (8.36)) 0 where + x2Q(x)y = 0. Eq. x—>±oo (8. % = {. ^ + 2 .36) in the form (coefficient of — x2 here chosen to be ~.38) can be obtained. 5„=4 Sl=o..37).58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]S0x2 + SlX + S2 In a. (8. Instead of dealing with Eq (8. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation. S\. we obtain equations from which the coefficients So. and demonstrate. Solving these equations we obtain = = 0. previously we had ^. i.144 CHAPTER 8.59) and equating to zero the coefficients of powers of x2.e. (8.^ + 1 ) . (8. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on.60) into (8.55) The large x asymptotic expansion can also be obtained from the differential equation.. • • • can be determined.
38).+ X M W = Q.38) in a variable for the solutions of approximated differential equations. In the following we are mostly concerned with asymptotic series in some parameter.b. (8.61) (the dominant terms of (8.z) is a confluent hypergeometric T.63) **W. 4~l~2 ' i i _i. / i N . Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before). n 22e"T JziL (8. Magnus and F.8. a z a a ( + 1) z2 which has the unit circle as its circle of convergence. p. are asymptotic series in a variable x.35) and (8. We. therefore.37).19).e.3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. The derivation of the asymptotic series from integral representations (cf. 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+ .91. T h e expansions (8. » 1.62) 4 \ 2 XiFi 2'2 .37) with (8. In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains. and it is clear t h a t higher order terms follow accordingly. . Oberhettinger [181]. do not enter here into their derivation and simply quote the result: W i t h \x\ . Nonetheless we shall need asymptotic expansions like (8. \x\ ^> \v\ and (a) for  a r g x  < D„{x) 3/4TT: e 4 x x i/(i/l) 2x2  v(y\){V2){yS) 2Ax2  . (8. i. etc.36) have correspondingly e±x / 2 ) . function. 2 X 2Wi. 2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a.
whereas e~x '4 increases exponentially.65) We observe t h a t the series of (a) and validity 1 1 7r < a r g x < 4 In this domain 'Qx > 9ffcr and so e (b) have the common domain of 3 3 7r.1 a2 an a0 + — + ^ + .4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion.+ — + ••• of a function f(x) for a given range of a r g x . If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion . Thus there is no contradiction^ between (a) and (b). top of p. Watson [283]. Whittaker and G. T. 8. . TimeIndependent Perturbation Theory (b) for 5/47T > arga? > n/4: Dv{x) ~ e 4 v x v{y\) 2x 2 V{y\){y1){yS) 2. N. 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 3.0. 349. 4 (8. Thus the series expansion (a) in this case is multiplied by an increasing exponential and the asymptotic equality " ~ " means an exponentially decreasing contribution has been ignored.66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases exponentially.4x2 (8.4x 2 27F (*/!)!  e^e^x""1 1 + (v + l)(i/ + 2) 2x 2  (i/ + !)(»/ + 2)(i/ + 3)(»/ + 4) 2.146 CHAPTER 8. E.67) Cf.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.
In fact. (8.69) and the consequent nonunique definition of the expansion. The equation to be solved is the equation H<S> = EI/J with H = H0 + eV. Chapter I. A critical discussion of the definition can be found in the book by Dingle [70].73) Vn + q/#> + e2T/i2) + • • • .68) is (effectively) simply a statement of the observations made at the beginning.5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) . possesses a whole chapter on asymptotic expansions.Sn{x))  = oo n—>oo (8. thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics. the divergent nature of asymptotic expansions together with a vagueness of definition (which is avoidable as explained by Dingle) frequently tempt mathematically prejudiced purists to turn away from asymptotic expansions. we set E^En * > ^ n = En°1+eEnV = + e2EnV + . lim xneW x—>oo = 0 (8.72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e. first published in 1902. which is assumed to be small.68) The definition(8. (8. .e. It may be noted that Whittaker and Watson's internationally aclaimed text on "Modern Analysis".8.70) It is assumed that the spectrum {En are known with and the orthonormality Smn = (lpm\lpn) = } and the eigenfunctions {ipn} of HQ W n = 4°Vn (871) / dx%l)*m(x)lpn(x). (8. 8. The definition is due to Poincare (1866).g. in particular with regard to possible exponentially small contributions to the expansion for which e. i.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.
As a consequence the state vl/) is a vector in this space.e. °° i J or 0 = EWJdx{rnVil>n). (8. (8. We obtain then the following set of equations: Hrj. With the help of the orthonormality condition (8.72) we obtain oo roo dxifc{H0 OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n.148 CHAPTER 8.78) into Eq.e.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n .n = 40)V>n. ^ ) = 0.e. i.76) The first equation is that of the unperturbed problem. i.4 0 ) M = / ^C(4 1} .n. The second equation can be rewritten as {Ho . TimeIndependent Perturbation Theory We insert these expansions into Eq. .V)i. so that we choose / • dxrpM^O. i.74) (8. as are also the states Wn )j Wn )i — We therefore write these vectors as superpositions of the basis vectors provided by the unperturbed problem. Ho^+V^n = EW^+E^n.70) and equate on both sides the coefficients of the same powers of e. J f°° dxrn E(^ 0) . (8.w » . (879) We multiply this equation from the left by ip^ and integrate over x. (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.77) The states \tpn) of the unperturbed problem span the Hilbert space"K.i4 0) )V4 1} = {EM . ( ^ .75) HoW + vW = E^+E^+E^. We insert the ansatz (8. (8.
84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / oo dxi.Viprt e (8.82) *> = ^ + E .^(EW ~ V ) ^ + E& / J—oo dx^miln . (8. In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution.83) makes sense only if Ef] ^ EnV for all i + n.Vil>n) (880) In order to obtain ipn .76) which we rewrite as (H0 . with En . i. we return to Eq.5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn. (8. We multiply the equation by ipm. ^ n ) + 0(e 2 ).V)^ Setting + 42tyn. and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ . (8. (883) Thus the procedure forbids the equality of any E\ '. (il)i. It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n. multplying Eq.4 0 ) M 2 ) = ( 4 1 } . (8. (8.84) (885) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above).a\^(E:S0) „• 4 0 ) )^ = / J—oo dxrl?m(E$V)1>n. the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates.e.79).m ^ n and integrate: oo /•oo / oo dxrmY. i. the coefficients a\ .i ^ 0.8.e. S ' ti^+Q^)• We observe that Eq. We therefore consider Eq.
^ ) + $& w £ (^fc.Vlpn] M _ P (0K 2 ^r .VV . (8. ^ n ) j&iift + ^n .V^n) .e.88) into xjjT we obtain similarly * r i + f ^ i^n (lpi.86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq. n) £f) (8.B i 0 ) ) ( £ f .2 (V> m .85) together with (8.150 CHAPTER 8.(1) (i&0) .^(8. (8.86) into the expression for En we obtain En = 4 ° ) + e ( ^ n .Vlpn)(lpn.4 0) ^+ e 2 E (lpk.Vlpn)E.Vlpn) E^ . En0) (8.81) we obtain (2) _ _ {lt>m.79) •/—oo oo / Using (8. TimeIndependent Perturbation Theory Setting m = n we obtain with (^„.V^j)(^j.88) (Q) Inserting Eq. FVn) + e 2 ^ ( V > n .^w .V4>n){lpn.87) (i>k.^j)(V .89) • • • where we guessed the term of 0(e 3 ). ipn ) = 0 i. (8. V ^ i ) X (Q) (Vi.£i°») 2 Hence +^£o (£«? .90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' . ~Ej E E.V'>pTl (E^Eny V'fc (8. Inserting the expression (8. j.
8.90) can be found in just about any book on quantum mechanics.90) is the term with only one summation J2 m the contribution of 0(e 2 ).e.89). One would prefer to lump it together with the other contribution.lpn+i) = ^ Cj5k. i.Vlpn) = y^Ci(lPk.n(x) = J2crtn+ii (8.5 RayleighSchrodinger Perturbation Theory 151 The expressions (8.8.7 is based on this procedure.n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0)... An ugly feature of the expansion (8.g. + l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w. in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i. sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e. We can therefore rewrite tpn ': The perturbation theory described in Sec. We then obtain (lpk.(0 4>k The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution. These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients. Merzbacher[194]). The functions ipn(x) are eigenfunctions of some secondorder differential equation.91) The coefficients Cj follow from the recurrence relations. . These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations. (8.
97a) and (8.97b) with ipn y£ ipm. TimeIndependent Perturbation Theory 8. (8. (8.93b) The lowest order eigenfunctions belonging to En. all n.96) (Ho . that of double degeneracy En = Em . (8. For example the magnetic states enumerated by the magnetic quantum number m. (H0 + eV)^n and (HQ + eV)Vm = Emym.6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '. (8.94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8.94a).94a) into Eq.96) are effectively the known equations of ipnitpm and so yield no new information..We now insert Eqs.e. . Equations (8.93a). I — 1 .152 CHAPTER 8. i.95a) (8. (8. is not uncommon.97b) are . .e. i.94b) with (ipn. (8. degeneracy.V)(c2l^n + ci 2 ^ TO ).97a) (8. . i. are allowed to be equal since otherwise the expansion becomes undefined.93a) (8. —I.e. so that En = 4 ° ) + eE™ + e2EP + ••• .ipm) = 0 = (^m. (8. Equations (8. . m = I. + c22^m). however. in a central force problem all have the same energy unless the system is placed in a magnetic field.93b). Em can now be linear combinations of ipni'ipm. (8.4 0) )</4 1} = (En1] ~ V)(cu^n (Ho ..Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• . Equality of eigenvalues.ipn).70). Em = 4 ° ) +eE$ + e2E$ + . In order to deal with the problem of degeneracy we consider the simplest case. (8. Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn.95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8. .4 0 ) ) ^ } = (Em] .
Vil>m)] V^m).H0il>W) = cn[EW . and ( ^ m .ipm ) the equations (iPn. ^ o ^ ) ) = 0.(lPn.cnVipn and tyi. .(^04°))41)) = EM^CuTpn (ipl.C 21 (^ m .6 Degenerate Perturbation Theory 153 inhomogeneous equations. We now multiply these equations from the left by ip* and integrate.C22(Vn. V^n).EJ®(lPn. Vi/>n)] . (8. H0lpV) .m). ^ ) = ( V .y>m. Vl/.V^n).98a) Setting I = n. But (8. £ aW^°Vi) .99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) . Then (^.^ ( ^ . t f ) = a k 1 } 4 0 ) " Eg>aU = 0. we obtain with (ipn. cn(ipm.H0^) .m.ci 2 (V„. V^n)\ .98b) + Culpm) + cuV^m). Wm. c2iV^n + c22V7pm). Vlf>m)] . ^ ) ) = CU[EW .£ < « = ag)^) . Equations (8.8.ipn n) = 0 = (ipm. (^m. (8. (^n.(Vn.99) and ( ^ ^ O ^ ) " ^ ^ . Similarly Wn./>£>) = C2l[E£> .^ ) ( ^ m .J#»a&> = 0j since ii4i = En (degeneracy).H0xl>$) . H0lP$) = C22[E$ ~ (Vm.
However.5).ci2.This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate. Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates. i. d>" E=^. Example 11. If one wants to perform a perturbation calculation in such a case. in particular in parabolic coordinates (see Sec.Vrl>n) . the secular determinant.100a) determine the coefficients Cn. 9} ip. It is then possible that a special perturbation. Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V . (8. 11..e.100b) determine C2i. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate. .Similarly Eqs.E)tp = 0.Vlpn) CHAPTER 8. It follows that ip and <j> are linearly dependent.^ ' V = — (•>!>'<t><t>'i>). the degenerate perturbation theory must be used.e.101) This secular equation determines the first order correction En • The equations (8. i. Let ip and < be two eigenfunctions belonging to the same eigenvalue E. Example 8. as we saw. (V>V</>V>) = const. i> 4> Hence there is no degeneracy. dx i. splits the problem into two independent onedimensional systems. 4.E™ (Vn.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n.154 and V (^m. (8. . In this particular case therefore nondegenerate perturbation theory can be used (cf.1: Do discrete spectra permit degeneracy? Show that in the case of a onedimensioanl Schrodinger equation with discrete point spectrum there is no degeneracy.e. \nip = ln(j> + c o n s t . just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6). there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r. TimeIndependent Perturbation Theory (l/>m.1. if)' (/>' — = —. A onedimensional system. ip = c<p. Then / > i>" 2— = V ip Hence 0 = V ' V . At x = ±00 the constant is zero.6.3).C22.^Vm) 0. like that providing the Stark effect (proportional to rcosO). An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8.
(8.V)(p = 0. V = V0 + v. Dingle [72] and H. B. The subdivisions (8. cf. i.V0)<f> = 0 (8. H.106) (8. We assume that V can be expressed as the sum of two terms. (8.8.104) "This method was developed in R. the method has to be applied in each domain separately. J. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term.102) can be written D<f> + (EQ .102) where D is a second order differential operator and E a parameter. of course. B. J. It is immaterial here whether D does or does not contain a first order derivative — in fact.e)<t>.* In various versions this method is used throughout this book. .e. W.7 DingleMuller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward. Muller [216]. Dingle and H. (8. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation. W. anharmonic oscillator potentials and screened Coulomb potentials. J. We consider an equation of the form Dcj) + (E . T For instance spheroidal functions can be studied with this method. It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way.t Since in general a solution is valid only in a limited region of the variable. Applications of the method are. Thus the method is a systematic method of matched asymptotic expansions.103) such that if E is written correspondingly E = E0 + e.105) (8.103). a potential. See also R.104) are chosen such that the equation Dcf) + (E0 . W.7 DingleMuller Perturbation Method 155 8. the considerations given below would be valid even if D contained only the derivative of the first order. and such that one can even obtain a recurrence relation for the coefficients of the expansion. (8. also possible in areas not of immediate relevance here. Muller [210]. Eq. with some restrictions on the function V. Muller [73].V0)<t> = (v. in particular in applications to periodic potentials. and the solutions then have to be matched in their regions of overlap.
108) with coefficients (m.+s ' (8. v = {D+EmVo)<t>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.e ) ^ = 2 ( m . We write the coefficients "(m. the function 4>(0) = <t>m (8. TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m. (8.111) s then (see below) 0(1 = E (m. m + s) which are determined by the recursion formulae.VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u .m + s)4>m+s s (8.) This result is obtained as follows.m + s)" in order to relate them to "steps" from m to m + s.E0\ (the latter in a restricted domain).e\ < \V0 .m + .V0)<f>m+a = 0.109) D(j)m+s + (Em . Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em . Then if ]e < \E\ and \v .V0)4>m = 0.110) If we write the next contribution <f>W to <p(°> in the form m+st (8.156 CHAPTER 8. (8. the equation Dcf>m+S + (Em+S .102) with E ~ Em. Considering (frm+s we have.113) . and so v (8.Em+s)(/)m+s. where m is an additional integral or nearintegral parameter (depending on boundary conditions). s an integer: (ve)(pm = 'Y^{fn.107) represents a first approximation to the solution (f> of Eq.Vo)(j)m+s = (Em . since D<f>m + (Em .V0)<f>^ . m + s)0m+s.
Eq. £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8.e. (8.119) is effectively the secular equation).m) 1^1^ IW . .119) follow a definite pattern and can therefore be constructed in a systematic way.n\ ^119) + . We then find altogether ^  l^m  Em+S) + y^y^ SJ. /01_.T? S The first expansion determines the solution 4> (apart from an overall normalization constant). (8.117) „ .116) m ^ v (m.„N This procedure can be repeated indefinitely.TP \(T? .m) E J^ ( m ~ Em+S) ' • fa.114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq. Thus (m.m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m.m) — —.m) — 0 (first approximation).8. (8. (8. We observe that the coeffcients of expansions (8. The factors (Em — Em+S) in the denominators result from the right hand side of Eq.113) must vanish.m + s)(m + s. (8. (8. (8. .Q (m.m)4>m in (8.110) and are therefore related to the inverse of a .102) to order (1) of the perturbation (v — e).. V ^ V ^ {rn.7 DingleMiiller Perturbation Method and so (m. .m + s)(m + s.m + r)(rn + r. .^ (Em .m) + > ^ .m + s) 157 (B\IA\ s^O: 1 cs = J m _ ^m+s . and the second expansion is an equation from which e and hence the eigenvalue E is determined (i.m + s)(m + s.Em+S) = 0 (second approximation).m + s)(m + s. the coefficient of the term (m.115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m.118). .
6q + 1)1 + 24(5? . .I2q + 64) + 256Z 3 (41 g . The result is the following expansion.38).e. as will be seen in later chapters. for instance. 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. Other applications can be found in the literature. as already mentioned after Eq.9) + 4096J4] + 0 ( l / g 3 ) . .1)( 2 + 64J3] — . (f>m in the above.[q(llq2 + 1) + 2(33g 2 .[ 4 ( 8 5 g 4 + 2q2 . § H. Kaushal [146]. 1 . for the investigation of the behaviour of the late terms of the perturbation series.2). An explicit application of the method (not including matching) seems suitable at this point.e. does not occur in any of the later contributions.423) + Z(2720 9 3 . (8.2: Eigenenergies of the Gauss Potential^ Use the above perturbation theory for the calculation of the eigenvalues E = fc2 of the radial Schrodinger equation dr2 + k2 1(1 + 1) V(r) ip(r) = 0 with Gauss potential V(r) = g2e~a r for large values of g2. This can in fact be done and will be dealt with later. has been treated by R. in which q = in + 3. One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders. as will be seen later.7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 . Imposing normalization later one obtains the normalization constant as an asymptotic expansion. this is — so to speak — the price paid for obtaining a clear systematics which is essential. Miiller [211]. An additional aspect of the method is the full exploitation of the symmetries of the differential equation. The eigenvalue expansion obtained is an asymptotic expansion. 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r .2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential. since it will be used extensively in later chapters. A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion. This means the solution is not normalized. J. . The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory. for instance. and is typical of a large number of cases. 2 . i. TimeIndependent Perturbation Theory Green's function. W.158 CHAPTER 8. ^The potential A r 2 / ( 1 + gr2). (8.* Example 8. The following Example 8. S. i. . n = 0 .120) n4 .
a)ip(a) + (a.a + l)if>(a + 1) + (a.2) where 1/ .6 and 16. multiply the equation by onehalf and set h = —a/g. The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: z2f(a) = (a. Setting q = in + 3 implies k2 + g2 ga(2l + q).122) where A is of order 1/g. (8. i/> with Dq = d2 . b.123) The first approximation (note the last expression as convenient notation) ^ . The solution i. We now insert this equation into Eq. . (8. We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf. 1 .b. 3\ 2 V ' 2.7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 . \2 .123) the contribution fl<°> Ah + . 11. .2a2 A.^z2) = i>(a) leaves unaccounted for on the right hand side of Eq. Sees.8. (. 2 The solution XQ(S) = $ ( a .1).• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a.. 159 •g'g'a'V U E Here we change the independent variable to z = y/2gar.l)ip(a . ._ „!2 ^ ( " « 2 r r 2 > .b.2 '(l + l) . . (8. .V ! 4 I z S *! V2 ipq(z).z2 is a normalizable function if a = —n for n = 0 . . a . Therefore in the general case we may set k2 + g2 = ga(2l + q) . Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2.0(z) = zl+1ez2^fa. 1 1(1 + 1) 1 . Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ . <«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ. „2 J„2JL.121). (fc2 + <?2) 4ga and b = I \—. S) is a confluent hypergeometric function. 2 .
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. fa. Hence the next order contribution to tp^ becomes oo i+2 r [a a . a . ( 5 m ( a . a + 111 r [a. when j = 0.a + j}i+1^(a + j). . (a. a]i + h fa. „ . a + r) satisfy the following recurrence relation: Smi(a. If we now evaluate the various coefficients in the last expansion we obtain the result (8.a]2+L ^ J1[a + 2 .a—lli. 0 = h[a.b = (g + 3) . The coefficients S m ( a .a — 211 . of course. o + r) + S m _ i ( a .2a = I + q. a) = 1.a + 2]i P .120). .S m (a.i ^(i)=£V+i i=0 J2 j = (i+2). a + r) with the boundary conditions Sn(a. As a check on the usefulness of the perturbation method employed here. So (a. TimeIndependent Perturbation Theory (g . In fact. 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. where [a.3). (8. a + r + l)(a + r + l . a ] i + 0(/i3) [a. a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a . the latter determining to that order the eigenvalue.^ 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. i. a + r) + S m _ i ( a . a ] x + L _ 2 J 1 [ « .120). This equation shows that a term fiip(a + j) on the right hand side of Eq. fa. Now we make the following important observation that Dqip(a + j) = jip{a + j). a + r — l)(a + r — l . 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 ] i + L ' ^ J 1 [ a + l . _ J [a . a + 1) = a = In general CHAPTER 8.a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps".j^0 ' + jU+1 J ^(a + j) with h[a. a + i) = 0 for i ^ 0.a]1=0.e.123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except. one could now use the RayleighSchrodinger method and rederive the result (8.a]i = . a + j)i/>(a + m).Z.l . (a.2.1) = a . a + r) = 0 for r > m. a + i)(a + r. a) = b .
e. We therefore begin with the appropriate classical considerations with a view to generalization.1 Introductory Remarks We saw in the foregoing chapters that in considering a particle system which is classically described by the solution of Newton's equation* is quantum mechanically described by states which in the position or configuration space representation are given by the solutions of the Schrodinger wave equation. and this means electrodynamics. light. 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. although not exactly straightforward.Chapter 9 The Density Matrix and Polarization Phenomena 9. the generalization to many particles. proceeds along similar lines. as we shall see in Chapter 27. 161 .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. except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets. 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. 9. Regrettably this has to be left out here in view of lack of space. i. and we shall then see that these lead to analogous results as in Chapter 2.
9.1. The Density Matrix and Polarization Phenomena potential A(r. This is the equation of an ellipse as indicated in Fig. (9. i. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0. t) = A 0J R cos(cji) + A 0 / sin(u. In general the constant Ao is complex. It follows therefore that A(r.6) . (9.i).2) (9. In the latter case one has the possibility of {A0R x A 0 /) • k > 0.162 CHAPTER 9.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0.1) Fig.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. 9. t) given by A(r. (9.4) (9.1 The vector potential ellipse. A 0 = A0R + iA0I.3) (9.e. When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized.t) = 3ftA.
a calcite crystal or a film of nitrocellulose. the vector potential A. We set therefore A 0 = Ai + A 2 . From the vectors A(°).e. i.y)) components A i . I=<l2x2>:=E40)*l*40)i. i. A2. not. (9. i.A«). See for instance A. the intensity of a wave. The intensity 7 of a light wave with amplitude coefficients A\' . but the polarization (state) and the intensity can change (the latter by absorption) from R(0)}  R ( 1 ) } In view of the linearity of the Maxwell equations. / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2.9.9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2. Here we are interested in observable properties of light waves — like. we recall. 17. determined with the help of a polarizer.A i = 0 = k . represented by a filter "F".e. p.A(°)).2 *i*4 0) (98) with coefficients F^. AW = £ fc=l. Then their connection is given by Eq.2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9.A« we can construct only the following scalar products with this invariance: (A(°\A(°)).2) the vector potential A is completely determined by its two (say (x. In classical electrodynamics. however.A2 is the sum of the moduli.AW). ie.e. . the direction of polarization. the fields E and B are observables. with k .11) 'This means we consider the polarization. (9. Observable quantities can only be those which are invariant under translations and rotations.t in which k and ui remain unchanged. Rae [234]. (9.g.A 2 . (9. as e.8). (A(°). (AW. (AW.7) We consider now measurements with polarized light. for instance. which leads to an outgoing wave which we give the label "1". the connection must be linear.k (910) We give the initial wave the label "0" and subject this to an experiment.
Thus (F) = J2FikPki = ^(Fp)a. We write the observable quantity as (F):=Y. i. Then according to Eq. if (F) is an observable quantity. (9. (9.13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments.16) i. But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables. To this end we observe that as a consequence of Eq.k l f c 4 ° \ (9.A( 1 )) = ^ ^ F i.15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik). (F)=Tr(Fp). (9. We use definition (9.13): j2^FikAk = Y^AiF*kAi = Y. so that (F) can be looked at as the expectation value of an observable F.e. (9.12) i.13).14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%. the connection between the initial polarization and the final polarization. (A(°).13) the quantity F must be an hermitian matrix and thus representative of an observable.k (9.e. i. pki = AkA*. (9.k i (9.A*kF£A* = E and hence F = Fl A p i tkAk. (9.A*FikAk.17) .164 CHAPTER 9. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment. it must be real and hence (F) = <F>*.e. i. Thus.15) in order to rewrite Eq.
In the present case of a polarized beam of light we have two possible. i.* For a lightwave travelling in the direction of z. 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. we have (9 19) ' In this case there is no preferential direction.e.18) The vector A describes a particular ray of light. i. The socalled pure case. is in general unpolarized. that of a pure state.21a) follows from Eq.2 Reconsideration of Electrodynamics In particular we have / = Tr(p). For a 45° polarized wave we have 1 .e. . b = 0 and o = 0.21b) This state is still a pure state. like that of a bulb. The density matrix for the pure state (9. b = 1 we obtain the expressions (9.15) with A\ = a.9. (9.(AAt) = ( £ £ ) . (9. it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions. This admixture is described by introducing matrices p. i. But light. 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip. mutually orthogonal polarization directions x and y described by the matrices (see also below) * . is given by a single polarized wave. <»^> (9.22) For a — 1.d "»=(o ! ) " In the case of an admixture with equal portions of 50%.e. "incoherently. A2 = b as </*) . 165 (9.19).( J o ) .
The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M . l l\ " i °\ ) ' ^925^ 9.g. (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. Pi35° = \ \ 2 ? )• (9. (926) where a is a parameter whose variation describes the continuous variation of the polarization from its initial state labeled with "0" and hence parameter value OQ (analogous to t or j3 in our earlier considerations of the density matrix) to the final state with label "1". Then according to Eq.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. Equation (9. as a consequence of absorption).8) A(1)=E^40)> k R = R(<*).23) 1 . We set in analogy to Eq. We shall see that the matrices appearing in the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics. 1 = . b = —=. (9. in which the intensity I = Tr(p) is not conserved (e.15) Pik — A i A k .166 and CHAPTER 9. (9.Z^ 3. 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.™ V 0 n kmAm / ^^jPim^mk' .
(9. the dependence on a is contained in F' = R^(a)FR{a).32) (9.4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation. i. In the case of Eq. (9.28) (9. that Io = Tr( / 0 (°))=Tr(pW) = / 1 .31).34) . (9. (9.29) where (i.30) Ti{FRp{®I$) = Tr(F.pW). the observable F remaining unchanged.q.9. We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)).30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9. the observable is transformed. In the case of Eq. The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size. (9.e. that the initial intensity Jo is equal to the final state intensity I\. The corresponding condition here is that no absorption of the wave takes place.31) (9. 9.e. on the other hand.4 The Liouville Equation 167 pW = RpMRl (9.t) in Chapter 2 in what we described as the Schrodinger picture there. here this contains the dependence on a.e.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq.e. and p(0' remains unchanged. i. (9.33) i. in other words without creation or annihilation of systems. The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations. in the state functional) in analogy to the time dependence of W(p.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR.
CHAPTER 9.36) (9. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda.M .46) suggests to define Poisson brackets of matrices by the correspondence {A.32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9.168 i. In view of Eq.40) can be considered as quasiequations of motion. = Tr(R^Rp^). j£ = i[H.p].40) Equations (9.B]. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H. (9. We set therefore M^:=iH. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1.39) and (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.e. the matrix M is antiHermitian: Mf = .21) and (2. Eq.37) (9.39) If the dependence on a is not contained in p^1' but in F' (cf. (9. .e.p(°\ao)]da. (9.35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da. (9.F}. (9. (9. (9.B}:=i[A. i. In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9. (2. Comparison of these equations with Eqs.35) Thus the matrix R has to be unitary.33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H.2) has to be taken into account.
Thus the expectation value with respect to the vector describing the state in the Hilbert space does not depend on this phase factor.1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real. Thus in particular we establish the uncertainty relation for observables in general. In keeping with our earlier notation the expectation values of dynamical variables are defined as follows: Postulate: The expectation value or mean value of an arbitrary function F(A) of an observable A in the case of a system in the state \u) £ "K is defined as the expression (F(A)) = (u\F(A)\u). (10. and consider timedependent perturbation theory. the Schrodinger.2 States and Observables Every state of a physical system is represented by a certain ketvector \u) which is an element of the corresponding Hilbert space "K. To insure that the 169 . we establish their Heisenberg equations of motion.Chapter 10 Quantum Theory: The General Formalism 10. 10.1 Introductory Remarks In this chapter we are concerned with the formulation of quantum mechanics in general. Heisenberg and interaction pictures.
which clearly does not hold in the case of operators. [qi. We can describe a system as one with a classical analogue if the Hamilton operator of the quantum system is obtained from the Hamilton function of classical mechanics by the correspondence •^i • Qi. and assume that A = A(p.170 CHAPTER 10. y.Pj] = 0. E.* a conspicuous exception being for instance a spin system. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator. (10.4) *Even in those cases where this is possible the operator correspondence may not be unique. \pi.Pi are called fundamental observables. we must have (u\u) = 1. Such a correspondence with classical mechanics is not always possible. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1.Pj] = ihSlj.3 represent the three Cartesian coordinate directions of x.g.Qj] = 0. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space.2.2) (F(A)) = {u\u) <"lf.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi.3) These fundamental commutators determine the commutators of arbitrary observables A = A(p.1 U n c e r t a i n t y relation for observables A. Here. such as for instance of the angular momentum operators Li. 10.B] = ih.2. we assume this now. however. . For these the following rules of canonical quantization are postulated: Postulate: The operators q~i. (10.q).y. B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. Pi • Pi. where i = 1. The phase space coordinates qi. classically qiPi = PiQi. q). Naturally one can also write (10.
5)) (for A\u).2 States and Observables then their uncertainties AA. B}= ih. = \\A\u)\\2\\B\u)\\2 (10..B\u) G IK as vectors) we obtain (for an application see Example 10. Thus AA = AA = ( i 2 ) V 2 .9a). (A{A))(B{B)){B{B))(A(A)) (10. (10. (10. Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u). (4.9) ((A2) + (A)2 . (10.. Eqs.(A)2)1'2 (1 °= a) ((A2 + (A)2 . so that (A) = (A) . Ai = = = ((A2) .B] = ABBA > ^h.5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A(A). (10.(A) = 0.(A)2 = (AA)2.. (10.12) the Schwarz inequality (cf. (10.12) where \u) 6 "K is a ketvector representing the state of the system. Applying to the expression (10.2A(A)}  {A)2)1'2 (10. Since A = A— (A). using Eq.7b) = = [A.13) AB = AB = (B2)1/2.6) B:=B(B). we have A2 = A2 .7b).2A(A) + (A)2. AB defined by the mean square deviation AC=((C2)(C)2)1/2.0)1/2 = AA. 111 (10.{A)2)1'2 {{A2) ..10) .B.11) (A2) = (A2) .8) and. (10.2(A)2 .1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2.10. C = A. Then [A.7a) (B) = 0. (3.
Zi (10. for a complex constant c.15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*. . (10. and with Eq. and hence for condition (b) 0 = = tSee Sec.e. AAAB>I=h.14) Hence (u\AB\u) = /^(AB where. (10.18) >R2 + I2. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih. (10. (10. We can therefore rewrite Eq. The condition (a) is satisfied when.4). The equaltosign applies in the case when (a) it applies in the Schwarz inequality.16) I = h. Zi Zi (10.172 CHAPTER 10.15) as (u\AB\u) = R + iI. (10. From this we obtain (uiJBu) 2 = i22 + / 2 .19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u). since A and B are hermitian.17) as had to be shown. and (b) when (AB + BA) = 0. (10. 4.3. (AB + BA) This means (AB + BA) is real. (10.13) (AA)2{AB)2 i. Zi + BA)\ + ^ih. A\u)=cB\u). Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq. R= (AB + BA).
e.e.7a). i. ALy = (u\L±\u) ^ 0. (A{A))\u)=i(Zc)(B(B))\u). we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li. i. in view of Eq. i. (Lx±iLy) ^ 0. What does the relation imply for states u) = \lm).e. = 0.Lj] = (LiLj + LjLi) + ih£ijkLk. i. (10. = 5R2 + G 2 > 9 2 = hi2(Lk)2. 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. + LjLi\u). (Lz)2^0. m\lm)). the following uncertainty h2{Lk)2.13) to operators Li — Li.y.e. . (Ly)2=0.3 OneDimensional Systems 173 Thus c = — c*. The observables like angular momentum are functions of p and q.) 2 > = x. Example 10. Separating the product into symmetric and antisymmetric parts.i relation holds: (ALi) 2 (AL. q) = ih.z. Ly are not "sharp". 9 = h{u\eiikLk\u).10.20) or. h2(Lz)2. such that Lz is "sharp". For these we have in the onedimensional case the relation [p. 9ftc = 0. (ALy)2{ALz)2 > ~h2(Lx)2. implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10. implying {Lx)=0. {u\LiLj\u) = R + iQ. Thus in this case Lx. for which one component of L is "sharp"? Solution: Applying Eq.3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier. (Ly)=0. we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2. (ALZ)2(ALX)2 > \h2 = (Ly)2. For states \u). ALX 10. (10.
p) can be expanded as a power series in q and p. (10.B]C + B[A.BC] = [A.qn] = ih^(qn).p)] = .21b) (10.21b) now follow with the assumption that A(q.A(q.23) = ihg(q2) (10.p] = ^ihp ~ Proceeding in this way we obtain [q. as remarked earlier.q). Q2] = [p. (b) its spectrum is necessarily continuous. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators). This is why the operator p requires a representation in the space of eigenvectors of q.p2] = [q.p)]=ih^^and \p.e. i.i h ^ ^ .C\. We have (cf. (3. Next we have a closer look at the operator q. i.e.21a) and (10.174 CHAPTER 10.p] = ih we obtain therefore [q. Then (see below) we can derive for A the following commutator equations [q. (10.34d)) [A. Using [q. for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable. and (c) the spectrum is not degenerate.p}p + p[q.A(q. . they do not have a common system of eigenvectors. We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function). Eq.pn]=ih^(pn).22) d ih—(p2). We obtain these equations as follows. Similarly we obtain \p.21a) Equations (10. A — A(p. q]q + # > v] = ~2ihq and \p.
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').1 how the delta function arises in the limit K —> oo. i. We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization). an operator. For the cases (b) and (c) we consider the operator U(a — pipa/h (10. U is unitary. 10.28) i. 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10.e.25) It is shown in Fig. U] = ih^= aU. .27) (10.26) it Here p is an observable.10.1 The delta function for K — oo. 10. we obtain [q. > We begin with (a). Replacing in Eq.3 OneDimensional Systems 6K(x) 175 Fig. and a a cnumber. (10. (10. = eipa/h = Uia)'1 = U{a).f^ ^ J—oo dke ikx —ikx' (10. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±.e.21a) A by U.
we can see that also p possesses only a continuous spectrum (from —00 to 00). that with U(a) = e~ipa/h = U.1 T h e translation operator U(a) We saw above. Eq. ^Observe that {iji\A{p. or qU(a)\x') = (x' + a)U(a)\x'). CHAPTER 10. Quantum Theory: The General Formalism qU = Uq + aU = U(q + a). Since this holds for any arbitrary value of a in (—00.3.31) . and that the eigenvectors do not have a finite norm and are normalized to a delta function according to (jPx\Px) = S (Pxp'x) With the help of the observables p.q)U^U\ilj). or qU\x) = U{q + a)\x) = (x + a)U\x).e. Thus all these values belong to the spectrum of the operator q. (10.00). 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. q any other observable F(p. Evidently every eigenvalue has only one eigenvector. (10.30) (10. which means that the spectrum of this operator is a continuum. In an analogous way by defining the unitary operator U(b) = eiqblh (q operator.29) This means: U\x) is eigenvector of q with eigenvalue x + a. i. (10. we obtain qU\x) = U(q + a)\x) = {x + a)U\x). q) can be constructed representing a dynamical variable. The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx'). the spectrum is not degenerate. 10. b a cnumber).e. cf.00).29).q)\i>) = (i>\U^UA(p. WU = 1.176 i.
(10. U(a)\x') = c\x' + a).e.x')\ = l. for instance.x")c(a.a).x')5(x" = (x" + a\c*c\x' + a) x').10.37) With the help of the latter expression we can calculate.32) we see that \x' + a)ocU(a)\x').32) Since U is unitary.x" . offdiagonal elelemts of the {x}representation of the momentum operator. (10.38) h 6(x' . It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a). (10.e./ (10. It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' .x') = = i.hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x").e) ~ 5{x' . (x'\p\x"). i.33) (10. (10. we have q\x' + a) = (x' + a)\x' + a). (10. \x') = U(x')\0).e. i.x") .3 OneDimensional Systems 177 Moreover. the operator U(a) acts to shift the value x' by the amount a.x" . it follows that 5{x" . i. (10.35) (10.34) {x"\U\a)U{a)\x') c*(a.37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' .x" .39) . The operator is therefore called translation operator.31) and (10.e) (x \p\x ) = . We choose the phase such that U(a)\0) = \a).e. \c{a.x") U{x'\p\x" % pz. In the expression of Eq.e.36) i. (10. Comparing Eqs. since q\x') — x'\x').5(x> .
H\uE(to))=E\uE(to)). (5. 5.to) exists which is such that^ m)) = u(t. For the way the state of the system develops in the course of time we make the following postulate: Postulate: A linear operator U(t. so that U{t.t0).43) (10.35).e. (10. .e.4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i. The operator U(t. Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E.° ° d(p) (ip) e"« = S'(x).178 CHAPTER 10. it follows that 5'(x) is an odd function of x. 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).4i) We assume first of all.e.42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / . to) with respect to t we obtain ihjtU(t. Quantum Theory: The General Formalism We mention in passing that.45) (10. since 5(x) = . .40) 10. 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)).to) is called the time development operator.27) and (5.§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian.44) (10.x") = (x'\p\x")*. (10.to)m0)). 6'{x) = !.f dpipeipx. This is in conformity with our earlier postulate in Sec. See Eqs.tQ) = eiH(tt<Mh. a conservative system. In this case the classical Hamilton function does not depend explicitly on t.4. Differentiating U(t. i.L fdpeipx.t0) = HU(t. i. (io.
In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues. (10. Evidently U(t. But obviously (cf.t 0 )^(*0)> U(t2.45) has very general validity. (10. = U(t0. From Eq.to) = 1. t0) = H(t)U{t.t0) or ih—U{t. Eq.48) with U(to.z / dt'H(t')U(t'.4 Equations of Motion 179 We now demonstrate that Eq.t0)]1 Let us set for small values of e: U(t + e.to) can be expressed as an integral as the solution of Eq.tiM*i.t) = 1.t).t)U(t.48): [/(Mo) = 1 .47) e.t0)U(t0.10. Ufa.t) = l. t0) (10. (10. i. (10.to) (differentiating we obtain immediately Eq.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t.41) we obtain ^(t 2 )> = = J7(t2. The operator H is d efined by this relation. (10.t) and therefore [U(t.t)l]U(t.t) = l^eH{t).e.t0)\^(t0)).t0) [U(t + e.44)) we have U(t. (10.*l)^(*l.*l)^(*l)> = I7(*2.48)).to).49) . (10. Then with U(t + e. t0) = E/(t 2 . Hence U(t.
41). (5. 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. In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities. Quantum Theory: The General Formalism We obtain the differential equation for the development of states of the system in the course of time by differentiation of Eq. (10. to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations. The eigenvectors of these observables are also constant in time. Since the only measurable quantities are moduli of scalar products.180 CHAPTER 10. requires the identification H(t) = H= Hamilton operator.t) = l % ^Hdt) \il>(t)) Hdt (10. Eq.30).(£ + dt)) = U(t + dt. Another such description is the Heisenberg picture. (io. the Schrodinger equation ju(t. it follows that U(t + dt.t0) m0)). the instant at which a measurement is performed on the system. Now let \tp(t)) be the state of the sytem at time 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.e. and these remain unchanged under unitary transformations. Comparison with the equation postulated earlier. whereas the physical quantities are represented by observables in "K which do not depend explicitly on t.51) Since </.e. t)\t/>(t)) =U~ and since H is hermitian. the predictions of different descriptions are identical.t0)\ii>s{to)) (10. i. Then \Mt)) = u(t. The operator U(t.5o) ihjtm)) = Hm)).53) .52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^). >(*)> = i. (10.
With explicit time dependence.56) U^[As.59) This equation is called Heisenberg equation of motion. = so that ihjtAH or ihjtAH = = [AH. Differentiating Eq. a scalar product) remains unchanged: AH(t) = U\t. (10.(tf HU)(rfASU) (10. even if As is not timedependent. (10. It does not always have to be the case that As does not contain an explicit time dependence. i.55) Thus AH is explicitly timedependent.HH]. so that a matrix element (i.t0)\ips(t)) = \ips(to)) = constant in time. for instance in cases of nonequilibrium or if part of a system disappears through absorption.55) and using Eq. t0)AsU(t. but here. we have to take into account the contribution dAs/dt.54) Observables transform correspondingly with a similarity transformation. {AH. i.p.57) and rf[As.H]U + iW]^U. in order to obtain the Heisenberg picture (description). As(q.4 Equations of Motion and \^H) = U\t.e. However the associated observables are timedependent. For As(q.p) in the Schrodinger picture dAs/dt = 0.48).H]U = = = U]ASHU AHHH — U^HASU . Here H is the Hamilton operator in the Schrodinger picture. 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. . (10. we consider only cases without explicit time dependence of As. we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10.10.58) HHAH (U*ASU)(U^HU) [AH. Thus.e. In the Heisenberg picture it is HH = U]HU.HH]+ihU^U. (10.i). if ^ = 0. (10.HH]. subject to a continuous change.e. t0). 181 (10. as a rule.
. as in the transition from the configuration or position space representation to the momentum space representation with the help of Fourier transforms.21a) and (10.H] = 0. 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. (10.HH}=0. This holds also in the case of the Schrodinger picture. since in the transition from one to the other the commutator remains unchanged.61) Thus conserved quantities commute with the Hamilton operator. Orfanopoulos [224]. i. we used in essence unitary transformations of matrices. in the case of the fundamental dynamical variables qi. if ffCH = 0.182 CHAPTER 10. Solution: We have for an observable A (A) = (i>H\AH\TpH). In particular. We also observe that the momentum pi is conserved when \pi. Example 10.e. Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture. 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. For an extensive discussion see B. (10.Pi): and where the expressions on the right follow from Eqs. We see that formally one obtains Hamilton's equations of classical mechanics.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10. [CH. Finally we add a comment concerning conserved quantities. An observable CH is called a conserved quantity.21b). We end with a word of caution. In our treatment of representations. In the treatment of the pictures we used unitary transformations of operators (and vectors).13) to derive the uncertainty relation for energy and time. In many cases the Schrodinger picture is more amenable for explicit calculations. A. provided the Poisson brackets in the latter are replaced by commutators.
4 Equations of Motion Then. i. at ^±AH>h. is displaced by A A in the interval A T = TA. (10. we can replace ([AH. (10. of A\<t>) = a<£)) are then called "good quantum numbers".e. if we set AH = AE we obtain TAAE (10.64) = 0.68) To put it The eigenvalues a of A (i.H}\f)\2. 183 dV ' \ dt >"•"»» + ( £ ) • (10. We defined earlier stationary states as states with a definite energy E and wave function * = V( r ) e x P ( — i E t / h ) .Such a characteristic time interval can be defined for any dynamical variable. i. (A).10. If r is the smallest such characteristic interval of time. We construct the following vectors using Eq.H]) = 0. > (10. (10.p) we then have 0=^M) = U[A.e. (10. This means that for a physical variable A(q. and TA O C (10. this is practically the same as at time t for times t with \t — t'\ < T.15): (AAf(AH)2 Using Eq.67) 1 d(A)/dt is oo and AE = 0. H] = 0. Let \ip) = \"4>H) represent the state of the system.\HAM\2) \(nA.63) so that with Eqs. H\tl>) = {H(H))\f).2 > h. the "centre of mass" of the statistical distribution of A.7a). then if a measurement is made at time t'. .e.62) and dAs/dt > \WAH\i.)\2 ( = \{i.e. a position or momentum observation is independent of t. dt in. (10. in a different form: The eigenvalues of an observable are called "good" if [A.62) The Hamiltonian H does not depend on time. since \IPH) is timeindependent. (10. AW = {A(A»V>. i.65) and rA AA d{A)/dt' . The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t.HH]) Hence we obtain by ih(A).66) This means.13) and (10. 2 d{A)/dt " and.
as is observed.5 States of Finite Lifetime When E is real and P(x) is independent of t and J dxP(x) = 1.71) From Eq.tyEt/hdt = = = / J—oo dt J dEJrl>E. Quantum Theory: The General Formalism 10. i.E1) 2irhfEo{E)4>E(x). t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime..Eo(x)eiEot/he'^2h9(t).(x)fEo(E')e«EE'Wh 2irh f dE'^E.69) and (10.69): oo poo p / oo *PEo(x. This means. t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10. /oo (10.t) — ipE(x. (10.£ ^ + n/2)^ o ( x ) e{^1/2+i(EE )}t/h 0 o =0 . (x)/ E o (E')5(E . i ) = J dE'^El^)eiEltlhfEo{E'). the particle density diminishes in the course of time (t > 0). Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei.70) by From Eqs. the wave function ^ ( x .72) = V'EoW 7/2h + 0i(EE )/h  oo .t) of such a state must be the superposition of states with different energies about EQ and a corresponding weight function / so that ^ ( x .)exp(iEt/h)). First we obtain from Eq. for instance. the wave function ipEo(x. On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum.184 CHAPTER 10. But the number of radioactive (thus excited) nuclei decreases exponentially.70) we obtain: oo / oo TPEO (x. (10. This means that in the case of decaying particles.e. (10.69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'. then fE{E') = S(EE') and integration with respect to E' yields again t()E(x. have around EQ the form ipEofct) = i. (10. The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could. states with sharp energy (AE = 0).H ) ( £ . Recall for instance radioactive decay. States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime".70) we can determine the function fEo{E) integration with respect to t. (10.
and is called "interaction picture" or "Dime picture". (10. (C/(0) unitary). We saw previously that if the state of a system is known at time to.105) and (20.74) Thus the solution of this equation is the main problem. VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula. i. Eqs. where U is the solution of Eq. .e.t0). every unitary transformation of states and operators defines a possible "picture"). 7 > 0. This description contains effectively parts of the other two pictures and is particularly useful when Eq. (10.10. that** EQ. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ. £/ = £/W(Mo)^'(Mo). (10.45) can be solved only approximately (in actual fact. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state. The state with lifetime r — h/'y is not a discrete state. 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 Prom the last two equations we deduce for E close to ipE0(x). We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum. They are called "resonance states".75) (10.e.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE). of course. 10.41).6 The Interaction Picture One more motion picture of quantum mechanical systems is in use. Let U^°\t. i. (10. 185 i.76) **For specific applications see e.e. to) be an approximate but unitary solution of this equation. i.45). Considered as a function of E. the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2. ^(*o))> then \xp(t)) for t > t0 follows from Eq. (10. (14.g. W)) = U(t.e. We obtained the result (10. ihU(t.e.toM{to)). i.11).to) = HU(t.
(10. with Hamiltonian H = H^ + H'. i. (10. Now let H^(t). i.e.e. i.e. . Quantum Theory: The General Formalism ih^(uWu') i. = HU<®U'.e.e. be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i.e.83) ih^U^ ishermitian. [/' has only a weak ^dependence (i.t0) 0 = l. (10. In addition we have (10.82) (10.78) (10.81) But [/(°) is unitary. with Eq.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0.79) ^'~°i.186 Then CHAPTER 10. varies only slowly with t). H(°Ht) = ih^uW.77) C/(o)t itiuWfLu^HUMu'ih^U' at or i. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0.h^U> = at ( io. £/(°)£/(°)t = 1. dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t).e.
78). i. (10.92) . (10.85) multiplied by U') on the right of Eq. Then ihjt\Mt)) = ih~u^\Mt)) = u(0)ihi^W and H'^it)) = U^H'U^U^\^s(t)) + ih(^) \Mt)) (10. and obtain from Eq.89) where ^s(i)) and As are state and observable in the Schrodinger picture. (10. (10.91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)). (10.e.78) at (10 85) (10.90) = U^H'\^s(t)).U^HU^U' H'fi'. lfe(*)}. U^H'U^U' (10. H = H{®{t) + H'. We also set H^U^H'UW. (10.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction.80). (10.88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^. (10. We multiply this relation from the left by and from the right by U^ and obtain umHU(o) = c/wt (^WtjW) + umH>Tj(°) i n u ^ ^ + U^H'U^.86) at = U^HU^U'.87) = We define correspondingly as interaction picture state iM*)> = t/ (0).10.84) U^ where H' is also hermitian. We now multiply this equation from the right by U' and insert the first expression on the right (of Eq. (10.85) = where we used Eq. U<®\Mt)) = \Mt)). (10.
the vector varies with time (different from the Heisenberg picture). cf.94) _Hf)Al 1 + lhdAl+AlHf).o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10.95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10.188 CHAPTER 10.e.e.94) Then we have (normally with dAs/dt dt (io=8o) — 0.93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation. ot i.89). i. Since H'j is assumed to be small (i. i. we differentiate with respect to time the operator defined by Eq. ^ (0) l^(*)>. (10. (10. (with ihjtAI(t) = [AI.90) and replacing the left hand side of this equation by the right hand side of Eq. (10. so that with Eq. dAs/dt=0) .80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)).60a) or (10.e.e. (10.60b) with iff0) instead of H.e.e. comments after Eq. (10. near zero and \tpi(t)) is almost constant in time. a perturbation so that H ~ H(°)). (10.92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ .55)) at + imm<Ms_u(o) at dt _umH(. H'j = UWH'UW. (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)). A^t) = U^AsUi0). Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq.H\0)}. we have i. (10.89)^0. In order to obtain the equation of motion of an interaction picture observable Ai. i. (10.
H0\ipn) = En\(fn).102) We assume that before the perturbation H' is switched on at time to. We obtain the operator U^ from Eq.98).e.97) The equation to solve is Eq.93). \Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10. i.87) and (10. (10.7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture. We have the Hamiltonian H = H0 + H'. (10. ^ \<Pn){Vn\ = 1n (10. (10. where according to (10.101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series.t0)H'uW(t. i. ih±\Mt))=Hi{t)\Mt)). Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')).100) (10.99) (1098) Instead of the original Schrodinger equation we now solve Eq.e. H' = H'(t). We assume again that the spectrum and the eigenvectors of Ho are known.7 TimeDependent Perturbation Theory 189 10.i0) = e<Ho(tto)/ft.96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion.80) as E/(°)(t. (10.94) Hi = uW(t.10. (10. Since . {ipn\<Pm) = Snm. i.t0).e. the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em. (10.
The corresponding amplitude or appropriate matrix element is n ( 4 ° V s ) = 5> m (i) n (v4 0) V4 0) >™ = m fl "W ( 10 . (10. This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„.107) Jto With this we obtain from Eq. 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. (10. assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^).104 ) Here we insert Eq. (10. n We wish to know the probability for the system to be in a state \ips ) n at time t after the switchingon of the perturbation H'.e.105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10. the time dependence of \ips } can be separated as for stationary states.88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (10.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm). (10.106) We insert this expression into Eq.103) for m — n and use Eq. Hence we set ^> = £°«(*)l4 0 ) >n. i.190 CHAPTER 10.108) . (10. Quantum Theory: The General Formalism HQ is independent of time.102) and truncate the series after the first perturbation contribution. we can write l 4 0 ) ) m = elHot/hWm) =e . (10.^ V ™ ) .103) The actual state \ipg) of the system is solution of ih\^s) = H\^s). (10.
Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En . One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions.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^. for instance.Em)t' (<Pn\H'(t')\pm) = Wnm(t). Thus we set H'(t) = H'0(t).109) the transition probability Wmn(t) = jp Jo . (10. 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 . This is the case. (10.110) With this we obtain from Eq. but that otherwise the perturbation is timeindependent. » — < (10.112) for a^0. where the momenta of the aparticles form a continuum. t o = 0. (10.8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum.10.Em)t} \{tfn\H'\ipv h 2h(En . A further example is provided by adecay. or if by emission of a photon with continuously variable momentum an atom passes from one state into another.111) Different from Eq. (10.En (10. < KO?t .109) 10. We assume that the perturbation is switched on at some time to.
Em). Fig. Otherwise.192 CHAPTER 10.io. though sufficiently small. Hence lim St(a) = 5(a).113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n . and no other state would be nearby. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken . In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En. we would have En — Em ^ 0.2. .114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. > We see t h a t this quantity has the same behaviour (in particular for a — 0) > as the function 5K(x) we considered previously. 10. if the energies belonged to the discrete spectrum.Em)t} r i 7 TTT.ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum.E7m)t} = (10.— = o (En 2h En 2h5{En . t—>oo Hence also (shifting t to the other side) sin2{^(ffn. Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig.2 The delta function for £ — oo. so t h a t Wnm would be zero. 10. (10.
116) The transition rate V is the transition probability per unit time./ t>oc TZ J_00 dx . in the article of W. pp. We evaluate the integral 0(0) = 1. Example 10. so that initial and final states belong to the continuum.r Z x 1 ^ ) = 4>{Q).W . If we are concerned with pure scattering. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT. 75. (10. t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e . Then.10. . then Eq. Schwabl [246].117) in many applications. F Jo . r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10. Fermi gave this formula the name "golden rule" . Fermi [92].3: Application of test functions to formula (10. Let p(En)dEn be the number of states in the interval dEn. (10. 266. Pauli [227].115) and (10. (10.118) where 8e is the separation of states around En which in the case of the continuum goes to zero.4). f See E.113).113) Use the method of test functions to verify in the sense of distribution theory the result (10. i.e.* In view of the usefulness of Eq. Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). p. sin 2 (tx) ax i—( * According to F. (10. the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t.113) this is* hm .8 Transitions into the Continuum 193 as equal.117) were derived by Pauli in 1928. In the case of the functional (10.t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation . for instance from a potential. _ 2irh AE > > Se. if p(En) varies only weakly with En. 142.117) provides the outgoing particle current (see Example 10.
. = ( — ) dk= 2ir (—  2irJ k2dkdfl. S. 491. formula 3. Ryzhik.. . L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. M. . . with n . 77^ + . = 0 . . Hence the result is 0(0) as expected. . dEk = h2kdk//j. The ingoing particle flux is therefore v / L 3 . Then p(Ek)dEk Since Ek — h2k2/2[i. .a 2 ' i that for a = b = t . p = 1 we obtain I = . 2 .tan p 2 + a 2 . and hence .. 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. and is obtained as ih.^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v .„I. Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) . Quantum Theory: The General Formalism with the help of Tables of Integrals. Jin § m = See I. E x a m p l e 10. E = h2k2/2fi. Then ^ . 2 T / ^ M. 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).. ^o n . .k. . . Hint: Use for ip„(x) a plane wave ansatz with box normalization (volume V = L 3 ) and obtain da from the ratio of outgoing and incoming particle fluxes.? 2TT / L \3k2dkdUn 1 /V(x). x . p..<p). iknx .4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10.117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n .194 CHAPTER 10.947. with v the velocity of the particles.b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn . 1 . L \3k2dkdUn It follows that T. Gradshteyn and I.e " 1 * 1 = 2 lim — . . * ^ = .( ^ V V • < ih 2 _ i k n . V +(ab)2 a _x 2p6 7 m —^ '. § We have r Jo \p\x dx sin ax sin bx  V./ cfc a n p 2 +fc 2 .( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<). K = k 0 .ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1.
148.t). The problem is to solve the timedependent Schrodinger equation ih^\i(>) = H\r/>). t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients. (21. (10.t)+ / n a{E.^°>(E. see the discussion after Eq. . 193.122a) (10. pp.10. 10.t) = ^On(t)^(x. Here S stands for summation over the discrete part of the spectrum and integration over the continuum with the orthogonality conditions J and fdx.9 General TimeDependent Method For a better understanding of the "golden rule" derived above.119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn)./continuum = <San(£)</4°)(x.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.t)^°\E^t)dE (10.x) = S(EE').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. (10. Schiff [243]. Thus we assume that the eigenfunctions (fn = (x</?n) of Ho are known. For a possible application of this result.69). by which we mean without leaving the Schrodinger picture. We expand the wave function V( x . we present another derivation using the usual method of timedependent perturbation theory.^ We start from the Hamiltonian H = HQ + H'.122b) S e e also L. Thus we set ^(x. 1 rfx^°)*(x)^)(x) = 8mn. (10. (10.x)^(E'.121) .120) In the case of Eq. specifically in the case of the Coulomb potential.
.x).„. (10. . Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x. t) = San(t)<pn(x)eiE^h..125) where an(t) — dan(t)/dt. V>(x. (10.x.E').. Observe that these conditions do not contradict Eqs.120).126) + H')<pn^)eiE^h}. (10.„„. H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn).123) We now insert these expressions into Eq.t) = eiEtlhy(E.127) i n c a g e Qf c o n t i n u o u g £. / *.iJ5 »*/Vn(x). We multiply this equation from the left by </?£(x) and integrate over all space.122a) and (10. . f 6kn in case of discrete tp's.E') = 5{E .127) we obtain from Eq. (10.129) .128) . into dip (10.. = e^EE'^h5(E Thus with Eq. since f rJ>(°>(E.196 CHAPTER 10..x. . dnptMrnW = { ^ _ En) (10. 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)rJ><®(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.122b). Next we use the orthogonality relations . i. ^0)(E. . (10..t)dx.x.e.126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10. (10. (10.i) = e. The second term on the left of this equation and the first term on the right cancel out.
to perturbation theory in the lowest order.132) These equations can be integrated successively.130) we develop a perturbation theory. We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip. Comparing coefficients of the same powers of A on both sides. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• .130) H'kn by \H'kn and at the end we allow A to tend to 1. (10. Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one. and we set ak°}=5km or ak0) = 8(EkEm). We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^.133) The numbers ak = const.130) along with the parameter A in front of H'kn. since it is equivalent to the latter.^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] .120).9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h.134) . (10. let's say am / 0. SH'kn(ty^E^l^\t) (10.10. determine the state of the system at time t — 0.e. SH'kn(t)e^~E^ha^(t). (10. they are fixed by the initial conditions. (10. 0. i. i.131) and insert this into Eq.130) This result should be compared with the Schrodinger equation (10. Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10. Thus A serves as a perturbation parameter which permits us to equate coefficients of the same power of A on both sides of an equation. (10. Thus first we have ak0) = const. In the first place we replace in Eq. We restrict ourselves here to the two lowest order equations. In order to solve Eq. (10.e.
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(9.198 CHAPTER 10.136) we see that both "amplitudes" are Fourier transforms and look similar. H'^y^"^^/hd£. i. i. Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum. The amplitude / B o r n is a Fourier transform in spatial coordinates.k . so that ihakl\t = oo) = 0.136) Comparing Eqs. ) ' X V(x')dx'. (10.ml2sin2{^frn (En .132).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. an expression of the following type AS?) = " 2 ^ 2 / e j ( k . ip) in the context of timeindependent perturbation theory. (10.137) . whereas ak is a Fourier transform in time (a distinction which is easy to remember!).Emy in agreement with Eq. We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential.e. In this case we obtain from Eq. (10.e.111).e.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. Integrating the second of the equations (10.135)? The quantity ak\t) is the amplitude which determines the probability that as a result of the perturbation ^H'km the system makes a transition from the initial state m (i. ipm) into the state k. What is the physical significance of Eq.135) We put the additional constant of integration equal to zero. (10. we obtain it*™ = f J—oo SH'kn{t')e^E^'lh5mndt'. ifm{kl) = f J —oo (10. Equation (10. (10. (10.135) and (10.
Z<i = Z) y(r) = _ M ^ . 199 (1L2) . 11. r=  r . Let r : = (x. The Hamilton operator is then given by w= jt + ji_^M. Both of these important potentials are singled out from a large number of cases by the fact that in their case the Schrodinger equation can be solved explicitly and completely in closed form. Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions.m. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics. the position coordinate and the mass of particle i with i — 1.1 Introductory Remarks The quasiparticle quantization of the harmonic oscillator is of fundamental significance in view of its role as the prototype for the quantization of fields. The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1. ( 1L1 ) We let pi.Chapter 11 The Coulomb Interaction 11.2 Separation of Variables.y.2. such as the electromagnetic field (the quantization in these contexts is often called "second quantization".i be respectively the momentum. 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).z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e.Ti.
4) P = p i + p 2 . But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J  = m0/pi2 2 \m\ p22 m2. /'Pi p = m0 P2 A v. P = pi + p 2 . Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc. mi + m 2 iTi'2 r2 = R ^ —r.200 CHAPTER 11. Solving Eqs. r = rir2 mi + m 2 (11. Z) and r = (x. Y. (11.3) and centre of mass momentum P and the relative velocity p/mo by setting r» .2/ mi+m2 The mass mo is usually described as the reduced mass. so that as expected we obtain the expression for p in Eq. The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • . r} representation they have the differential operator forms Setting R = (X. mi + m 2 va\ (11. p are such that as operators in the position space representation or {R. (11.4). (H4) \mi m. . so that as expected according to Eq.5) The momenta P . 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 ' . y. where m 0 = mim 2 .U7&.3) for the individual position coordinates we obtain ri=R+ •^ —r. 2Pip2 m\m2 pip2 mi + m 2 . (11._d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc. z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2.
pj.16) Although we write ri. The canonical quantization must always be performed in Cartesian coordinates.13) * and = M ' 1V1 P = 0. p on the other are completely separated. with M := mi + vn^R = ^ P.12) (11. The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line. ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii. [Ri. we mean n = rx s x.+ ^(r) 2(mi + m 2 ) 2m 0 (H.2 Separation of Variables.Ri. (11.8) and so W =777 v + 7 T . .15) In accordance with Eq. etc.2 It follows that p2 2 P2 n2 n2 n2 (H. Hence Hamilton's equations apply and we obtain.+ ^2. (1114) m0 We observe that the equations of motion in R. Pj (which in the following we write again ri. Pj over to operators ?i.11. p = W. 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).z).= ^. Angular so that Momentum 201 ^ ^ . r = —. for which we postulate the commutator relations* [ri. 2(mi + m<i) 2mo 2m\ 2m. pj.Pj} = ih6ij.y. (11. P on the one hand and those in r.10) we set H = Hcm + H.+ P.j = x.10) One can convince oneself that the transformation from (ri.Pj).pj]=iMij.e.pj. (11.9) ft = ftcm + ^ = centre of mass energy + energy of relative motion. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj. (i. r) is a canonical transformation. (11. Ri.Ri.il) ""Op' i. (11.r2) to (R.
V(r) = V(r).24) . (11. 0 < r < oo.Q 2M' + V(r). as in the case of the Coulomb potential.20) r2MAR_ $(R) = ER*(R).r). where Hcm$(R) = and Hip(r) = with ^total — ER (11.22) If. In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ . i. = rcos9. 9. r ) = J E t o t a i*(R. —n < tp < ir.21b) + E. Thus the twoparticle problem is practically reduced to an effective oneparticle problem.23) y — r sin 9 sin tp. 2mo (11.21a) L*+v{r)] ip(r) = Ei/j(r) (11.e. (11. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11. 0 < 9 < ir. The Coulomb Interaction nc H = 2m. ip: x z = r sin 9 cos ip.r) = *(R)V(r).19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R. (11.202 CHAPTER 11. it is reasonable to go to spherical coordinates r.18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R . to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r).17) (11. (11.
j. k = 1.1 it is shown that pr is hermitian provided lim n/>(r) = 0.j.)In Example 11. is the operator of the relative angular momentum. = 0.5r = . n ( . that the above expression follows from l:=rxp=ift(rxV).pr]=ih. i( i.i ' .A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq. (11._0di})\ x We now demonstrate that 1. (l + . otherwise. where (r x p)j = ^eijkrjpk. k an anticlockwise permutation of 1.2 Separation of Variables.31) .3. = (p2 + ^ V (r + 0). 2 1 9 / 1 9 .32) (11. Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«.33) i. (11. = —1.26) However. with tijk — + 1 . if i. i—>0 (1L25) (11.2.11.3. the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r. (11.12 — I2.27) We can now write (see verification below) p V = h2Ar^ where 2. (11.2.k a clockwise permutation of 1.2. _ h2 \ . We have I2 = (r x p) • (r x p).24)) 2 + M\ d dr2 J = tfL^(r^A\ r2 dr\ dr J d2ip (11.3. . j . i.29) .28) . (11.e.
15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj.40) (11. (11.ih5jk)pk (11. (11.k Using Eq.j.39) + ih).43) r2(p2p2).ihr • p ) .44) .ih)pr and x 2 2 9 9 2 2 Q O (11. (11.35) j.(r • p ) 2 + 3ih(r • p) .42) (11.204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11. r— = r • — or or (11.37) . This means that l2 = so that 9 9 1 (11. ih6jkrjpk (11. Eq. * ' S Since r =xz we have (cf. (11. (11.38) + y* + z .ihr • p r2p2(rp)2 < + ih(rp).34) W i t h this we obtain — keeping in mind the ordering of r and p — I 2 = 5 Z eijkrjPkeij'k>rj>Pk' = ^{rjPkrjPk i.25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p .k ~ rjpkrkPj) (11.41) o = r pr.ih) — rpr(rpr With Eq.36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 . The Coulomb Interaction = &jj'&kk' — Sjk'Skj'.27) we obtain r{prr)pr = r(rpr . (11.
<p))ril>(r.l2} = 0.e.2.<p) (11.48) .[ i J With partial integration with respect to r this becomes (<l>.26).e. <p) d °° . r := r : lim rip{r) = o i . S o l u t i o n : Set d£2s = sin. ¥ p)r 2 drdn s = (p r 0.e. The Schrodinger equation of the relative motion can therefore be written V1 V 2m.6l. 6. ip 6 X>Pr C 7i.e. r—>0 il>(r. Angular Momentum 205 Comparison of Eqs. <p)r2dfls = . (pr4>. h f l d —(r<l>*(r.2TT]. <p)*rip(r} 0.Pril>) J / r= dQs rcj>(r.d. 6. pr = . 9.(r.[ 4>(r.<p))drdns. 8. 9 € [0.11.ip) = (<£. H 1 2mo pi + K) +V(r).e'iP^} . <p)*Pri>(r. Then {<j>. or J .! /').7r].. 0.47) Example 11..6. (11.1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3).<p)drdna " r r=0 J \ll§r~(r4'(r.<p)*?(ri. i. /'(r. (11.6d0dip.Q ' 2mor2 with the condition (11. 11. i. prip) = = I <j>{r.46) (11.2 Separation of Variables.<p G [0.d_ r dr the operator pr is hermitian.e lim rip(r) = 0. Pri/Oi V0. (11.<P) = Etl>(r. <p)* .e.30) I2 is to be identified with the square of the angular momentum operator. reR3.45) oe Thus in Eq. <p))r2drdns % J r or r<t>(r.— (riP(r.1 Separation of variables We deduce from the expressions of H and l 2 that [H.30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11.28) and (11.
49) By cyclic permutation of x.lz (or l . g i = x. ?3 = z). In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily. i. (11.lz]=ihlx. 11. (11. H and l 2 possess a common system of basis eigenvectors.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions. These operators here play a role analogous to that of the quasiparticle operators . /z> *xJ = l <i\}ylx + tx'j/Jj [h.53) Either of these operators is the hermitian conjugate of the other.e.e. [lyA = 0.pz]x = ih(xpy . 2 2 2 = 0. We therefore first search for a complete system of eigenfunctions of l 2 .l2] = 0.q2 = V. we have [h.3 Representation of Rotation Group Our objective now is to develop a representation of the rotation group in analogy to the energy representation of the simple harmonic oscillator. The Coulomb Interaction i. Since 1 = r x p .52) This means that l . i. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0.ypx) = ihlz.lx] = ihly. for any two of the components there is no common complete system of eigenfunctions. z we have altogether [lx. [lz. (11.qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l . zpx] + [zpy. they are incompatible variables. (11. 2 .e. z]px + py[z. y.50) Thus the components of 1 do not commute pairwise.l2} and correspondingly [lx.206 CHAPTER 11. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions. xpz] = y\pz.ly]=ihlz.lx or l . 3 .2: Verify that [h.xpz] = [ypz. Example 11. [ly.51) (11. ly] = [yPz ~ zpy. zpx . However. so that [lz. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously.
57) i. y . (11. (11. in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo. (11. (11.r) (11 = {l02)h{l)2\l) (11. (11. (11. We write or define: / . since lz is hermitian.l)hl\l). In the present context the operators are called shift operators for reasons which will be seen below.Y\l) = (lQr)h(lY\l).54): lzl\l) = {llz . (n.* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq. i + ] = [l2.55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states.49) we obtain the following relations: [iz.61) But lo is by definition the largest eigenvalue of lz. (11. (11.52): 0 = [ l 2 .h~lz(1156) (11. l\l) is eigenvector of lz with eigenvalue (IQ — l)h. With the help of Eq.60) Similarly for l+ with the help of Eq.11. assuming a finite dimensional representation space. Hence we must have Z+I0=0.59) = 9) h(l0 r)\lr). [ig. and with Eq.58) {l0 : max. Therefore (lo + l)h cannot be an eigenvalue of lz.r) := (Z_)rZ) so that lz\l .3 Representation of Rotation Group 207 A.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i).+ ll.lh)\l) = {l0 .54) On the other hand we obtain from Eqs.).51).54) l 2 = \{l+l~ + 11+) + ll = l+l. (11. (11.e. [i+M = 2%iz. A^ in the case of the linear harmonic oscillator. Similarly we have lg{l)2\l) and more generally lz(l.i] = m.i+] = m+. (11. .Z_] = [ l 2 .
60) and (11.56) CHAPTER 11. i. (11.l2\l) = l0(l0 + l)h2l\l). l\l ...56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll.64) we obtain on multiplication by Z_: lVl\l) and more generally l2(f)PJ> (11.e.h)\ln) {(lQnf(lQ. (11. (11. (11. l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 ...\ll).66) It then follows from the second of relations (11. \l). (11.+ 11+) + l2z\l) = i+i.\lr) terminates at some number (let us say) r = n (n a positive integer).n)}h2\ln).55)) we have 12Z_Z> = l.60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i.e.208 Thus for l 2 we obtain: V\l) ' = 11.63) + l ) ^ 2 . IQ(IQ 1 0 (11.From Eq.54) (n = 57) (11. (cf.e. The Coulomb Interaction (l+l.64) i. Eq.y\i) l)h2\lr).Z_] = 0.67) i (11 60) = But according to Eqs. [12. i. (11.e. Since l 2 commutes Hence Z) is eigenvector of l 2 with eigenvalue with l_. (11.65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz.68) . i.65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l .n) (1 = 60) {l)n+1\l) = 0.n).+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i). (11.
1).n .m y/lo(lo + 1) — (m — l)m\l.67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo . § See e. The operators l+.  U o ..1). lQ = .70) they satisfy the eigenvalue equations (with m = l0Jo l. m + l)h.m / + l)h . . lo = l0n. M.. y/(l0±m + l)(l0^m)h.g. Messiah [195].65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2. (11. The nonvanishing matrix elements of 1 = (l+. We therefore write the 2/o + 1 orthonormal vectors: \l.l_. *D.l. II..l0): l2Z.n)} = (l0 .2.69) Prom this it follows that IQ must be either half integral or integral. (11.m).3 Representation of Rotation Group so that with Eq. 24.lz transform the vectors \l). (11.. R. —loh.. p.e.m + 1)(Z0 + m)\l.11. y (lo .m) with — lo<m<lo2 (11. where lz\l.70) According to Eq. Vol. (IQ — l)h.m .{lo . i.....m) = = mh..m) (l. As in the case of the harmonic oscillator we introduce normalized basis vectors. We write these* for integral values of IQ: \l. p. A. (11.. we skip this here.. 17.. The normalization factor in Eq.m) = mh\l.n){l0 . The phases are chosen such that§ l+\l... In the following we consider primarily the case of integral values of IQ.m).Z 0 ).63) and (11. lz\l.m) : Mo).m±l\l±\l. lz) are then: (l. The dimension of the representation space is therefore n + 1 = 21Q + 1.m).m) = IQ(IO + l)h2\l. \l — n) into one other..71) can be established in analogy to the method used in the case of the harmonic oscillator. (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator.l)h \/(lo + m + l)(l0 .l)h. Satchler [38].m) — mh\l. The basis vectors \l). (11.m\lz\l. According to Eqs.m) l\l.n) 2 .m)\l.m) = = = = y/lo(lo + 1) — m(m +l)\l. Brink and G. \l — n) are eigenvectors of lz with eigenvalues loh..
2. and m = —IQ. $ oc eim*. l_\l. for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip). with i2Y£{e.210 Then. 9. We now perform the analogous procedure for the angular momentum or.yp. . ip... as remarked earlier. as above.e.76) d2 (11. . 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. mhY£(9. In these coordinates we obtain: h l± so that h2 = xpv . ip) = e(9)$((p).77) $((p) = m2<b(y).<p) = i)n2Y^{e^).<p).With the ansatz Y(9.72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r. . /o. where IQ = 0.73) 1 d d sm0 06 \Sm9dd) + (11. CHAPTER 11. for the simple "rotor". 11. d2 sin2 9 dp2 _' (11.75) the variables contained in the expression (11. ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ . as this is sometimes described. (11. (11.74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9)..l0)=0. —lo + 1 .74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation. (11..1. We . i. The Coulomb Interaction Z+Mo> = 0.<p) = i0{io + lzY£(0. .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.
t»'\Y7n((l'. These functions are defined in such a way that they are normalized to unity on the unit sphere.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.ypx = tfrK(11. orthonormalized system of eigenfunctions.(5cos 03cos0).11. (11.4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs. (11.76) and (11.=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™.0) is real and positive.<p) Jo = 6mm'5w.(p) are called spherical harmonics.* ? ^ . The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9. (11. (11. so that im(ip±2ir) The functions Ylm(8.r o ) ! YT{e. and that their phases are such that Yt0(0. (11.^ = *(n . / 3 / 5 J — casO.82) .79) .\ = W .78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}.i». We mention in passing that in spherical coordinates d lz = xpy . y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—.<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?.80) y? Furthermore Y°l = V^fcosO). This is the case if one demands that l 2 and lz possess a common complete. Uniqueness implies immediately that m has to be integral. is given by (2Z + l ) ( Z .V>)Yir'(8.n>).77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability. m > 0.
x2)y" . respectively. For the solution we use the ansatz of a power series.l)xK+i~2 i .1.2xP[ + 1(1 + l)Pi = 0.m(m + l ) ] i f = 0. exp(imip). Instead of the values I = 0.2(m + l ) x i f + [1(1 + 1) .90) . (11..212 CHAPTER 11.2. of Ytm. Consider the equation for Pi(x).88) we obtain for every value of x ^2 ai(K + 0( K + * . (1 . (11.g.m = 0. Setting y = (lx2)m>2Pr(x) m . i.p..d. i=0 (re lowest power > 0)... f. (11. It is instructive to pursue the appropriate arguments. i.e. (11. i.h.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 .2xy' + 1(1 + \)y = 0.87) we obtain the equation (1 .e. 22 (11. we would now have to search for those values of I for which the solutions are square integrable. the designation s.85) (11. is also in use.e. (11.1(1 + l)]xK+i = 0.x2)y" . in the replacement ip —> if + 2K.x2)Pf . i.86) y = 0.x2)P[' .2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function.89) Inserting this expansion into Eq. oo (11.e. The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 . If we did not know yet that I = IQ = an integer > 0. The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed..Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) .88) y = ^aiXi+K..
We have therefore (multiplying Eq. they have to be polynomials (recall the orthogonalization procedure of E.95) Vl = rxiThe hermiticity condition for pr then requires that (cf. Schmidt). (11. first line of Eq. (11.E Xl(r) =0 (11.Qr2 with (cf. Eq. (11. ai(re + !)« = 0.*" 11. For \x\ < 1 we can obtain from this the coefficients of a convergent series. p.26)) yi(r = 0) = 0. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) . when the series terminates after a finite number of terms. We see that this happens precisely when I is a positive or negative integer or zero. (11. > 1. l(l + l)h2 _2mo + 2m. (11.1(1 + 1)]. This is precisely the case.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.94) ^For further details see H.1. l 2 and lz are the solutions of the Schrodinger equation of the form (11. Eq.91) from which we deduce that K = 0. (11. Margenau and G. For the solution in accordance with (11.92) This is a recurrence relation. Murphy [190]. these are of little interest to us here. However.96) + V(r) .29)) 2 fe2l 92 Pr = & r 7Tororz We set (11.94) from the left by r) h2 d2 h2 2 + H! + \)IL.11.1 we obtain by comparing coefficients QOK{K 213 — 1) = 0.46)) p.78) to be square integrable.93) where Xl(r) is the solution of the radial equation (cf. but also those for a different series for a.5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0. 93. M. .2 + 2mo dr 2mor V{T)E Viir) = 0 (11. .
100) with (11.99) e ip(r) = Eip(r). r m0 = meMp me + Mv (11. s = 1 + 1. The Coulomb Interaction (11. (11. He + . since it satisfies neither the condition yi(0) = 0. Equating the coefficient of 1/r2 to zero. L i + + . This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential.97) / r2dr\Xi(r)\2 = / \yi(r)\2dr.96). (11. 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. Z\ = Z2 = 1 in Eq. and others. an analogous consideration applies to hydrogenlike atoms like. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 \ ).e. oo) of a onedimensional space.101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. and substitute this into Eq. We assume that V(r) has at r = 0 at most a singularity of the form of 1/r. (11.98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11. Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0. In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq.96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0.102) . which behaves like 1/r1 in approaching r = 0. for instance.1.1)). we obtain the socalled "indicial equation!'' s(s .e. nor the condition of normalizability for l ^ 0. The equation also has an irregular solution. We now consider the hydrogen atom with V(r) — —e2/r (i.214 with CHAPTER 11. Obviously the irregular solution has to be rejected. In the first place we investigate the behaviour of yi in the neighbourhood of the origin.1)+1(1+ 1) = 0. i.
Inserting the ansatz into Eq.a * = 0. see e.103) °y moe2 ( = 0. for E < 0 they > decrease exponentially. p.g.+if>x/2 vi(x).6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo.x ) $ ' .11. behaves there like xl+1) extended to an exponentially decreasing branch at infinity. *The factor 47reo appears in SI units. (11.e. (11.6 11.105) 11. .529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11.108) This equation has the form of a confluent x$" + (6 . d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 .104) W i t h these substitutions we can rewrite Eq. W. (11.88. Oberhettinger [181].107) equation.105) we obtain the equation for V[(x). We can see from Eq.e. (11.e. Magnus and F.6. hypergeometric (11. It remains to determine the function vi(x) for a complete determination of the solution. T This equation is known as Whittaker's equation.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.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. We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT. We set therefore Vi X J. (11.v) vi(x) = 0. i.105) t h a t for x —• oo there is a solution behaving like exp(—x/2). (11. (11. i.100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0.
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.! ) ! = imlFI' c a l led inversion or reflection formula. ^ ( 2 ^ ) ! = 2 2z z!(z . (11. However. called dupZication formula. particularly pure mathematicians. and hence the *Some writers. The function T(z) is defined by the integral /co T(z) := / eHzldt. meromorphic function. more precisely. 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. (5)r(n + l ) = n ! . (11. + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z. and for both cases. with many of these around in some calculation. (4) r ( i ) = V5F. it is convenient to use only the factorial notation.216 CHAPTER 11. (2) r ( z ) r ( l . (11.z(zl)\ = z\. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 .z) = ^f^.J.(0_i)!(6 + n l)!n!' (11.112) The gamma function is a nonvanishing analytic.e. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases. the factorial for an arbitrary argument.109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function. .106). since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq.1/2)!.111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). (z)!(z . 02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + . i.
The quantum number n' is called radial quantum number. 1=0 (11. every n. One defines as the principal quantum number the number n. Eqs. the orbital or azimuthal quantum number I can assume the values 0.2. n' = 0 . for z.. (11.e. (11.113) This is a quantization condition.» • » • » .V))T(v .11. This is achieved. which are also described as its nodes. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin). Thus the infinite series must break off somewhere. .a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ . which is thus seen to be a consequence of demanding the square integrability of the wave function.e.) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq.I) r(vl p)sm{n(i/ I ..114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2. . (11.6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable. 2 . = n = Z + l + n'.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.v) T(l + lv) _ ~ K sin{7rQ/ . The degeneracy of the levels En is therefore of degree* n—1 1=0 n—1 V(2Z + l) = 2 V Z + n = 2^(n . if the series terminates for certain values of u. With (cf.1) + n = n2. . . For every value En.101) to (11.e. i. in our case: a = 0. and so to the number of different values of m.. i. must be a polynomial. • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression.1) + n = n(n . d = 1. i.104)) 1 n =v = K. 1 .. .1.110) as follows: T(l + 1 + p .
1 = 0 (called principal series. (b) The spin of the electron was not taken into account.. / = 1. I = 0 to n = 2. The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically.\ I = 0 or 1.d.1 = 1 to n = 2. The spectrum of the hydrogen atom thus has the form shown schematically in Fig.'.1..§ One also writes n2l+1l.1 = 0 (called sharp series. I = 0.1 Hydrogen atom energy levels. . 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. One describes as Balmer series the final state n = 2. .4. 0. nd.. 11. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11. ' O n e describes as Lyman series the final state with n = 1. from an initial state n > 2. and from n. hence d).l = 2 to n = 2. from n. respectively. nf . where the superscript 21 + 1 at the top left of I gives the degree of degeneracy.1. from an initial state n > 2.. The following spectroscopic description of states is customary: nl —> ns.. (c) The nucleus was treated as pointlike and not as something with structure..'. ne.. 11.218 CHAPTER 11.1 = 1 (called diffuse series.p.. . hence s). 2 . np. stand for I = where n denotes the principal quantum number and s. . and hence the fine structure of the energy levels is excluded. hence p). .
S.p)\(2l + 1 + p)\ The spectrum (11.r = 0 . . • ( . Appendix B. (11. k + l. • fc. formula 7. We distinguish between different definitions in use. in fact. (u.117) The function *Lk(z) is a polynomial of degree r. .r f .= t . Apart from factors. (11. p.110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials.115) is effectively the associated Laguerre polynomial normally written Lk(z). for instance.414(3). Gradshteyn and I. 2 . a + 2:*) = t v .115) we can reexpress this function as Lr{z)  r\k\ p=0 \* p\ (rp)!(fe+p)! (U 118b) ^This definition of *LJ? is used by A. M.114b) yields the binding energy of the states. so that one always has to check the definition. The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me. For clarity and later use we denote the function used there by its normal form L^. Unfortunately the definition mostly used in mathematical literature differs. see. p. Messiah [195]. the polynomial contained in y\ of Eq. We mention here three different definitions used in the literature.6. however.z) = (r + k)\ Lkr(z).2l { + 2. . Vol. W.1. 84. (11. Each can be recognized by reference to the generating function. "1'f„+11" P] „. and I. Ryzhik [122]. Oberhettinger [181].{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z).118a) where Lk. several different definitions are in use. l .2.2 (11. Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr). With the help of Eq.115) ' =0 (ri .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. 11. it is given by^ *Lkr(z) = ^ ^ f 1 *ir. 3 . • (11.11.x). Unfortunately. I. . 844.6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'. . Magnus and F.
** The associated Laguerre polynomial here. .164. .1.220 CHAPTER 11. The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'. Messiah [195].122) "See A. I. ^ F o r this definition and the following equations we refer to I.120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk. Ll(z) = 33z + ±z2. It is said that this is the definition preferred in applied mathematics.z . 163. L\(z) = 2z. r r Z r=0 v tr k k (11. L2(z) = l2z + z2.119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1. Appendix B. **I. For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials. The generating function of the associated Laguerre polynomials — obtained by searching for an integral representation of the solution of this differential equation which is of the type of a Laplace transform — is for i < 1 and expanded in ascending powers of t (cf..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. N. pp. Sneddon [255]. is defined by^ dk (—'\\k(r^\2 (11.2. There at some points n + 1 is misprinted n + 1.(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11. (11. p. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 . and Ll(z) = l.„M„ z"*L. 162 . Vol.3): e2t/(lt) (1 . Sneddon [255]. Example 11.. which we write ^Lk.3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function. N. . L1(z) = l .„\*Kk.121) In Example 11.d z—^z + {k + l~z)—+r Lkr{z) = 0.
4 + 2z.124c) Example 11. *L\(z) = .. The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11.. and obtain with this the normalization condition (11.% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp.3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq.118a) and (11. ^L\(z) = .. (11. i L1(z) = l .123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !. Then.121). (11. '"''Comparison with Eq. .1 .107) implies k = 11 + 1 and r = n + I.120)) it)k& zt/{lt) ~*P p=k fc (11. with y'{i) = — / ps(p)e~ptdp ty'(t) = t f ps(p)eptdpp2s(p)ept s(p)e Pt J ^(ps(p))eptdp.z .119): ty"(t) + (a + 1 .1 8 + 18z 3z2. (11.119) derive the generating function of associated Laguerre polynomials (11. (11.124a) (i*) pi The orthonormality condition is (cf. N d rk 0.11.120).8): (11. (11..122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z). The limits will be determined later. (11. and *L\(z) = . and partial integration. ^L2(z) = 2 . Example 11.124b) Comparing Eqs. Jfpd>2s{p))eptdp.4z + . Solution: We consider the following equation of the form of Eq.6 The Discrete Spectrum of the Coulomb Potential The differential equation of these associated Laguerre polynomials is** z 221 ~r^z + (k + lz)— + dz dz d2 n .
Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a. e ^ t '?/( 1 '?) = coefficient of qb in /"OO / = = / Jo Ll{t)Lac{t)tae~tdt taet{P/(lp)+q/(lq)+l)dL l t I * * [" (2TT.a = 21 + 2.] from the above equation: C(p) := p(p + l)s(p)ept = 0. Then the preceding two equations (inserted into the original differential equation) leave us with I eptdp d {p(p + l)s(p)} . since for n > b : (b — n)\ —• oo.^ . p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo. t > 0.( a + l ) p s ( p ) + 6s(p) dp 0. p and one has (as polynomial of degree 6. the only possibility is the last.) 2 J J p 6 + i ( l . s bK> (lq)°+i (11. It follows that 1 d i I _i.. this is the rejected irregular solution. 2TT_ Jq=0qb+l(lq)a+l 1 . (11. with requirement of a finite solution. (3) around p = — 1 if a + b is a negative integer." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt. . (2) from —1 to oo.p ) « + i q c + i ( i . and (4) around p = 0 if b > 0 and integral.124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials. Hence the expression in square brackets vanishes..118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <*>" (a + 6)! n! —(.p ) o + i 7 t = 0 . agreeing with Eq. We demand (to be considered later for integration contours) the following condition. 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.222 CHAPTER 11. j p = (P 0 "V?. which removes [. we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* .p + 1 = 1/(1 .q) and P~ t ( . Case (2) implies the exponentially increasing solution ~ e + t . The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp.li I u {p(p+l)s(p)} : p(p + l)s(p) dp a + 1 b p+ 1 p(p + 1) a + b+l p+ 1 b p Integrating the equation we obtain s(p) and hence y(t): a+b+l^b ln[p(p + l)s(p)] = In ip+ir+o+'p (p+l)a+b f (p+l)a+b The condition C(p) = 0 now implies for possible contours of integration in the plane of complex p: C(p) = p(p + l)s(p)e'pt („ _i_ na+b+i = ^ ' e~pt = 0. (11. Thus in the case of Eq. n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1). case (3) similarly. Then dp = dq/(lq)2./ ( l — <?) —— = G(t).119) with a and b as there.
(11. is introduced for dimensional reasons.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.127) .112).(x). Jo (11. where Fnl(x) = yi(x) (11.—: <f h_nJ_[ dp = (a + c)\ T^ if c = o. /•OO / Lg(t)£?(t)taefctt: (a + c ) ! .111). I.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.125a) and the factor a~ 3 ' 2 with a = Bohr radius. (11. The wave function of the state (n. m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O.—i.?). so that I = (c + a)! 1 f . and 0 if cj=b. . 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. Clearly only the second form is sensible.11.c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11.3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En. and obtain r taet[P/(iP)+q/(ig)+i]dt Jt=0 With this the integral I becomes = . Eq. f a p)(i 1) r + 1 l (1P9) J The coefficient here required is obtained from the expansion as .6. 11.
130) We see that the mean value of r is the larger the larger n is. 2 (11.224 CHAPTER 11.7. 11. thus on the average the electron is the farther away from the proton or nucleus the larger n is. a (r)is = ~a. For the ground state (n = 1.3(2Z + 1) + 2} 2 {2(n + 0 .2. The Coulomb Interaction One finds in particular (see also Example 11.(2Z + 1) + 1} = ^[Zn21(1 + 1)].n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ . = 0) we have ) r /u =.(l) n+l. This behaviour is indicated in Fig. (11.128) These expressions permit us to obtain the mean or expectation values (1/r) and (r). (11.131) ..1 + 1) (11.2Z . and for the explicit derivation of the second case below Example 11. + .3fc + 2).6(n + Z)[(2Z + ! ) .v (6n2  6nk + A:2 + 6n .8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ .! ] + (21 + l ) 2 .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 . 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 + /„.
S. Problem 5. v ^ V2n + 1 (11.20.j .2 1 A r = V (^2) . In the case in which I assumes its maximal value n — 1.l ) n ] =  [ 2 n 2 + n] an[ n + (11. Sneddon [255]. pp. 173/174.11..2 Hydrogen atom wave functions. In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton . / ! J — . i.133) for large r.6 The Discrete Spectrum of the Coulomb Potential r. 11.^ We obtain therefore for the quadratic deviation (r) . In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 .(r) = x W 2 n + 1 =. = l a nl n N'nll rFnl.( n .134) *These are taken from I. we have and tp becomes particularly simple. with n +1 replaced by n and 21 + 1 by k.e.J n 1s 2s 3s *r Fig. This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o. \f{r ) ~ n a.2 I 3/2 Kl r 225 \2 1 R„.132) (11. or (r ) ~ n a 2 2 ^ [ 3 n 2 .
identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*.3. i. in which periods are obtained from the BohrSommerfeldWilson quantization condition.r. \/v vT 7 v^ See also Example 14.4 the relativistic equation in cylindrical coordinates. but rather like classically in a plane (i. and separating this in cylindrical coordinates r. Since 6 is a cyclic coordinate. This means. In order to obtain an impression of relativistic corrections in a simple context.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). Thus.226 CHAPTER 11.e. Example 11. i = 0. the states Ytl are predominantly concentrated in the xyplane).135) This expression becomes very small for large values of n. . The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. r.e.— > where ampler is the length of the semimajor axis of the elliptic orbit. = ngh. However. the electron remains practically localized within a spherical surface of radius (r). relativistic corrections are of the order of v2/c2 ~ 0(En/moc2). Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r. „ _ 'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler . pJ = —— and j> Pidqi = nth. we consider in Example 11. Solution: We have H = yjm%c4 + pIn cylindrical coordinates. determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory". p2 = p2 + — pg • eV. 6. we have pg = const. 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. (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded.6. The Coulomb Interaction or TT = ^ = (11.
it is possible to separate the Schrodinger equation in other coordinates.T).6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16.136a) (11. 11. 11. 2 2 Therefore the energy is no longer a function of n = nr + ng.9. ip are the spherical coordinates. ( v c + i ) . Whenever there is degeneracy. z=^rj). where ip = 0. y=y/%qsanp.2.(p .6. (11.d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—». which correspond to a new choice of polar axis.3.4 Hydrogenlike a t o m s in parabolic c o o r d i n a t e s The Schrodinger equation of a hydrogenlike atom can also be separated in parabolic coordinates £. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 . <p = <p.136b) £ = r ( l + cos#) = r + z. y — 0: V ^ c o s tp. p. .. the "fine structure obtain the ordinary nonrelativistic energy) Z2ei c2h2[nr + ne^jl 1/2 E — mac + 1 Z2ei/c2h2n2] 1/2 VTIQC 1+ •+ ng^la Z /r. r\ = r ( l — cos#) = r — z. as also indicated in Fig. where r = ^/x1 + y2 + z2. The deeper reason for this are symmetry properties of the problem. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines.V . It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'. where the constants are identified by comparison.* See L. which are related to Cartesian coordinates in the following way. 86. Schiff [243]. (p). In the case of degeneracy with respect to the magnetic quantum number m (as in the present case) one can find linear combinations of hyperspherical functions Yj^(9. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137.11.
The Coulomb Interaction Fig. so t h a t the variables can be separated. .22) becomes — as we show in Example 11.137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip. We rewrite the first of these equations with the help of the relation M (w ^  m + e) . = 0. Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A. and r\ = const. a separation constant. Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 . . In terms of the coordinates (11. one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o.3 Parabolas £ = const.136b) the Schrodinger equation of relative motion (11.3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip. d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike. . for uniqueness of the wave function. with zaxis as axis of rotation. (11. 1 . 2 . 11. there is really only one basic equation to solve here.228 CHAPTER 11.
.114b). . . . .105)) Hence we write I = (m — l ) / 2 . from Cartesian to parabolic coordinates. Eq. p = /?£ and cr = /^r?. i. (11. In the spherical polar case we had (cf.' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a).6. . . 2 .6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0.± ( m + 1) = 0 . . 1 . m0Z1Z2e2 ^ 2 = n (say).e. (11.i n +n = n i + n 2 + m + l = we obtain the energy eigenvalues given by E = n i = n' \(m + 1) = 0 . Adding now / . T h e n by comparison with the spherical polar separation the answer is as in Eqs.106) and (11. To obtain a closer analogy with the separation in the spherical polar case. We leave the calculation of the degree of degeneracy to Example 11.125b) (dividing by p 1 / 2 ) £ oc e . E x a m p l e 11. . 2 . " . T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) .o. . ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11. 1 . we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". ds2 = dx2 + dy2 + dz2. n 2 = n" .11. . as well as the Laplacian A = V and the Schrodinger equation.5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric.
W. etc.. In the case m ^ O w e have n\ + 77. We then sum all possibilities and select the coefficient of ojn~1 in the result. for m / .* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n . > . last chapter. Solution: For the wave functions ip = £(£)J\f(r])&(ip). dx = d{ y £77 cos ip) = . Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly). / — cos ipdl.2d? dr)}. see Eq.[df + dr]2 . as t h e two signs of t h e exponential indicate.g. 2 + ^dV . Thus we want to obtain the coefficient of wn~1 in £2 . We consider first the case m = 0.. .tx2 . (11. all cross terms cancel in the sum of 7 the squares. 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11. Observing that the highest possible value of m is n — 1 (i. We use the method of selector variables.1 in this expansion is seen to be n. (n _ !) n = n2 §See e.l)a + (n . when m = 0 = 77. Thus dz2 = .e.2 = n — 1. \— < / — cos ipdr] — \/£r] sin ipdtp. {g^Y.2 . Oberhettinger [181].e.136a). for instance.l ) ( n .e.2)d} = (n. are elements of length. and attach a factor to to each term introduced in the wave function. The Coulomb Interaction Solution: We begin with z of Eq. Thus. 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n . n the double sum ]T]2 is TI — m. N2 where g^dS.E E n l „ . = e±im^e("+^'2{par'2L^{p)L^ (a). 712 and m which give n. Similarly we proceed with dx and dy. (11. / J„^2 . 2 _ = (9id02 + (gvdvy + /„ .2)d/2.2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 . and so dz = (d£ — drj)/2. Eq. </. 0 there are two.e.g<p 9 \ .114c). Beware! For m = 0 there is only one mstate. we have to find the number of values of the quantum numbers n\. " . i. (11. 2 2V? V^ One now performs the square of this and obtains similarly dy2. and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j. Here we have (see preceding text) 711+77.?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr. n 2 (1UJ)2 The coefficient of a . The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr. i. . z = (£ — rf)/2.6: 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.230 CHAPTER 11. + l 4 {^^)drl f i + v \ . 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. . i.1) ~ . Magnus and F.2) and summing over m.136a).
which will then give the change in energy of that state to the first order (i.(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I. fc — m. 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.1 (cf.r. with z = r cos 9. (11. ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny. In fact. Z2 —> e. . { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } .g.e. Sec. .6 The Discrete Spectrum of the Coulomb Potential Example 11.128) if one makes there the substitutions n —> n\ (or 712) + m. Ikr+k.6). . we can relate the expressions Iq to those of Eq. if we expressed the perturbation in spherical polar coordinates.e..eF / nh2 \ {v)=3(n1n2)^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8. We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£.™(p)] 2 By contour integration (see Examples 11. with RayleighSchrodinger perturbation theory).11. (11.) = ^eF(p . Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e. eF{z). we would have to use degenerate perturbation theory and obtain the same result.•* n (l) = / Jo It follows that . L ni\ ny.e. = j (2ni+m+l).124c). 8. s 0/ . . 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by. and determine the average value of v for a given state.{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—.9) one finds that _ = %2 jm+2 {n2+m)\ (ni + rre)! +1 : > in.r+k(l) = / Jo e~*xk[(k + r)\Lk(x)f = [(/= + r)\flk.7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz. these expressions agree with those of Eq. .a). (11. {p — a) — = dxe~*xkvLkn (z)] 2 . where F is the electric field. with Eq. Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function). with the connection » (11.8 and 11. Thus e. However.—r~n = 3(mn 2 ). I. i.127) and setting n = r + k. With the help of the wave functions we derived above. i. with Z\ —> Ze.128).drid<fi oc (p + o)dpda.
3: /"OO I := Jo et[L^{t)]2tm+1dt. 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 .n+2 in the expansion of i n c o e f f i d e n t o W (m + 1 ) 1 ( 1 . In Example 11.( m + 2 ) is equal to the coefficient of (qn — qn_1) in (1 — pq)~(m+2\ The coefficient of qn in the expansion of (1 — pg)~ M is pn(n + fi — l)!/rt!(/i — 1)!.1 ( n + m)!"[ _ (n + m + 1)! (niy J n! (ra + m)! _ (n + m)! (n .1)! n\ This verifies the expression I™t in Example 11.p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n . (11. c positive integers.^~T^——^ ) = coefficient of (pn .pK — — —L .9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq.a. .b. (11. L™(t) = coefficient of qn in G(t) e tq/(lq) (1 . The Coulomb Interaction E x a m p l e 11.9 we consider integrals with an arbitrary power of t.124d) of the alternative integral representation obtained in Example 11.232 CHAPTER 11.x ( n + m)! ( n . s = at.p ) ( l .128). the integral arising here becomes with Eq. tt„:_ (1 .8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11.0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( 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.111) /*oo /*oo „ — sam+l / e~attm+1dt= (m + 1)! ds = am+2 I ' Jo For use in the following recall that Jo «m+2 {i + X)n = ~ ^2(ir . dt = ds/a.p ) ( l .p ) r + 2 [(lp)(lg)]"l+1(lpg).9 )« Solution: Using the generating function for both associated Laguerre polynomials.7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11.g )(l.3).3 by one power of t. E x a m p l e 11.9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11. (n + r . and is a special case of Example 11..l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l. ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) . the integral / is the coefficient of pnqn in the expression _ / . _ „ .l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t .1 ) ! (n .
e. c + b odd).Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01. (lp)(lq) + a(l q)(l . — (cr)!r!(6r)! Kv.p ) ] . we have 2r = 6 + c = even.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 f + c)\(vr)\(l + ayrP „ • ^0^r(cr)\r\(a •• Ha) a\ f dp ac(c + a)\ ^ =%*??* c!a!(l + a ) ^ c + l 9 ( p ) = g J c+*"«a> where with /3 = a + c + l + b (observe that for a = 0 a term with c + b — 2r = 0 remains) „_ = 2 ^ 2 ^ K «. (a) j / = 1. = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a. so that = (q+fc±£) . Using the contour integral representation (11. . i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q .e. c + 6 — 2r = 0 (i.^ ( l . . ° ^ + a ( l . In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 . c + b even) and/or j / = 0. we can reexpress 1(a) in the form: /(a) = (2^)2 J J p t + l ( l . c + b — 2r = l (i. i.6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£.q)(l .w(r) = and / a ) = r\(P ~ r .a t [Lg(i).a _ 1 = [1 + a ( l . 7£. we have to have y + (c + b — 2r) = 1.?(t)t a e.e.p) .f ].(i) = i ! ( . i.r ) ! a ' .L. where (a+**£)! c j n Jo = .11.™( r > (1 + a)P r £^0y^w Jc+6_2r = .124d) of an associated Laguerre polynomial.l ) * x coefficient of a* in the expansion of 7 ( a ) .111) of the gamma or factorial function.(*) from the coefficient of a1: />00 233 1(a) = / Jo dte.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.
Taylor [267]. + (!±<)«. ^A very readable source to consult.7 Continuous Spectrum of Coulomb Potential In this and the following sections we consider the scattering of a charged particle in the presence of a Coulomb potential. we consider the traditional treatment here and refer the interested reader to literature in which it is demonstrated that the derivation given below finds its rigorous justification on a basis of distribution theory. The Coulomb Interaction b + c\ 2 J (b¥)\(Pb¥w)\{a) ($±£ w)\w\(J3 ^±£1)1(111.7. . Since such a rigorous treatment is beyond the scope of this text. R. In view of the slow falloff of the potential at infinity.)!' f^)(. however parallel to the Coulomb case.^)(.234 and (only w = 0 and 1 arise in the sum) CHAPTER 11. 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. also with regard to earlier literature. These results are seen to agree with the expressions 1^. is the work of J. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. One should actually consider a screened Coulomb potential* and consider quantities like the partial wave expansion (to be defined in a later section) as distributions. b\ V cited in Example 11.1%^ 11. . the standard principles of nonrelativistic scattering theory.~* 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 . Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r). 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11. t We have the Schrodinger equation of relative motion of particles with charges Z\e.
i. (11.z) + W2(a. b.z)).b.b. . a(a +1) z2 ] (11.138) becomes 235 ^A + k2 . (11.108) and (11. (11. 86 .^ W ) = 0.142) This equation is of the form of a hypergeometric equation. . z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a.z).z) = Wi(a. (11.. In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr).M) = l + .11.144) *(a. with the asymptotic expansions (11.+ ^  + .. for instance. v = ik(r . z = rcos<9.b. „ az 1.e.140) The simple Coulomb potential permits particularly simple solutions. where ^. pp..140) and using (11. Oberhettinger [181].z). what is the > behaviour of 3>(a.109) we can therefore write the solution ipr = Aeikz$(i>y. Our question is now: How does ipr behave for r — oo. (11. for r — oo) • ipr = Aeikzf(r .87 under Kummer functions. that we encountered previously. A = const.7 The Continuous Spectrum of the Coulomb Potential Then Eq. (11.e. (11. (11. W. Magnus and F.136b).146) See. i. According to Eqs..143) and v = ikrj. writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0.145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr). .141) Inserting this expression into Eq.
(11.l.151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)].153) The asymptotic solution represents t h e stationary scattering state (E = const.) of a particle with incident momentum hk in t h e direction of z for large distances away from t h e scattering centre. i.(11. We now define a quantity f(9) by t h e asymptotic relation lb O < ^(. > A = Aeikz[W1(i1.a) (z) na)"~ r(la) T(ba) n\ ' (11.154) . (11. The Coulomb Interaction W2(a. if tbr were „ikr ibr ex el"z + f{6) r (11.152) Since z = r c o s # .148) .^l*7 l + O (11. Without t h e logarithmic phase factors.z)T(l .i 7 ) (11.ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n. A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr. J\") ci(kr~/\n2kr) (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) .z) + la)T(n + b. a comparison of Eq.7 ln*(rz)) k(r .ik(rz))} =^ + ^d. with 2.kz+l^ktrz)) C .150) ^(rz)]1"^ 1 + 0 T(«7) i. (11.e.149) (11.l.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo. = r cos 0.b.236 n\Zzab CHAPTER ^T(n '£ fo 11.151) with Eq.e.
also called phase anomaly. the effective range of the Coulomb potential is infinite.4 Variation of the observed differential cross section with 0. 11. We define as current density the quantity h J := 2imo [^(v^)v(v^n.156) . the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain. The quantity /(#) 2 would then be a measure for the scattering in the direction 9. that we encounter here is a characteristic of the Coulomb potential. 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. As a consequence of the masslessness of the photon. z — — oo. (11. In other words. we see that this expression is the exact analogue of a current density in classical considerations. 2im.Q hk mo v0.11..1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity. do dQ K 12 K Fig.7. the current density is * h . The problem of the logarithmic phase.n .155) Since the gradient is proportional to the configuration space representation of the momentum. This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles. For the incoming wave ipi = exp(ikz). (11.
That this formula retains its validity here in quantum mechanics is something of a coincidence.e.153): 4fc2 s i r  The expression /(#).159) as the Rutherford formula which can also be derived purely classically. Correspondingly the other part of (11. This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V . The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later). 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. is called scattering amplitude. the square 7 2 ). In reality this divergence does not occur. This implies that scattering takes place for the attractive as well as the repulsive potential. for 0 —> 0.152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11. 'screened ZlZ<2e2 cr/r0 ° . in the current this would in any case contribute only something of order 1/r. In Eq.157) The differential scattering cross section da into the solid angle element dVL is defined by the ratio da ir which in the present case is with \f{6)\ obtained from Eq. We recognize the result (11. i. We see that a depends only on the modulus of the potential (i.152) as 1.e. since in actual fact the Coulomb potential occurs only in a screened form. (11. We obtain the total scattering cross section by integration: „2 — /£<" = £/777*V sin 4 « 1 »«°) We observe that this expression diverges in the forward direction.154) we chose the prefactor of the incoming wave in (11. (11. which yields the scattering cross section. We make a few more important observations. thereby screening itself off from the surroundings.238 CHAPTER 11.
t). 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. of the equation r fi.t) = e . i. t 0 )^k(x. of continuum wave functions. (11163) (11. 11. the wave packet in the absence of a scattering potential. be given by ^ f dk AL.i i f (''°>/Vk(x.11. (11. The particle velocity v is v = — . 11.2 fr2 . 2mo V + F(x) k (11.164) Ek~Eko + {kk0)v0h.*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * . t 0 ) . whose maximum moves with the particle velocity.8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet.t).e.e. _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) . 2 (11162) where Vk(x) = ^ ( x . and if H is timeindependent one has the stationary wave function ^(x. m0 m0 and V.yJ ^ j A k o W 727)3/2 e 0.166) Let the incoming free wave packet at time t > t$. / dk 7^340(k)4(xxo. i. which in the case of free motion (zero potential) are simply plane waves. (k). The wave packet is a superposition of waves ip(x.4.8 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.e. t0) = e. v0 = .x° ' r j .k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0.161) where Ak0(k) ~ A(k — ko) and ^ k (x. (11.^ ^ % ( x ) . i.165) ^ ( x ) = £ fe Vk(x). t) = U(t. and which provides the uncertainties arising in quantum mechanics. t 0 ).
(11. The Coulomb Interaction In Example 11. can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf.ip) is the scattering amplitude with respect to the incoming plane wave.x °Vk(x). From Eqs.154)) that for x — oo.167) by replacing the plane wave exp(zkx) by this scattered wave.ik .169) The wave packet in the presence of the potential is obtained from the free form (11.165) ko • (k .173) . (2TT)3/2 eik* + —fv(P.240 CHAPTER 11. i. this solution > has the following asymptotic behaviour in which f(6.167) we have ^k0(xx0.171) The phase of fk(0.x 0 ).<p) + Ol Akr l (11.172) h h k0 {rv0(tt0)} (11. Then with k = ^(ko + k . With Eqs.e.170) A ko (k) e .t o ) / V k ( x . (11. ip) could be carried along but we ignore it.166). (11. ^k(xxo) ^k(x) ~ = e.k0) ~ k0 + and the expansion of E^ in Eq.x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11. k = kez.xo.^ ( * .169) and (11.k 0 ) 2 ~ Jk20 + 2k0 • (k . (11.161) and (11. i.163) we obtain </>ko(x .i) dk (2TT (11.iB *(**°)/ R °'tj + 1 pikr e * . (11.e.10 we show that the solution V'k(x) of Eq. the stationary scattering wave.t) = J ^ A k o ( k ) e .ko) k0 (11. Eq.
8 Scattering of a Wave Packet so that with Eqs. Adding a suitably normalized solution of the free equation. (11. A particular solution of the Schrodinger equation is — J G(r.x 0 . '. Solution: We define the Green's function G ( r .z Fig. r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^.*0ko (x . (11.r ' ) . i. T h e scattering amplitude is t h e same as t h a t given by Eq.e. 11.^y /k0^. with the energy of the maximum momentum of t h e wave packet. r ' ) = 47r<5(r . (A + fc2)t/.r'). E x a m p l e 11. however.5.r')V(rW).r')U(r')ip{r')dr' /4T7 (as we saw earlier) .(«)/ ipko(x . and determine the Green's function G(r. Here ^ (x' — xo.167).174) r /k0(^.11. *. . U(T) = ^ V ( r ) .r')l/(r>fc(r').10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation.y)eiko'Xoe'[fcor£. t) f. t) .to). (11.171).(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r. the solution can be written iMr) : (2TT)3/2 ^dr'G(r. r ' ) by the equation (A + fc2)G(r.5 Scattering of a wave packet with scattering centre O.154).x 0 . 11.fco(tto)/ft]^0o)(x/xo.to) has t h e form of t h e wave packet in radial direction as indicated in Fig.^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11.
Thus we consider the contour integral taken around the contour C+. 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.O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist. .242 CHAPTER 11.fc ) oo dfc' 1 d nik r 7rr dr / . we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k). dk' ik'*r 2TT2 fc'2 .6 The contour C+ in the upper half of the k'p\ane.e \ r dr\ k ikr 1 ikr 0 With the choice of an analogous contour C— in the lower half plane. 11. Imk . 11. one obtains another Green's function given by G_(r)= r eikr. e > 0 small. r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r . which is such that the contribution along the infinite semicircle vanishes.fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j. These are the socalled outgoing and ingoing Green's functions. displaced slightly away from the real axis as indicated in Fig. « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' . Hence we write G(r)G+(r).6. The Coulomb Interaction We find a set of Green's functions G(r. 9{k') = 1 1 S(T) = j G(r) 1 ^ A j dk'e i k ' r .Rek' kfc/2K Fig./2 dk'd(—cos 6)dip iki.=k+ie/2k 1 d f.
0 + W2(l + 1 + iy.179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij.11. (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7. the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems. Before we introduce the general form of the socalled partial wave expansion. 21 + 2.9 Scattering Phase and Partial Waves 243 11. b. where Mo is a parameter in Chapter 16. (11.z2 27fc 1(1 + 1)' yi = 0.145)) F(a. (11. (11. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. z) = W^a. z).g. i. .178) (11.21 + 2.2l + 2. 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. i.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.2Z + 2. b. 2ikr)] [Wi{l + 1+11. b.140) (previously we solved this only for the bound state problem) § . Thus we proceed as above. z) + W2(a. Then £^<fc + (2l + 2 £ = ~2ikr. (11. e. under the stated conditions 2/07 = Z\Zie2 = —Mo. (11. Eq. If we proceed from the radial Schrodinger equation. whose continuation to infinity is again given by (cf. Then yW (11.9 Scattering Phase and Partial Waves We observed above that the Coulomb potential is associated with some special effects which one does not expect in general.176) " 0—A ~ (I + 1 + iy)<t>i = 0. Eq.£).175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi).e.e.
Then e2i5< = Reie Re~ J2i9 9 = Si.181) so that eldlT{l + 1 . (11.147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 .ry) = e"^r(Z + 1 + ry).146) and (11.244 CHAPTER 11.' .183) r(Z + 1 + 17) := Reie.i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 .ii)' (11. +ikr—iry In 2fcr—i7r(Z+l)/2 _i_ — iSi— ikr+i^ln2kr+iTr(l+l)/2 ! We set I^TiT^rsin(J:r"7ta*"^+4)'<1L182) 6 = arg T(Z + 1 + ij) (11.j syyll iy il( _ .n) • e „—ifer+i7ln2fcr _eMr(i7Jl)/2 r(z + I +17) T(2Z + 2) /2 • ikriy In 2fcri7r(J+l)/2 r(z + 1 . (2kryr{i" V 7 . The Coulomb Interaction Using Eqs. T(2l + 2) (2fcr)*T« (Alliy ikr .i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?. .180) ^ ' (11.«7) „ .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 .l e*"" + T(l + 1 + H ) T(l + 1 . Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5.
as we saw. The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element. i. 245 (11. (11. i.188) in the form y/E. considering I as a function of E extended into the complex plane.1.e._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity). i. a positive integer.e. For instance the poles of S are given by Z + l + z7 = . Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ .9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7). One defines as regular Coulomb wave Fi(j..185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function). where the continuum of the spectrum starts.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. . indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type.u\ respectively with the following asymptotic behaviour: u(±) r^o e±i(fcr_7ln2fcr. kr) the regular solution with the following asymptotic behaviour: r 0 (11. (11.187) contains almost the entire physical information with regard to the Coulomb potential..n ' . In the present quantum mechanical considerations I is. i. . However..e.184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase. considering I —• an(E).114a) reveals).2. (11.e. That the energy E appears in Eq. with Eq.139) n' = 0.
190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V. We mentioned above that the logarithmic phase does not arise in the case of shortrange potentials. The Coulomb Interaction and plotting these for different values of n.7 A Regge trajectory of the Coulomb potential. 11. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig. For the Coulomb potential a typical such trajectory is indicated in Fig.7. (11. (11. We shall return to these in Chapter 16 in the consideration of screened Coulomb potentials or Yukawa potentials and indicate other important aspects of these.191) .189) where as before the incoming wave in the direction of z is ikz ipin = e (11. 11." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections. and thus in general.246 CHAPTER 11. we obtain the partial wave expansion of the scattering amplitude as follows. 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. In these cases. Singh [252]. we obtain trajectories which are known as Regge trajectories.
W.192) Thus the incoming and outgoing spherical waves have a definite relative phase. 21 .k)=me2iS^k\ in which r\i^\ indicates absorption.i f c r .195) with the scattering matrix element S(l.196) "See e. pp.r ^ V ^ Wostf). When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg .i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—.~ J > Z + l)Lil*e~ikr . . (11. The mathematical expression can be found in Tables of Special Functions (there cf. (11.g.^2l + l^S^k) lik (11.22.193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave. r e. Oberhettinger [181]. The potential disturbs this phase relationship (by delay or absorption or redirection).^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY. Magnus and F. (11. 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.9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9.eikr\P^cos9). 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. 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 ' . 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„ ' .11.
This page is intentionally left blank .
g. namely the instanton (or more generally pseudoparticle) method. These newer methods will be presented in detail in later chapters. Similarly difficult is. M. one of the main and now wellknown methods to perform such calculations. in texts on quantum mechanics is rare.Chapter 12 Q u a n t u m Mechanical Tunneling 12.e. Razavy [236]. was developed only in the last two to three decades. For various general aspects and applications of tunneling we refer to other literature.* *See e. therefore. which in one form or another. 249 . In fact. In particular we consider cases which illustrate the occurrence of tunneling and that of resonances. as — for instance — for trigonometric or double well potentials. It is very instructive. can be found in any traditional text on quantum mechanics. It is not surprising.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. that the explicit derivation of such quantities in nontrivial contexts. that quantum mechanically a particle may have a small probability of being in a spatial domain. to explore first much simpler models in order to acquire an impression of the type of results to be expected. which a particle in classical mechanics would never be able to intrude.e. i. i. Such a vital quantum mechanical phenomenon is socalled tunneling. the computation of finite lifetimes of a quantum mechanical state. states of a finite lifetime. however. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts. in general.
2) we have = o + ~v[(vv>*)V'V>*(W)] = vj.250 CHAPTER 12.2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x.t).t) that the one particle concerned is somewhere in space at time t. i. According to Eq. t) was developed by Born in 1926.t)\* (i2 2)  With the help of the Schrodinger equation ihiP(x. so that p = T/>2 is the probability density for a spatial measurement at time t..1) as the probability described by ijj(x. .t) = HiP(x.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. r (12. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ.e.t)\2 =1 (12.4) (12. (12. Quantum Mechanical Tunneling 12. ihr(x. M^ Then dt + Vj 1=0.t) (12.
(12. If ^>(x) and V^>(x) are known at every point x.2a6(x)]il> = 0. It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous. pp.33. the problem is analogous to that of the scattering process from a potential in a onedimensional case. However. 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. the equation can be integrated at every point. ~hr> k = n^ (12 7)  . 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. with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. If there is no absorption. as we mentioned above. Tarrach [119]. Delta potentials in more than one dimension do not permit bound states and scattering.12. Gosdzinsky and R. This effect is known as "tunnelincp. 12. From this follows that both p and j have to be finite and continuous everywhere. higher dimensional delta potentials do not have properties like that in one dimension. 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. In much the same way that we define there reflection and transmission coefficients.3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x. 2m E 2 V0 > 0.2 . 'Maybe the reader dislikes being confronted again with the onedimensional case. Nonetheless. t For the singular delta function potential a modification of these demands is necessary. which is the case we consider here. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. where „ mpVo V(x) =+V0S(x). The Schrodinger equation for this case is* ip" + [A.3 The ShortRange Delta Potential It is very instructive to consider in detail a potential V(x) = VQ6(X) with the "shape" of a onedimensional delta function.6) . This case plays a special role. 1 0 ~x o a= f See L. 29 . we can do this also here. since this potential implies a discontinuity of the derivative at its singularity (as will be seen below). even for discontinuous potentials. as shown by P. Schiff [243]. although in general small. Thus it is possible that the probability for the particle to be in such a domain is nonzero. Such a region is for instance the domain where V is larger than the total energy E.
[<//]0+ .9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo. _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x . Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity).: (i2iQ) Ce for x > 0. i. a — ik £Qr < { eikx a eikx •t" . .e.e 1 + B = C.8) (12.ik(C .14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes.13) ~A R= for x > °(12.e.ikB) = 2aC or 2a(l + B).= 2a^(0). i.1) . (12.252 CHAPTER 12.2 a J 5{x)4>(x)dx = 0. 2 (12.e.9) for ij. e Q (12.11b) 2aC. (12.(ik .' : It follows that ik .ikC = i. The continuity conditions at x = 0 are therefore for V : and according to Eq. (12. except in domains of the potential V. that the derivative of i/> is discontinuous there: W\U + k2 J i>dx .11a) ikC . a 2 — a 2 a — ik a2 + k2 ' . (12.[^']o. i.12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—.12a) (12. Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see. 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.
Omnes and M.e «x = J \ CeKX for for x > 0. at k2 = —a2.15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential. 'Note that the physical pole has negative imaginary part. possesses simple poles at a ± ik = 0.K C . See e. z<0.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R.12. K = a. We write the solution ^ so that ^(0+) = ^ ( 0 . ( 1 2 . Newton [219].(K)C = . i. when normalized to 1. as coefficient of the outgoing wave. .e. poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. the quantity T. i.ip'(0) = .17) da#(s) 2 = 1. 2m 0 (12.i 8 ) .16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW. oo (12. and C follows from the normalization: OO A.2 a C . / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0.K 2 < 0. Proissart [223] or other books on scattering theory like R.2 = .g. = . R. (12.e. . Thus ^ = Ce.a l s l.2 o ^ ( 0 ) . or h2 E = —a2. f Ce~KX c.) and V>'(0+) .19) Thus the delta function potential supports exactly one bound state.
or in the lectures of E.22) + De Feikx + Ge~ikx *We follow here largely F. 12. 58. oo]. The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0.254 CHAPTER 12. p.a].4 Scattering from a Potential Well We can immediately transfer the above considerations in a similar way to scattering from a potential wen* as indicated in Fig.20) for for \x\ > a.21) The solution of the Schrodinger equation then consists of the following parts: Aeikx + Be tp{x) = { Ce~ iKX lkx iKX with with with xe[—oo. (12.1 The square well potential. —a]. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics. p. \x\ < a. Quantum Mechanical Tunneling 12.1. 55. Schwabl [246]. . 12. Fermi [92]. We want to consider the case of stationary states for E > 0. The latter contains also specific applications in nuclear physics. (12. x G [a. V(x) V=E a v=y Fig. We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12. x€[a.
(12.4 Scattering from a Potential Well 255 The connection relations obtained by equating ip(x) and its derivative from left and right at x = — a and a yield the equations Aeika + Beika = CeiKa + Df.f )e" t\ K\0—tKa—ika . (12.ikBeika = inCeiKa + DiKe~iKa.. Fikeika . (12.26) K^gifoa—i/ca fc.12.23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca.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 \ .e. (12.24) yields F ^ G. ) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12. I (12. fc Af(a) = .iKa ? ikAe~ika and . ).27) v ' Similarly Eq. .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.ikGe~ika.25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 . M(a) . (12. ( V C n vD ).
= < Ce~lKX + DeiKX . i.^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. we obtain A = F( cos 2na — .36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave.256 CHAPTER 12.32) [cos2«.i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq. (12. Quantum Mechanical Tunneling Hence with the complex quantities IK (12.oo].e.31) we have A AT(a) = )MFo B ^r)sin2na .e.4T(£)e ifci. .(1 + e 2 ) sin2 2na 1 .34) into Eq.e sin 2na IJlika e B = r)F sm2na. (12. (12. i.37) .35) with with with x G [oo.34) and as transmission coefficient (note e is complex!) \T(E)\2 = 1 .22) we obtain ' Aeikx + AT(E) I sin 2Kae~ikx T/>(X) (12.a . xe[a. the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12. (12. \R\2 = T(E)^sm2Ka (12. x € [a.—2ika [cos 2/ta + \e sin 2«a]e~ (12.33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.a].a ] .
for VQ < E < 0. cot(fca) = i— or tan(fta) = K K .39) 2 i.38) as expected.34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0. where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)]. i. We see in particular from Eq. (12. they have solutions.44) .e. and k real.12. nor when both parameters are purely imaginary. formula 406. (12. (12.31): 1\T(E)\2 257 1 .43) Thus we set k — —i ' H .e.42) These equations have no solution for K.\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12. p. (12. 2\k K. which is satisfied if either / = g or / = — g'1. an equation of the type f~1f = g19.12.40) we obtain cot(na) — tan(/«j) — i K IK .e. 83.41) i. B. However. (12. Li ( — H— I sin 2na = 0. in the domain of A imaginary and K real. (12. i. .4 Scattering from a Potential Well On the other hand according to Eq. (12.35) and using Eq. 2m0\E\ h? ' (12.e. Dwight [81]. (12.
t&n(2Ka)ER = 0. The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — .65.g.+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n . ± 2 . (12.42) which then leads to bound states with associated even and odd wave functions.c o . We now return to the scattering states. i. i. "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1.35)) for E > 0 sin2Ka = 0.20). Also. . = 2/rn (EER).48) We expand T{E) around these energy values.' We observe that T(E) has poles at those values of E which supply binding energies. The amplitude of the transmitted wave assumes its maximum value where (cf. and is itself infinite there.46) This condition implies the eigenvalues E^ER = n2^^ V0>0. „ . 2m0\E\ Ktan(«a) = — y——^— . (12. (12.e. 2Ka = nn. . (12.47) These states in the continuum are called resonances. around the values given by Eq. 2m0\E\ and Kcot(Ka) = y—^—L. (12. 64 . F. pp. The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave. ± l .l . Then according to Eq.e. n = 0 . At their positions cos(2/«z)£. This can be understood.50) See e. The remaining wave part becomes exponentially > decreasing. k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12. i. Eq.e.fl = ( . .45) These are the equations that one obtains from the connection relations for the case of Eq. Quantum Mechanical Tunneling . t2 2 (12.l ) n . (12. (12.49) Taylor expansion of the denominator in powers of (E — ER) yields . A = 0.34). from Eq.258 and obtain the equations CHAPTER 12. (12. (12. Schwabl [246].46). . T(E)e2ika = — r .
The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i. . = 0. T(E) has poles at E= ERi: (12.' ' 5a \V / Fig. We restrict ourselves to the period — la < x < 2a. This result is physically very plausible. A potential of this type is known as a PenneyKronig potential.^ l / .52) Here ER is the energy of the resonance and 2/F the lifetime of the state. 12. 2a<x<2a.51) i.5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig. a 3 a. V(x) = 9(x + a)VQ6(ax).' ''' 5a 3a \ a ~~~.12.e. 12. Since we have reflection as well as transmission there are states with repeated reflection and finally transmission. s \ V0 /• ~\ 'X . V(x) x=2a x= =2a .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. . 12. These are precisely the resonance states.2 The PenneyKronig potential.2.
and we choose (12. In view of the periodicity we have ^(x) = ^(x + 4a). (12. i.2 < a . i.1.. (12. x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall. This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r..a ) = 0 = ^(a).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. = EVip(x).. E^En = ^ ^ t n = 0.e There is the additional boundary condition (for Vo = oo) ^ ( .53) *w«w^( « j)'*^{i i ^ . the differential operator is even in x. with V as parity operator. .55) The Hamilton operator of the problem. . VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x). Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\.. HV^(x) = H. 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. Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo.2 = W h 2 _„2 = 2mo(E .260 CHAPTER 12. We set .Vo) n..e.2. and hence for = Eip(x).
— a]. This wave function is no longer restricted to the interior of a box. ^kAcos{k2a). (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. At x = ± a we now have to connect ip and ip' of neighbouring domains.59) Asink(x + 3a) with x & [~2a.63) = KBe~ F = . we shall not exploit here. We > obtain even and odd functions ip± by setting ^ = ±1. VV(°) = °. V+(0) = const. This means that i>±(x) = <  = ±1.61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka].. — a]. At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K. (12.5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a. (12.62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] .12. of course.2a].2). so that (W meaning Wronskian) W[tl>+. ±Asink(3a — x) with x€[a. BeKX±Be~KX with xe[a.. We expect. that these are such that for VQ — oo they approach those of the previous case. which. are ^_(0) = 0.58) D sin k(3a — x) with xE[a.Be Ka (12.a].a]. We therefore have different boundary conditions. xe[a. (12. = Ka ±Asin(k2a).60) The boundary conditions that ip± have to satisfy.2a.2a].il>]?0. (12. however. 12. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ . The central region of the solution is the classically forbidden domain. V'(O) = const.
the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—.nir) + OU2ak .even. It is plausible therefore to use appropriate expansions. and is — as we see — less than that belonging to the lower sign case. 12.nir)2}} = 1 ± 2e~2Ka.Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt. say A. (12.e.67) The value of k belonging to the upper sign.nir + 0[{2ak . i. (12.3 Level degeneracy removed by tunneling. In general one now rewrites the equations in matrix form. = AA. is the one belonging to the even solution. in the present case it is easier to continue with the above equations. the odd solution with.— (1 ± 2e~2A + O (—).2 a . Quantum Mechanical Tunneling Fig.+ • • • 1! cos^(n7r) 2ak . F ^odd (12.64) we obtain ~[(2ak k .mr)2}. (with reinsertion of nir/2a for k) 2ak ~ nir .68) .4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) . k = fc0dd. we set tanhx = l . However.e.. For large values of Vb. i.262 CHAPTER 12.2 e .e. + 0 ( e .e. for VQ — oo. (12.65) and we expand tan(2afc) about nir. Taking the ratio of the equations in both cases. where Akoce'2™. 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. i. (12. » we expect the eigenvalues to approach asymptotically those of the case Vo = oo. say. i.66) Inserting these expansions into Eq.
v = 0. (12. ^v + ^[E.^ 0 = °' Hence in dominant order •00 — exp . 57.12. Compute this probability. '.V{x)\i.70) xo = (a + b) 2' . . w(x0) •• = = mogy(x). but quantum mechanically there is a finite probability for this to succeed. b > a. Example 12. Solution: The wave functions i/>o>Vv in regions V = 0.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E.e. exp r rb rb / •J a 2mo (V(x) . y(x) = sin and y(xo) = 1. as indicated in Fig. 12. 0. The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation.3. V ^ 0 are given by the equations ^o + " J .2 dxy/V(x) (12. Classically the car is unable to overcome the bump. 12. . a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig. 12. 1pV 2ro 0 4>o '.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. and is a reformulated version of a problem set in lectures of Fermi. 2mn.4.** " W ~^7~ Fig. p. .5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states. . Fermi [92]. 2m0E „ 2mnE ..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.4 A car meeting a bump in the road.. exp . i.
l .72) Assuming a mass of 1 ton of the car (i. : 2 ^ B ( ^ .2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x.384.055 X 1 0 _ 3 4 J s .17)). We can now evaluate the probability.e. and assuming a length of b — a = 100 m of the base of the bump.1. Ryzhik [122].4 x 10 3 9 ]. Emde [143]. p.**. 14. which give /•TT/2 dx sin'M . dx sin ' x •• Looking up the value of the factorial in Tables.621.807 m s . Jahnke and F.2 .1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0. M. 950. . p. 1000 kg) and using the following values of the natural constants: g = 9. p. n I .1.y) is the beta function (see Eq. (15. 369 and 8.9191 ~ 1. formulas 3. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12. Gradshteyn and I. S.71) The integral can be looked up in Tables of Integrals. Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w.2.264 we have ro / dxyfV(x) Ja CHAPTER 12.1 / 2 ) ! = v^r) TT/2 /. we obtain ( . **For instance E. one evaluates an approximate probability of exp[6. (12. h = 1. ^ ) = 2 ^ . We find™ (with ( .
Example 14. 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. in Chapter 15. This is the problem of Galilei in quantum mechanics.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. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves. Thus. 144. we consider the linear potential in three space dimensions that forbids the particle to escape. 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. This is the quantum mechanical version of the problem of Galilei in one space dimension. each of which propagates with the classical momentum and energy of a Galileian particle.* In Chapter 14. 265 .1 Introductory Remarks We distinguish here between three different types of linear potentials.3. p.Chapter 13 Linear Potentials 13. the last case allows only a discrete spectrum. we consider the corresponding case of a particle above the flat surface of the Earth. 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. Then considering the probability distribution determined by the squared modulus of the wave function. * A highly abridged treatment of this problem is given. Siissmann [266]. 13. for instance. whereas the first case permits only a continuous spectrum. Finally. in the book of G.
(13. 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. i.1). c> 0.266 CHAPTER 13.= l V 27T J_oo 1 dkeikxijj{k. x > 0. (13.1) where g is the acceleration due to gravity and we can put Vo = 0. and yields rriQX = mog.t) = m0g with potential V(x. We see " that the case g < 0 can be related to the reflection x — —x. g > 0.t).4b) .= l V 27T Joo dxelkxip(x.2) which is the case with a discrete spectrum. Recall the classical treatment of the freely falling particle. (13. The equation of motion follows for instance from the derivative of the constant energy.3) In general the momentum representation is of little importance.t) = c\x\ + V0.t).t) = .e. In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x.t) = . (13.t) = mogx + V0. Therefore we use the Fourier transforms il){x. The present case is an exception and a good example of its applicability. A different > "linear potential' is the socalled "confinement potentiaF V(x.4a) f°° ${k. V(x) = —mogx. We consider here the case of Eq. (13.t). from d_ m x2 0 dx + V(x) 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. 13.2.
(13. multiplication by a test function and subsequent integration). • fd ~(' m0gt %h . Applying the total derivative d/dt to this function.7) In the wave function on the left hand side of Eq. (13.t).e.6) the argument k is then replaced by this. d \ ~f1 dkr[ k + m0gt ^r't (13.t) h2k2 d ~J(k. (13.2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k. The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i.t).4a) d ~ f I dk ikx dk 2vr (im0g) iP(k. To this end we consider the following substitution in Eq.< (13.13. 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. We rewrite this equation in the following form th (13.6).t) the expression (13.6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13.9) .t) dketkxiP(k.e.3) yields us now by inserting for ip(x.7) applied to every quantity in Eq. i.6) the latter equation becomes d ~( m0gt ^(^M<=+=?. (13.d m +imo9 .8) With this result and the substitution (13.5) m+tm9Wcr{ktt)=2^^ktt)t (13. (13.t) dk nikx k=oo '2K (im0g)il)(k. Equation (13.t) m0gx /2TT m0gxip(x. we obtain the differential operator on the left of Eq.t)im0g^k.6) and wish to solve it.
10) k by m0gt back : k —> k h The solution of Eq. Thus we have first (the second term arising from the upper integration limit.11) *[k?f.6) is then $(k.6).t In Example 13. (13.i) . We have thus obtained the solution of the transformed equation. (13.1: Verification of momentum representation solution Verify by direct differentiation that the solution (13. we have to reverse the substitutions and replace in (13.6). Our next step is to reexpress the solution (13. (13. E x a m p l e 13.268 CHAPTER 13. On the other hand the second operator expression implies dtj>o imog^—ip(k.t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13.= k rmt.e. Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential. (13.gh [ d t ' j f c + ^ ( t ' .11) in a neater form.t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc.t ) ' U ( f c .11). the third term from the integrand) > / mog \ o exp dk h J h dt y ' 2mo k2i>(k. Thus addition of both indeed leaves the expression on the right hand side of Eq.1 we verify t h a t this solution indeed solves Eq. To obtain the solution of the original equation (13. We define the wave number kt as kt.t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13.6). t ) . the following result h 2moi tp[k\ ^—. Solution: We apply the two operator expressions on the left of Eq.10) Note the amplitude function ipo(k) in front.11) solves the partial differential equation (13.6) to the solution (13.12) .t). (13.6). We observe t h a t t h e solution is actually a function of kt — k — mogt/h. i.
t) = . we obtain rp(x.13.11) can be written j'A ^.2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13.13) rn0_t+2 + mlg2t3 . m0g (13. we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * . _ k t  h2 3 Inserting here the explicit expression for fct.gfci + 2 g2m0 2/i 2 ' (13.16) Next we evaluate the Fourier transform of ip(k.19a) .17) = J_/ C X ) h2 [It(k) . we have ip(x. m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13.4a).t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With . Inserting ip(k. 27T ^ . .14) (13.o o 2 —  exp iktx h mZgH' .o ut = 2mo —kf 2mo A. .18) h =k mogt p h S' h ..} fA ' _ _tkt 2 k+ t) 2 k . (13.I0(k)].15) we obtain / = .t) into this integral and recalling the property (13.t) defined by Eq. (13./„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential.11).= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13.2„2+2 mfaH 3^3+3 .+^f (13.— [It(k) .
2.gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 i + ft2 fcr 2mf)g 03mg £3 .v = kh/mo.t)\2 for t — oo (distant future or past).19b) h3 3 we can rewrite ip(x.18) into a • different form in order to be able to perform the integration with respect to k. 13. These wave vectors vary in accordance with Eq. To this end we convert Eq.o o Jo This result can be interpreted as follows. the relation (13.23) .20) expresses the wave function tp(x. 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. Such a decomposition is possible only for homogeneous force fields. (13. (13.21) In other words.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 .t) =y= dkip0(kt)) exp i< fctx V27T J . The fcdependent terms in the argument of the exponential are T:=ikX^{^fk2+r^k 2rriQig [ n nr (13.19a) exactly as for a falling particle. Linear Potentials f dt'.t) as 1 I00 ip(x.The associated kinetic energy of the particle is mv+ = huJf. 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.ut> = Jo t2 g2m0 i 3 k t .t) as a decomposition into de Broglie waves.2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x. 7 u>t'dt' (13.20) (13.270 and CHAPTER 13. fm0gt\ \ 2h ) (13.
t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13. 2h h 24ift ' mogt^ r m. Then ip(x. /m0gt\ \ 2ft J ° ) I m09H3 ft J 6ih m0g2t3 _ m0x2" 2Aih 2hti m gt IT / dk^oik t) exp oo ht k2mo« ~fti 2ft~] (13. It follows that ip(x. We evaluate the Fresnel integral in Eq. fi > 0.18) thus becomes ip(x.13. rnogt m0g2t3 tX^r~. (13.t) = 1 1 oo 271 27T /.29) . integrate and then take the limit fi —> 0). m0xcl ht m0gt 2h gt O v: °° 2hti (13./oo dkip0(kt) exp l x ixI exp oo . (13.t) 1 exp r .25) The amplitude ipoi^t) then becomes for i — oo: > 4>o(h) \t\—»oo 4>o[ k \ . being now independent of £.QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i.26) with the formula 1/2 (13.27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics. ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f. may be put in front of the integral.QX 2 ^ 0 7 (m.e.24) We now set fit / 2m 0 V" 0 V m0a. (13.26) d£e ie £ kt^ m0' 0. TOO#A ft and.28) (for a verification replace a by a + ifi./m 0 a.2 The Freely Falling Particle: Quantization The integral (13.
1) has only a continuous spectrum.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 . kt and ut 13. Linear Potentials Replacing here x by the large time value x c i.t) ~ W —exp i V irlt m ktxd J cot. 13. These are incoming and outgoing waves.dA 4>o(h (13.3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0.31) and obtain from Eq.1 A turning point at x = —XQ. For x exponentially increasing and decreasing solutions are possible.t) = eiEtlhct>{x) (13.30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i. (13. (13.3 Stationary States In this section we show that the Schrodinger equation with the potential (13. this expression can be brought into the following form: if *\ 0 ip{x. V(x)=m 0 gx V=E Fig. xc\). . 1/2  ±^(2mlg/h^x^ / 2i (13. For stationary states we set xl){x.272 CHAPTER 13.e.33) oo for E <C \mogx\.
at which the classical particle bounces back. i. H2k3 4>{k) — cexp Ek (13. so that ih dtp{k. i.3 The TimeIndependent Schrodinger Equation 273 Classically the particle coming in from the right with energy E cannot penetrate into the potential barrier since (where V > E) with conservation of energy E — V = p2/2m.1. and we can establish for the resulting continuum solutions the orthonormality and completeness relations. is reflected.4b).36) m0g \ 6m0 with c = const. probability.35) It follows that /W tm0g \ 2m0 i. cf.e.4a) and (13. The Fourier transform is given by the relations (13. which is not possible for real particles. exponentially decreasing.t).6) that also for 4>{k. at which E — V = 0. t). 1 f01 dketlcx^{k.00].13. E)dk. Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00. (13. its momentum p would be imaginary there.t) = dt (13. we .34a) Then d &k 1 f°° ih—il>(x.e. The point — XQ. However quantum mechanically this is possible with a small. but also d 00 dt Eijj(k.e.e. 13. J \ „ (13. t) = — = / V27T J . In view of the simple form of the potential in the present case the equation can be integrated in the momentum or k representation. i. i.34b) It therefore follows from Eq.0 0 dkelkxE^(k. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function. Fig. Akx (13.t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m.e.t). is therefore called a turning point Quantum mechanically this is a point at which periodic solutions of the Schrodinger equation go over into exponential solutions or vice versa. and hence is continuous and extends from —00 to +00. i.e. t) = Eil)[x.
38).L dxeikx(f>E(x).38) we have for <j)E{k): OO /"OO / oo (13. </>E(x) = .(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13.L [ dkeikxMk). 13.e. / dE(j)E(x)(t>E(x') = ^ s .e.43) y=2^x+2^E. (1340) i.41) J—oo Next we insert (13. (13.39) dk4>*E{k)4>E. h? h2 . / <>~ikx VZ7T J (13. Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf). apart from a constant phase which we choose to be zero.36) (with c replaced by (13. V 2lT J Mk) = .37) (eikxdk. the relations oo /*oo dE4>E(k)4>E(k') = 6(kk').40)) into (13.39) we obtain dkcc*exp m0g It follows t h a t cc = &(# .a : ' ) . i.£') = <*(#£')• 1 27rm0g' = /o .274 have / Using d(x) = — one verifies t h a t with CHAPTER 13. (13. so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s).36) into (13.. (13..M *K Joo Setting dt cos ( ts n Jo st). In a similar way we can verify the validity of the completeness relation.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 .
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.45) One can now derive normalization and completeness integrals for the Airy functions. the function 4>E{X) i and setting s y = — = 1 . For a further solution Bi of Airy functions we refer in Chapter 14 to Tables. whereas the other is of exponential type.44) is then expressed in terms of the Airy function: . and in Chapter 15 in the computation of energy eigenvalues for the threedimensional linear potential. In the next section we derive the important asymptotic expansions of Ai(s) for both positive and negative values of s.J * .13. and we shall see t h a t one branch has periodic behaviour. so t h a t the integral becomes 4>E{X) = 2vr ^/mQg / n oo dtexp oo h? * 2m£g y \ 7 5 * " 3 M' Choosing /2mfo\1/3 A= ^ . . (13. (13. For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z).^ . / s M .46) This follows since with Eq.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./ 2m 0 E (13..^x + ^ E (13.
48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 . (13. (13. i. (13. Linear Potentials 13.51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0. o (13. (13. .r 3 sin 30 ). 13.. .\zz. Thus we must have sin30<O.49) (13.f dzJ*W 2TT JC with z = reie.276 CHAPTER 13. i.e. We want the integral to exist and so desire that lim e**^ = 0. (13.r 3 cos 30 J + i ( sr sin 9 .51b) .TT < 30 < 0.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig.4 The Saddle Point or Stationary Phase Method The integrand of Eq. or n <6 <TT o for r > oo.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i .43) contains the phase (with x — z): > *(*) := sz .e..TT < 6 < 0 for r > oo.2 The contour C for s < 0 with ends in the angular domains (a) and (b).
Inserting (13. Such a piece can be found in the neighbourhood of the socalled saddle point zs.55) into (13. (13. 13. oo] as indicated in Fig.e.e.54) i ¥~s)3/2' (13.55) s < 0. Then Ze TeC : V^se±i7r/2 = ±i\fzrs. provided its ends at infinity satisfy these conditions.57) $"(z s ) = 2% s.4 The Method of Stationary Phase 277 The integral (13. and the main contribution to the integral comes from the a region around the saddle point. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral.z3s and V^s + .48). we choose zs = iy/^s. Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(.S )V2 (13.45) exists for the contour C. (13.2.58) .54) where a is the new variable with —oo < a < oo. —i ${zs) = szs .52) = exp(±i7r) Let s be real and s < 0. Then 6S = ±7r/2. so that exp(±2i9s) — 1 and — s — rl.51a) and (13.53) Since we shall have the contour C in the lower half plane. 0 = &(z8) = s~zt s = za „2„2i6>.51b). (13. at which the phase becomes stationary.56) 2n(s)^Jc But I daefr"?*"^. i. (13. where the derivative of the phase function vanishes. i.13. and we set with a real: z := zs + a (S)V4' (13. We now choose the path of integration C at fixed O z = —iy/^s with a e f [—oo.( \ / ^ ) 3 2z. (13. we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13. Then the ends of the contour lie in the domains specified by Eqs.
2 we show that for s — oo: > Ai(s) 1 ..44) we can convert Eq.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13.45) 4>E(X) ~ Ai(z). (13. According to Eqs.32) therefore becomes dz2 + zd) = 0._ 1 .44) and (13. Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here. M.63) The Airy function can be reexpressed in terms of cylinder or Bessel functions Zv(z). as one can see from the following equation^ ri' + Ffz •2^2. S. This is an important aspect in the WKB method to be considered in Chapter 14. but would increase exponentially for imaginary values of a — thus describing a surface around that point very analogous to that of a saddle. Sec. 2 s 3/2 . Gradshteyn and I. (13. 8.56) decreases exponentially around a — 0 for real values of a.60) The reason for describing the point zs as a saddle point is that the integral in Eq. V^s / 4 V3 (13.42) into an equation with a more appealing form. (13.62) This is the Airy differential equation with one solution 4>{z) = Ai(z). For (—s) — oo the integral converges towards T/TC.278 CHAPTER 13. Ryzhik [122]. where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}. With the help of the substitutions (13. > so that Ai(s) 2 v ^ ( .491.2/32u = 0 (13.32) with the solution (13.59) In Example 13.64) tSee I. (13. cos . (13. .s ) V 4 exp \(s) 3/2 (13. V 2h2 (13. We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction.
* " ( z ± ) = =F2Vi.. with *"(z) = 2z. we obtain f° Joo dp=—[a+ i\a\]da.59) and (13.13. Eq.z ± ) 2 * " ( z ± ) . Hence the saddle points are given by 0 = * ' ( z s ) = s — z2. Thus there are two symmetrically placed real saddle points on the real axis. Then._ . Choosing the contour C to lie along the real axis. P Introducing a new variable a by setting p = ai\a\. 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 .51a) and (13.^ e ^ / V2 4 [°° Jo dae'2^.s 3 / 2  .43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s). 27T Ic Jc 3 This time s is real and positive. Example 13. the conditions (13. (13. changing to the variable p. kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = .4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (1365) With expansions (13.2: 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. . It follows that for s —> oo: Ai(s) ~ . cos  .51b) are satisfied.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.
.e. i. 13. z o = n . Thus here / ( z ) = — z + n l n z . Jo Jo \ . and hence integrating as described above through the saddle point indicated in Fig.3 The saddle point at z = n + ix. since f'(zo) = 0 and (z — ZQ)2 = —x2.280 CHAPTER 13. (11. one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en. n!= / e~zzndz= / e~z+nlnzdz. / " ( z o ) = —n/z2 = —1/n. ~ 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.111).dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point). Then. ^/2KJW1.3: 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. the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix.3. Linear Potentials E x a m p l e 13. 13.Be 2 Fig.*. Solution: Briefly.
1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method. It goes without saying that such comparisons are very instructive. H. 292 . Froman [99].§ 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. Kramers and Brillouin. Jeffreys [144]. B. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i. named after its first promulgators in quantum mechanics: Wentzel.g. N.* Descriptions of the method can be found in most books on quantum mechanics and in some monographs.317. i. Wentzel [282]. A. Froman and P.e. Kramers [153] and L. and the role played by h in our considerations here. where *G. the linkage of solutions above a turning point to those below. been developed previously by others in different contexts. Brillouin [37]. will there be that of a coupling parameter. ' T h e most prominent earlier expounder is H. ^ The central problem of the method is the topic of "connection formulas". 281 . A critical overview of the history of the method.e. Dingle [70]. 316 . however.295. O.* The method had.Chapter 14 Classical Limit and W K B Method 14. *R. 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. § See e. for misunderstandings pp. The WKB result is therefore frequently described as the semiclassical approximation. of misunderstandings and other aspects may be found in the monograph of Dingle. For the history see pp.
(14.282 CHAPTER 14. so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14.t)\2 = \iP(r)\2. 14.g. It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition.t) = eiEt/h7P(r) and \V(r. for which the observation of position and momentum is independent of the time at which the observation takes place.7.* We illustrate the use of this condition by application to several examples.V(r)]^(r) = 0.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh. Classical Limit and WKB Method E — V changes sign). We start from the Schrodinger equation ih <9* =HV at with H=£— p2 2mo + V(T). • {VA + %AVS [ h B iS(r)/ft +^{vSVA+^A(VS)2 *See e. In this equation we set. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles.3) .2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0. the function S being called phase function. For stationary states ^/.1) (14. the energy has a definite value E with V(r. A(r) and S(r) being real functions of r. It is for these states that one obtains the timeindependent Schrodinger equation A^(r) + ^ [ E . V>(r) = A(r)eiSWn. the derivation in Example 18.
(14. (14.13) .4) (14. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m.t)}2. V(r. (14.14. (14.t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r.t)eiS{r't)/h.V(Y) + and 2VA • VS + AAS = 0.5) Apart from these we also wish to consider the equations that follow in a similar way from the timedependent Schrodinger equation. We can identify Eq.11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *.* VAVS+AAS + V(r)A.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq.6) 2mo dt now with A also a function of t. These equations are equivalent to the Schrodinger equation. h h (14.Q A 2i 8A 1 (14. h2 ih— = H$ A + V(r) * . (14.10) m 0 ^ . We set in this equation.10) with a continuity equation by defining as probability density p:=*** = [A(r.1) and separating real and imaginary parts.7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14. (14.12) VA A (14.+ VAVS + AAS = 0. we obtain the following two equations: E .
m0 But now (14.9) for the phase function S.20) With this we can rewrite Eq. %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq.21) .9) as 9S 1 mo^ 2 + T.+ AVA • VS + A2AS = 0.19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS.14) A2VS. + h2 AA V = — ^ . the implications of this equation are also those of the equation of continuity.17) With Eqs.18) ot ot This is the equation of continuity that we encountered already earlier.10). (14. (14. Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS.n.16) BA 1 moA—.15) and (14.01. Next we investigate the significance of Eq. (14. ie. (14. mo <™> (14.284 CHAPTER 14. Since we obtained this equation from Eq. we have mo—(A2) + V(A2)VS + A2AS = 0. (14.10) we obtain = — [V(A2) • VS + A2AS}.16) this becomes m 0 1 rr + m o V . TTIQ (14. Since J = — A2VS = PVS. . ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real. ^ + V • J = 0. (14..J = 0. (14. (14. (14.
V ) v + V 7 = 0. i. The particles have no interaction with each other. This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2. (14. we obtain ^ V 5 +  v ( v v ) + V V = 0. v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. d — = (14.23) (14. This means.20).e.1/h.2) for the probability amplitude. V • J = 0. Such series are typical for expansions in the neighbourhood of an essential singularity.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. and hence''' ^(m 0 v) + m 0 (v. where.14. This result is very general. One should note that in no way do we simply obtain the classical equation of motion from the Schrodinger equation in the limit h2 — 0.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ. which for very small values of h yield the dominant contributions. that the quantum theory implies an essential singularity at H = 0 (i. they are asymptotic expansions. r o t v = ( l / m o ) r o t g r a d S = 0.24) d 1 dr d •— = d Lv _ . However. with the help of Eq. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). (14. dt 2 Constructing the gradient of this equation. > The motion of the particle or rather its probability is altogether given by the above equation of continuity. > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents. We deduced this equation solely from the phase S in the limit h — 0. We see therefore that in the limit h — 0 the particle behaves > like one moving according to Newton's law in the force field of the potential V.22) (14. This expression with h in the denominator shows. so that V ( v • v) = 2(v • V ) v + 2v x rotv. a singularity different from that of a pole). the wave function \I/ describes effectively a fluid of particles of mass mo. 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv. In the first • place we require the ansatz (14.e. . The equation of continuity then describes the stationary flow of a fluid.
28) The calculations we performed at the beginning imply the following equations (compare with Eqs. Sec. 318. similar to our earlier procedure with A and S real functions. or h2/2mo(E — V). is an asymptotic expansion. There are also other more specific descriptions. The expansion in powers of h2.3 14.286 CHAPTER 14.3. We consider the onedimensional timeindependent Schrodinger equation y" + ^.e.2. . z (14. p.[EV(x)}y = 0. such as "LiouvilleGreen method' and "phase integral method'. §We follow here partially A. h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V). the method is widely familar under the abbreviations of the names of Wentzel.$ 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. Messiah [195].26) We observe already here that an expansion in rising powers of k. We set. (14. This means that the WKB method can make sense not only where one is interested in the classical limit but also for E — V(x) and hence E large. Kramers and Brillouin.27) (14. i. B.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.30) ax *For details see R. (14. Dingle [70]. As remarked at the beginning. (14.4) and (14. Classical Limit and WKB Method 14. i = eiW(x^h.§ y = A(x)eiS^'h where W(x) = S(x) + . 6.5)) and 2 ^ f dx ax + All=0.In A(x).
. (14.10)).V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• .32) is therefore (S'0)2 + 2H2S'QS[ . (14.35) can be integrated.31) \a. Li This expression can be substituted into Eq.34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E .36) Equation (14.32) s" = sz + h2s'( + •••.3 The WKB Method 287 Equation (14. (14.4 (S>)2 1 S'" 2 S> (14.35) . + (14.30) corresponds to the equation of continuity (i.29) becomes (note that this is still exact!) (S')2 . r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq.33) 3 (S")2 .A = const. 3 / 5 .\ In S'.e. Eq.2rn0(E V) = h2 lS'A 2S'0 (14.14. Since the expression (14.2mQ(E .28). (14. Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 . (14.V) and (14. But first we note that: A dA dx2 2 = c{S')ll\ = A 15^ 2 S' + ^ = A ax 3 (S") 4 (S')2 2 2 S' With this Eq.
39) where a and /3 are constants. (14. y/2mo(V(x) .33) to make sense as an asymptotic series.± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2). and instead is asymptotic which we shall not establish in detail here.and So = ±h / — — A J k(x (14. (14.e. y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13). i. For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====.27) we obtain now with Eq. 1 S e e R.40) rx dx v "7 W)\. .E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14. it is reasonable to set (cf.41) We now ask: What are the conditions of validity of Eqs. (1.I In S' + In c h 2 2 2 exp ^{S0 + h Sx) . we expect that in any case So » ft2Si. E > V(x).39) and (14. Dingle [70]. B.38) h y/2mo(EV)' (14. p = h/X) k{x) :Then S'n +.' For the expansion (14. The solution is seen to be periodic provided k(x) is real. (14. (14.31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S . (14.288 CHAPTER 14. Chapter XIII.41)? We note first that the expansion in powers of h2 does not converge.19).37) For Eq. Eq. Classical Limit and WKB Method has the form of a momentum.
14. E > V(x).45a) . (14.42) [(^). Eq.V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14.3 The WKB Method We have 289 S0 = ±h and Si follows from (14. Since k(xy 1/52\S'0 (14.3/2 s&T + 3 {So Il« we have (cf.36). Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A.1 / 2 ]" [§0SS). we obtain 1/2 1/2 2S[ and hence ±2hS[ = [kL/2]"kL'2 = lk~1/2X = Al/2 [lAl/2A//_lA3/2(A/)2lAl/2 so that It follows that (14.43) The condition So 3> ^2<Si therefore implies that /f>^i/^)k = H/y/2m0(E is sufficiently small.
45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l . (14.2 Turning p o i n t s a n d m a t c h i n g of W K B solutions Let x = a be the value of x at which E = V(a). Roughly speaking one therefore requires E to be large in both cases. h = 1) . In the neighbourhood of E = V(x). around the classical turning point. In a similar way we have for E < V(x): l'(x) <C 1. In terms of X the original Schrodinger equation is y" + = y = 0. i.3.44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh. at which E — V{x) changes sign? We note first that in this case near x = a: E . (14.e. 14.V(x) = V(a) .48) The (dominant) WKB approximations are exact solutions of this equation.290 CHAPTER 14. the WKB approximation is in general invalid (see below). Classical Limit and WKB Method we can rewrite the inequality (14. We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o. V(x)) (14. *2 A(z) = V / 2m 0 (£ . How do the solutions of Eq. and hence (with UIQ = 1/2.V{x) ~ Or .a)V'{a).48) behave in the neighbourhood of such a point called "turning point'.46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14. (14.
3.e.14.1 The regions around the turning point at x = a.1. since this requires different approximations there.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a. As indicated in Fig.3 after stating them first here.1 there are WKB solutions on either side of the turning point and some distance away from it.a ) 9 / 4 ' 291 (v^r 1 (x — a>2 These relations imply that Eq. 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. On one side.1. 14.48). The transition is provided by matching relations.3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x. We shall not derive these conditions here in detail but rather make them plausible in Sec. (14. 14. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for . in the domain E > V the WKB solutions are oscillatory. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq. and on the other side. in the domain E < V the WKB solutions are exponentially increasing or decreasing. 14. which connect one domain with the other. i. (14. We assume we have a turning point as indicated in Fig. 14. V=E Fig.
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 / —.1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14. has the same asymptotic behaviour in the domain i « a a s Ay\. This means the asymptotic part suffices only in t h e exponentially decreasing case.H— J =^ v/exp + m (14. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a . (14. The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right. 14. (14.292 CHAPTER 14. 14. Note also the extra factor 2 in yi. (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0. in order to make the formulas independent of whether the potential is rising or falling. however with the limits of integration x.49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a .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.50) x S> a. a interchanged. as in Fig. and E ^ V for a ^ x the same formulas apply with the same direction of the arrow. P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi. Thus.52) For a potential rising from left to right. one could simply replace everywhere /•a / pa • • • b y / Jx J x .
2 1 X (14. (14.2 The overlap regions around the turning point at x = o.3. If higher order contributions are to be taken into account.3 The WKB Method 293 In that case Eq. y" + Toy = o.3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq.54) . 14. the equation V approximately linear around turning point V=E Airy solution WKB .46). the equation becomes approximately y" = (xa)xiy. is singular at a turning point x = a. as we saw.Airy overlap* Fig.48). i.51) — always with the decreasing exponential on the left — is valid in both directions. (14.14. Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14. (14. the matching relations have to be altered accordingly.e. 1 ~ (x . Therefore we consider now this original equation in the domain around x = a. 14. whose exact solutions are the dominant WKB approximations.a).53) Since at x ~ a. One should remember that the matching relations above are those for the dominant terms in the WKB expansion.
^ 4 I 7T x — a <C 0. (13. as > may be seen by referring back to Eqs. — a ^> 0. (14. In the following we require the asymptotic behaviour of the solutions Ai(—z). i I 772 \3 + > for ia>0. Abramowitz and I.57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0.59).a)3/2 ^ /2 =^ 3/2 o Q. . With the substitution z = (x .57a) and (14. > for a. A. B.Bi(—z) for \z\ — oo.60). (14. (13. formulas 10. This is given by the following expressions. Dingle [70]. 448. Stegun [1].4.55) the equation becomes the Airy differential equation d2y dz2 zy.57b) UV2 3 11 by J dx . (14. p.56) We have discussed some aspects of solutions of this equation in Chapter 13. (14. or to the literature:H 3/2 2^{zy/* Ai(z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e K*) n for ~. p.5960. R. J (X (i4 58)  Replacing in Eqs.a)X\/3. or M.57b) To the same degree of approximation as the equation y" = (x — a)xiy. In particular we encountered a solution written Ai. (14. Cf.291.1 / 4 by \^.294 CHAPTER 14. A second solution is written Bi. Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant. we have r x * f x ' /2(x a)i/2d*=^(x .
3 The anharmonic potential well. As we approach a limit of this interval in going away from the turning point. At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions. F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig. In this sense we have now verified these relations.1: Quartic oscillator and quantization condition With the W K B matching relations derive for the case of the quartic oscillator with potential V{x) = x 2 + x4 the BohrSommerfeldWilson quantization condition given by rb dx 1 fb i / — = . As an illustration of the use of the WKB solutions we apply these in Example 14.l.. We summarize what we have achieved. however. These solutions. we enter the domain of validity of a WKB solution.2. by Ai(z).g.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution.59a) . In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation). (14. n = 0. and hence must be proportional (in view of the uniqueness of the solution there). are only valid in a small interval around the turning point at x = a as indicated in Fig. Example 14.51) and (14. In this way different asymptotic branches of one and the same solution are matched across a turning point.2. 14.(zf2 by J we obtain the matching relationsX (14. e.l v/2m0(EV(x))dx 7r = (2n + l). 14. 14.3 The WKB Method and 295 •.14.1 to the quartic oscillator. In the direction to the left as indicated in Fig.52). The latter domain is a small region where the solutions from either direction overlap.
14. .51).l. j/i. (14. In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a.e. 1 .(x) = c ' v Acos I / — I for a < x < b. 2 . Let E = V(x) at x = a and x = b. for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the BohrSommerfeldWilson rule. (14. /n = s u ^ TJcos^ya TJ+cos^a T J . Then around these points the potential would be well approximated by a linear potential (as we saw above).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 .296 CHAPTER 14. The condition for this is ya{x)=yb{x) for xe(a.2. i. .^ . b are the two turning points. Classical Limit and WKB Method where x = a. yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b. . ^ A = (2n + l ) .44) this implies large values of n.60) Using the matching relation (14. (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. n = 0 . The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained. whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum.61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique). regions 1 and 3 in Fig. I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ . the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J. We argued earlier that the WKB method is suitable in the case of large values of E. 14. .3. . The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin. Hence we search for solutions which are exponentially decreasing in the far regions.. Consequently only the lowest eigenvalues would be reasonably well approximated by those of the harmonic oscillator. Here we are interested in the discrete states (the only ones here). In order to see this we consider r dx TT\ _ Jx T " 4 J ~ COS / r" dx_ _ Via ^""1 ^~4J" COS Ua I + 4 ~ i ( .. One can also express this by saying that the linear approximation of the potential around the turning points must extend over a length of several wavelengths and is therefore valid for large values of n. as in the case of the harmonic oscillator itself. According to Eq.b). extends only over a length of very few wavelengths and is therefore valid for n small.59b) Solution: We consider an anharmonic potential well as depicted in Fig. (14. Sometimes this distinction becomes imprecise.3. (14.y o dx TT T provided 6 _ 4 /. 2 n = 0. In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator.
.4 BohrSommerfeldWilson Condition 297 14. Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0.e.1. pdq = n + 1 — .x number of turning points h. .1. 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. with n — 0.x number of turning points TTH. and knows of no published form or script. . The corrected form* is. (14. This relation is sometimes wrong. Wilson by supplementing classical mechanics by Planck's discretization. In this socalled "old quantum theortf (i.63a) or. (c) V Fig. equivalently. n = 0. 2.4 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.4 Three cases with different pairs of zeros.64) (14. . Dingle. ri2 I Jqi :pdq n + 1 — . A. Sommerfeld and W. B.63b) *The author learned this in lectures (1956) of R.14. 2 . Bohr. . 14. (14.
where n —. We have the case of trigonometric solutions. 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. this is an exponential zero. the wave function is to vanish at points q = q\ and qi with the function /(g) remaining negative in between as illustrated in Fig. (14.4(b).e. where /(g) is negative. In quantum mechanics we have dq2 + Ki/j(q) = 0. 14. i. i. (14.66) J<1\ Thus in this case the socalled old quantum theory is correct. / • n an integer > 0. this means dqyf—f = (n + l)7r. 0(g) ex .e. Case (a): The case of two trigonometrical zeros.4(a). There are three cases. In this > case the wave function for g > go is .67) Case (b): Next we consider.e. i. 14. where /(g) is positive. together with q — — oo. Jqi ^ fpdq J =2 pdq = (n + l)h. Before we continue we recall from Eq.65) We use the knowledge of these solutions to find the eigenvalues of the equation. —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0. if /(g) is negative.e the relation (with a shift of TT/2) 1 2/V4 exp Jqo f9^W)dq 1 1 1 4 (Z) / sm Jqo (14. this is a trigonometrical zero.298 CHAPTER 14. as illustrated in Fig. (14. i. Since sine and cosine have zeros spaced at intervals of TT. the case of tp(q) — 0 at q = gi.
i. As illustrations we consider Examples 14. i. ^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r. the wave function for q < q'0 must be proportional to ri'o (Z) / 1 4 sin J V W)dq+^ Z i • oc (7) 1 / 4 sin / y/~f{q)dq . (14.63b). Since both cases refer to the same region qo < q < q'0.2 and 14. (14. and • • for the wave function to be exponentially decreasing to the right of q'0. Hence we obtain the condition I'o Vf(l)dq 90 (n + 1) IT.63b) to obtain the eigenenergies of the quantized harmonic oscillator.14.67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa . 14.70) We can therefore summarize the results in the form of Eq.3.69) pdq = (n + l ) " 4 h.2: The harmonic oscillator Use the relation (14. E x a m p l e 14. ri\ / J do y/f(q)dq (n + 1) 1" (14. For the wave function to be exponentially decreasing to the left of qo and using Eq. the sines must be proportional. Therefore their arguments differ only by (n + l)ir. Hence we have pdq= i n +  . Thus in this case the socalled old quantum theory is wrong.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.68) (14.e. or Pdq (n + l ) " 2 h.63a) or (14.77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1).4(c). Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig. this is the case of two turning points.e. Solution: As in the case of the quartic potential. (14.
5.2 3 J x (n + 3 / 4 ) 2 / 3 . Thus there is one turning point at E = mogq. :: V=m0gq Fig.V(q))dq The result is E/u. . 14. t Hence we have the case of the integral J> pdq (n + 1 )  Inserting the potential this becomes fqo=E/™0g 2 / dq^2mo(E Jq: Jq=0 Evaluation of the integral yields the energy 1/3 2/3 — mo m) • (n + 1) En = ( — m0g2h2 n+ • (14. g = 9. Example 14.807 m s . so that E = hu{n + 1/2).5 A particle above a flat Earth.63 x 10~ 3 4 J s . Calculate its quantized energy. one obtains E ~ 2. as illustrated in Fig. Solution: This is the case of one trigonometrical zero of the wave function ip at q = 0 and an exponential zero at q —> oo. permit this although with rapidly diminishing probability. in principle.28 x 1 0 . and h = 6.300 With potential V(q) = 2ix2mov2q2. 14. whereas a turning point does.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. The WKB Method this requires evaluation of the integral with qi = ^' E/2K2mQu2 / 2 m 0 ( B . where g is the acceleration due to gravity.71) For m 0 = l x 1 0 _ 3 k g . ^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it. 4 / x CHAPTER 14.2 .
73) Jx! rp2a dx I^<E 2 v) = ^ n — = .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 £ = .2. h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' .5 Further Examples 301 14.= — <*— = h(n+\.AE(q0 2 = 2n + 1) = — exp 7r dzy/E zn + V(z)/h where q=zo are the left and right barrier turning points.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 ( — = . In the case of screened Coulomb potentials (cf.72) \\2dxp=Qdx^{Eh Hence. E x a m p l e 14. d dE V)={n+)n. 2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V).75) which verifies the formula immediately.) M0 N = l + n+1 : + • 2/B Hence dE K 2 d dE[ M0 2VB.14. Chapter 16) one obtains the quantization relation (for Mo = const.l. ra 2/ = 0. 1 .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 . Solution: The proof is contained in Example 18. and u> is the oscillator frequency in either well. (14. (14. with p = modx/dt.5 Further Examples E x a m p l e 14.f V ft h JXl h 2TT (14.( 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 (14.7. 9 .
i. q = y+ a . (14. where 1 2 • 1 2 y = Q. which determine the imaginary part of the energy.302 and T CHAPTER 14. Eq.a . 14. Example 14.<?2i<?3 for E > 0 given by E — V = 0.77) Determine (a) the turning points. and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u.6 The cubic potential. The q term in this equation can be removed by transforming the equation to a cubic in y. 14. It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2. Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig.y3) = 0.y2)(y . where .~ \ (14.6: WKB method applied to the cubic potential Consider the cubic potential V(a)=1b\\a*~q). The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14.133)). Fig. (11.e.76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf. (14.6.79) . 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 .78) (a) We determine first the three turning points at qi = <?i.
_ \ / 3 — tan(9 2 V92 .k 2 2tan# \/3 + tan! . (14.g i ) ( g 2 .§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14. W i t h t h e o t h e r i n t e g r a l from T a b l e s we o b t a i n (K(k) b e i n g t h e c o m p l e t e *L.g i < 1. 93.1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ . p . W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 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 . F r i e d m a n [40].80) I n s e r t i n g t h e r o o t s yi.81) m a 91 . (14.0^6O°. 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 .1 / 2 .gi = 1. 92 = a2[l . H e r e cos(<9 ± 120°) = .08.85) T h e angle 9 = 60° c o r r e s p o n d s t o E = 0. e > 0.{ V 3 e } > 0. g 2 ~y{3}>0. 91 < 93 < 9 < 92(14. 14. ' P . tan60°=%/3. D w i g h t [81]. M i l n e .04.c o s ( 0 + 12O°) + c o s ( 0 . B .9 2 [ . B y r d a n d M . D . 6 = 0 i m p l i e s E m a x of E q . f o r m u l a s 236.82) 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. p .< 1. 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. 0<6O°.93 2a2 [ . ui=sn 1 \l— — = l. p . V3 + t a n 0 2 .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 . M .9 l ) f f / du s n 2 u en2 d n 2 u .= s i n t 3 v 3 (14.ui=K(k).3/2. 80 a n d 361.83) 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 . 212. (14. § T o h e l p we cite H. we set 0 = 60° — e. V cose' + sin 8 (14.79) a n d t h e n s e t t i n g m o = 1. .9 3 — = = 2 " 3 " [ . q3 ~  . 92. we h a v e qx = .2/3 i n t o q = y + a 2 / 3 .2 c o s ( 0 + 1 2 0 ° ) ] .T h o m s o n [196]. 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 .q)(q .14.e. 78: sin 120° = v ^ A c o s 120° = .' : ± 6 0 ° w h e n E = 0. gi ~ y {V3e} < 0.5 Further Examples 303 T h e r o o t s of t h i s e q u a t i o n a r e k n o w n * a n d for 4 P 3 > 27Q2 a r e r e a l a n d e x p r e s s e d in t e r m s of a n a n g l e 6 given b y (observe t h a t w i t h t h e choice of t h e m i n u s sign. (b) W e see t h a t gi a n d 93 m e r g e t o 0 w i t h e — 0.2 c o s 0 ] . 37. (14.cos 0 + cos(6> + 120°)] = a2 92 .93).120°)] = . F .84) 92 . a n d t h e a n g l e 0 = 0° t o E = £ m a x . i.a 2 [ l .c o s O + cos(6» . q3 = a2[\ .6. (l_fc' 2 a + fe'4)l/4.2 c o s ( 0 .\tan0^ v/3. h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q .120°)] = a 20? 2a2 3 2a 2 • cos 0  91 .78)) c o s 30 •• 4P3 ( ' \ aeb2} .
2(fc2 .= —^ = k a2[cos0+^] v ^ + tanfl V=".qz) (14. Eq.) . Whereas the integrand of the above barrier integral is effectively a momentum.k). as in (14. Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —.91 F(TT/2.2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it. (14. (14.89) This expression will again be obtained later with the help of configurations called "bounces" (cf.fc'2 + fe'4)"1/4. with K(k = 1) = oo. 2 2 . E(u = 0) = 0.?)(?3 .00.84) we obtain Carrier = 2a 2 sin6> 2 b( .2)K(k) + 2(fc4 + k'2)E(k)) k ^i — . E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 . 91 < q < q3 < 92 bJc. sinV = q3 qi ~ = 1. The WKB Method elliptic integral of the first kind.32)). Next we evaluate the required integral across the well from 91 to 53 at energy E. (14.sm t %/3 £\.S . the integrand of the integral across the well is effectively the inverse of this momentum.86) 15 tan. G(fc) =86aS (ITS£ li[fe.304 CHAPTER 14. Friedman [40].87) Inserting G(fc) together with the expressions for the prefactors into (14. . E(u) the incomplete elliptic integral of the second kind./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q. we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! .92) ''P. Thus Iwen=lgF(ip. (24. p. 72.~ y ^ ) a< [ COS 5 + .90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = .^ + ^ ) (l + fc'2)2 .85). formula 233.91 . g= 2 . V 93 . . V?2 . Byrd and M.91) a2[cos0^] y/3tanfl ? L . We can evaluate this integral again with the use of Tables of Integrals. D. E(u = K(k)) = E(k). F.^ V3 / V V3 y a / c o s e J. A:') = # ( * ' ) • (14. 1+72 = and J o s 0 + ^ = (1 . V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d . 0 (14.
3 and 8.96) Expanding the coefficients of the elliptic integrals in Eq. pp. Thus one obtains from (14. To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2).63b).85) (14. (14. so that even in the case of the ground state the zero point energy will contribute to the prefactor.88) in rising powers of fc' . 906.! With these expansions one obtains ^barrier = 15 ahb 2 + 8A . formulas 8. With q = 1/z the integral (14.2. . With algebra — which it is impossible to reproduce here in detail — one can derive expansions in ascending powers of fc' (which is small) of all relevant quantities.4 2n + 1 fc' = 4 — ^ — .89) can be expanded to exhibit a simple pole at z = 0 allowing evaluation with Cauchy's residue theorem: 2 dz 2Ez2 a2b2 %2dz Ez2 2nib\ o?E o?b2 2KE ab tt I. 32 (14.l. (14. (14. The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14. we obtain > .** However. we obtain E a6b2k'4(l + k'2). M. Ryzhik [122].113. in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1. and hence at E = 0. we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1. (14. 2a 5 b K(k) + 2 E(k)..114.3.5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14. we obtain 15 .93) where w is the harmonic oscillator frequency in the well. a°b n = 0.80).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. to represent a physical decay rate.99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab.94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq. 905.2 IE 4 fc' ln 8 (" «V iU^fc' 4 . Gradshteyn and I. S. . (14.14.95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) .
8 Bfc .. _ exp[2/barrier] >. (18.102) . /25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l .103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq. so that we obtain 4 . +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e .96) for the imaginary part of the energy as in the Breit. 2n + 1 — / 26a5b \ .178)) call the Furry factor set equal to one.. oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s.73). The WKB Method For k' —• 0 the first and the third terms dominate.Wigner formula (10. . . n ~ V27T .100) 'barrier = ~0°b In [ ^ — j .2 7 b a r r i e r ] Z Z7T where (cf.— (14..e. i.> tf = = 2/wfiii = ±tv8a°b——P. V (14.e x p [ .ba / 25a5b\n+2 ±l_l eT5 a f >. this has not yet been done.11)) dq 2TT y/2§?(Ev) It would be interesting to derive the same quantity with the perturbation method and to compare the results.„. 7 ~ —.( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) .101) (14. the result agrees with the ground state path integral result (24.fc'2 + fc'4)1/4 * ± t ^ .J + 0(VoJb).j . (0) i hw En = E\> . 15 a .7 . i. (14. also Eq. (20.e.105) As argued in Chapters 24 and 26 one expects this W K B result to agree with the one loop path integral result using bounces.306 CHAPTER 14./^~^l^ab ^aub . We observe that with \j2je as 1.. It follows that (for one complete orbit back to the original turning point) . _ T T • (14.
that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. There are numerous. 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.1 Introductory Remarks The particular linear potential we considered in Chapter 13. now known as quantum chromodynamics. 307 . Rosner [269]. Lane and T. K. that this is indeed the case. [233] and H. Gottfried. after the realization that quarks as their constituents might not exist as free particles and. In the present chapter we consider briefly the threedimensional potential V(r) oc r = r.Chapter 15 Power Potentials 15. [232]. i. the classification of the various states of nucleons and mesons and other particles.e. This potential became widely popular in the spectroscopy of elementary particles. Yan [82]. [231]. C.M. T. Rosner [230]. Kinoshita. L. at least indirect. indications from the nonabelian generalization of quantized Maxwell theory.g. K. led to a spectrum which contains only scattering states. Eichten. L. Thacker. The study of this potential.* In the present chapter we consider various aspects of this potential and extend this consideration to ' T h e s e investigations became very popular after the discovery of the heavy charmonium bound state ^ and were naturally extended to sufficient complexity to permit comparison with experimental measurements. C. Quigg and J. See particularly E. which permits only bound states. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e. the Coulomb potential. B. along with inclusion of angular momentum and spin effects and their interactions. describing free fall under gravity. in fact. D. Quigg and J.
308 CHAPTER 15. In order to have a clear starting point. we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r).6) = (I + l)u(r)/r = R ~ rl. (15.1) i. These particles therefore cannot be separated by any finite amount of energy.e. f dr{u(r)}'2 1. we return to the Schrodinger equation in three dimensions. i. where \x is the reduced mass of the two particles of masses mi. ~ (I + l)rl (15. v > 1.m2.2 The force The Power Potential F = .5) u'(0) u[r) r=0 = R(0). Power Potentials general power potentials.4>)ru{r) = Ylm(9. and so u'(r) .e. and the normalization [ dr\y(r)\2 = 1. mi + m2 and r is the relative coordinate. and u'(0) > [u(r)/r]0.4) Instead of considering the linear potential. and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15. (15. (15. R(r) = ^ . A > 0.V 7 = const.£ ] * ( r ) = 0. h2 2n V 2 * ( r ) + [V(r) . We leave consideration of the logarithmic potential to some remarks and Example 15.1.r acting between two particles is constant and directed towards the origin (of the relative coordinate).u ~ r i + 1 (/ + l)R(r).2) tt(r) = Ylm{9.3) with the boundary conditions''' u(0) = 0. For a central potential V(r) we write (15. </>)R(r). 15. m\m<i M= : . This force maintains this value irrespective of how far apart the particles are separated.
(15. (15.10) Here rc is the turning point and n — 1.8a) ^ JO = (2Ar + l ) £ .8a) we can write the quantization condition (14. . (14.9) 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!).. We also assume that V'(x) > 0. We therefore proceed as follows. which has been reduced to an effective onedimensional problem in the polar coordinate r.2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14.63a).7) k(x) = h/y/2m0(EV(x)). N = 0. d5. .2.V(r)) = (2nl) +1 N.2.63b): 2 and V(0) = 0 (15.63b) for the onedimensional Schrodinger equation.9) into the threedimensional case in the following form: / Jo C dry/2n(E . and the turning points are at a — — b := — xc."MA.. is to be interpreted as the principal quantum number in three dimensions. Then k(x) = k(x).4) in the threedimensional case. Onedimensional case: We consider first briefly a particle of mass TUQ in a symmetric onedimensional potential with the symmetry V(x) = V(x) without loss of generality. . x > 0. In order to to obtain the . so that with Eq. We now want to find a quantization condition for the threedimensional problem.. 3 . and there identify the Swave states with those states of the onedimensional case whose wave functions vanish at the origin — these are the nonsymmetric ones with N odd — as required by Eq. (15. odd integer ^=(nll)vh. This means that we transcribe Eq.1. or f l H " ^ ) ! . (15. (15. (15.15.8b) This condition implies that we have only one pair of turning points.
E A fv.14) Prom Eqs.17) The integral on the right is the integral representation of the beta function B(x. . c (nj\irh.11) rc = (15. _ —vr" *dr.13) we obtain _ X_ „ _ so that A£_ (15.y) defined by (xl)\(yl)\ y ^ (x + y . s.6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E .'(1*)1/2= J1F\lt)y1dt. 1W or Xv \E I It follows that (I) / ^""^(l*)1/2. dr = . Now.15) I ' = ^ 1 (( 1) B{x.12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15.y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/. (15.12) and (15. Power Potentials eigenvalues E —• En for the radial potential (15.13) dt . (1516) (15. (15.310 CHAPTER 15. (15.\rvc = 0 .
^ '"Compare with Eq. 15. The case v = 4 is that of the pure anharmonic oscillator also discussed by M.20) These are precisely the eigenvalues which one obtains from the vanishing of the (periodic) eigenfunction at the wall of the square well (of course n — 1/4 has t o b e replaced by n since the W K B approximation is only accidentally correct for small values of n ) . for instance simply graphically as indicated in Fig. i r ( i ) = r ( l + i ) = ( i ) ! .18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15. and n — — ]7r/i. Weinstein [281]. One can convince oneself now. $T(z) = {z.1 Approach to square well with v —> oo.2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/. 15.55). (12.19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig.1)!.15. t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well. (15. T(z + 1) = z\.* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15. .1.
15.g — —2.oj2/2. (6. eigenvalues (JV + h)tkj — (2n .21) (4nl). discussion at the beginning of this section). the case of the linear potential. (15.25) "Comparison with Eq. (6. since these are the ones we are interested in (cf. But for our present purposes this applies only in as far as the solutions (j)(x) which are exponentially decreasing at infinity vanish at x = 0. we obtain En = (n . the harmonic oscillator.41) for the .> r ( 2 ) A V 2 in \)*\W \it^T& (15. Power Potentials In the case of v = 2.23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function. For the following we set in Eq.1 + \)hx> = {An • l)£fiw. In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig. (13.oj = ^/2X/JX.24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative). It is therefore necessary to demand its continuity there in the sense of equality of the first derivatives from either direction (apart from the equality of the values of the functions there from either direction). we obtain E„. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 . Physically it does not make sense for the probability amplitude to have a discontinuity there.2.312 CHAPTER 15.//L This result agrees with that for the onedimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E\r2)cf>(r) =0 (15. In the case v = 1. x > 0. The potential V(x) = X\x\ (15.2) implies A s fj. > and with Eq. = (n>r(§)A 2/3 3TT \h n 2/3 L r(§)v^7^ J (15.22) 0 which selects from the usual and with the boundary condition 0(0) eigenfunctions the odd ones.61) h2 = I.mo = \.
and for s —> oo the trigonometric behaviour of Eq.e. Ai(s) 1 2v^(s)V4 exp x —> +oo. Eq. (14.57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E . W i t h z :— E — x this is d24>{z) + z<f>{z) = 0. i.x)\x=0 = 0. 15. i. i.x).27) i.60) applies. dz2 According to Eqs. (13.e. E > V or s > 0. (13.29) . s = E.2 Behaviour of an odd wave function at the origin.2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig. i. \(s?» (15. this wave function has the required exponentially decreasing behaviour at infinity.26) For z = s := E — x —> —oo.e. (15.59) applies. Ai(s) "v^ G ' cos s 3/2 _ t (15.e.56) and (14.15. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E . In the domain —a < x < + a .28) In particular at x — 0. (15.e.
Table 15. (15.28) is really only valid for s large and s — oo.03914 11.. It is interesting to note that in the dominant approximation both methods agree. (15.00852 11.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large. Rosner.02137 10.93528 12.1 demonstrates the quality of the WKB approximation by comparison with the exact values. Quantum Mechanics with Applications to Quarkonium. copyright of NorthHolland Publ. n = l. so that n — 1/4 ~ n.e.e. The expression (15. L.31) (an odd function has an odd number of zeros!). p.29) implies that cos (  ^ .32025 4.52056 6.82878 1 2 3 4 5 6 7 8 9 10 2.3.78445 7. 201. Quigg and J. *This Table is reprinted from C. with permission from Elsevier. = im — (7r/4).1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 .33811 4.94249 9. of course.78671 7. but for large > values of E Eq.. (15. the eigenvalues En are determined by the zeros of the Airy function.2.93602 12. to obtain the zeros of the Airy function numerically.30) ! ^ / 2 .08181 5.e.314 CHAPTER 15. Table 2. i.^ = !(2nl). (2/3)E3/2 or 3 / 1 2/3 E — En — ..02265 10.51716 6. It is possible. . i.04017 11.82814 i.En from Ai(£„) = 0 2.94413 9. Power Potentials Table 15.. Co. [231].08795 5.00767 11.= ) = «.
r <53) 1 3  dr cos 2 () (15. ee.• •) by a mean value. See also J.e.3 The ThreeDimensional Wave Function 315 15.15. The decay widths T.g. which are inversely proportional to the lifetimes.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. of course. have only a finite lifetime and decay into other particles such as electronpositron pairs. .3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs. D. in Chapters 24 and 26. dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (. The resulting masses agree in most cases very well with those extracted from experimental observations. the charmanticharm meson ^ . e. e. can be shown to be proportional to the modulussquared of the particle wave function at the origin. Jackson [140]. We saw previously (see for instance Eq. We shall encounter WKB normalization constants at numerous points in later chapters. i. P.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. In order to be comparable with experimental data these investigations. Weisskopf [239].g. van Royen and V.e.e. *Note that this makes N a W K B normalization constant. We indicate briefly here — without entering into further details — how these quantities can be calculated. (14. cos2()^— / 1 fr° 2 > w .34a) **The most frequently quoted reference for this result is R. In general these quarkantiquark pairs. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions. i.** r ( # ^ e e ) o c *(0) 2 . F.
38) N : sin V ^ Jr dr' 7T (15.e.Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( .34b) VMEvy Variation or differentiation of Eq..316 In accordance with t h e relation CHAPTER 15.39) in / .4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 .2) and (15. (15.32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15. A = (15. Power Potentials this averaging implies a numerical factor like 1/2.2. (15.rcu w X{r)dr = Kh . (15.37) (15. i.36) \i dEn irh2 dn According to Eqs.l N n .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. From Eq.. so t h a t approximately l N2 / Jo k(r)dr .35) u T (•T=2TT/OJ dt cos 2 cot 2' (15. & (15. (15.. dE a—— / dn J0 so t h a t with Eq.1 )n .40) VWY . (15.\ \ * n = l.35) N2 .
(15. J. S. Kaushal and H. K. 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).43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1.* Using the result (15.2. Bose and H. 3 . Thus only theoretical investigations performed immediately after the discovery of the particles \I/ and T are basically of an analytical nature* and therefore have not been performed largely with numerical methods and the fitting of as few parameters as possible. for instance. E. § S.44) (15. Hence Eq.1 ^ ( 2 ^ „ ) 1 / 4 . Bose. MiillerKirsten [29].l ) n . 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. (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here. K. The case of arbitrary positive power potentials as above has been considered by R. Itt^pocn2*"1^"^. the expression (15. Hite and H. W.3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e. . (15. MiillerKirsten [28]. W. .45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. and hence with Eq.37) we obtain: 1 (2M^)1/2 47r ft. G.g.23) for the energy levels of the linear potential.38) implies 317 (15.15. (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15. . S.41) I*(0)2=_L^)^. for various potentials: (a) For a confinement potential of power v. W. J. . we have ^'(0) = ( . MiillerKirsten [147]. J.
(11. .g ^£+(<*Pl»riy = 0. i.75). (15.j .2. . Power Potentials *(0) 2 oc . MiillerKirsten [29]. = (I + 1/2) 2 . S.^ ( 3 ^ + l)55p^«(3^l) + . ip = exp[(z — c)/2].L2 h4 = 4/3exp[(2a — /3)/f3]. (15.^ +•••. and there Eq. {S = ^ > 0. rewrite the equation as ^ z + [ . D.1. K.— oo < z < oo... 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 . Lifshitz [157]. W.c = —a//3.. .e. M.1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro).133). (14. We observed there in the case of the Coulomb potential that the period is proportional to n 3 . Example 15.318 (c) and for the Coulomb potential^ CHAPTER 15. J. in agreement with Eq.5. Landau and E. n = 0. Eq = i q . U(z) =  dz zf3e2a/?e2*. Solution: Details can be found in the literature. *=^f..I l n f ^ p .46) and the classical Kepler period of Eq. Bose and H. = 1(1 + 1)." 1 11 L . E' = E + gln(r0).46) We can now link the value of the wave function at the origin to the oscillation period T using the arguments of Example 14.L 2 + U{z)]<f>. q = 2n + l. 7 With the substitutions r = exp(z — c).
In reality. that no matter how far the particle is scattered away from the source of the Coulomb potential. Historically this potential arose with the realization 319 pVr .Phenomenologically the screened attractive Coulomb potential may be written V(r) = g2pr/ro . r in which /i has the dimension of an inverse length or equivalently that of a mass in natural units. so that like charges are repelled and unlike charges are attracted by each other.Chapter 16 Screened Coulomb Potentials 16. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric. 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. The result is a screening of the Coulomb potential which thus attributes it a finite range ro. it is. it will always notice the latter's presence. and in fact. A very similar potential is the Yukawa potential V{r) = g2 — .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. This implies. r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state. This effect sounds unphysical. It is clear that for ro — oo the > potential becomes of Coulomb type.
ii k .e. i.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.3) where for the real potentials we consider here. Yukawa potentials and superpositions of such potentials play an important role in nuclear physics. and that. Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons. (16.1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ). and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory. Heisenberg [133]. from the relation Pi + P 2 + M2c2 = 0. 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. The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies. 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 . however.* 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. in fact the parameter [i represents the mass of such a spinless meson. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator.320 CHAPTER 16. one frequently resorts in calculations to the expansion of the potential in rising powers of r. (16.1) We encountered Green's functions earlier. Regge trajectories — which we encounter *The Smatrix theory of strong interactions was actually initiated by W. the mathematical expression with an exponential is not so easy to handle analytically. . all coefficients Mj are real and independent of the energy E = k2.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. Since. who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory.
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. Kinoshita [21].e. if>oo.16. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes. i. Potentials expandable as in Eq. (16. C. S. Froissart [223].e. Lovelace and D. Bethe and T.1 in the above. i. (16. however.^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum.4) where E = k2 and h = c = 1 = 2m.k.§ 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. (16. Cheng and T. to be of the order of 1/k. (16. physical) values of angular momentum as functions of energy E. Omnes and M.e.3) have been considered by various authors. Masson [180]. l+n+l = ^P. A. Mandelstam [185].r)=0. and hence that the confluent hypergeometric series there obtained has to break off after a finite number of terms in order not to destroy this behaviour. J. for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R. See also the other references below.r). Frautschi [97] and E. bound state eigenenergies and the 5matrix. Naturally it is easiest to familiarize oneself with these by studying solvable potential models. . .5) where K — ik and A n is an expansion in descending powers of K. ''The approximate behaviour of Regge trajectories for the Yukawa potential has also been calculated by H. Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1.Q. Squires [258]. .mo being the reduced mass of the system. They are usually written I = an(E). T. 2 . $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq.3) in the radial Schrodinger equation for the partial wave ip(l. S S.{l. only for the cases of n = 0.k.e.1 Introductory Remarks 321 in this context — are functions which interpolate integral (i. "These were first considered by C. . Wu [47]. £+**PVV t/.1. 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. i. f H. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential.
Section 11. J. o rV(r) I dpp\V{peie)\ regular at r = 0. (16. Miiller and K. Schilcher [203].ndp.9) the cut starts at E = k2 = 0.k.)/j. In the following we follow mainly the last two of these references. 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. < oo for all 0 < vr/2. J. for all n.e. in the present case at fc2 = 0 or k2 — Mi = 0. 16. starting with the power r _ 1 of the Coulomb potential. one can assume an expansion of the potential V(r) in ascending powers of r. (16.322 CHAPTER 16.4) to z = —IKv and set TP(l.6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H. A.2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential.z) Then x is a solution of the equation VaX= (16. W. as may be seen from Eqs. Muller [201] and [202].19).z) = ez2/2zl+1X(l. Under these conditions the S'matrix is meromorphic (i.k. Longoni and T. H. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis. tf H . Regge [31] and E. *See in particular A. < const. Squires [258]. W. Bottino. M. Proceeding now as in the case of the Coulomb potential we change the variable of Eq. J. 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. .t Considering such superpositions of Yukawa potentials with /•oo / a(/j. Screened Coulomb Potentials as a function of the energy. W. Muller [204]. The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r.** The energy is later obtained by reversion of the resulting series.15) and (16. J. 'In the case of the Coulomb potential (cf.
1).10): Sm(a. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a).16. (16.* Recurrence relations for coefficients of perturbation expansions for Yukawa. (a.9) we obtain m l)$(a .b. We also observe that the ansatz (16.L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq. (a. By a repeated application of the recurrence relation (16. so that to leading order we have VaX{0) = 0. Sharma and H.0) = 1.a + + l). J. (16.b. all other So(a.9) (16.11b) ^ The associated boundary conditions are: So(a.j) may be computed from a recurrence relation which follows from the coefficients (16. (16. a) —b — 2a.11a) The coefficients Sm(a.awhere (a. j)Smi(a.z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a. and all Sm(a. a + 1) = a — b+ 1. W.10) zm$(a)= ^ j=m Sm(a. anharmonic and cosine potentials have been derived in L.j) (16.5)) a = + 323 .z) = Ha).7) is seen to be of order 1/K. The function <&(a.a)${a) + {a.a + j)Smi(a. where $(o.j +(a + j + l.j l) + (a + j.2 Regge Trajectories where d2 Va = z^ and (cf. .i 0) = 0.a+ l)$(a + 1) + (a.j) for \j\ > m are zero. K. MiillerKirsten [249]. (16. which means that the hypergeometric series breaks off after a finite number of terms.5) is equivalent to a = —n. Eq. {0) X = Ha.j)$(a + j).j) = (a + j l.a + j)Smi(a.a — l) — a — l. as in the case of the Coulomb problem.
13) one obtains the quantity An(K) and hence with Eq.12a) where [a.13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion. [a. Va+n = Va .aj 2 (2K)A _ [a. the latter can be written .324 CHAPTER 16.l.<&(a + n) on the right hand side of Eq. i R »] = 2K[a. #(a + ra) fjL$(a + n) V. [a. (16. (16. for the coefficients M^> of the expansion Details can be found in the references cited above. (16.a\4.7). This follows from the fact that D a $(a) = 0. so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i .a + j]i+l = MiSi(a. (16. This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0.7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except. of course. [ a . when n = 0.aj 2 \ [a. a . + (16. (16.n.5) the Regge trajectories I = ln(K).[ a + l. oo _ . Any term /j. Mn The coefficient of the sum of all the remaining terms in 3? (a) is then set equal to zero and determines to that order of approximation the quantity An(K).o]i = M 0 .e. 1 .' — . i. Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq.j).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. and hence Va$(a + n) = ra$(a + n). 0 < \j\ < i.a + l ] 2 [ . We first cite the final result . . Evaluating the first few terms of the expansion (16.An{K).[a>a + J]i+i*(a + 3).a]l^a) + ^2(2K)i+1 Y.1] 2 .
i. 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. However. Obviously the expansion — which. 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 .2.2 Regge Trajectories 325 and then demonstrate the calculation in Example 16.§ This is an important observation which indicates the equivalence of the methods.5).l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n .1) + 3M3M02 + M22n(n + 1) (16. With numerical methods one can achieve more.1.15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16. Ahmadzadeh.^ 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. The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n . Inserting this expansion into Eq.1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] . G. it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential. (16. Burke and C.14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) . 16.16. . Thus the expected appearance is that shown in Fig. we obtain the expansion for the Regge poles. .( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n .1 by evaluating the first two terms. is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero.l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n . (16. P. Tate [6]. Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A.^ p [ 3 M 4 ( n .e.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8). of course.
0) = (a. 16. and the lifetime At of the resonance satisfies the relation TAt ~ h. a] 2 = M i Si (a. just as a decay width T represents the width of the resonance in energy.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. .13). Frautschi [97]. 0).O) = Other terms vanish since So (a. p. this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig. For a bound state a j = 0 and the orbit becomes permanent. 16.0).2 a = In See S.1 Typical Regge trajectory for a strongly attractive Yukawa potential. (16. 0) and Prom Eq. However.a) = b . In fact. For a resonance with a long lifetime. The conjugate variable to energy is time. a] 3 = M 2 S 2 (a. Hence Si (o. one can expect a resonance with the lifetime determined by this imaginary part. satisfies the relation ajAO ~ h.12b) we obtain [a.^ . From Eq.1)." Re I 1 0 1 2 Fig. Example 16. so the imaginary part ai of I — an represents its width in angular momentum. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown.326 CHAPTER 16.11b) we obtain Si(a. through which the particle orbits during the course of the resonance. and the angle A#. 114. where the imaginary part of I is very small.a)So(a. C. ±1) = 0. Similarly the conjugate variable to angular momentum is angle. ai is small and A6 is large. (a. [a. Solution: We evaluate the first three terms of expansion (16.—" + In = — . . (16.
i ) again from Eq. using the Cauchy formula (6.n ( n + l Inserting these expressions into expansion (16.0) = (a .2 a ) . .1) Si(a.6 + 1 ) .^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 . (n+l)U+^pJ . a)Si (a. 0) + a S i ( a .65) 2m dz V'A + 2Bz + Cz2 • Z— 0. 1).6)Si(a. 1) (a . .0) Si(a.1) + (a .2 Regge Trajectories 327 Analogously we evaluate S2(a.13) we obtain 0 (Mo . a)Si (a.1. Solution: Ignoring higher order terms in r. (16.6 + l)a + A.al)S0(a. 14.11b).An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 .4) with mo = 1/2 and around the classical orbit the integral in the following relation I h = <b ^J pdr: dr •K2 Mo {l + \)2h2 1/2 ^classical orbit The integral is most easily evaluated by the method of "poles at infinity". A" An An A" = = = (a.1 ) + (6 . we have to evaluate (cf.0)+ 0 + 0 = (a1). a)Si (a. 0) + (a + 1. We obtain the quantities S i ( a . (a. One sets z = 1/r so that.(ra + 1) [n + (2K)3 A n+1+• n K K (M0 .2o)Si(a.16.65) 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 . Thus (with terms which are 0): Si(a. Sec.2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6. (a.a)5o(a.b)(a . .1 ) + (a. 1) Hence 52(a.A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 .0) + 0 + 0 = ( 6 .00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I.0) (a .0) + 0 + 0 = ( a .a + l)S0(a. 2 : E i g e n v a l u e a p p r o x i m a t i o n b y " p o l e s at infinity" Verify the dominant behaviour of the Regge trajectories of Yukawa potentials by contour integration of the BohrSommerfeldWilson integral in which 1(1 + 1) is replaced by (I + 1/2) 2 .
In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq. *J. Boukema [34] and [35]. (16. by expanding the right hand side in rising powers of v and evaluating the individual integrals. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case.15). Solution: For details of the solution we refer to papers of Boukema.328 Thus here dr K' CHAPTER 16. Example 16. evaluate — with 1(1 + 1) replaced by (I +  ) 2 and V(r) = —Mo/r + v(r) — the following equation dr K2 Mo (l+\? 1/2 . tSee Chapter 11 and V. (16.2.t First. This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 . i.16) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z .* 16. ±2HK in agreement with our expressions above.14) for An(K) in order to reexpress the latter in terms of I so that i+i+ A. The turning points in this case are at r = 0 and r = oo. .e.(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16. I.1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ .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. Thus we can write down the S'matrix for the present case by exploiting the limiting case of the Coulomb potential. Screened Coulomb Potentials Mo . n = 0. Singh [252]. Here we do not perform this procedure.5) together with the expansion (16. however..3: Calculation of Regge trajectories by the WKB method Use the WKB method to verify the first two terms in the expansion of A „ for the Yukawa Regge trajectories.l. 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).
We observe also that the S'matrix is unitary as a consequence of the result (16.0[(KZ + Mx)'6} (16..l)n(n + l)(n + 2) + 2M2M02 . Now reversing the expansion (16. Paralleling the case of the Coulomb potential in Chapter 11. the behaviour of the scattering phase in the domain of high energies. W.4 The Energy Expansion It may have been noticed that in the above considerations we could have combined the constant Mi with the energy into a combination K2 + M\. .16. for instance. i. and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\.l (two additional terms are given in the literature).19) we obtain the energy.2. n = 0. § See H.4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16. Miiller [211].* 16. we can now write down the S'matrix in terms of the scattering phase 5i as the expression TV/ I 1 I ^l(J<l) S ( W = e r(/ + i^#>)e ' (16 18) ' We observe that the poles of the S'matrix are given by l + l + ^p = n. W..18) of the Smatrix now to explore further aspects. J. which are precisely the expressions yielding the Regge trajectories of above. Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16.17) and is therefore real for real values of the potential coefficients Mj.g. such as. This has been done§ and one obtains .14).17).19) +2M 3 M 0 (3n 2 + 3 n .e.. the following *See e. J. M 0 2y/K2 + Mi n(n + l)M 2 4 ( # 2 + Mi) 3 /2 (2ra + 1)M 0 M 2 8(K 2 + Mi) 2 + 16(if2 + Mi) 5 / 2 3M 4 (n . Miiller [204].l. One can use the explicit expression (16. H. .
l)n(n + l)(n + 2) .e. p. B. Frautschi [97]. C. Mendelsohn [135] and E.l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n .2)(n . Large coupling expansions derived in H.10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n . Warburton [279]. N.l)n 2 (n + l) 2 (n + 2) 1 + • • • (16. Zauderer [286].3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n .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 . W. i.f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + n . k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r). 'According to S. E. 107.20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature. the transformation was introduced in the form given below by G.< 3 M 4 M 0 ( n .1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n . A. Screened Coulomb Potentials expansion.^ 16. J. VahediFaridi [212] contain as special case the expansions of the first pair of authors. 282. Sommerfeld [256].11) + 2M22M03(9n2 + 9n . . MiiilerKirsten and N. in the following expression with an "See in particular G. Analogous expansions for specific energydependent Yukawa potentials have been investigated by A. J. Iafrate and L.* We recall from Chapter 11 the definition of the scattering amplitude F(9. p.5 The Sommerfeld—Watson Transform The basic theoretical tool for the exploration of Regge poles is a representation of the scattering amplitude given by the socalled SommerfeldWatson transform.330 CHAPTER 16. 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) ^ — . Watson [280] in 1918 and later resurrected by A.1) + 2M2M.
. 16.22) = \F(9. c = 1. (11.23) f(l. ~:' sin nl (16.187). n = 0.1.k) — e 2i5[(k) ~" p—iirl (16.2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i. 16.iJ 11 —[S(l.. \x\ <C l .k) = 2zfc L " v "'' v .fe)P . dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig..24) Si(k) being the phase shift.''' The idea is now to consider each term of the expansion (16.k)\2.21) The scattering amplitude determines the experimentally measurable cross section o given by da (16.5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1. .187)) (16. Actually 5j lrr/2 as in Eqs.TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0. z=o in which (cf.k) 0ikr (16.184) to (11. (11. (cos 9).2.23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral. so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n).k)~l] = ~ k" PiSiW sin 5i(k) and S(l. For this purpose we consider the following integral taken along the contour C shown in Fig.k) = ^T(2l + !)/(*.kmcose).26. I = n + x.2 The integration contour C. Eq. The scattering amplitude possesses the partial wave expansion F(0.25) Setting Note the minus sign in the argument of the Legendre function..
For the momenta indicated we set (with metric +. be elastic scattering of a off b. 16. as indicated in Fig.27) a(p) © b(q) Fig. d(q') (16. 16. a + c —> b + d. 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').—.—) s = (p + q)2. t = (qq')2. u={pq')\ (16.3 Reaction channels and their respective Mandelstam variables. Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig.332 CHAPTER 16. Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem.3.—.28) . a + d — • c + b. we obtain m® = ± I dl^±^f(l. 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. n=0 which is the usual partial wave expansion.n) oo {l)1 Pt{cosO) v ' (16.3. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics. 16. Thus the direct reaction may.k) Zi Jc TT{L . in which case c = a and d = b. In order to see the relevance of Regge trajectories in a reaction of particles. for instance.fc)P„(cos0).26) P((+COS0) = 2 ( 2 n + l)/(n.
which is an analytic function of the variables s. F. Treiman [25].^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories. For a Lorentzinvariant normalization of the initial and final states.5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo.0s). (a) Mandelstam's hypothesis: All three processes (described by the s.u are known as Mandelstam variables.30) with partial wave expansions (observe the noninvariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98).16. One now makes two hypotheses. Symbolically we then have the situation shown in Fig.t. e.29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s. Thus. B.t) = VtFt(t. (16. L. Ft = £ ( 2 J + l)F. In a Regge theory on the other hand. one then has A(s. one can write the amplitude as depending only on two.31) i l Taking all particles to have the same mass mo. Mandelstam [184] and R..t. Chew [48]. Goldberger.t.2 ^ ( 1 + cos6 s ). . except for poles and cuts which characterize the reactions in the three channels.u are not independent. etc. N.u) (if any of the external particles has nonzero spin.4. s = 2qf(lcos0t)(16. Khuri and S. A(s.(<fe)^(cos 6t). the dynamics is described by the exchange of quantum numbers. N.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s. Blankenbecler. if s describes the square of the total energy in the s channel. M.g. In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions). t.The variables s. the momentum k there would correspond to qs) s = ~4(ql + ml).t) = y^Fs(s.t). if we had been considering the s reaction originally. T See e.9s)\2.g.et) (16.c o s 0 s ) . the quantities carrying these quantum numbers are the Regge trajectories. u.32) t = 4(q? + ml). (16. the variables t and u would describe momentum transfers in the crossed channels. § See G. the only kinematics we require in the following is given by the relations (e. there will be several such amplitudes which together describe all three processes).§ Since the three variables s. S.t.g. A(s. u = . 16. t = 2q£(i .
5 The Lehmann ellipse. However.334 Yukawa CHAPTER 16. the partial wave expansion is not valid in the physical region of the (i. But > > wave expansion X)(2Z + l)F. Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories. 16. the Legendre polynomial Pj(cos 9) does not possess a cut.e. Lehmann [162].4 Yukawa versus Regge theory. Screened Coulomb Potentials Regge oc(t) Fig.)e~ •fj. So what can one do? 1 H.5 for Yukawa potentials V{r) u: lm cose dr a(/j. It is known from the Mandelstam representation that the amplitude possesses a branch point at cos 9 = 1 + TOQ/2(^. .r 1+m 2 /2q 2 o ^t "" Re cost: Fig.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig. 16. 16. For s — oo : cos 9t = 1 + s/2q^ — oo. in the domain of schannel physical values of the kinematical of the ichannel). But for integral values of I.
e. and there is only a finite number N + 1 in the domain of 9W = .5 The SommerfeldWatson Transform 335 T h e answer is.1 / 2 .6 The integration contour C". Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n.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.16.33) . B Iml 1 \ i All \ r\ 0 / closed .6. 16. N F(6. all poles of Ql > 0. 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.^ B 1 A Fig.t). contour C / / D ^ ^ . Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l. to use the SommerfeldWatson transform. sin nl [16. „ i.26) but for a different choice of the contour. i. the contour representation (16. Moreover. 16.e. Thus consider now the same integral but taken along the closed contour C shown in Fig.
The > result demonstrates that the high energy behaviour of an schannel reaction is determined by the leading Regge trajectory in the crossed ^channel. for high energies in • > the schannel).e. W.32)) and finite. Apart from more refined details. Regge [229]. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics. 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.t)&B(t)sa°V.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. K.e.336 CHAPTER 16. at high schannel energies. MiillerKirsten [29]. if ao(t) is the Regge pole with largest real part. Limic [177] and E. for plots and discussion see R. as we discussed briefly in the above. i. Screened Coulomb Potentials Since for s — oo : cos Qt = 1 H n ~* °° > 2?* and (from Tables of Special Functions) farMoo: Pa(z)* ^+l]\(2zY. (16. Regge trajectories have. we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i. Bose and H. Omnes and M. Foissart [223]. we have (16. .35) This result is valid for s — oo and t negative (cf. of course. Predazzi and T. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N. J. i. (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential. this behaviour has been confirmed in high energy hadronic reactions. the amplitude can be represented as a sum over ^channel Regge poles.e. Eq. It follows that under these conditions. 42. 16. p. In particular.
g. called "Regge cuts". J. like a meson or a baryon. For an overview see e. Singularities appear also in the form of cuts in the plane of complex angular momentum. and is composed of quarks. The way to do this is similar to calculations in Chapters 18 and 20. a composite particle being one with structure. The above treatment of screened Coulomb potentials is incomplete.6 Concluding Remarks 337 important parameter in string theory.16. . Related aspects are discussed in Sec. and also as socalled "fixed poles" .** Whereas the former are related to absorptive properties of a reaction.1. MiillerKirsten [209]. H. W. the latter are related to the distinction between "' elementary and "composite particles". The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated. Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum. 20.
This page is intentionally left blank .
19 and p. 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. Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6. example 14. motivated mainly by the regularities of the crystal structure of matter. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type.g. the most immediate candidate is a trigonometric form like that of the the cosine function. the ellipsoidal wave equation reduces to the trigonometric form of the spheroidal wave equation.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms.) 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).0 < k2 < 1.Chapter 17 Periodic Potentials 17. M. are not much harder to handle than trigonometric functions. p. since many analogous relations hold. 25. In essence these periodic Jacobian elliptic functions.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other. is effectively the Mathieu equation whose solution has for a long time been considered as being very difficult. The Schrodinger equation with this potential. called elliptic modulus.* These elliptic equations involve Jacobian elliptic functions like sn(a. depending on the parameter k. See F. K being the complete elliptic integral of the first kind. Arscott [11]. and are themselves periodic with period e. 2K or AK. The reason for these difficulties is. 339 .1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered. Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers. Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip\x2. however.
1) where h2 is a parameter and — IT < z < IT. or Sw&ve Schrodinger equation with elliptic potential. 8. The first of these conditions is already tMathieu equation: R. need not be afraid of them.2 . J. The Mathieu equation.340 CHAPTER 17. which does not possess any polynomial solutions. i. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. 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. . Dingle and H. 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.2h2 cos 2z]y = 0. ''The reader who encounters these Jacobian elliptic functions here for the first time. See also the discussion in Sec. the Lame equation possesses 2 n + l polynomial solutions. Muller [73]. This is a significant difference compared with the Mathieu equation.1. it is this difference which implies in the case of the Mathieu equation the existence of the parameter function v called Floquet exponent^ The Mathieu equation can be obtained from the Lame equation in the limit of n — oo.. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A .7. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc. We might mention already here the much less familiar but more general Lame equation. J. (17. each with its own characteristic eigenvalue.. those of integral order. [205].e.4. 17. cosine and tangent. (17. W. Miiller. In fact. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0. W. A main objective of this chapter is the calculation of these level splittings following the original calculations of Dingle and Muller^ with the perturbation method described in Sec. The most important formulas for our purposes here are collected in Appendix A. and to relate these to the weak coupling bands or regions of stability. Periodic Potentials vide periodic potentials like trigonometric functions.2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K. the three Jacobian elliptic functions s n z . Both of these domains are important for a host of other considerations. B. which one looks up in books like that of MilneThomson [196] or Tables when required. c n z and d n z are handled with analogous formulas. Like the trigonometric functions sine. which is conveniently written ^ + [\2K2sn2z]y = 0. Lame equation: H. For large coupling. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions.1.
these concepts are rarely explained as such there. we compare the For interesting related discussions see M. 17. i. since the solution there is of the form cos vAz. a bounded function. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials. i. we see that for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability. we see that in a plot of h2 versus A. In classical mechanics stability is treated for instance in connection with planetary motion. Although the mathematics literature on the subject uses the physical concepts of stability and instability.e.1.2 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. Salem and T.2. 17. . Vachaspati [242]. the case of h2 small.17.^ 17. Thus the solution of the appropriate Newton equation is essentially a periodic function like cos 8 (in the case of planetary motion this yields the polar equation of an ellipse).e.1 The Floquet exponent We consider the given Mathieu equation with argument z and compare this with the same equation but with z replaced by z + ir. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre.e.e. h2 = 0 belongs to the domain of stability. Adding the negative term — 2h2 cos 2z to A. Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms. as in Fig. Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to. the semiaxis A > 0. and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r). i. i. y" + \y = 0(h2). Looking at the Mathieu equation for h? —> 0. It is this additional parameter function appearing in the solution of the Mathieu equation which attributes the equation its reputation as being particularly hard to handle.
We set (here and in the following frequently apart from contributions of 0{h2)) with constants A. Setting z = 0.B and a.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(z). sin V Az or linear combinations. b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz. d2y(z + 7r + [A . a = const. Periodic Potentials equations ^ p + [A . 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. 2/(0)' yfr) . ° ^ [1 + 0(h2)}. i. y+{z) .2/i2 cos 2(2 + vr)]y(z + vr) = 0. d(z + TT) cos 2z We see that there are solutions which are proportional. y~(z) = 6sin vXz. (17. — • . y(2vr) 2 a = —7—^ = 1.2h2 cos 2z]y(z) = 0.342 CHAPTER 17.e. i. 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). (17. TT.e.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. (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z).
oWX] . y'(z + n) = a[ay'+(z) + py'_(z)}. This expression is equal to the coefficient of a2 in the quadratic equation for a.4).y+(7r)j/_(7r)} = 0.e. 7 y\z) = ay'+(z) + (3y'_{z).y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\.W or fy+(n)aa J/_(TT) = W ^ ^ J a ^\ = n For linear independence of a and /3. we choose these with the following set of boundary conditions: y+(0) = a. and by'+(?r) = —ay/Xy{n). The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4. the determinant of the matrix must vanish. with constants a and /3. y'_(Q) = bV\ with Wronskian W[y+. [y+(v) . \y+>y\ = y+{ir)y'{ir) . 2/_(0) = 0.{ay'_{ir) + bV\y+(ir)}a Now from Eq. This means.aa][y'_{n) . yV(0) = 0. y{z) = aty+(z) + 0y(z). we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa]. (17. 2/'(7r) = a ^ W + ^ . We can write therefore.2 Cosine Potential: Weak Coupling Solutions 343 The solution y(z) is an arbitrary solution. Setting in these last equations z = 0 and using the previous pair of equations and the boundary conditions.a2b2\ a± ~ 2a6\/A . (17.3) we obtain y(z + ir) = a[ay+(z) + @y(z)]. whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) .y{ir)y'+{K) = abVX in agreement with its value at z = 0 above.y+(7r)y_(7r) = 0 or abVXa2 . Using Eq. i.17.
8. (17. we have <7± — CHAPTER 17. Eq. Periodic Potentials g= 2aby/\. <7__cr_ = 1.= — . (5i/ 2 + 7)/i 8 vz + — .22)) the boundary conditions of periodic solutions. i. 2 _ hA 2(A1) (13A25)/i8 32(Al)3(A4) 12 ift (45A3 . where the constants are usually taken as unity from the beginning. _ / ± v 7 ^ V . <T__ + <T_ = 2y + (vr) .7. we have for the sum of the roots a+ + <7_ = 2 cos TTV or" C O S T T I ^ ^ ^ C O S V A T T . cr+ = 1/<T_.+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58. .1. 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. This is the easiest application of t h a t method since only simple trigonometric expressions are involved. .2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ .344 Setting / = 2by/\y+(Tr). T h e result for A is the expansion A = h4 2 .6) We can derive various terms of this expansion perturbatively with the method of Sec. Setting cr+ = e .0 This series may be reversed t o yield the Floquet exponent.5) a The parameter i/ is known as Floquet exponent.5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2). 5 a Prom this we see t h a t (as one can also verify explicitly with some manipulations) one root is t h e inverse of t h e other. This calculation is demonstrated in Example 17. (17.e. this implies y+(Tr)/a = ± 1 .. Prom this condition we determine later (cf. i.1169)/i 12 64(Al) (A4) (A9) 5 2 + 0{hw) r 16 (17.8) "Observe that for integral values of v (in lowest order of h2).455A2 + 1291A . ** An important consequence of Eq. **This point is not immediately clear from mathematics literature.e. (17. (17.
— 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0. Example 17. With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17. as will be done later. = J—^ + o(hr).11) or smvz or e±ivz. rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12).e. (17.2 Cosine Potential: Weak Coupling Solutions and so 345 (8 . It follows that in these cases i? = n2[l + 0(h2)]. W. Schiifke [193]. we must have a2 — 1 and hence e 2iv/A7r = 1} i.1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs. n an integer.9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A .7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small.3).jl2. . (17.10) A = n2.35A + 15A2)/l8 /.5) and (17.12) S e e J. Meixner and F. (17. (17. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2.„„ x n. 124. p. (17.35A + 8 .1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec.2 how the expansions for small values of h2 may be obtained perturbatively.9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A. Solution: We write the eigenvalue equation A = v2. For these integral cases of v we demonstrate in Example 17. which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17. VA = ±n.17. 8. cos TT\/A 32A(A1) 2 . (17.1) 15A2 . These cases therefore have to be dealt with separately.
The recurrence relation can also be obtained by substituting the right hand side of Eq. Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z.v±2) = l.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq. so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) .v + 2)yu+2]. v . Proceeding in this manner we obtain the solutions ce.2) _ __ _ (y. where except. v)yu + {v. se and me as* oo i y = „«>) + yW + i. siniy + 2)z + sin(i/ — 2)z.14) = = = cos(y+ 2)z + cos{y— 2)z. (17. i. h2).14) therefore leads to the following nextto leading order contribution (v. see Example 17.. and (v. v) = 0.11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0. for instance. by J. e ± i ^ + 2 ' 2 + e±i("2)z. as shown. me±v(z. = . h2(u.346 CHAPTER 17.16) *These smallh expansions are convergent. We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration.e. . i.e.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes. (17.2) which applies when v is nonintegral.2)j/„_ 2 + {v.3 and 19. (17.h2). or A = 0(/i 4 ). .2. Schafke [193].e. since j / 0 ' = yv leaves unaccounted R\. so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17. Next we treat terms yv+a in j / 1 ' in a similar way. The right hand side of Eq.2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions. (i/.16) directly into the original equation in the form of Eq. . v . We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i.v— 2) 2 ^ + yu+2 Vv 2 y (D = h _2(2u2) 2(2i/ + 2 ) ' Then up to 0(h2) the sum j / 0 ' +y^ is the solution provided the remaining term in Ri. 17. up to 0(h2) we have A = 0. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 . With our perturbation formalism — as demonstrated in cases considered in Sees.e.15) Hence a term fMy„+a on the right hand side of Eq. W. i. ("• ^ + 2) (o) 2(2^ .„ + £ h2i £ P2i(2j)yv+j (17.11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v.h2). sev(z. Meixner and F.!/) = 2A We now observe that DvVv = 0. (17.3. (17. when a = 0 or 2v + a = 0. '. (17.(2) + .11). of course. (17.
we obtain immediately 0 = 2Ah2+ 2(^T)+' which verifies the term of order h4 in Eq.1 we now have the situation that the contribution y\ to the entire solution leaves uncompensated on the right hand side of the equation the terms amounting to R(°li = = 2 h 2 ( A + cos2z)3/„=i =2h2Ayi+h2(y3 h2[(2A + l)yi+y3]. Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• . Following the first few arguments of Example 17. Solution: We start as in Example 17. + y~i) (17. * Note."2). Example 17. W. thus determining the quantity A.2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A.R{ .2 ) v(y2.. v) ' ( 2 v + 2) (17.u) + h4 (".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. . For v = 1 we then have yi = cosz = y~\. . fo^ + 2) . „ . (17.2—r——— (i/ + 2. The higher order terms naturally require a little more algebra.17. p. Meixner and F. .1 with A = v2 — 2h2A but with v = 1.. Schafke [193]. 2 ( 2 i / .g. is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero.7). one has to deal with each integral case separately as in the following example.14). (17. 0 = h2(u..e.15) (but with v — 1 and a = 2) we obtain the next to leading contribution to the solution as (i) = h2 =_fe2 V {2(2v + 2)}„=1y3 8 y3' This contribution leaves uncompensated the terms amounting to R\ = ~ — 2h (A + cos2z)j/ 3 o = —[Ay3\ 4 \ 2 Hence the next contribution to the solution is 4 V 8 2(4)(2 + 4 ) / 4 V 8 48 Proceeding in this manner we obtain as the complete solution the sum y = y(0)+yW+y(2)+.17) Inserting 1 for the stepcoefficients of Eq.. J. (17. . however. See e. We consider the specific unperturbed solution yv = cosvz which is the dominant contribution j / ° ) of our solution y.18) Again applying Eq. This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2. i. the solutions are not yet normalized. In the case of integral values of i>.. 121. These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature.v)\ .
The reverse situation is later of importance. In the above we considered the case of v an integer and calculated the eigenvalue A. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf. see below!) one obtains the additional terms given by Meixner and Schafke [193]. Schafke [193].348 From this we obtain 2h2A = h2 CHAPTER 17. Eq.* The associated solution is ym+ym. We introduce a normalization constant c and rename the normalized solution y then yc. (17. i. 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]. With normalization (not to 1. 123) (there the solution is called cei(z. Meixner and F. *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn .19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai). h2). and v is to be found in ascending powers of h2. h2)) except for the overall normalization which implies that in our (still unnormalized) case above there are no contributions cos z in the higher order contributions. . The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z. J. h4 (17.n integers. 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 .e. W. 120. m. Periodic Potentials h4 r ••• . p. the case of A an integer. We consider this case again in an example. 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 (17.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p.24c) below) .
Substituting these expressions into Eq. We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + .4 ) + .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. MiillerKirsten. J.3 5 A + 8)TT2(A4) _ 3 / / 64(A .9b) about A = 4. (17.3 1 A + 8) ( A . (11A2 .('108\/E\2/ Expansions of u for integral values of vA have recently been given with a larger number of terms:^ l//i\4 3 el 2J 1 fhy 15 1 2 / + \/A = 4 : v § 311 4320 V 2 ' + 1555200 V 2 + 137 (h\° 305843 (h 12 27000 I 2 ) + 680400000 V 2 133 (h 12 + ••• R .1) (A .4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A . on the left hand side of Eq.Q . (17.17. so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above. we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i . We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq. .. fe47T2(A4) 4%/A(A .1) .4)Av A2v A ~ 32A(A . (17. Hashimoto [124]. o 1• • • . H.. Gubser and A. W.= l 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • . ^S. Liang and Yunbo Zhang [189].1)2/A TT2 +h8 (15A2 .3 1 x 4 + 8) 274233 (. 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 . .2 Cosine Potential: Weak Coupling Solutions Example 17.l ) 2 (A . J . Manvelyan.9b) v — 2 + <5.• 1 + 5n h 2 s 2932 Setting in cosm.9b) and considering the approach A — 4 gives > (A4)2 + . S.
i. "These conditions.21b). bV\ y+(0) 2/40) 07. in F.W4W=UtWSta±I(.10) applies. These are defined by specific boundary conditions. Periodic Potentials Our next immediate aim is to specify the solutions for which Eq. pp. for instance.29. M.6. 28 . (17.e.21a) and (17.. = and s+w 2tW±I. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding.4) into account. 17. Arscott [11].350 CHAPTER 17. may also be found. We now derive these boundary conditions.2 Four t y p e s of periodic solutions The Mathieu equation allows four types of different solutions — briefly: Even and odd solutions of periods TT and 2ir.1 Boundaries of domains of stability.w^. Taking in particular Eqs. 17. Fig. as also conditions (17. (17.2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir. We verify these for the large/i 2 solutions in Example 17. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r.2. . solutions for integral values of v.
2 Cosine Potential: Weak Coupling Solutions We select the equations with lower signs and put z — TT/2.21a) Since (with the help of the Wronskian) *W4)MM?)^ we also have y+(v0 _ . /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.y V+ \ 2 7"^ V 2 fla^+U a cos V A  ] 6^A + a&( sin\/Aj ^A = abV\. we obtain y+(7r) _ i  2y_(f). . Then y+ y 351 U . ^\y±M /vr\yL(7r) . which after some rearrangement can be written y+(vr) _ . f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ .17. /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. 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+. .4(f) (17.
the case of ai with solution cei (of period 27r).24c) 8 h2 sin 4z \ .1. Thus for these solutions the right hand sides of these equations imply the vanishing of the functions y± or their derivatives at z = ir/2.22) * a2n+i(h ). y > se2n+i=qo.. A 2 y • ce2n=q0i.352 CHAPTER 17. (17. This is shown schematically in Fig.. sei = sin 8 64 /i4 h6 a\ = 1 + h2 — . Mr7nA (17. We see that there are four possibilities and hence four different types of functions.' fh2 V2ceo = l .1).24b) 8 h2 z —— cos 3z + • • • .21b) must be ± 1 . y > se 2n +2s 9o +i The functions so defined have respectively period 7r. These may be subdivided into classes depending on whether the integral value n — 0.24d) v 12 ' **To avoid confusion we emphasize: In Sec. 17. Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example.21a) and (17. A *• b2n+2(h2).23) We can now see how the domains of stability and instability arise.3.2): hA 7h8 29h12 + Y l28^04. 19.T + . 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 . These are defined by the following boundary conditions. and whether the function is even or odd. in this order. . 2K. (17. (17. ce± = cos 8 64 h4 5h8 b2 = 4 1 . The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0. . 17. (17. 27r. There the second term on the right is nonzero.• a2n(h2)..— — — • • • . (17. of v is even or odd.+ .2 we require this relation for nonintegral values of v. was treated in detail in Example 17. IT. se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • .24a) ao = hA h6 b\ = 1 — h —— + — — • • • . (17.. A > b2n+i(h2). Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq.2..5) that the left hand sides of Eqs. (17.1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec.. y > ce2n+i=qo.
of course. . The cosine potential cos 2z depicted in Fig. In the case of the cosine potential.1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig.2 The cosine potential. 353 . We can see from the inverse of Fig. J.3.3 17. B. 17.> (17. 17.z3 ce2 = cos 2z .2 suggests that for very large fluctuations the potential can be approximated by a number of independent and infinitely high degenerate harmonic oscillator potentials. 17.3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 . Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator. Thus in Fig. .1 that the boundaries of regions of stability merge to lines.24e) 17.hz — 12 .17„. Miiller [73]. Dingle and H. the walls are not infinitely high and hence tunneling occurs from one well into another. 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.2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R.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.2cos4. 13824 . 17.17. . 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. W. 17.
3.354 CHAPTER 17. 4 (17. simply the onedimensional Schrodinger equation for the harmonic oscillator. of course. Taking the remaining terms of the cosine expansion into account.27) . We therefore wish to expand it about these points. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± . 8.2. Ion (17. Thus in this case the quantity on the left is not an exact odd integer. Periodic Potentials minima at z — ±ir/2. n 0.26) 17..2/i cos z V = 0.3) and is.26) into the original Mathieu equation and obtain y" + 2h q + rrxr . In the case of finite heights of the potential barriers.1. but only approximately so.2 The solutions We insert Eq. (17. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1. as in the case of the cosine potential. (17.25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf. we therefore set 9 A A = 2h2 +2hq + —.2 and 8. Sees. and we set this equal to q.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. where A / 8 is a remainder.... there is tunneling from one barrier to the next.
3 Schematic picture of Mathieu equation eigenvalues.17.29) is known. z).+2h': 2h level splitting oscillator 2nd excited state associated solutions A^9 h=4 oscillator 1 st excited sate ^1 oscillator ground state A^Oh 2 large h 2 =0 Fig. (17. and substituting this into Eq.29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) . (17. Choosing y(z) = A(z)e2hsinz.30) D A): COS « = 4z + 1 iq sin 2. 2 (17.27) we )A = 0.31) We make the important observation here t h a t if one solution of the type (17. the linearly independent one. 16/i/ (17. another solution. (17.3 Cosine Potential: Strong Coupling Solutions 355 For h —> oo the solutions of this equation are y ~ exp I ± 2h / Writing y = obtain J 4exp(±2/isin cos zdz I = exp(±2fasin z). 17.28) A" ±4hcos zA' + 2h[ q =f sin z \ A. can be .
(9._ 4 .4 ] .(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17.34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq.e.35) But since and so a term iiAq+n on the right hand side of Eq. 1 + 4 ) ^ + 4 + (^. (9l)(93). (17.36) Then ^4.9) = 0. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout. ^ ) A + ( ? . (17. 2[(92 + l) + A]. With some algebra one finds that 16^ + where (9.30) amounting to Ri ] q = . (9.33) (17. of course.94) = (9 + l)(9 + 3).32) cos^'+i) [\K . Aq+4 2Jh .Q)Ag + ( ? .4)A. (17. ? .94) \ — Aq4 (17.^ 6T [(«>« + 4 ) A ? + 4 + (Q.4 ) ^ .356 CHAPTER 17.9) (9. 2 A A = («.35) not yet taken care of is set equal to zero.g + 4) . D{q%Aq+4i = 0.\z To lowest order the solution A is A^> = Aq. (17. . when i = 0 — the terms not involving Aq in (17. The solution Aq(z) of the first order equation Dq Aq(z) — 0 is c o s ^ 1 ) ( \K+\Z Aq{z) = sml(9+i) ( ivr + \ z ) COS 2 9 ( D I\n+\Z (17.9 + 4) = (9. i.35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g. 2 /i D^Ag = 0. ? .30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except.
<J+4 Aq+& + 1 2 1 . q..39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout.4. except those involving Aq. ) ^ 4. q.4) (q . h) = A(z. (17.g + . Thus 0 = (g. (q. (17. . Then for A = A(0)+A(D+A(2) + .. an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17. can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q.38) Aq8 + 1 2 Clearly we can continue in this way. (q.30) and hence y = Ae2hsmz to be a solution of the Mathieu equation.4) (q . q. (17. q .37) Here all the terms.8)A.30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ .q + 4) (g. q . q (17. A(z. +4. to satisfy Eq. h).4 .39) which is the equation from which the quantity A and hence the eigenvalue A is determined. (17. h) = A(z. q.17. We note here again that if one solution y = A(z)e2hsmz 2hsmz is known. .q 4)Aq+4 q + 4 1 q + + + + + <^(.g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17. the sum of the terms in Aq in Eqs. h)._ 8 (17. .)+ & ! = « ( .4.q 8)Aq+8 (o)' ^±V(q 1 4. Thus in particular A(z. A.37) and so on — left uncompensated — must vanish.3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation.. Q .4) 1 1 (q. In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq.g4) 4.q + 4)(q + 4.g + 4)(g + 4. q .35) and (17.40) .
67). We return to Eqs.) i ( Q .44) where D(B) 9 = dw2 $_ !)• (17.46) 2s(«. .41) so that the early successive contributions decrease in magnitude. (6. a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw). it is found that D(B) z+ — ]B = 0. i. 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. The differential equation there is Eq.43) 9 2eh .28) and write the upper equation for a solution y = Bexp(2hsinz) in new notation B" + 4/icos zB' + 2h( qsin Changing now the independent variable to w{z) = iVhcOS I 7T + Z I . Note different definitions in use. (17. this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0. We observe that for the large values of h we are considering here. and discussed there. (17.d2B w dw2 + w(l w dB_ dw d dw 1.e." Hence in dominant order the solution B is JB^0) = Bq[w and it is convenient to set Hei Bq(w) = 1 (91) H (17. 2 w2 + H* (17.358 CHAPTER 17. where Hen{w) is a Hermite polynomial when n is an integer. Before we study the eigenvalue equation in more detail.i ) We encountered Hermite polynomials earlier in Chapter 6. we show that two other pairs of solutions with a similar coefficient structure and the same eigenvalue expansion can be found which are valid in adjoining domains of the variable z.42) (17.45) {B) We recognize Dq as the differential operator of the Hermite equation.
34).e.. 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).51) . 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.sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17.49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ .17. (17. 1 COS I .e. also the same coefficients in the case of functions C below.3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17. y — Cexp(2hsinz) and Cexp(—2/isinz). A{wz (17.. The solution with contributions B^°\B^>. Finally we obtain a third pair of solutions.e. i.qi)Bq_4. which is an amazing and unique property of the Mathieu solutions — cf.47) where the coefficients are the same as before. i. the expansion for the quantity A being identical with that of the previous case.q + 4 ) £ g + 4 + (q.3 ) \He* ! (9+1) H.7 T H 1 2 Z 4 <1.e. i. (17. q)Bq + (q. forms a rapidly decreasing expansion provided that \w (z)\ « Vh.. with the equation C" + 4h cos z C + 2h ( q .d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q. (17. . as in Eq.4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized. 16hJ (17. i.48) This shows that this solution is valid in particular around z = ir/2.50) C = 0.
. each of them associated with one and the same expansion for the eigenvalue.52) and the coefficients are again the same as before.(w2 + A)Hen.54) where the proportionality factor is complex. show that the perturbation remainder of Eq. ie 1 COS I . (17.4. (17.A Hen .44) can be linearized with the same coefficients as in the case of the trigonometric solutions. q)Cq + (q. q + 4)C 9 + 4 + (q. The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i. (17.q. 17. (9+3) (2n 2 + 2n + 1) + .4: Hermite function linearization of perturbation Using the recurrence relations (6. Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq. These domains of validity are indicated in Fig.q.n2(n l)Hen2 .4)C g _ 4 . Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged.h). (17. 2 J where n=(ql). Example 17. we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w). can be reexpressed as 72 = (n + 1 ~(q + l)Hen+2 l)He. We have thus determined three pairs of solutions of the Mathieu equation. A closer look at that expansion is our next objective. He'n(w) = wHen(w) Hen+i(w).360 CHAPTER 17. We note that C(z.7 T 1 2 Z 4 <1.(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) . 2' the remainder 72.44).q.w2)He'n .59d) of Hermite functions Hen(w). (17.h) ocB{z. Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q.53) This solution is valid in particular around z = —7r/2. Again a change in sign of z throughout yields the associated solution y = Cexp(—2hsinz) with C[u»(z)] = B[w(—z)] and domain of validity in particular around z = ir/2.
(q. (?) 1 1 + (9. q + 4) = (g + 1)(? + 3).q + A){q + A. 17.g.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).)Bqi(w). q)Bq(w) + (q.57) . . in the following compact form with coefficients Pr(q. 2)Aq_8} + • (17..0 and —4. q . q) = 2[(q2 + 1) + A]. (b) The allowed moves are +4. (q. the solution y = exp(2/i sin 2)^(2). with (. ) .. r j=r.j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V .q + 4)Bq+4(w) + (q.3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K . no intermediate move to or from q being allowed.g4)(q4.56) (94.g) 1 1 and so on. ^ 1 Hv" '^ ^' ^ ' .q4) (17.55) M2 = M. We see that (a) every coefficient Mr results from a sequence of r moves from q back to q.q + 4) (q + 4. We can write the solutions obtained previously.l)(q .3. 2 + ^ M 3 + (17.1i . l)Ag4 + P2(q.4) = (q . In addition (c) each move except the last has to be divided by the final displacement from q divided by 4.' 361 in 2*f*1>[i(gl)]! = (q. e.l)[l(q _!)].17.i ) ( w ) 2l(. .3). The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation. (q.j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q. ~l)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q.q 4.
g + 4j)P r _!(g. (17.j). g + 4j)P r _i (g. j) +(g + 4j + 4. Miiller [73].1) + (g + 4j.e. P2(q 2) Pl(g. Thus we have.g + 4) ^3(9.59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q.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 + 1) (17. as jP r 2 (g. Dingle and H. j) = (g + 4j .g + 4)(g + 4. and the contribution from a sequence of r moves from q + 4j back to g. Periodic Potentials PlM)=(M+i).g + 4)(g + 4. (17.j).j).g + 4)(g + 4.j ^ 0) = 0.4. i.g + 4) 2 1 (g.362 Clearly CHAPTER 17. 2) 1 2 ' ' ^ ' ^ 1 (g. . From the coefficients (17. p = (<?><? + 4 ) ( g + 4 >g + 8) . l ) . 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.l ) . the factor j removing a duplication of factors in the denominator at the junction.g + 12) 1 2 3 (g. J. .i (g. for instance.j)^2(9.60) M2r+l = = *R.g + 8)(g + 8.j)] = 2j]jPr2(g. x)_ (g.g + 4) + 1 1 1 and so on.j). M2 = P 2 ( g . g + 4 j ) P r .58) with P 0 (g.P 2 ( g .j).g + 4 ) ( g + 4. j .j) = (l)rPr(q. W._1) = Mzil.3) ^3(9. r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g. and in general* M2r = J]j[P2(g. B.j)Pr+1(g.34) we deduce that Pr(q.g + 8)(g + 8.g + 4)(g + 4. 0) = 1 and all other P0{q.
h — —h. [137].62a) f R .61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above. of course. W. These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig. [116].4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions. We observe that there are regions.2 * +3) + ^ 2 (9g4 .5 the following proportionalities there: B[w(z)] = aA(z). as a result of tunneling between wells. In fact we demonstrate in Example 17. q is only known to be approximately an odd integer qo = 2n + 1. It will be seen in Chapter 18 that in the case of anharmonic oscillators. Ince [136]. a= (8/i)3(9x) [1(91)]! 1 .^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. the solutions and eigenvalue expansions again have the same type of symmetries. t Terms up to and including those of order l//i 5 had been obtained by others and by different methods. 17.4. Miiller [73]. Its precise deviation from qo. J. Goldstein [115]. B. Dingle and H. Up to > > this point.(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) . where solutions overlap.3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 . S. remains to be determined next.284g + 57) + (17. . *E. as pointed out earlier.17.* We observe that the expansion remains unchanged under the replacements q — —q.92q3 + 70g2 . This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness.^ . L.( 3 ^ .3. 17.
and determine the constants a and a.1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ .1 ) ( 7T + . a .32). Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17. Example 17.62b) *z domains of overlap Fig.Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) . 17. The first term of the expansion of A(z) is the function Aq(z) given by Eq. We write and expand this function as follows: . Solution: We begin with the solution containing the function A{z).1 4 .364 and C[w(z)] =a~A(z).7T H Aq(z) = = C O S ^ .5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA.COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 .(17.4 Domains of solutions and their domains of overlap._ ^ l CHAPTER 17.
g. we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e 2/isin. with e.46). (6.3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •. we obtain Bq[w(z)} H. s . we have to impose the appropriate boundary conditions.22).284g + 57) + • The other relation is obtained similarly. Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2. (17. with (8h)li''1) (3g2 . (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. For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73]. the coefficient of the dominant factor c o s ( .63) In Example 17. i.2.2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed.2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 .!)(.36).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 (.61) for the connection and Eq. (17. Since the solution with A is obtained from that with A by a change of sign of z. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2.3) 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.6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec. these are conditions imposed on even and odd solutions at the point z = it/2. (8.53) for the expansion.g + 4) (q + 4. This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq. As we saw there. 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 .17.i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a. 17.e.i ) 1 (g. A*1) given by Eq. These are the conditions of Eqs. _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa.92q 3 + 70q 2 . \ql)/2(w) 2i(9D[i(.
63) to the domain around z = ir/2 by using the proportionalities (17.66) then follows from the recurrence relation (6. Thus y+ = ce and y_ = se. i. The second of relations (17.22) and to correlate the solutions to those for small values of h2. 80. W.66).59d). Then y± oc ^ M e ^ i n .62a) and (17. The functions defined as in Eq.» a \dw _ C[w = 0}c_2h a w=0 = Q (17. He'v(0) = Hev+1{0). He2n+1 (0) = 0. He'u{w) = wHev{w)He„+1{w).366 CHAPTER 17. .65) For the evaluation of these conditions we require the following expressions involving Hermite functions which we obtain from Tables of Special Functions:^ ^ ( =0) = 2TT2I//2 Fi(^ijii' He2n(0) = ( Ur w=0 •W+m' (17. Magnus and F. one obtains the first of relations (17. With the help of the duplication and inversion formulas of factorials. i.g.62b). a ± C[w{z)}c_2hsinz^ a (17. and then replacing 2n by i>. (17.64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17. give the following expressions for polynomials with argument zero: ~19)"(^)!. 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. Periodic Potentials We can extend the solutions (17. Oberhettinger [181].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\ ! ' .e. p.66) ^E.e.
1) 7T . B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17.67b) <f ~ 82g^ .g0)3].67d) Considering now the second of conditions (17.67c) 4 2 \dw)w=Q C[w = 0] [\(q .69) If the right hand side of this equation were exactly equal to zero.87 4 2 ¥w (17.Qo) + 0[(q .67a) g 106g 87 2 14 /i 2 (17.( g .82g2 .7.17.\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g . we have cot j ^ ( g ~ 1)} = ~\<(l .39 2~^h? (17.11.! ) ] ! [ .68) Inserting here expansions (17. (17.106g .» 2 6 /i the condition reduces to the equation cot{ .136g + 9) + • • • .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 . and using the reflection formula (8. f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 .67c).3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative. from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b. the solutions would be given by q = go = 3. Hence expanding the cotangent about these points.39 (17.40g3 + 18g2 . and similarly for C and its derivative.67a) and (17. .
70) + ^ 2 (9?o .e..73) We now return to the eigenvalue X(q) and expand this around an odd integer qo. Thus we have 2 (16/i) qo/2 T2 4.. (17.408g0 + 1089) 2 /i +• (17.136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg . .5. The remaining condition (17.70) except that now qo = 1.9.528gg + 3307Q .136g0 + 9) + • Clearly the last of conditions (17.(17.368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17.120g£ + 467g .40gg + 18gg .40gg + 18gg ..40g3 + 18q2 .65) yields the same result except for a change in sign. .136q + 9) + .65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{(ql)\ 7T ..65) leads to the same result except for a change in sign. i.72) This leads again to the result (17. we use \(q)~\(q0) + (qq0)(^J . Considering now the first of conditions (17. = _a a 6 _4h (17.h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ . Periodic Potentials ' " Vn [i(gol)]' 3(g02 + 1) 26/i .71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 .
^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w .3 we call this difference 26E„n. example 3.17.3.A_(g0) ^ . The tunneling effect. The dominant contributions were also found by S. Dingle and H. Goldstein [116]. (17. See also F. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential.1.. + ^ 2 (9<?o + 89o " 78g02 . quite apart from applications.87) = Azpteo). W. . results from the boundary conditions.74b) The above results for the cosine potential are in many respects important. just as any solvable problem is important for the insights it offers in a concrete and transparent form. (17. i. though approximate. (17.3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq. We see here that *R. M.3. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels. and that for the odd Mathieu functions se qo+ i and se9o (lower. 9o = 2n + 1. n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h. The complete.2. plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig. Miiller [73].61). B. 20. *In Sec.. and not the separation of these as a result of tunneling.. p.74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) . Arscott [11].88q0 . 17. In the literature that we follow here* in all cases several more higher order terms have been given.e.t From Eq. i. minus sign). In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion.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. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics. n = 0. J. .e. 120. In the literature one finds frequently the statement that the splitting is a nonperturbative effect. (17. this becomes 369 = (16/i)^+1e4h l .
results from the imposition of boundary conditions. Then we obtain first y±(z) = yA(z)±y^(z). M±TT) . 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 exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft). which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion. Naturally one expects the solutions there to be related to those of the periodic case considered here. and one can compare the methods.§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues.74b) is also important for various other reasons. evident through the exponential factor exp(—4/i) in Eq. yA{z) = Aq(z)e2hsin '. in effect. The splitting can also be obtained with some methods of large order of perturbation theory. y^{z) = Aq(z)e~2hsin *. The explicit derivation of the level splitting (17. We begin with the solution containing the function A(z). We consider for simplicity only the dominant contributions and replace throughout —1 to some power by exp(i7r) to that power and combine such terms into cosines and sines. 17.2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . . The splitting can also be derived by the pathintegral method as will be shown in Chapter 26.63) — 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. From Eq. The associated or modified Mathieu equation with cosh z instead of cos z is another important equation which we shall study in detail in Chapter 19 in connection with a singular potential.2. As mentioned at the beginning. Thus we do not reproduce every step. Example 17. Periodic Potentials this nonperturbative effect. The exponential factor represents. 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). the Schrodinger equation with the cosine potential is a special case of several other more general cases. (17.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). (17. y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs.370 CHAPTER 17. . ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)). T y'_(0) y'A±*) y{z).6: Translation of Solutions For the largeh 2 even and odd solutions (17.74b).
17. y'_ (0) = \/2(4h .1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials.c o s  ~(<1 T 1) f y . the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r). Similarly we find y(z±ir) = = and by putting z = 0: V.4ft) sin j ^n(q .7 r ( g ^ l ) U + ( z ) q : i s i n { .4.4ft) cos i TT(<J + 1 ) 1 .q). ( . (l^ifr^i^O + Cl^ifo^j/aCO cos  . A' (z = ±ir) = (. A ?q( Z = ±TT) = . where Aq(0) = Aq(0) = V2.4 17. since A q (0) = \/2 = Aq(0): y+(±n) = ±isin^(9=Fl)J3/+(0).• Then y'+ (0) = 0.yV+\z^ = [j/+(0)] 2 (<? .31).3 The Ellipsoidal Potential Replacing VA.y]z=„ = [y+yL .%/2(g . Replacing sines and cosines by the functional expressions.T ( ? =F 1) p + (0). = ±ir). and y+(z = ±TT) = V2(q ..( 2 ) . which is easiest by comparison with the foregoing treatment of the Mathieu equation.g A ? ( z = ±TT).4ft) ~ 8(q .VA DV 371 1 V±.z ) .17.4ft. Aq(z) The derivatives can be handled with the help of Eq.( ± 7 r ) = c ° s  . but also to enable a brief familiarization with the Lame equation.[A'qM + 2hAq(n)] = iV2(q . 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+. (17. The Lame equation has not been a widely known equation of mathematical physics so . we obtain the conditions stated at the beginning. — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) . we obtain — with Aq{z + 7r) = (1)^ ^ 2 A .1 )  .) cos < 7r(q 1) [ = .2hAq(n)] .Ah) ~ 32ft.7 r ( q = F l ) } j / . etc.( z ) From this we deduce for z = 0. Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) . y+(0)~2V2.
Ince [138]. the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940.0.* and the work of both prepared the ground for presentday investigations.0 and ..1. so that Floquet's theorem cannot be immediately applied . Arscott [11] (p.4sn4u]y = 0. 194) remarks.5.fi2A. Periodic Potentials far.0. and was continued by Erdelyi. p. J.5 The potential sn2(z. . M.Q. L. This line of investigation.. § F .. the nonoccurrence of the Floquet exponent. In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials. If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here).. 19. but also cubic potentials. however. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A . f E . Liang.0. 194) remarks to parallels with the Mathieu equation:". 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 .. *A. as alluded to at the beginning. k) for k = 0.75) ' T h i s means double well.75. inverted double well and cosine potentials (J." Fig.K2sn2u . Erdelyi [85].. H. W. Arscott [11]. MullerKirsten and D. 17. (17. for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane.5 < z < 5. But more recently it has been observed to arise in various contexts.372 CHAPTER 17.* As Arscott [11] (p..98. Tchrakian [165]). H. The most conspicuous difference compared with the Mathieu equation is.
(17.76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K. In comparisons with the Schrodinger equation. and of the derivation of the level splitting'. (17. The first step is to write the eigenvalue A as A(q. (17. (17. If we put fl = 0. Jianzu Zhang and Yunbo Zhang [206].\k\ < 1. is the elliptic modulus of the Jacobian elliptic functions snu. MiillerKirsten. for very > high barriers (harmonic oscillator approximation around a minimum of the potential). 17. where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k.K) = qK+^±. For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects. the usual factor —h2/2mo.e. Although the results have been calculated for the ellipsoidal wave equation. The function sn2u is plotted in Fig. i.77) where q —> go = 2iV + 1. (17.5 for several values of fc.3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2).2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions. . in the case n2 — oo. we consider mainly the Lame equation.e.78) We follow in this brief recapitulation the description in H.2k.17.76) 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. the equation ^  + [A . Eq.2 = n(n + l)k2. . in front of the second derivative has to be kept in mind (mo being the mass). J. 17. . 2 . K being the complete elliptic integral of the first kind.K2sx?u}y = 0. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential.1. cmx and dnu. In order to distinguish the above equation from that with the Lame potential. i. W.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system. .75) reduces to Lame's equation and one writes K.77) into Eq. The second step is to insert (17. N = 0.4. The range of the independent variable u is 0 < u < 2K.
81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed). /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. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k.77).K — > A(u)=A(u + 2K). i. . dnu± cnu K Thus one can construct solutions Ec(u). Periodic Potentials . one pair in terms of Hermite functions of a real variable. C replacing yl. A into equations in terms of the variables z(u) = ^—{k' .374 where CHAPTER 17. determine their proportionality factors) in domains of overlap (their extreme regions of validity). by transforming the equations for A. A. . Thus in the third step two more pairs of solutions B.. \/8K (dmz = cn« \ ' F . the other in terms of those of an imaginary variable. (17.q.K). (17. but solutions B. C which are valid for dnu±cnu <1. A second solution is written y = A(u) exp < / nsnudu >. which are respectively even in u (or snu) or odd. dnu =F cnu .e. more precisly for dnu = cnu F 1 » . ^ (17.e. i. one obtains again the same expansion for A.80) The domain of validity of these solutions is that away from an extremum of the potential.e.q. A are derived.K)=A(u.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. (17.82 .Es(ii). B and C. A(u. _ Solving the resulting equations iteratively as before. one has to derive new sets of solutions there and match these to the former (i. / z J . k' = Vl~ k\ \dmi±cnu/ 17.
C = a~A.384n2k4(q2 + 1)} q r {(l + A.3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a. Oberhettinger and F. J.^ ^ { ( 1 + k2)5(63q6 + 1260g4 +2943g2 + 486) 8k2{l + k2)3(49q6 + 1010g4 + 1493g2 + 432) +16fc4(l + fc2)(35g6 + 760g4 + 2043g2 + 378) 64fi 2 fc 4 (l + fc2)2(5g4 + 34g2 + 9) + 256f22£.38g2 . who obtained this expansion for the eigenvalues of Lame's equation (i. (17. . Magnus.17. (17.75).63)} + ••• .3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions. as well as \duJ2K 11 (17. a of _ _ B = aA. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0.2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} .6(5g4 . F. (17.83) 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. p. W." 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) . *A.e.4. the ellipsoidal equation.85) The first three terms of this expansion were first given by Ince [138]. 64. Miiller [205]. W.e. Erdelyi. Tricomi [87]. £2 = 0).86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H. 17. Es(2f0 = Es(0) = 0. i. G. This expansion is found to be in the more general case of Eq.
i Here the upper sign refers to Ec*> l or Ec£°. from which the difference q . 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).88g0 .90) + 3.:J {3(1 + k2)2 (9q* . + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ . Muller [205]. W.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)*.Ec£ 0_1 and Es2° of periods 4K. . 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 .go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) .g 0 ) ( ^ .78g^ . and the lower to Es*>+1 or Es*5.376 CHAPTER 17.89) one obtains the eigenvalues from which the level splitting can be deduced. Finally expanding A(g) ~ A(g0) + ( g .87) (17. 2K.91) tt H. One finds^ A(?) A(g0) T :(l 2K 2 2 fl + k 7T V 1 — k 2 njk 8K k 2 yo/2 I 1 Hqom 25. 2K and 4K respectively.l n 2^+ i J + 0(«i) (17.j 26K2 (1 85) = A(g0) + (q . Periodic Potentials These conditions define respectively functions EcX°. go being an odd integer.Es^ 0+1 .4k2(q2 + 2g0 + 5)} + • • • ].2 /c +256fc2g0(g2 + 5)} (17. (17. J.1 / 2 (2TT) 1 /22« l + (l_fe)^«( i .g0 is obtained by expansion around zeros.136g0 + 9) n 3.40g03 + 18g02 .
K = 2/i. that calculations with the Schrodinger equation are not only easier. W. A = 0. (17. A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq. (17.2h2 = A. (17. 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.90) one has in this limit: /l±±\ ~K/U = e .4 h ( i e M 9 0 7 2 \. V. A . it = x ± ^ . . in which they dubbed the classical configurations Lame instantons. however. Thus in the case of the dominant contribution of Eq. J.86).f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«. but also more easily generalizable. Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite. Dunne and K. Rao [79].92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues.* Apart from the choice of notation.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. . One can see.74a). the conditions (17. + E. Miiller [206].93) (17. 17. Replacing u by x ± IT/2.3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods./2i and hence A^ A/ A AU I2 . L.Ah2 sin2 u}y = 0. h .87) agree with those of Ince. i.4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation. Ince [138]. K2 = h2 (say).4.J1 O. Hence the conditions fi = 0.* M G . (17.2h2 cos 2x}y = 0. then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A .e.17. this equation becomes y" + {A .2h2 . *H.
**Y.g. Cho.** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19). M. See e. H. See Chapter 25. W. Liang. A. N. Quian [123]. H. de Veigy and S. H. S. J . W. for instance. Gu and S. 1 1 See Z. all basic potentials. This is an interesting problem. D. J.^ The equations considered above also appear in diverse new problems of physics. § . A.W. Ouvry [276] and Jianzu Zhang.e. Kan and K. Rana [287]. J. A host of related elliptic equations has recently been discovered and studied. . Thus. 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. i. MiillerKirsten and D.Y. Ganguly [103]. MiillerKirsten and J. Sukhatme [148].5 Concluding Remarks The potentials we considered above will reappear in later chapters. A generalized associated Lame potential has been considered by A. Shiraishi [49]. Tchrakian [165]. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. Periodic Potentials 17. 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. In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine.§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation. double well.Q .378 CHAPTER 17. inverted double well and cubic potentials. Khare and U.
* 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. ' T h u s A. The recent work of R. Achuthan. T. Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1. [19].. J. This seemingly simple problem revealed extremely rich internal structure . W. Wu. Bender and T.1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory. Bender and T.Chapter 18 Anharmonic Oscillator Potentials 18. T. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems. J.t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest. W. D. M. H.. Weinstein [281].Q.". in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another. MiillerKirsten [163]. f C . MiillerKirsten and A. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. Wu [18]. Priedberg and T. both of which make the calculation more difficult. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C. In particular the investigations of Bender and Wu. Liang and H. 379 . This lack of popularity of the calculation of complex eigenvalues even in texts on quantum mechanics may be attributed to the necessity of matching of various branches of eigenfunctions in domains of overlap and to the necessary imposition of suitable boundary conditions. Wiedemann [4] and a revised version of parts of this reference by J.
1 The three different types of anharmonic potentials.1. (3) Complex eigenvalues with tunneling: In this case. which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation. Case (1) is obviously the simplest with the anharmonic term implying simply a shift of the discrete harmonic oscillator eigenvalues with similarly . 18. described as the case of the double well potential. 18. V(z) = \\h*\z2 I 2i 4 \cr\z . V(z) = \\h*\z* + \\<?\z*. with the potential described as an inverted double well potential. and thus lead to very different physical situations.380 CHAPTER 18. To avoid confusion we specify first the potential V(z) in the Schrodinger equation dz2 + [EV(z)]y(z) =0 for the different cases which are possible and illustrate these in Fig. Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution. *~z (1) (2) (3) Fig. 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. These contributions may be given different signs.
Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side). 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. 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. (18.86) below. The level splitting will be rederived from path integrals in Chapter 26. The shift of the eigenvalues is best calculated with straightforward perturbation theory. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined. We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way. 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. (18. it will not be possible to obtain the exponentially small contributions with convergent expansions. except for a change of sign of c 2 . The boundary conditions are nonselfadjoint and hence the eigenvalues are complex. 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 do not dwell on Case (1) since this is effectively included in the first part of Case (3). that the general applicability of the method becomes evident. and this behaviour is that of asymptotic expansions our treatment largely terminates this muchdiscussed topic. Thus we are mainly concerned with the double well potential and its inverted form. . as is sometimes hoped. since the potential decreases without limit on either side of the centre. however with decay as a consequence of tunneling. The result is an expansion in descending powers of h?. In the case of the double well potential our aim is to obtain the separation of harmonic oscillator eigenvalues as a result of tunneling between the two wells. The eigenvalues of this case are given by Eq.18. In this case our aim is to obtain the aforementioned complex eigenvalue. Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics. Since these exponentially small contributions are related to the behaviour of the late terms of the eigenvalue expansions (as we shall see in Chapter 20). The imaginary part will be rederived from path integrals in Chapter 26.175) below. Case (3) is seen to be very different from the first two cases.1 Introductory Remarks 381 normalisable wave functions. We begin with the latter.e. The eigenvalues of this case are given by Eq. i.
18.1 refer to the two cases.2. (18.2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature.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.z Fig. a point which has to be kept in mind in comparisons. Anharmonic Oscillator Potentials 18. v(z) = ^h4z2 + ^c2z\ (18. This implies t h a t results for THQ = 1 (a frequent convention in field theory considerations) differ from those obtained here by factors of 2 1 / / 2 . (18. .2 18. We take here h = 1 and the mass rriQ of the particle = 1/2.2. we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei.2 The inverted double well potential with (hatched) oscillator potential.3) "1/2 harmonic oscillator h8/25c2 . 18. 18. The potential in this case is given by V{z) = v(z). If suffixes 1/2. h 4 2 2h1> ~1/2 2c?.1) for h4 and c 2 real and positive.382 CHAPTER 18. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0.1 and more specifically in Fig.
\h?\ — oo and c finite. the harmonic part of the > potential dominates over the quartic contribution and Eq.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..d2 .2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18.e. and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig.2. (18.. 18. y{w) oc Di.5) becomes . The result will be that derived originally by Bender and Wu. (18. (18.. and a variable w defined by setting E= b* 1 + A A and w = hz.2.2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr. n = 0.4) we can rewrite Eq.5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite.1.7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w). _i\(w) and q = qo = 2n + l. The problem here is to obtain the solutions in various domains of the variable. /i). (18. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different. (18. then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue.1/7  Vg(w)y(w) = o(J^j. to match these in domains of overlap.8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small.. i. 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. 4 .18.
In order to arrive at solutions we set in Eq.10) into (18.14b) .11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18.12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^.4(2.2 We are concerned with the equation d2y(z dz2 where ^ 1 .2.13) + 2qh u2 A + W A{z) Later we will be interested in the construction of wave functions which are even or odd around z = 0. •A~2 hrz" 2^\ c^z V2 —— + (18. This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point. (18.13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout. This observation allows us to define the pair of solutions yA(z) = .14a) 1/2 VA( ) = A(z)exp (18.9) z =0 these Thus these c2z4 1/2 + • 2 (18. Anharmonic Oscillator Potentials T h r e e pairs of solutions 18. y 4 2/i Here again q is a parameter still to 2 determined from boundary conditions. We observe — before touching the square roots in Eq.384 CHAPTER 18.10) E= qh2 ^ j.) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (18.+ ^ . Before we return to solutions we derive a new pair which is valid in the adjoining domains. (18.2 A (18.9) we obtain 1 h4z2 c2z4 A L2 (18. h) is obtained from the perturbation expansion of the eigenvalue. these solutions are not valid around z = 0. 1 d r A!{z)±iA{z)±{ 0.11) V = 0+ dz2 + 2qh + W The solutions in terms of parabolic cylinder functions are valid around and extend up to z ~ 0 ( l / / i 2 ) . Inserting (18. as encountered and explained earlier. as we shall see. be and A = A(q.
/ c z Aq{z)+0 1/2 .15) where A{z) is the solution of the upper of Eqs. /H 4 Z h +~C Z 21.16) For large h2 we can write the Eqs.13) TzA'(z) T ^A(z) + qA{z) = O We define Aq{z) as the solution of the equation zA'(z)(ql)Aq(z)=0. (18.19) We see that one solution follows from the other by replacing z by — z.34) below) — to be derived in detail in Example 18. 1 (18.Aq(z) approximate the solutions of Eqs. One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq.e. Clearly Aq(z).18.4 . We take the square root by setting z2h^1/2 +. i. (18..20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h. Since these higher order contributions are of little interest for our present considerations.zh2 =()*2"(18. 1 2 4 2 Aq(z) + 0 (18. (18. (18.2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z).13) and we can develop a perturbation theory along the lines of our method as employed in the case of periodic potentials.17) Aq(z) 1(91) Z2 (Z )V4 exp 2 We define correspondingly w dz (z2)l/2 (18.20b) . 385 (18. Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2. (18.1 and as a verification again in connection with the solution y# — as the other solutions. we do not pursue their calculation.18) M*) zf(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18.13) and A(z) that of the lower of these equations.
1 .2 ( * ) 2 and from this or separately A"(Z) : 1 (q . z)). . 2 .! ) ^ .20b). W i t h proper care in selecting signs of square roots we can use the solutions (18.1) 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). (18. (18.f .. procedure.^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl. (18. to obtain in leading order the eigenvalues E.18) we obtain A » = H^rL^z) 2 z . 04 = 5 .e..17) and (18. a2 = 2 1 .20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q. A A _ 3 . one observes that with £>. Anharmonic Oscillator Potentials i. 03 = .21) Example 18. (18. h2. i. z) ± yA(q. 128 8 16 A = i. A = .:=*£ +i(5l). (18. away from the central minimum. As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z).386 CHAPTER 18. ft = . i. h2.20a) and (18. c*i = 1 1 . X » / A q + 4 i = Mg+4i .• Using Eqs..l)A(z) 1 = tfA"(z) A ..2 .20a).[yA(q.e. „ . .13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q .v ' = o (« . * = — . zA'(z) where the expansion coefficients are given by ao = l. h2 —> —h2).1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A. „f 1 \ S o l u t i o n : We rewrite the upper of Eqs. Vq+ii = Vq + 2i.e. we write > V±(z) = .
[1(9 + 1)]! ..w — > > ±iw) or functions R B w q( ) . We return to Eq. w = hz. and the coefficient of terms with i = 0. (18. (18. of course.. 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). give an equation from which A is determined.(q + 3)(q + l)Aqvri(q '^ ' .22) are parabolic cylinder functions D i . _1J±w) and D_\. Rq '. Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 2 { 4/i2^ . when i = 0. 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.23) . c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B. i. In this way we obtain the next order contribution A^1' to A^°>.5).'2H9i) ^faDfr™) and . ^f( g + i)(±H2^ + 1 > \ — 1.. Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^3)^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = 24 1 2 ^ + 1>2^+°G9A= . The solutions yq(z) of the equation Vq(w)yq(w) = 0.2 The Inverted Double Well Potential Thus a term /j.—77 [i(9l)]. those in Aq(z).22) is invariant under the combined substitutions q — —q. +1J±iw) (observe that Eq. r ( CJw) = ^ .Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except.. (18.e. (18.18. = •>— r.
q + 4j}yq+4j. (18.3. Tricomi [86].25) where in the case i = 2: SA{q. (18.0). l(q2 + l).( < ? . W. Actually these factors can also be extracted from W K B solutions for large values of q.24) The extra factors in Eq. Vq+4j = Vq + 4j.4j)yq+Aj. (18. As an alternative to Bq(w) in Eq. since Vqyq = 0. Oberhettinger and F. (18.388 CHAPTER 18. we also have Vq+^yq+tj = 0.23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form. j=i (18.28) (A + c w )yq(w) E 2 4 1  2 ^ [q. G.3)y g _ 4 . and so Vqyq+Aj(w) = Ajyq+Aj(w). .±8) 5 4 (g. pp.Aj)(18. (18.^ For higher even powers of w we write i w2l V<i = Y. S2i(q.26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q.l)2/?4. See Example 18.±4) S4(9. F.27) Now. 115 . Comparison with our notation is easier if this reference is used.29) ' A . Magnus.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 + . 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 . (q±2)(q±S).0) = = = ±(q±3)(q±7). and for j ± 0 : [q.q}=A + c2S4(q. (18.123.q + 4j] = c2SA{q. Erdelyi.
so that the entire expansion is invariant under the interchanges q .2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq. .—A(q + 1) .32) Evaluating this expansion and inserting the result for A into Eq.e.(!) («. .) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained.^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J.31) Proceeding in this way we obtain the solution y = yW („. . and c 2  replaced by — c 2 . p i(9»9 + 4 J'WK7». (18. This type of invariance is a property of a very large class of eigenvalue problems. i.q]=0.1. (18.) + j.—v*+*j ( 18 .34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1. 2 .35 ) .18.27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j.) ty^^w) to be a solution to that order we must also have to that order [q. n — 0.e. A = ^(q2 + l)c 2 + o(J^. (18.34) We observe that odd powers of q arise in combination with odd powers of 1/h2. Thus the next order contribution to yq is y H = ie 2^ —z. and even powers of q in combination with even powers of 1/h2.10) we obtain E(q. (18. We can now write the solution y(w) in the form y{w)=yq{w) + Y. i. (18. Equation (18. ( 18 . h ) = qh .[rp ) Y.3 °) For the sum y{w) = y(°)(«.• q. In Case (3) the parameter q is only approximately an odd integer in view of tunneling. h2 —> h2.
—. and for j ^ 0 all other P0(q. their recurrence relations and the solutions of the latter we refer to the literature. We thus have the following pair of solutions VB(Z) VB(Z) = B w . w — ±iw.37) For further details concerning these coefficients.)]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. q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t. there is another solution y(—z).e. q + At] (18. i. q) = = 1. g ± 4 ] T4 ±4 and so on. (18. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1. / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz. > "p. We observed earlier that these are obtained by making the replacements q —> —q. Achuthan." Since our starting equation (18. q + 4j) = 0. 2 4tPi{q. ^VO(<L<?) ^i(?. we may infer that given one solution y(z).i 1 + 4j) in complete analogy to other applications of the method.g±4] ±4 ±4 [g.11) is invariant under a change of sign of z. MiillerKirsten and A.argz=0 (18. W. Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q.q±A) = MM!i±MM+[M±8][g±8. ? T 4 ] [ g T 4 . Again we can write down a recurrence relation for the coefficients Pi{Q.38) [yB(z)]argz=7T = [?/s(2. J. H.390 where for instance v J CHAPTER 18. Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument.36) with the boundary conditions p o(q. Wiedemann [4] . o.
.g.5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles. h2) in which odd powers of q are associated with odd powers of h?. in fact. Thus in the transition region some become proportional.4)). one has to stretch each by appropriate expansion to the limit of its domain of validity. 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 order to be able to extract the proportionality factor between two solutions. like w of Eqs.h^ih 391 (18. In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant. with normalization since the normalization constants are also asymptotic expansions.2. h2 — — h2 as long as corrections » resulting from boundary conditions are ignored. (18. the solutions of type A being valid away from the minimum.g.40) "Variables like those we use here for expansion about the minimum of a potential (e.2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0. Integrating and expanding as follows since h? is assumed to be large. Vc(Z) = [V!B(Z)}q^~q. (18.z2 .18. so that the eigenvalue expansion remains unaffected by the interchanges q —> —q. 4 . see e.** First we deal with the exponential factor 1/2 exp exp dz z z h* + c z J2c z 2 2 . 18. Mauss [192]. we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18.The solutions yc.q + 4j) as in ys.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q.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. Such contributions arise.h^ih. Aq(z)) does not appear in any of the higher order terms. J.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\ = 0.39) are with the same coefficients Pi(q.
45) . We do not require this at present. 2 ) i ' ir! i iHim 1 3 j argw < 7T.ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw.43) W + 1)]! w^+V A . y^(z) we see that in the direction of z = 0 (of course.42) we obtain for the solution ys(z).0x ! [ £ ( g .44) In the solution y~g(z).w yB(z) ~ B.(«. Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=. (18.41) with the solution ys(z) of Eq. 4/T 4 Prom (18. Comparing the solution VA{Z) of Eq. > n. eh* /12c* e\z*h? (18.4 i .7 r .43) (since z — — 2 implies > argz = ±7r).2 « .392 CHAPTER 18. . we would have to substitute correspondingly the expression (18.l ) ] ! ( .42) H W2 (ql)e$W I „ .44) we see that in their common domain of validity yA(z) = ~yB{z) a (18.2)i (27T) 1 /2 e f(9l) 1 eb Uq + m (18. not around that point) VA(Z) VA{Z) = ehVl2c 2 e\z*h* z^) + o(± zfr+i)+o(±\\. (18. 41) The cases of the solutions of types B and C require a careful look at the parabolic cylinder functions since these differ in different regions of the argument of the variable z.. The function Di. with z in ys(z) replaced by — z.z). Anharmonic Oscillator Potentials Considering the pair of solutions ju(. ^Jw) has a similarly complicated expansion for 1 5 — — > argw > . 2 ^i![i(g4il)]!(2«.) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (18. (18.
large probability for the particle to be found thereabouts. We proceed similarly with the solutions yc(z).50) will be seen to imply qo = 2n + 1 = 1. (18.4 B o u n d a r y c o n d i t i o n s at t h e origin (A) Formulation of the boundary conditions The more difficult part of the problem is to recognize the boundary conditions we have to impose.46) However.2.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 . At z = 0 the solutions of type A are invalid. the ratio of yA(z)iVB(z) i s n ° t a constant.2 — we see that near the origin the potential behaves like that of the harmonic oscillator in fact — our largeh 2 solutions require this for large h2..7. (18.42) into y~c(z) w e obtain Vc(z) = (h*. Looking at the potential we are considering here — as depicted in Fig. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18.y'_(0) ^ 0.49) d yc{z) Again there is no such simple relation between yA(z) 18.41). Recalling the solutions y±{z) which we defined with Eq.9. hence we have to .18.48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18.e.50) and y+(0) ^ 0.5. 18.yc{z)pansion (18. ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18.1 ) e 12c 393 Kgl)]^"1) l + O h? (18..47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq.21) as even and odd about z = 0. (18. and the second qo = 3.11. The first of the conditions (18. Thus the boundary conditions to be imposed there are the same as in the case of the harmonic oscillator for alternately even and odd wave functions. we see that in their common domain of validity yA(z) where a = =Vc(z). — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0.. i.
 ( 9 + 3)]! s i n { . e.s i n { .g. 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1. + 1)]!' D ' j . IT 4 dio In fact.394 CHAPTER 18.u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(.( g + 3)}.2: Evaluation of yB(0). [ . one finds that [ 1 IT (q + 1)1! = ./o^Tpif (q+i) . M.(0) + ^ ( 0 ) l z—>0 2.SxW) = \ i»fl(0) . Expressions for C 9 (0).56) D 4(. Example 18. 1 [7(93)]! n «(°) = ..52) # 1 = ^ y' c (0) a and 2/B(0) = a (18.1)('») + ^ r + l)]! ^.(iu)].58) diu [  ( g .2. Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i. y'B(0). we obtain i V7T24( g .^{(9+1)}.F ? .i)H = e.54) dio q(w) . A.! ( .51) and 0 = V(0) = Jim i»xM .(0) (18. C. (18.l)(°) Thus with the help of the reflection formula cited above: D B. Stegun [1].(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)}.l ) ] ! [ . with the reflection formula (—z)\{z — 1)! = IT/ simrz.i ( g + 3)]l (18. y c ( 0 ) .fc(0) Thus we obtain the equations (18.55) Solution: Prom the literature.( g + 3)}.53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin.) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18.. Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin.(0) = .(0) = and htoi)(°) ^—'.' '.."*(«!) ^ ' ~ [ . . We leave the detailed calculations to Example 18. Abramowitz and I..(«.i ) .57) [i( 9 _l)] ! 2 i(9i) B. = = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18. [C. Z (18.5 ( 9 + !!)(*"')• i ( 1) .oi(. — C.
W .sin{^( 9 . (18.46). Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g . using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z . + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(.3)}. 4 r T (18.56).(0) = *>i(„+i)(0> 2 .i .sin{^(g+ 1)} = J .i ( « + i ) [ _ i ( . ~ ^ 1 5(9+1) 2i(''. _ l)]!2l(91)(/J2)(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„. (18.60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18.2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18. we obtain 2 i(»+i)[_i(.49).+1iH]«=o = .62) .6 ) } ^ — i t a n < — (q + 3) >. + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}.53) in dominant order and insert the appropriate expressions for a and a from Eqs.1/2)!.+i)(°) [ _ I ( . A __f7r \4 'J We rewrite the right hand side of the derivative equation again with the help of the inversion and duplication formulas and obtain _ (^)g/4(_l)j(g+l) e /> 8 /6^ _ p (/i4)g/4(_1)(g+l) _^_ . isin{f(g + 3 . 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs.18._l) From this we derive C. (18.3)} (h?)faV[\(q [Uq U + l)]\2*(q+V _£ e 6c^.)[i(g3)]! and ^i(.
With a Taylor expansion about go the left hand side of (18.63) and (18. (18.64) the right hand side is an exponentially small quantity..64) In each of Eqs..396 CHAPTER 18.. Anharmonic Oscillator Potentials Then the derivative relation of Eqs. Thus we have to determine these conditions first. (18.i .11.9. This is Case (1) of Fig.64) about go — 3. 18. .63) Proceeding similarly with the second of relations (18. this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions. .^ . we obtain cos((g + 3)) = ..5.65) Expanding similarly the left hand side of Eq. + 3)} = ..63) vanishes for q = go = 1. . (gg 0 )^±^ y e5?. In fact the left hand side of (18.64) for q = g0 = 3. i. (18.9o) ^ cos  .9.2.1.63) becomes (? . the vanishing of the wave functions at infinity).5.2) with potential (18..53). 7 . z^ ein/2z._ e =*.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..e.( g 0 + 3)  + • • • ~ (g ..L ^ i . .. equivalently. Recall > the original Schrodinger equation (18. For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo. z2 > z2. (18. by the rotations E > einE = E. 18. (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.go) ^ (1) It follows that we obtain for the even function with q — go = 1.^ >^_>m e J*. and the left hand side of (18. 1 1 ..1). ^ .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. we again obtain (18. (18. 7. The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or. the energy E is real..53) becomes 8 m(I(.
y{z)~exv\i[) 7T dV'V. we set z w r = \z\eie. 6 6 In the case of the inverted double well potential under consideration here (i. or — < 6 < . in the domain — TT/2 < 36 < > 7r/2.68) .cos 36. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — .18.e.66) z3 = \z\3em. 1/2.e.+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18. 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. 2 c2>0. —}.e.e.2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations. the resulting wave functions vanish at infinity and thus are square integrable. Now.67) Rotating z by vr/2. we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0. r / c 2x 1/2^3 = exp{ ± i — —\. i.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0. we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase. (18. Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s. the case of complex E). replacing sin 30 by " 0 + ^) 1 IT = . It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling. (18. i. Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36.e. Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied. if TT/3 < 6 < 0. i.
(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.Z\. Looking at Fig. We do this extension with the help of WKB solutions. We therefore explain our procedure first. Our procedure now is to continue the even and odd solutions (18.e. and there impose the boundary condition (18. i. This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator.3 The inverted double well potential with turning points ZQ. 18. We have to remember that we have various branches of the solutions y(z) in different domains of z. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0. (18. 18. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq.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.398 CHAPTER 18. Eq. Anharmonic Oscillator Potentials V(z) Fig.68) for c 2 > 0. This is the boundary condition also used by Bender and Wu [18]. Thus we have to match the solutions of type A first to solutions to the left of z\ and then extend these to solutions to the right. we match the WKB solutions . for z > Too.21) to + infinity and to demand that they satisfy the condition (18.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero).Z\. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure.21)) were defined in terms of solutions of type A which have a wide domain of validity.
f R .4.B. Vol. (22) or A. 6.e.4) for E and ignoring nondominant terms) t o .71) We now come to the algebra of evaluating the integrals in the above solutions. . 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18. The distant turning point at z\ as indicated in Fig. z2h4/A > c2z4/2.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i. I [195]. equations (21). Dingle [70].69) The W K B solutions have been discussed in Chapter 14. and then use the W K B procedure (called "linear matching" across the turning point) to obtain the dominant W K B solutions beyond z\. ^ 2 4 •K + —(r.. (18.3 is given by (using Eq. ^2? 1 2qc2 /2? . B. 291. t To the right of the turning point at z\ these solutions match on to (r. Messiah. 291. 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 .^ T rz\ 1 + yh* .e.70).2. *R.4 h + c z 2 1/4 < . Dingle [70]. 18. In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken.2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A.z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ . A J24 1 ? 2 • "^ •z h \—c z ~ — q h . p. 4 2 T ' i. (18. p. z\ ~ .^ 4 1/2 + icV 7T Jz + (18.*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) . We begin with the exponential factors occurring in Eqs. i. Sec.18.70) where z < z\.zi)/ \ 1/4 ywKB\ ) z — g/i* . h2 (18.
(18.73) .72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf. Thus the above factor yields '2c2\l/2 h4 so that (cf.i c V 8c T 2 1/2* ti>_(2 ~ exp 8c \3 2 r< 2cV 2c2 z2 i24£Y'2u h* 3/2 >.40)). Anharmonic Oscillator Potentials E± = exp = exp . (18.2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ . Eq. (18. CHAPTER 18. zi 2q 1 c 2 [1 .40)) E± = exp _/^_2 f ± In A 2c2 . We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\). (18. In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c.211/2 1 \ /j.±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18.4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq.^ ] V 2 and so /^2f 8c2 3 1 2cVl3/2 h* f q fZ\ ck h^y/ E± = exp T 2 jJ Z ^ [ g ..^ i ^ ^ c 2 .2.2 \ 1/2 .\qh2 + \z2h± . .2. Eq.^ I 2 2 1/2 zV = F 8c2 Jz [i .e.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.
2Z 4 C S ' + ( ± ) e x p I i\ cz / cz3 V3\/2 U^2 h6 12c 2 12c2 J (18.76b) apart from factors [ l + 0 ( l / / i 2 ) ] . we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .20b). > » Returning to the even and odd solutions defined by Eqs.18.Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18.77) Now in the domain 2 — oo we have > \qh2 .71) and these into (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) (18.79) +S_(±)exp . h2 — — h2 is maintained. CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18.74) Comparing these solutions now with the solutions (18.(J*i) VA 0 ) = PVwKB (Z C.20a) and (18.\z2h4 + J Z\ nl/2 \M fZ .77) we can rewrite the even and odd solutions for 9te —• oo as (by separating cosine and sine into their exponential components) 1/4 y±{z) .*l) (18.76a) P P r 2/1 2 " 9 ( . (18.1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18. In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q.75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18.21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J.78) Inserting this into the solutions (18.21 1/2 .
i.^ § ( 4 g 2 + 29) + O ( ^ ) . Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq./i2) = E{qQ.3.65). (18. (18.76a) for (3 and /3. (18.h2) + {qqo)(^pj = E(q0. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( .. we see that we have to demand that 5 + ( ± ) = 0../ ? = 0. (18.6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues.e. E(q. 18.h2) + (qqQ)h + (18.68). Eq. (18. i / 3 ± .82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1.80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq.82b) This is our second condition along with Eq. (18. (18.85) + .5.^ j ..e.34). i.2. (18. .402 CHAPTER 18. this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J ...81) Inserting expressions (18.^(q2 + 1) . h2) = \qh2 .84) Expanding about q = qo we obtain £(g.
2 .h6/6c2 E = E(q0. * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/. Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0. S W K B ( 2 ) = PVA(z). h2).h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o .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. i. o 2 + l)^(4.e. In the above we did not require the matching of WKB solutions at the lower turning point ZQ to the solutions yA(Z).yA{z).~p{q. Example 18. h2) of the relations KWKB (Z) =PVA(Z).3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q.l ) ] ! > + = ^ 2 .2 The Inverted Double Well Potential 403 Clearly the expression for (q — go) has to be inserted here giving in the dominant approximation / h6 \ <?o/2 . where the superscript {r. zo} means "to the right of the turning point ZQ" . [19]) for ft = 1 and in their notation 90 = 2 ^ + 1.18.4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z.g ( . 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.g. ^ 2 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 . in comparison with the work of Bender and Wu [19]) we deal with this in Example 18.86) will be dealt with in Chapter 20. Since this is not without interest (e.3.ZQ) .36) of their Ref. above) (1. ft6 ^ = e. !*WKB(Z) ~ 2 . The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18.
J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20. and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 . Expanding the square root occurring in ^ W K B W ' ^ W K B ^ ) ' Igft2_I22h4+lc2z4 2 4 2 and we see that 1/2 = 7 ^ 1/2 2. N. cf.g In other words.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. Anharmonic Oscillator Potentials where zo > z. f x Zn 4) h2 2.zo)/ \ qh2 2 ZO + z2hA 4 2 . the (approximate) solutions* 1/2 exp • ± i ( ../ ^ 2 X 1 / 2 exp(ifi.O.] { hz combined with the particular constants contained in y ^ ' ^ g . 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.404 CHAPTER 18.ic224 2 4 qh2 + i^fe 4 .zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e . .1/2 h? \z{z zlf' \z \n\z 2 2 2 + {z2z2Y/2\ 2q h2 4 • In (V2zh V s/q Then to the same order of approximation . On the other hand in the exponentially behaving W K B solutions the quartic interaction term acts as a correction to the harmonic term close to 20. Schwind [247]. 2 z 2 ) / 2 e \ ." / 4 hj (ft2 2 2)( 9 +l)/4 J'W'KB (Z) 2\1/2(fe222)^1)/4 h) exp(±h2z2) ^2ey/4 * Solutions of this type (which are asymptotic forms for q — 00) have been investigated in t h e > literature.
We can reexpress these relations with the help of Stirling's formula.89) . as in the case of the Mathieu equation.3.1 The Double Well Potential Defining t h e p r o b l e m In dealing with the case of the symmetric double well potential._ . We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0. h4 > 0. In this form the WKB solutions reveal their similarity with the solutions VB i VB > Vc i Vc a n c ' demonstrate that the factors which we inserted (e. in Eq.J ' i J .kyil(. (18. The last expression can be made acceptable for q = odd integers by applying Stirling's formula in the denominator of the second of the previous pair of expressions. (j .23)) appear quite naturally. Then . ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18.87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(. 4" ) (h?z2)^q+1) for q j£ odd integers.3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3.z ± ) = .^ 1 .g. J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e . (18.yA{z) with expressions 27WKB i J'WKB ' that the factors p.3 18.88) + [EV(z))y{z)=0 (18.13) of Bender and Wu [19].5 "l _ o f .18." / 4 ( ? « / 4 .e.zo). But there are significant differences. „ / M 1 / 2 i .DfJ»v we see Comparing now the expressions (18.41) for yA(Z). i.1)! = ( 2 7 r ) 1 / 2 e . we shall employ basically the same procedure as above or. (18.(r. in fact..
the previous equation (18.V(z±) l{z z±fhA + 0[(z .95a) .z±). Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2.406 CHAPTER 18.94) (18.93) w± = h±(z .i>5.90) V(i)(^)=0. (18. . (18.z±) y = 0. 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 . vW(z±) = ±Qch2.92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^.4 The double well potential.91) (18. (18. (18.z Fig.91) becomes T>q±(w±)y(w±) = o(^)y.
1.e.2 T h r e e pairs of solutions We define our first pair of solutions y(z) as solutions with the proportionality y(z) = exp \h\ f uV2{z)dz (18.18. Eq.. 18.2. Thus again.3.97b) Our basic equation.98) Evaluating the exponential we can define these as the pair yA{z) VA(Z) = A(z) eyip = A(z)exp + —z V2\3 4c'' Ac' (18. (18.. (18.95b) with the equation of parabolic cylinder functions u(z) = D„(z). qo = 2n + l.93) for E into Eq.95a) and (18. where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)]. is again seen to be invariant under a change of sign of z.96).i 4 f W » = o. we conclude that in the dominant approximation q± is an odd integer. (18.95b) By comparison of Eqs.96) dz2 + 5 « 4 +  . we obtain another by replacing z by —z. d2u{z dz2 + v+ 2~r u(z) = 0. (18... given one solution. we obtain d*y (18.3 The Double Well Potential where ^ 407 d2 •« 2 2  d8. (18. (18. i. n = 0.99) 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) = 0.100) . Inserting the expression (18.97a) U(z) = (zz±)2 + 0[(zz±)3}.87).
z) = yA{q.101b) To a first approximation for large h2 we can neglect the right hand sides.101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18. (18. (18. h2 h4 2c' with g+ = q. yA(q. Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —. We leave the calculation to Example 18. z) may be obtained from the solution yA{q. ~h2.102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q. (18.^ A"{z) + 2c V2 *A(z) (18.408 Since CHAPTER 18. The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0. (18. i.101b). Aq(z) = Aq(—z) = Aq(z).101a). h2.106) . h2.^q(l7q2 6 4/i + 19) + (18. (18. n± = V2h2.h2.e.105) "8 + 4 ^ .c W + 1) . h2.1 6 ) .102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18. Integration of Eq. i.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c).h2) .c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t . z) = yA{~q.e. we observe that the solution y^iq.4. but the solution is a different one. these equations can be rewritten as + (qz+z)A{z) = .104) Both solutions yA{z).yA(z) are associated with one and the same expansion for A and hence E. Vq = {z2+z2)— + (qz+z). z). 8hw (18.103) Looking at Eqs. The result is given by A and E(q.
± 1 . z • .. .107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z .e.103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials. <4q+2 — z — z+ Aq+2iAq+2j (18. (18.*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18. J = 0.109) We observe some properties of the function Aq{z) given by Eq.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 .114a) . (18.112) From these we obtain. We know the first derivative of Aq from Eq. First. we reexpress A'q in terms of functions of z multiplied by Aq. such as periodic potentials. Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (18.qz+)2 . however.106).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.108) We wish to rewrite this expression as a sum y coefficientiv4. ± 2 .qz+) [2z2 . (18.101a) with A replaced by A .110) Aq+2+Aq2 9+2 A.z\) + 2z(z .111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 . as a linear combination of terms Aq+2i.3 The Double Well Potential 409 Example 18.103): . .102).z2 ) N t. . (18. Solution: The structure of the solution (18. . + Aq+2 + Aq Aq+2 .Aq 1 + 2^± 2 Aq+2 (^M^f (18.113) 2 . (18. The first such step would be to reexpress the right hand side of Eq. .18.2 z\) • Aq$4 £Aq + Aq—4. (18.105) and (18. .4zz+q+ z\{q2 + 1)].{z2 .+ 2 ^ ± l + 2^±i + . = Aq+2i+2jAq.(l + + 2  z y* . for instance by componendo z z et dividendo. 92 ZZJf (18. A Differentiation yields » = ^^w (18. i. Thus it is natural to explore analogous steps.
Anharmonic Oscillator Potentials A 1 .410 and from this = Hence with Eq.l ) ( g .114b) . A(°~) = Aq.2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18. we obtain 5 K = ( ^ f ) [ ( 9 .115) and (18.2\2 (z* .119) It is now clear how the calculation of higher order contributions proceeds in our standard way. (18. q) = 0. In particular the dominant approximation of A is obtained by setting (q.4]j.q + 2)Aq+2 (18.z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 .e. (17.q)Aq + (q.9=F4) = (g=Fl)(?=F3). q + 4)Aq+4 + (q.115) (z*z%) 2 i2 (z 2 .112).117) 4(q + l)2Aq+2 Here the first approximation of A. (18. (q.2Aq + Aq4] 1. QA?+4 _ 10 A + 6 .101a) the contribution R . (q. _ 4 .116) [Ag4 .34). cf.116) into Eq. where the lowest coefficients have been determined above as (9. (18.108). (18. Eqs. i. ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential. Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+.+6 + • • • ] .4 ( g .33) and (17.<? +{q. leaves uncompensated on the right hand side of Eq.4A ? _2 + 6Aq .118) 2s/2c ~^j[(?.aAg+S 12+ 16 (18.q + 2) = 4{q + l ) 2 . 1_4^±^+8^±±12^±^ +.q + 6)A<.l ) 2 A .4)A„_ 4 + (q.(0) V2 ' 2c K + h4+Ao . _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4. .z\Y Z2 + {~)\Aq+42Aq 1 224 Aq.4 4+* + . .112) CHAPTER 18. <?) = 6(<j2 + 1) + z: 2 2A +72"' (18. (18.3 ) A . Since VqAq = 0.q2)Aq2 + (q. (18. •••].2A g + A g _ 4 ] .4Aq+2 + Aq+A Inserting (18.
y'+(q. y'_(q.(0).105). i. (g. (18.2 ) (0) (18.0)=0. (i) _ / V2c\ [ ( g .h2.122) The solutions yA(z). g + 4)(g + 4. z)\ = 2^A^ h2 '>z)± VA(Q. where Rq is the sum of terms left uncompensated by AW. h2.94) 8z+ (9. 9 + 4) ^+48z+ (18.101a) the contribution Rq . A' (0) = q/z\) y+(q.94)(g4.4 ) ( 0 ) ( g .h2.g2)(g2.0) = ^ . g . h2.h2.yA(z) derived above are valid around z — 0. h2 2y^ Age? h4 (18.3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except. g) 8z 4 . of course.h2.e.121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g. (18.124) .Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9.z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A. Terms jj.g) 4z+ {q q)+ 2r (9.g ( 1 7 g 2 + 19).g + 2)(g + 2.e.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.{^ h2. i. for i = 0. (18.0)=0. *)] (18. \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A.92) ig4 ' H ~ 23/X4 iz+ An2 H + (9.g) Az.g) 8z+ (g. in the domains away from the minima.z)±yA(q.123) ^[yA(q. g . 9 + 2) Aq+2Az+ (9.h2. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g .28.0) = y^(q. thus yielding the next approximation of A as given in Eq. The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq.
412 CHAPTER 18. 18.93) and setting w± = h±(z — z±). (18. (18. (18. This means. Inserting (18. (18. and another VB(Z) ± 25^ch3wl + c2wi .yB(z) is obtained around a minimum of the potential. Bq[w±{ ^ .2A y(w±).91). yB(z). Since w±(—z) = —h±(z + z±).23) with appropriate change of parameters to those of the present case. but w±(z±) = 0.95a) and (18. in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum. in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq. (18.95b) that the solution there is of parabolic cylinder type.^ % [i(?l)]!24(«. (18. We see already from Eqs. Moreover.Evaluating the exponential factor contained in VA{Z) of Eq. 2(9+) .125b) It is clear that correspondingly we have solutions yc(z).yc(z) with complex variables and Cq(w) given by Eq.3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z).95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2). Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2).1 ) D (18. we > . Anharmonic Oscillator Potentials Our second pair of solutions.126a) hjW + l)]! VM = Vc(z) = Cq[w±(z)] (18.yc(z) providing a pair of decreasing asymptotic solutions there (or correspondingly y~B(z)iyc(z)). + 0(hl2).99) for z — z±.125a) = VB{~Z) = Bq[w±(z)] + 0(h±2) = Bq[w±(z)] + 0(h±2) (18.126b) These are solutions again around a minimum and with yB(z).3. the equation is — with differential operator T>q as defined by Eq. we have w±(—z±) = —2h±z±.We draw attention to two additional points.
97b) ~ exp . . (18. 24(9D[i(gi)]! hj (18.3 The Double Well Potential (cf.• • " . (18.^/w + ± 1 .« + ) 2 (9+1) e (18. . (18.129) . We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4]. . .. Allowing z to approach z+ in the solution IJA(Z). and comparing with Eq.128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e. . a= ^ . 2 An+z+e 4w+ (18.97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18. Consequently the above integral from z = 0 to z± differs from that from z = 0 to a turning point (as in expression /2(C)) of Eq. Eqs. (18. (18.^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq.164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18. Later we calculate the coordinates of turning points ZQ. ein+z+e in+(z 22+(<?+i) .127) Recalling that around arg w± ~ 0.172) obtained later. (18.172)) we add here comments on this approximation.frW.18.147)).z±)dz .125a) we see that (considering only dominant contributions) in their common domain of validity e ^+z+ l + O 1 yA{z) = yB(z). we have \z ~ Z+\^Q ^ l f e 2 2 2 _kh2( ~ i .i*l(«^ 4'"+^+ 4"'+v ( « .146). Thus for h very large these turning points are very close to the minimum at z+. Here z± = ±h?/2c are the positions of the minima of the potential.• )2 (^)itoi) ( 2 z + ) 5 ( 9 + i )e ' yA(z) z+> lft2 2 _ I .ZI and find that these are given by z+ + 0(l/h) for finite q (cf. the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4.
Thus 1 E . de Deus [66] and W. Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(z_. The involvement of these WKB solutions leads to problems. Bhattacharya and A. K.\h\)^+l\\{q l + l)]\ 1+ 0 J_ .414 and CHAPTER 18. g .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. but it is clear that none of the above solutions can be used at the top of the central barrier. even above the turning points.131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima. and this means at both minima. We have to impose boundary conditions at the minima and at the origin. Bhattacharya [23]. Since it is more probable to find a particle in the region of a minimum than elsewhere. D. we naturally expect the wave function there to be similar to that of the harmonic oscillator. The pair of solutions of type B is defined around argz = 0. The solutions in terms of parabolic cylinder functions have a wide range of validity. R. 18. K. Furry [102] .130) Therefore in their common domain of validity a a = dte+vM' {2z+)^.(18. S. Rau [22]. since. 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. The next aspect to be considered is that of boundary conditions.+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z . and are both in their parameter dependence asymptotically decreasing there and permit us therefore to define the extensions of the solutions y± which are respectively even and odd about z = 0 to the minima. S. P.3. J. H.7r and the solutions of type C around argz = ±7r/2.z+)]fr+V ' (18. Various investigations^ therefore struggle to overcome this to a good approximation. Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point. basically. they assume large quantum numbers.
y'_(z±) = 0. We have y±(z±) = ^VA{z)±yA{z)]z^. y(z±)=0. We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0.132). 18. However. and y'±(z±) V'B(Z±) ± =y'c(z±) (18.Zii yB{z±) a ± a yciz±) (18. (18. as indicated in Fig. ^(*±)#0 (18.135) Hence the conditions (18.136) .133) y+(z±)^0.5 Behaviour of fundamental wave functions.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18.3 The Double Well Potential 415 the most basic solution would be even with maxima at z±. as indicated in Fig.132) Fig. y{z±) £ 0.5. (18.134) y'+{z±) ± 0. an even wave function can also pass through zero at these points. and y+(z±) = 0. 18. At dtz — ±oo > we require the wave functions to vanish so that they are square integrable. 18. as indicated there.18. The odd wave function then exhibits a correspondingly opposite behaviour.5.
. Thus sin {^ +1) } = ^^>^30fe^i.134) this equation can be written Now sin  ^ ( 9 + 1 ) 1 oi sinl j(q0 +I) \ + ^(qq0) ~ (_l)^(9o+i) ( g _ g o ) for cos I ^(q0+ !)[ + ••• 50 = 3 . (18. (18.«*{=(.5.2 we can rewrite the second of (18.416 CHAPTER 18. also (18.141) and cos<J^( 9 + l)> 4 ~ cos  ^ ( g o + l)  ~^(qqo) (_l)*fo>i)( g _ g o )?[ 4 for sin   ( g 0 + 1) J + go = i. ..54) +l j 2^[I(Ql)]![I(g3)]!sin{f(g+l)} ^^ ^ r i r ( T(y+me'm' (18 137)  where we used first the reflection formula and then the duplication formula. 1 1 .. (18. Eqs.138) Using formulae derived in Example 18. _ 8)} s . . . (18. Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs. 7 .136) as  M ..9.136) the dominant approximations we obtain (cf.142) (<Z9o)J(l)^°+1)(l)1/2 . (1.m Using Eq.„*{!<. +1)} = 4 m.
» ^ B ^ ' .3. We do this with the help of WKB solutions.1 " <813 1 4)  In Example 18. we matched them to the solutions of types B and C which are valid there and hence permit the imposition of boundary conditions at the minima.144) Thus we require the extension of our solutions to the region around the local maximum at z = 0.z\.18. We emphasize: We needed the solutions of type A for the definition of even and odd solutions. we must also demand this behaviour here along with a nonvanishing Wronskian. y'+(0) = 0. y+(0)^0. V(z) — • z E=V Fig.3 The Double Well Potential Thus altogether we obtain 417 ^ • . We deduce . 18.5 (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.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. 18.6 Turning points z$. Since the type A solutions are not valid at the minima. y'_(0)^0. (18. Hence we have to impose at z — 0 the conditions j/_(0) = 0.
We also note that the height of the potential at the turning points is h8 qh V(z) 20.144). (B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18. As a consequence.^1 25c2 + V2 2 Thus again we see that for large values of h2 the turning points are very close to the minima of the potential for nonasymptotically large values of q. in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0. i.146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18. one finds that . to the left of ZQ where V > E.148) ("f * "The superscript (I. (18. h8 25c2 (18.418 CHAPTER 18. In the domain 0 < z < ZQ.\qh\ + \h%U{z) (18.e.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i. the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ qh\ x exp + h%U{z) .145) \qh\ + \zW Using z+ = h2 /2c.24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18.147) Since the minima are at z± — h2 /2c with /i 2 very large. we see that the turning points ZQ and z\ are very close to a minimum for reasonable values of q ~ go> the latter being an odd integer.e. + £\h%U(z)=Q. ZQ) means "to the left of 20" . Anharmonic Oscillator Potentials from Eq.
i (2q ln (18. Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1.S2 ) dz rzo 1/2 'Ihl + ^Uiz) I 1/2 / Jz dz (z~z+)  w + 1/2 1 = ±lh+ 1/2 ZQZ+ Z — Z+ 2 i^$) 1/2 ~4 >'^. (18. 20 dz \qh\ + \h%U{z) .^ e therefore have to consider the exponential factors occurring in (18.148) and (18.^ + ) 2 + igln2(zz+).3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I.149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+.(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ . . (18.149) In order to be able to extend the even and odd solutions to z = 0. )+^*+)2lr} rnD h\z'2)}z Q )\ = . 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) .^ j O e .y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) . we have to match yA(z).150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z .* 4+ .151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing.18.
In a corresponding manner — i. using Stirling's formula (and not the inversion relation) — we have —(1. Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr.126a). We see therefore.154) .420 CHAPTER 18. since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q.125a). However. Since correspondingly 1/4 2 l/2 \*% we obtain + 4/1W*) (*^)i/'(M)i/*' 27T 1 / 2 {h\)^{\{q1)]\\2"+ *+ 1(91) ~hi l2 q/A e \{zz+)*hl for \z — z+\ > m 2q \ (1 ' (18. the results necessarily require adjustment or normalization there in the q—dependence. This is the aspect investigated by Furry [102]. l/2 ~h\U{z \(zz+?h\ (18.ZQ) . The only way to relate these solutions is with the help of the Stirling formula which converts t h e product or ratio of such factors into factorials such as those inserted from the beginning into the unperturbed solutions (18.125a) and (18.153) This expression is valid to the left of the turning point at ZQ above the minimum at z+. of course.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. \ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (18.e. 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. we can write the exponential as exp (27T) ("f 1/2 I dz Z+ 1 qh\ + q/A 1. t h a t the Stirling approximation is amazingly good even for small values of the argument). 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.
155) (2z. we obtain in leading order (i. The relation (18.e. 27T1/2 9/4 1VA{Z).K?+1).143) which was obtained with our perturbation solutions from the boundary conditions at the minimum. . 7 of t h e matching relations y^NKB\z) i.158) Applying the boundary conditions (18. (18.159) 7 i(«i) r(9^3) ^2~ j(9l) .123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18.157) is used in Example 18.99). (18.3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large.e. multiplied by (1 + 0(\/h\))) the proportionality constants 7 .144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18.154) with the typeA solutions (18. [\{q — 3)]! are really correct replacements of [\q}\ only for q large. linearly matched) W K B solutions. and is shown to reproduce correctly the result (18. this result is somewhat imprecise.n('>2o) 3/WKB(Z) =1VA{Z)I (18. it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions. However.103) and (18. \hl.143) T^^*)) 27T (18.e.156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4. . Returning to the even and odd solutions (18. (18. Comparing for z —• z+ the W K B solutions (18.104). as the results also seem to support.18.153).5 for the calculation of the tunneling deviation q — go by using the usual (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. (18.157) Since the factorials [^(q — 1)]!.
Friedman [40].148). (18. 60 and 361.160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18.149) near z = 0.164) / 2 (0) z ca [(1 + k2)E(k) .3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18. (18. F.162) T h e integral appearing here is an elliptic integral which can be looked up in Tables. 3G2 12V2c ^See P. . D..163b) The integral hi?) .. . formulae 220. p.163a) (18. G2 K [i 4.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.161) we can rewrite the integral as r I2{z) = = dz^{a2 .z2)(b2  z2). p.05.19. Byrd and M. We have 1/2 dz Jz \qh\ + \h\U{z) dz Jz 4 ^? lfh 2\22c dz ¥+ A4 l 4 3 1/2 1/2 dz Jz — cz h4 CZ qh\ dz4 qh\ <Wc h+ 1/2 Jz zo 2V*\Vc h 4 ) T Jz 2\/h ~~2 /2(^2 —pz I V2 dz ^?^ h+ Z 4? + ^f ^2<«2 (18. n + uY 2c „ ^6 8V2c2 (18.iV2 ! i . 213.422 CHAPTER 18. K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\. Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18..k'/2.
r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 . /2n + l 2 ± — .166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o . ± ..e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref.. where the result is shown to be (with q ~ q0 = 2 n + l .( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n .3 )l ! ..1)]' Gfo. .§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs. Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 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 .± ' 3G 2 ' T ( ^ (18.( ft 7. . 1.1.1 " e " O T (18 169)  Y^V / 4 tfo. . (G\ . n = 0. n. [167]. [4] misses an n—dependent power of 2 in the result.2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( . ^ ^ . .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 \. which implies small G2.2. The nontrivial expansions are derived in Example 5.165) in agreement with Ref..
. (18.^ W ^ / 4 t o . 18. 2 .173). . (18.»W. Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18. h2) the split expressions 29o+1/i2(—W2 h6 (18.143) with qo — 2n + l. (i*™) We obtain the energy E(q... (18.h2) + (qq0)^=.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.175) ~E0(q0.174) . . 18. and obtain ( * .172) into Eq. (18.3.3.i ) ] ! e ~ s A ' .6 Eigenvalues and level splitting We now insert the replacement (18.157).172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z.106). 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.h2) and hence the splitting of asymptotically degenerate energy levels. we can therefore impose the boundary conditions at z = 0 by making in Eq. (18.3 concerning the exponentials.n = 0..1.. (18.424 CHAPTER 18. Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs. with the help of Eq. .143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A. Inserting the expressions for h+ and z+ in terms of h. we see that the relation is really an identity (the preexponential factors on the left and on the right being equal) with the exponential on the left being an approximation of the exponential on the right (which contains the full action of the instanton). (18. we obtain for E(q0.
h2)E+(qQ. 1 u2 c 2 (3 9 2 + l) ^ « b ..176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB. which evidently supplies some correction terms.e. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k. K.V2^\ \n+ i .3 The Double Well Potential where £0(<7o.106)..e. .18. The result (18.^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 .178) "In S.7 =( e N n+1 .2 = 2 . can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i.2. c 2 /2 <> A — given as E0 = 4\ h(2fc) 1 / 2 o y J H q2 4/c > ^— + q ^ 2s c2 V2 . T \ *n ! 1 ' (18.fr2) is given by Eq.^ 2^ h* . (18.e.143) with (18. (18.h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1. for finite h2..174) and (18.q2 7. 2h4' i.e.n = 0. i n U n / ! Thus AE(q0. the difference between the eigenenergies of even and odd states with (here) qo = 2n+l.1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^. i. Combining Eqs.. 1 7 425 2 . in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency.h2) = E_(q0. the pure WKB result of Banerjee and Bhatnagar [14] (i.e. the level splitting. the following expression for large q ~ 2n + 1.1. i. that without the use of Stirling's formula and so left in terms of e and nn) being this divided by the Furry factor f l / » : = \ .157).
(24) there is k 2 (i0n+i3)/4 //c3/2\n+1/2 V2fc3/2 6XP n\ ft \/nn\ in agreement with AE(q0.** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2). (18. we would have obtained the result with E. K.176) together with Eq." Of course. (66). as is also explained by Bhattacharya [23].6) to yield an improvement of WKB results for small values of n. the Furry factor corrected WKB result follows automatically. The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation.3)).4/4 here.h2 <> — h2. (18. 2/z4 and 2c2 respectively (see Eq.181) "W. h = 1.2X2 V(z) = \z2tL) . supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn. h4 and c2 replaced by 2E. and harmonic oscillator frequency equal to 1 given by \/y/2~n and independent of n as explained by Furry [102] — see also Example 18.180) h8 E0(qo. Furry [102].178). H. ** Comparing the present work with that of S. and A there with c 2 / 2 here. .h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o».. h2)\mo=i = . Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0.2 / 7. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2. Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula.^ 2 + ^h2 If in addition the potential is written in the form A / . we identity the parameter k there with ft.6 \9o/2 e~h&l^\ with (m 0 = l) (18. This. We see therefore. A>0.h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq. Eq. Bhattacharya [23]. (18. that if this symmetry is taken into account from the very beginning.426 CHAPTER 18. however. Then the splitting A ^ W B of Eq. as we do here. (18. these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q. 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.
11). D.88) h4 = 2/x2. MiillerKirsten [186].15). Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1. 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. (2.181) therefore implies the correspondence rj1 <> A/2.18. P. K. Lee [98]. ZinnJustin and U. Liang and H.3. first paper. .* A similar correspondence is also found in the case of the cosine potential. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions. and g2 is given as g2 = r]2/m3. these are presumably not of much interest in physics (for reasons explained in Chapter 20). (18.^ *See E. §J.3 The Double Well Potential 427 a form frequently used in field theoretic applications. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation. Bhatnagar [14].h>) * 2V2^^^y/2e^3/3A.§ 18. See R. Equation (18. M. H. D.34) and (E. Friedberg and T.180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods.W. The comparison with Eq. M. (18. ^To help the comparison note that in J. J. D. Gildener and A.—Q. Jentschura [293]. formula (4. ZinnJustin and U. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions. J. m 2 <• p 2 / 2 . Jentschura [293]. Patrasciociu [110].c2 = A/2.182) implies AlE(l.^ Thus for the case of the ground state Eq. so that by comparison with Eq.7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator. (18. J. W. Mansour and H. Eqs. however. We considered above only the symmetric twominima potential. Banerjee and S. 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. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26.
In fact. 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. [132] and Loeffel. Mahapatra. can be found in an article by Coleman [54]. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. though incomplete list of references in this direction has been given by Garg [105] beginning with the wellknown though nonexplicit (and hence not really useful) ground state formula in the book of Landau and Lifshitz [156]. There is no reason to view the anharmonic oscillator differently.[131]. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. in both symmetric and asymmetric form. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle. 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. 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]. for instance. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. The double well potential. Very illuminating discussions of double wells and periodic potentials. P. 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]. Ashbaugh and Harrell [12] and Harrell [130]. A reasonable. Martin.g. though not exclusively numerical. mostly in connection with instantons. Pradhan [182]. Santi and N. we cite papers of the Uppsala group of Froman and Froman [100]. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations.428 CHAPTER 18. As references in this direction. B. B. . as he discusses. like those for anharmonic oscillators. [19] — that the expansions considered above are asymptotic. 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. Perturbation theoretical aspects are "See e. Simon and Wightman [178]. has also been the subject of numerous numerical studies.
Saxena. 1+ u : Gy/2q. T h e r e f o r e it is convenient t o deal w i t h this here.\G\ + • and 1k2 =2Gv/2q4. q ~ 2n + 1.18. Biswas.3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14]. Example 18.2G^2q + 4G2q . Cooper and Strottman [221]. Bender.^ + • (18.184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 . Srivastava and Varma [24] and Bhatacharya and Rau [23].G2q + . (18..190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18. Datta.187) We now reexpress various quantities in terms of u. 2 a. [122] as 13 1. (18. which is assumed to be small.188) T h e following e x p r e s s i o n a p p e a r s f r e q u e n t l y in t h e e x p a n s i o n s of elliptic i n t e g r a l s . 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.192) 'F 12 16 .191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' .u)1'2 = V2u{ 1 . Thus k' and hence k' = v ^ ( l . k! (18. We obtain the expansions from Ref. Gutschick.185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±. We have 4 ~k' 2u(l .5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'.183) S o l u t i o n : We have k2=1^./4 E{k) = 1 + • l n In (18.186) (18. X +2 + ' ^ (18. + (G 3G2 3^ + 2lnU 2 + i(2n + l)ln(f)+I(2n ~~ 3G 2 ' 2 l)ln(^±i q 2 1 ^ln q/ 4 2n + l (18.189) =2u 2u2 (18. fc ~ 1 .f + • • •) 2u\ .
8 .13)} + ~u3(u .3 u .3 y .196) Now consider the last line here without the factor 2.u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l .3u(6« .u)} .u ) (^){« + 5^ 1 )} + 5.3) .u) 2 {8 .u) x2 2/1 u2(lu)2 H K 42l We can rewrite this as — Analogously we have uK(k) 6J ' — {4 + 3u(l .u)} + — u ( u . up to and including u2): u(u .u){4 + 3u(l .195) we obtain E(k) 1 + ln uK(k) .u){24 . (18.3u(6u . Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + .^ L ) H[(i _ U ){4 + 3u(l .3 . 8 + 6 u 2 .195) With (18.6w2 + 13u} + 12u 3 (« .3u(6u .« ) ^ { 2 4 .13)}.3u(6u .u){24 .6 in the denominator and in the last step pick out the lowest order terms in u (i.u) + 2}] 1 H 1 o u(u .«) + 2} + V [ .4.1)(1 .u)} u{u(l .2{«(1 .193) 2u(l . (18.1)(1 .3 « .430 and CHAPTER 18. (18.3) .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 . 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +.u 13 \ 1 ln(± k> fc'2 +  (18.18u 2 + 39M} + 12u 3 (u .1)(1 .u) + 2u} + i u 2 [ 2 + (1 .e. . 8 + 9u 2 .194) ln '^J + 2 In i/2u )+H l + 2u(l .13)} + ±u3(u .3).u)(u .1)(1 .u 3 ( u .. 1 3 « 2 = .3) 1 + In ^ .u ( l .3 u ( l .{ M 2 ( 1 .u ( l . ) J 4 (18.u){24 .13)} + 12u 3 (u .197) .a = (l«)(jl 2u/ 2 ) .{l u ( l u) / W 2 + 2} .194) and (18.u) + 2} + i u ( u .3) = = ~ .— ) .u) 2 {24 .« 2 + 3u] In [ .
l2 = ^ ( l + uf/2[E(k)uK(k)}.128) and (18. (18. V^LJ 4 2 16 ^ .198) 9 «. where w±(z) — h±(z — z±).156a) we obtain a and 7 and hence the ratio (always in the dominant order) T2~ ) w i t h h += ^ h • Since (cf.u H I ..199) We now return t o the expression on the left of Eq. (18. (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• .e. 2 .uK(fe) ~ 1 + In I — = Mu2(3u.196) and (18.197) we now obtain 4 1 • E(fc) .200) Remembering that u = Gy/2q.148) (for particle mass 1/2 and in dominant order) is given by H.4u .it)} . / to + 3GH~ \ 8 .e.J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs. + iu^Zu2)^~u+ — +0(u3) l .3). From (18. we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 .«){4 + 3u(l . this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / . (18.195) with (18.] in (18. VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V .155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z).2 l_ln(1 JL\ V {2. i.3) .18.4u .^HfHGH Example 18.196). Inserting here from the above expansions the contributions up to and including those of order u2.173) t o be evaluated. i.3 The Double Well Potential Now consider the bracket [.2 /o 2 431 (18. . Eqs.128) and (18. (18..2 (18.. ' 8G 2 q = 4 2 q ln f 26/2 \ o . [(1 .2{w(l u) + 2}] = u(3u2 .6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18.zo) HWKB normalized _ ( h* y^eM. we have.
y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji.g.35)).432 CHAPTER 18.** oo / dwD\. Oberhettinger [181]. i. (18. ° \Edzy/\EV(z) V(z)V4 VWKB. (18. Magnus and F.*o)/ ^WWKBOO (18. normalized = We see that the normalization constant is independent of a quantum number. Then ((r. and obtained as Eq. 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 [ . Example 18.203) Different from above we now continue the solutions (in the sense of linearly matched W K B solutions) across the turning point at zo in the direction of the minimum of the potential at z+.zo).204) ./ . Solution: The turning points at zo and z\ on either side of the minimum at 24.143).146) and (18.e.7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go.l)]![i(0 . Eq. (18. W. 1 7(l. 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 .147). 93.3)]!23<«D [5(91)]' >A.f hi\ VWKB.Aw) = s/2i 2("!) V (41) and [(9 .hA/2 = h^/4 this implies . i. by using the periodic WKB solutions below the turning points. We start from Eq.158). are given by Eqs. oo . The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf. (18. zo) meaning t o the right of ZQ and note the asymmetric factor of 2) 11/4 y±(z) ^ W K B (z) ± ^ = % K B ( Z ) : \qh\ 1/2 \h\U{z) ±— cos 7 **See e.e. pp. 82. Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions. (26.
207) (18.2.3 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 . (18. A.^r + + . this implies sin cos sin cos Thus e. Then at any point z € (zo.iK h\U{z) + / 4 (18. 1/2 .133) at the minimum z+ and obtain the conditions: 0 = — sin 7 and 0 = — cos 7 Hence £ + 1/2 dz(±qh\\h%U{z) 4_ ± — cos 7 />G^ 2 2 +i4c/(z)) 4 (18. 6.206) !/2 ' + J*+dz(^qh2+±h%U(z) 1/2 7T' 4 1 =p — sin 7 f*+dz(^qh%h\U(z) ^ TV + 4 (18.r + ^=co S [r. from z\.g.. approach the minimum also from the right.209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left. i..e.205) We now apply the boundary conditions (18. of course. . i.6.Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = .i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds. We could. \h\U{z) 1/2 . ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1.132) and (18. Sec. z{) we expect tt (18. Messiah [195].e...18. 7T\ + ^ 4 7r\ dz . i.210) Choosing the point z to be z+. JV = 1.3.2. from ZQ.g..e.V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) ..( .J^Wj +(18.212) tt See e.208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18.211) 7 z l .
Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±). (18.209) and (18.. .exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18. P. 9. The formula thus agrees with that in the literature. .27 9 . and L. 5. K. Bhattacharya [23].434 CHAPTER 18.. where u> is the oscillator frequency of the wells. Bhatnagar [14]. in fact (with our use of qo).205). + l ) j } ^ 1_ 27' (18. We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18. (18.e.Zi / r. they become c o t { ( .215) which results from the factor of 2 in front of the sine in the WKB formula (18. K. Landau and E.90 — =F— for q0 = 1. .217) exp ( V [ ° dz^/E JZQ + V{z) This expression represents the WKB result for the level splitting and may therefore be called the WKB level splitting formula. M. D.214) {«+"i} 0 for q = go = 3. i. Banerjee and S.146)).215) we can insert for 7 / 7 the expression given by Eqs. 7r"y in agreement with Eq. W "Cf. One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients).143).215).213) The integral on the left can be approximated by 1. and cot< (q + 1) '!}» for q = 50 = 1.2 .5. . 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq.215) we can expand the left hand sides about these points and thus obtain . (18. The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18.216) Inserting this into Eq._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ . the level splitting is given by AE(q0. We now see that Eqs. 7 . 1 1 . (18. The prefactor 2\/2h2/n is.149). ' /^i . 2IO/K.3.h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 . In Eq.210) assume a form as in our perturbation theory. Lifshitz [156]. (18. (18. (18.X++4 7T We rewrite the quantization condition in the present context and in view of the symmetry of the potential in the immediate vicinity of z+ as r J z0 1/2 dz qh [\ +\h+u^ \ 1/2 •K (18. S.148) and (18.
H. Aly. W. 435 . 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.* However. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. Giittinger and H. J. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. there is no need to have only field 'Generally a potential more singular than the centrifugal term in the (3 + l)dimensional Schrodinger equation is described as singular. f K .1 Introductory Remarks Singular potentials have mostly been discarded in studies of quantum mechanics in view of their unboundedness from below and consequently the nonexistence of a ground state. M. 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. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction. M. W.* 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. n > 2. MiillerKirsten [9]. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful.Chapter 19 Singular Potentials 19. Case [45]. Feshbach [198]. also in P. However. in an early investigation Case^ showed that potentials of the form r~~n. are not as troublesome as one might expect. Morse and H.
S. In fact. D referring to Dirichlet boundary conditions) in 10dimensional string theory. J. 19. K. H. G. some also permitting further research. It is therefore natural that we study this case here in detail. In the concluding section we refer to additional related literature and applications.436 CHAPTER 19. S. MiillerKirsten and Jianzu Zhang [226]. D.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. Tamaryan. Manvelyan. but also that the Smatrix can be calculated explicitly in both the weak and strong coupling domains. J. Maldacena [41].e. 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. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. J. our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. N. for instance. Liang and Yunbo Zhang [189]. R. This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. Callan and J.2. . Gubser and A. S. the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions. Singular potentials arise in various other contexts.Q. W. Recently. Singular Potentials theories in mind. W. (19.2) § C . § 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). H. Hashimoto [124]. that with the hyperbolic cosine replacing the trigonometric cosine). Park.2 19. MiillerKirsten.
19. using the method of Dingle and Miiller of Sec. W. h2 = ikg. \[\ in the correspondence is nonintegral.3a) In the case of the attractive potential (as in the string theory context referred to above. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 potential.4) first for small values of h2 and then for large values of h2. 8. Subsequently we derive the same 5matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential.2 The Potential 1/r4 — Case of Small h2 437 where E = k2. and as we have in mind in Sec.19. ig h \j g (19. The radial Schrodinger equation then assumes the form 2h2 cosh 2z .4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation.28) below). It is clear that we draw analogies to our earlier considerations of the periodic potential. This has advantages compared with higher dimensional cases. It may be noticed that since I is an integer. (19. (19. We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively.mo = 1/2 and h = c = 1.3 below) g2 is negative and hence h? is real for E > 0. The following substitutions are advantageous: y = r1/2cf>.Q. R. 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.g. J. We observe that with the replacements p2 z^iz. (19. Manvelyan.3b) 7 = % j. MiillerKirsten. H. .7 that we employed also in previous chapters. (19. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17. Eq. Liang and Yunbo Zhang [189].^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e. r = 7 e z . In the following we study the solutions of Eq. J. h = ei^y/k^.( I + 0. This is rarely possible and therefore this case deserves particular attention. and so =4 5 and n ez = r = ~ = e^4fir.
7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation..438 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.Nv(w). J. VahediFaridi [10].I . (19. and H„ (w). —Zu — (Z„_i + Zu+i). The zerothorder approximation = Zv[w) (19. Proceeding along the lines of the perturbation method of Sec. MullerKirsten and N.2. . Aly. where £>„ : = —_ + i _ + J l . 8.g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions.Hv (w). (19. w 2 .1.2 CHAPTER 19. i. W.^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+2]. — Zv+i).2 A / i 2 .9) leaves uncompensated on the right hand side of Eq.e. Eq. First we make the additional substitution w = 2/icosh z. Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively.7.11) We follow here largely H. H. 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 = 0.6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0.i l . dwA w dw { wA ) (lg. (19. (19. (19. where these are the Bessel function of the first kind. H. (19. The solutions Zv(w) are written Jv(w).4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + .5) so that Eq.6) terms amounting to 1 „ 2 d2Zu A V (19.
12) (19. can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except.6). v . ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a. The terms (19. (19. v2)*{y2.10) in terms of functions G„ defined by Gv+a = —nZ v+a .10) is now particularly simple: i?i°) = h2[(v.19.20) 2)*ZV_2 + {y. v + 2)*Z„ +2 (19.2 ) * ( z / . where We observe that DUZV = Q.17) Thus a term fiGu+a on the right hand side of Eq.u + 4fZu+A}.18) Now </>(°) = Zv left uncompensated Rv .2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19.10) as a linear combination of various Zv. = §.i/4)*Z 1 / _ 4 +(u.2)*Zi.u + 2)*Z„ +2 ]. v + 2)Gv+2].2 ) * i ^ 2 + (!/. v . However. .2.13). of course.v + 2)*{u + 2. it is more convenient to use these relations in order to rewrite Eq.21) +(i/. (19. We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately).2)G„_ 2 + (y. . I / ± 2 ) = 1. (19. and this means in (19. when a or 2v + a = 0.16) ( I / ._2 + {v. The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z.14) (19. where the starred coefficients are defined by (19.13) therefore lead to the first order contribution jfU = h2[{v. therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ . v)Gv + (y. v + 2)*(v + 2. z ..13) (19.!/+ 2)*4°J 2 ].i/) = 2A. Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19.u (19.15) (19. wz The expression (19. (i/.
i/±2)*.455a + 1291a .25)/i 8 32(al)3(a4) . P2*o(0) = 0. j=i. .v)] (19.9) 16 + + ^ j " liy '2° This expansion is seen t o b e familiar from t h e theory of periodic Mathieu functions where (Z + l / 2 ) 2 represents the eigenvalue.22) where p2(±2) p4(±4) = = (i/. (19. v ( 4 5 a .v + 2y{v + 2.v) + {v. . i/ ± 2)*.g. subject to the boundary conditions P2i(2j) = 0 for  j  > i.1169)/i 12 .17)). v + 2j) (19.v + 2)*(i/ + 2. Evaluating t h e first few terms.2. p4(±2) = (i/. v) +(i/. we obtain t h e expansion . v — 2)*(v — 2. Adding these terms and setting the coefficient of Gv equal t o zero. ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 . i/)] + • • • .440 CHAPTER 19.v) + hA[{v. + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9.26) iy — 2. Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e. in Eq.4 + 58. i/ ± 2)* (i/ ± 2. .25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv . ..23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j . and p 0 ( 2 j ^ 0) = 0.I/±2)*(I/±2. v + 2)*{v + 2.2 + 2 9 ) ^ 64(^ 2 .I/±4)*. 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v . . Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + .j^0 (19. (I/. . (19. Ri. . (19. i/ + 2j). i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j.24) P2i2(2j + 2){v + 2j + 2.v2)*{v2. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j. p 0 (0) = 1. etc.l)(z/ 2 . we obtain 0 = h2{v.2)(v + 2j . we obtain 9 h4 = a 2(al) 3 2 (13a .4)(»/2 .
30) £>„&.2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4.19. so that J2 (19.34) .cosh 2z)4>v = = 2h2A(/>„ .3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions.33) = = cosh(^ + 2)z + cosh(^ — 2)z. Thus v is real for small values of  h21.7) into Eq.An(u + n). (19. (19.i . = 0. Substituting (19. 19. Thus to O(0) in h we have (jr ' = 4>v — cosh vz. v)(f>v + {v. ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z. we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A .h2[(j)u+2 + 4>v_2] h2 [{u.4). 2 . apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1.2. v . and is therefore real for both cases g2 > 0 and g2 < 0.31) It follows that Dv+2n4>v+2n = 0.29) sinh vz or e±uz. ±2.2)<t>„2 + (v. £>„ = — .v2(j> = 2/i 2 (A . — We return to this question later but mention here that this problem has already been dealt with in Chapter 17. (19. The solutions 4> of the modified Mathieu equation are now completely determined.v + 2)<t>v+2]. We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 . (19. . we can rewrite the latter as j ~ .^ + e±(v~2)z.cosh 2z)4>. 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 . (19.32) (19.
Arscott [11].2. The use of the symbols (v. .7 — we refer to Meixner and Schafke [193]. W. 8. 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.J/±2) = .38) (f> with (\>v = exp(vz) — Meu(z. we prefer to adhere to one equation with different solutions. In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2.h). etc.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter.35) The form of i ? ^ is seen to be almost identical with that of the corresponding expression for solutions in terms of cylindrical functions. (19. 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. > These solutions correspond in the periodic case respectively to the solutions and me^. Schafke [193].I/) = 2 A . In particular the following notation is used for the solutions obtained above which we characterize here by their first terms: <f> with (j>v — cosh vz —• Ceu(z. CHAPTER 19. (19.1 . In order to avoid confusion arising from the use of different equations. > (f> with </v = sinh vz — Seu(z.I/±2)*.h2i *=1 J2 j=i.h). F. Meixner and F. M.h). Singular Potentials (I/. v ± 2) etc.442 where (I/.h) = <(>„ +Y. Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z.J7t0 p2i(2j)<f>u+2j. The solutions in terms of cylindrical functions are written *J. 19. • (19.37) where p M (±2) = (I/.4 Notation and properties of solutions We now introduce standard notation as in established literature.
39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant. h). Mi. (19. h) = exp{vniri)Me„(z.h). (19.h) and oo i M^\z. this normalized solution possesses contributions exp(i/z) in higher order terms. 17. Me„(z. Magnus and F. introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z. — oo cv2r{h2)e^+2r>. (19. (19.e. as Mev{z.h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(2hcosh z). Meixner and F. 16.MP(z.h). i.45) T Below we frequently write the normalized solution as in J. p. As emphasized in Chapter 8.h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19. .h2) oo = Y. h).44) implies.h) = exp(vz) + Y.40) does not. Schafke [193]. i.g.42).h)±Sey{z. h). 4>v = Hv (2h cosh 2) * M^\z.h).40) Cev{z. W. h) are therefore proportional to each other as a comparison of Eqs.(2hcosh z). h) = exp(inwn)MP(z. (19.44) (19. 443 (19.e. For the relation of Hankel functions used below. 'See e. we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv.43) and hence similarly M^\z + inir.h). (19.19. W.41) i=l We see immediately that Meu(z + niri. h). {2h cosh z) • MJ?\z. ' (z. h) = av{h)M^ (z. which. see this reference p.42) The solutions Meu(z. Also from the expansion of Jv{z) in rising powers of z. 4>v = Hi. <f)v = N„(2hcosh z) MJ?\z. whereas the — as yet — unnormalized solution (19. as we remarked earlier.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) . Oberhettinger [181].
h) . p.\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0. H™(z) = e^HfXz).h) Using further properties like (19.8. we have the following for a change from v to — v.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. .h2) = ] T cv2r(h2)e{y+2r)z. (19. the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions.e. i. p. Meixner and Schafke [193].4) we obtain oo £ r=—oo [(i/ + 2r) 2 4. Singular Potentials where clearly MeJQ.444 CHAPTER 19.h) can be shown (cf. r=—oo (19.e. Meixner and Schafke [193]. h). These points are important in our derivation of the S'matrix below. h) = M™(z.49) Inserting this expansion into the modified Mathieu equation (19.e.50) E. As mentioned earlier.h2): oo Mev(z. oiv{h) = (19.48) The series expansions of the associated Mathieu functions M„ (z. ^ .43). 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19. Mi 3 . 130) to converge uniformly for all finite complex values of z.exp(TM"r)MW(. 2M™ = eiuvMi3) + e~ivlrM^\ .h) on the other hand can be shown (cf. i. Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ). 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z.46) H^z) = e^H^(z). MJ>>(0.h) . s (19.g. For later essential requirements (i. the first relation is obtained as follows: 2M<£1 = M^l+M^l. The functions Me±v{z.
Alternatively taking the inverse of Eq. . W.^ + 2r + 2 ) 2 ] .^ p 2 2 ^ or c 2r+2 2r2 1 (19. p. h~2[\ .(/^. (19.^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 . Thus now c 2r . Schafke [193].51) we have 2r (19. Eq. p.54) 2 C2r 4 ^ h. + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(. 117. \ ./ i 2 C ^ . (39).53) /i2[A(^ + 2 r ) ] . 2r (19>51) 2r For ease of reading we give the steps in rewriting this. Meixner and F. + 2 r ) 2 ] .* For 'Actually. 122.[X(u 2 + 2r2) } " c 2r2 This continued fraction equation can again be used to obtain the explicit expressions of coefficients of normalized modified Mathieu functions.2 [ A . .(v + 2r)2} / z .^ C 2r+2 (19.19. see J.T22r[\ A .2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] .52) in agreement with Meixner and Schafke [193]. eg = c~v = 1. \) /.
and also R.( " + 2)2] ~ h2 h. (19. W. (19. Manvelyan. J.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. MiillerKirsten.52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(. + i). Spector [257].C + l) 3 (i/ . t For this purpose *We do this in the manner of the original derivation of R. J . but follow our earlier reference H. Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19. .8 ^^> 1 hi(v2+4i>+7) 32(^+2)(i/l)(^+l)2 4 ( i / + 1) /l 2 4 ( i / + 1) which is the result expected.[v^. H.+0^"> < 19 . i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) .+4)!ll_.55). Aly. Here we insert for A the expansion (19. 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ ._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(. VahediFaridi [10] (this paper contains several misprints which we correct here). M.^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10). + 0(/. J.. W.12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy. H.5 D e r i v a t i o n of t h e S .27) and truncate the continued fraction after the second step.1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq.1 for the evaluation of a typical case): hb + 0(hw). Then di 2 1 t" + 2(^1) + • • • .. Solution: We put r = 1 in Eq. Liang and Yunbo Zhang [189]. H.2. MiillerKirsten and N.55 ' Example 19.Q .446 CHAPTER 19. .l ) 19. 128(i/+ l) 2 (i/ + 2 ) ( ! / .
59) which tends to zero with r. \w\ S> 1 and — IT < arg w < IT is known to be given by* (19. 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19.3a) and (19. and then the continuation of this to infinity.j7L0 P2i(2j)exp i[v + 2j + 2 2 . W . I / + 2 2 l\TT +E ^ i\ E j=i. 22. We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0. Magnus and F. (19.5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo. and considering the propagation of this wavefront. The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\. we have W r = ip + ut — 7r/4' . In the case of the repulsive singular potential here..h) 1/2 w) (19.2 The Potential 1/r4 — Case of Small h2 447 we require first the regular solution yreg of the radial Schrodinger equation at the origin.56) where by Eqs.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 . Then y reg = = r rV2M®(z.57) ^ 1 . p.19. 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. this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves). Oberhettinger [181]. . (19.
AV . 178) to be convergent for  cosh z\ > 1.61) In fact. Oberhettinger [181]. In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z.3b) ln • 7T Q"*4' r fg ° Vk (19.448 CHAPTER 19.j¥=o Ki(2j)exp(Tmj) (19.60) Using the above asymptotic expressions for the Hankel functions. Magnus and F.57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1. We require therefore > > the continuation of Mi. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m.62) Also with Eq.h) oo j 2 — rV flp^ + ^fc* £ i=l j=i. 4%r g >0 4r^r ' rg See W.#. (19.A1"—2 z '+"2 A~~ n or rA .j^Q p2. '2"4 .3a) and (19. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo. Singular Potentials so that when t — oo : r — 0.(2j)<gH • (19. ^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz).A14A. 2h and the square of this expression implies 2 2 ~ ' A. h) through the entire range of !&{z). The series M^ (z. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo. 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. p./i).h) = exp(Mr^)il4 4 ) (z. Meixner and Schafke [193]. 17. p. (19. This means that the origin of coordinates acts • > as a sink. '(z. Now from Eqs. but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z. .V + § 9 2 > 4hh* ./l ~ 2 ~ 2z + " 4 > 0.h) can be shown (cf.
1. Originally we chose the sign of m / 4 as in Eqs. 19. (19.r2_) > 0 with r£ = rg(2 ± >/3). (19.rl)(r2 . (r — r+)(r + r+)(r — r_)(r + r_) > 0.19. which has to be bridged by using another set of solutions. z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo.62).e.3b) and (19. p. 130). This interchange in the > solution My interchanges y reg with y+. is satisfied only for r < r_ < r__ or r_ < r + < r. r 7r for 0 < r < ro.1 The domains of solutions.e.3a). A suitable set is the pair of fundamental solutions Me±v defined by (19. Meixner and Schafke [193]. . we have to assign different signs to the imaginary part of z in the two distant regions.ijt/4 Rez . the domain r_ < r < r+ — as illustrated in Fig.ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig. These solutions converge uniformly for all finite complex values of z (cf. 19.49). (19. and in order to maintain this symmetry.63) Thus there is a gap between the two regions of validity — i. Since the real part of z changes sign at r = ro. i. 449 This condition. i Im z z = In r/r0+i7t/4 .2 The Potential 1/r4 — Case of Small h2 Hence (r2 . The one further point to observe is that the variable w = 2/icosh z = (kr + ig/r) is even under the interchange z — —z.38) or (19. Thus we choose .
(19. at this junction we require also the imaginary part of z to change sign. (19. Hence we match the right hand sides of Eqs.67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_. i.z=—in/4. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19.v{z)\ . h) = Me^u(—z.^)] dr z=in/4 dr (19. Singular Potentials Then starting from the region r ~ 0. At r = ro the real part of z vanishes and then switches from negative to positive. 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.66) a = (3 .l/2Me_i/(T.65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z. (3 = a . i.e. as just explained. . and we determine these coefficients from this equality and the additional equality obtained from the derivatives.B^0.64) 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_. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me. to r^2[AM^ + BM^\ A.450 CHAPTER 19.(2)]z=w/4 + 0Me. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3.v(z)\ —m/4) + ^[r^Me. 131).[ r l/2 M. Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}.)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+. Thus. p. h) (as one can check or look up in Meixner and Schafke [193].
(19. for instance. (3.70) by Me'_u (the prime meaning derivative). we use the notation W\f.3a) and (19. (19. A and B. dr BM^(r)].65).19. (19. (19. ^ ] = ^ + ^ ^ .g]:=f(z)*&g(z)WZ) dz dz Thus.68) the derivatives with respect to r by this relation.2 The Potential 1/r4 — Case of Small h2 451 This solution can be continued into the domain below r = r+ by matching to the right hand side of Eq. (19. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point.2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)]. Thus we obtain from Eqs. (19. (19.71) We now determine the coefficients a. Then.3b) we infer that (z = l n r + const. (19. with the replacements of Eq. the nonderivative parts (from the first term on the right of Eq. and we are left with relations expressible only in terms of z.) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ .67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri.68) From Eqs. (19. multiplying the first of Eqs. (19. = aMev(z) + 0Mev(z). (19.65) and (19. and the second of Eqs.69) Thus if we replace in the derivative relations of Eqs. (19.70) by MeU and subtracting .69)) cancel out in view of the nonderivative relations.68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}.64) the relations M&Hz) jz[M^\z)] and from Eqs. (19.70) = a±[Meu(z)}+P^{Me_„(z)} A^[M^{z)] +B^[M^(z)}.
7T W[Mev. W[MJ?\ M^] W[Mev.M^] W[Mev. (19.] W[Mev.77) av I 2ism VK \ ctv av ^Thus.M^]= IT e~iv*av. we obtain immediately W[Z.3] = [1. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a.(0. 170/171). and the second is obtained similarly: W[M?\Mev\ W[Mev.Me^} (19. (19.47) we obtain (av defined by Eq. W[M^\Mlf)]W[Meu.46)) W[Mev. (19.58) and considering large values of \z\.e.46).Mevy W[M?\Mev] W[Mev.* « * r r z H . h)/M^\o.Mev] = = = av. 169) M0) = e«^Ma) _ isin U1X M(A) _ (19. .j] given in Meixner and Schafke [193] (p. 7T (19.\ \ ( av 2iwKav ^ v .74) Moreover. Me. i..u}W[Me. or obtainable by substituting e.u. (19. B = J 2 L _ e .73) We now use Eq. (19. and Wronskians W[M^\MIJ)] = [i.M^} W[Me„. i. Singular Potentials the second resulting equation from the first resulting equation. .^ 47 9? MJ>4)] MeV] [3.e. TV W[Mev. we use the following circuit relation which can be derived like Eq. M^4)] + W[MJ>3).g. we obtain the first of the following two equations. (19.57) and (19. using Eqs.^ ^ a „ Q _ . Meixner and Schafke [193].A] = W[Mi3\z). av{h) = Me.452 CHAPTER 19.MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>. (19.71) we obtain in a similar way A B = = W\MP\ Me. the leading terms of the respective cylindrical functions. p.4] = .Mev]' P l ' From Eqs.76) .. 7T [1.. = IT <*„.75) With these relations and Eq.M^] J™aV. h).48) (or cf. (19.4] = .
(19.85a) . (19.82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#).79) where the second expression follows from Eq.h). (19./) = (l)1/2R2^2i^)ei^)U^.46) together with the relation Mev(z. (19.f c . f c r . Z ) e ^ ] ./ ( .2 The Potential 1/r4 — Case of Small h? The regular solution in this way continued to r ~ oo is then y reg ~ ~ 453 Tl'2[AMiz\z.h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM). as we recall from earlier considerations (e. (19.80) 2^fc (1981) where \7r/ i t sin VK /(*. (19.83) The Smatrix element is therefore given by ifc*"* . eii/7r/2 (19.19.l) ' We note already here that with the substitution R = ei7r7. Then f' . B=tfD'"".(ite™)" 1 { 8 J ' f(k. Z ) 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 .78) we obtain Ay reg « _ .h) + BM^\z.h) = Me_u(z._ [ / ( f c .(l)le^}Pi(cos 9). 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq. in Chapter 16).g.
V(r)=g2/r Fig. 19.Smatrix in the form s i n 7T7 Si = CHAPTER 19. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u). 9?. cf. 19. R — — = 2i sin ivy. we can write this (multiplied > by 2ismvTT. we can define respectively as amplitudes of the incident wave. (19. Fig.2.85b) We shall obtain the 5matrix in this form in the case of large values of h2 later. R 2i sin TXV. We can rewrite Eq.454 we can rewrite the . (19. h) (see Eqs. (19. and recalling that yreg is proportional to a function Mc> {z.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. Thus in terms of the (3) variable z.2 The repulsive potential and the various waves. (19. Singular Potentials r(i+k) sin 7r(7 + u) (19. (19.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves.58)) 1/2 2r e~ 2hir cosh z.58)) in the limit r — 0.56) to (19. and with the left hand side following from Eqs.86) .56).
(p. (19." \R~Ji\2 1  tO _  2 i s i n VK\2 [  . H.2COS2VTT = (R. therefore we cite it from Meixner and Schafke [193].e™*/R\2 ' ' . l=—oo 1 1 lite*"* .6 E v a l u a t i o n of t h e S .2. Manvelyan. (19. however not here.Q.79). in which v is complex. (19. i.In Meixner and Schafke [193] (p. W.88) where the factor Me„(0. Liang and Yunbo Zhang [189]. but in the 10dimensional string theory context. By inserting in Eq. (19. Hashimoto [124] and R.„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* .e. unitarity is preserved.( i J e ^ ) " 1 ! 2 = R2 + ^ .e. J. The latter is what happens in the S'wave case of the attractive potential.. Eq. Jv+2r.88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence. i. 178). The expansion is: 4 r ( / i 2 ) M S ( z . MullerKirsten.e.±)2 + (2sinj/7r) 2 .89) . this expansion is given as 1 oo MJ> Hz.h) is (cf. unity minus reflection probability = transmission probability. 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.84) of the ^matrix.** 19. 180). (19. This implies basically the evaluation of the expression R of Eq. J. Gubser and A. (^)l4lu{h2)Jir(hez)J±u+i+r(hez). i. / i ) = Y. Eqs. i.h) The function Mo (0. (19. S. h) serves the use of the same coefficients as in the other expansions.m a t r i x Our next task is to evaluate the expression (19. " S e e S.28)) and R = ey real.39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind. M^l)(0. and then exploit it for the evaluation of the quantity R.e.19.2 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. y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles. 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). if R becomes a pure phase factor and eil/7r a real exponential. (105) to (108).
3 9 ^ . (19. /n ^ . ^ (19./i) with the help of Eq. Singular Potentials 4r (h )M$(0. (19. The formula (19. r = 0 and 2. r = 2 as a check). one obtains *.6 2 ) m 6 (^_i)3(z. (19.e.90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0.28)).90) and choose r = 0 (one can choose e.49) from which we obtain for z = 0: Me„(0.g.. As a matter of introduction we recall the definition of the coefficients with Eq.90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq. i.90). Eq.h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••. we obtain M&(0. Next we evaluate M^1}(0.g./>) = Co> 2 ) £ 7 " = — OO j ± > . (19. (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. Here in the case of the repulsive potential. (19. e. by allocating different values to r. h2 is complex but v2 is real (cf.2_4)(l/2_9)iv2y) + <AA. h) in many different ways.49). u»^. /=—oo (19.91) Inserting here the coefficients given in (19.h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h).55).94) . + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ . setting r = 0 in Eq.456 so that in particular for z = 0 oo CHAPTER 19.
111) / / i 5 4 3 1+ 4i/ 2 + +' (z.4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' ..19. 4(z/ T H ^ 3 .l) 5 (19./I) h u .1) 2u(4u + 15i/ .96) With this explicit expansion we can evaluate the S'matrix.1 ) 2 ( I / + 1 ) ( V 2 .95) These expansions imply for the quantity R: I/! 2v ri1 i . one obtains M£!(O.23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ .9) V2 1 fh ±v 1+ 2 + • (19.12i/2 + 64u ..97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves.32i/ . _2_(h\2 .7) F " (i/±l)2(^^l)(i/24)V2.. 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu .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. In this procedure our attention is focussed particularly on the quantity called absorptivity in the case of the attractive potential.2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19.] 2 \v i + {v2 .!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ .93) and the coefficient expansions (19. We observe that this approximation involves v and h4 and hence is real.2 _ ^ 4 ( ^ 2 _ 4)2 (19. From expansion (19. .4)(i/ .55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v). + # "+" ( I / .2^ 2 ± 59^ .
eiun/R2)(eiu*n ~ (19.2.101) where / is complex and g is real.28) that v is real.^ i 4 2+^ # 1 / •.101) is > zero.458 19. Then cos 2TT/? = 1 .28)).9 / . 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3.103) Then SiSt (1+5) i . + ( i + s w (E 4 . (19.84): Si ST 1 (11/R2)(11/R*2) . (e iun e~iv*^/R*2)' (19. Singular Potentials Calculation of t h e a b s o r p t i v i t y We consider the absorptivity in a general case.104) We now consider the case of a — 0 implying that g of Eq. Eq.102) (19. In this case R — R* and so 1/i?2 ~ Ofji4). This is the case of real . _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2).98) (cf. (19.100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19. where n is an integer and a and (3 are real and of 0(h4) We have with 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.7 CHAPTER 19. (19. Thus we set v = n + i(ot + i(3) = (n — 0) + ia.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. and sin 2TT/3 = ft/. (19. (19.R*2 ^ * #2 \ 11 tf2 R* (19.
Oberhettinger [181]. Here in t h e case of the attractive potential (as remarked after Eq.28).61n2 . 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 .97) this can be written A 4TT2 [v\{v . Liang and Yunbo Zhang [189] (there Eq.3b)) h2 is real for E = k2 > 0.1)\} 2 h \2 iv 1 + Au {y .19.* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 . p. S. MullerKirsten.SiSf » 4 sin7rz/ R (19. one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 . ( I l l ) ) in agreement with a result in S.102) we obtain Sis.Q.86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)].1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19.577.e. f W . "This is the result effectively contained in R.105) sin7r/3 (19. and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0. In all comparisons with these papers — which consider predominantly the 10dimensional string theory context — one has to keep in mind that many formulas there do not apply in the threedimensional case we are considering here. Gubser and A. We can also evaluate now the amplitudes (19.3. W. (19.^ ) ~ 2 2 459 1+ 9 / • & ( ! .61n/i} + ~9~ 0(h8) (19.2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z). (19. Hashimoto [124]. (19. i. H. Hence with Eq. .6C . 1 29/ ^ (1 . J. J. Magnus and F.106) and evidently violates the unitarity of S.l ) 2 2 \) +0{h«) (19.107b) = Ah2[l + 0(h4)}. = 1 . With t h e help of Eq. At = 2i[l + 0(h2)}.107a) In this result* we now have to insert the expansion of the parameter v given by Eq. (19.107c) Here h2 is actually Vh4. Manvelyan.
Park. h) for large h and upwards I = 0. Aly. Tamaryan.h) (19.4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z .e.3 The function q(l.3.a }tp = 0. H.H. we again make use of the replacement (17.4 Fig. . 2.460 CHAPTER 19. J. VahediFaridi [8] and G. 19. S. 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. W. W. 2 2 a = [l + 2 (19. MiillerKirsten and Jianzu Zhang [226].3 19.h) 10.3. Wannier [278]. Singular Potentials 19.109) ^We follow here mainly D. H.$ q(l. i.26) and hence set a1 = 2hl + 2hq + A{q. (19. K. where the part of interest here is based on H.1. H. J.108) For the large values of h2 that we wish to consider here. We are again concerned with the associated Mathieu equation given by Eq. This is therefore a highly interesting case which found its complete solution only recently. N.1 The Potential 1/r 4 — Case of Large h2 P r e l i m i n a r y remarks In the following we have in mind particularly the case of the attractive potential (h2 real) and rederive the 5matrix obtained above by using the asymptotic expansions for large values of h2. completely different solutions. Miiller and N.
to which we have to return in this subsection. See also Eq. which is even about z = 0. Therefore the second term on the right is nonzero.3. y(0) = 0.111) we have W[y+. (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. (17. (19. y± oc A{z)e2h sm * ± A(z)e2n sm z .21b).113) 'We are considering nonintegral values of v. We begin.§ a W[y+. (19. i.5) that this is given by the relation cos iris = ^ M . The behaviour of q as a function of h for some values of / is shown in Fig.l + 2y+(l.e. The replacement (19. we defined the Floquet exponent v and observed with Eq.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively.h*y (19. 19.3.h*y_(l.19.112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq. with an essential mathematical step.21b). 19. i. With the boundary conditions y+(0) = l. however.e. (19. and a is its normalization constant. J4(0)=0. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v. We also found the boundary condition (17.yJ\ — 1 and cos TTV . y'_(0) = 1.3 The Potential 1/r4 — Case of Large h2 461 where q is a parameter to be determined as the solution of this equation. (17.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case. and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively.110) a where y+(7r) is the solution evaluated at z = TV.63). which therefore cancels out or can be taken to be 1. .y} where VF[y+.
1 1 Or see R.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 .117) a a Referring back to Eq. Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h .39 2 14 /i 2 + (19. (19. B. and hence with Eqs.112) from evaluation of Eqs.32). (19. Miiller [73].43) we see that z = 7r/2 implies W(TT/2) = 0. (17.116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq. We now obtain the expressions needed in Eq. J. Hence we require (obtained from Eq.118a) "We emphasize again: For integral values of v the normalization constants are different. in the second expression 2h is obtained in addition from differentiation of the exponential factor).115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0).111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. . (17. B. i. See R. This paper contains the normalization constants of large/i 2 periodic Mathieu functions. Miiller [73]. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h. Dingle and H.114) at z = 7r/2. (19. J.114) a a Our first step is therefore the determination of the normalization constants No. Dingle and H.A(z)e2h sin 2 sin z sin z (19. (17. W. W.e.462 CHAPTER 19. 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 . which is \/2 at z = 0. (19. K [ (z)}2hsinz (19.
3 The Potential 1/r4 — Case of Large h2 and 463 as dz )z=n/2 dw (dw\ o\dzJz=n/2 1C%2 .119a) This formula — derived and rediscovered in Park.62b) (see also the explicit calculation in Example 17.19. Tamaryan.119b) COS TTV + l ire This result gives the leading contribution of the lavgeh2 expansion determining the Floquet exponent v. (19.117) are known from the matching of solutions in Eqs. . MiillerKirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193].62a) and (17.[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 . (19. Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq. a in Eq.87 2 14 /i 2 [ i ( g .! ) ] ! [ .112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q .2g + 3) 27/i ( 3 g 2 . p. 210.! ) ] ! [ .h . (17. (19.3).118b) The factors a. (19. With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! .8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (19.i ( g +3)]! 1 24/i {2hl>2).1Q11QM cos irv + 1 ~ •==— e . Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others.\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)].
of course. h.z) or as tp(q.h).124) Since this function differs from a solution ip by a factor k(q.3 Following the procedure of Sec. .120) The resulting equation for the function A(q. With the solutions as they stand. In the following we require solutions H e ^ ( z ) . h. „ 1 A „ —A cosh z——H .h.e.3.3. i = 1. h.z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V2ih k(q. z). 17. h(q. We define these solutions in terms of the function Ke(q.3.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. the expression (19. h.h) =?==. z) = ip(—q.z)exp[±2hismh z\.123) We again make the important observation by looking at Eqs.4.122) that given one solution ip(q. we can obtain the linearly independent one either as ip(—q.108) and set ip{q. (19.s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19.h.. . —h. dA 1.464 CHAPTER 19. z) = ip(q. —z).121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i. each with a specific asymptotic behaviour. h. V— 2in (19. —h. Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19.120) to (19. h. 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ . Vcosh z \ 1 .122) Correspondingly the various solutions ip(q. —z). (19. it is still a solution but not with the symmetry property ip(q. Then straightforward integration yields > inq 4 „.z).109) remaining unchanged. z) can then be written as .z) i/>(q. h. for h — oo).h. tp(q. z) are TP(q.2.z).h)iP(q.2 we insert the expression (19.? s m h z) Vcosh z (19.109) into the equation (19.z) = A(q.h.h. .h.
the Bessel functions contained in the expansion of the associated Mathieu function M{.127) sin z T /'1 M TT\ — <>{ 4 . (0. (z. : 7r/2.19. One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO. 2j<. * ** = > rojKe( ". in performing this cycle of replacements the function picks up a factor.e.<5±1) (19 125)  (the expression on the far right in leading order for h2 large). (19. 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]. i. . [ \ . It follows t h a t S e t t i n g s i n h 77 = t a n 0 . av{h2) = ^(0J ^ Mi i . Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V. Me„(z. we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo.^ s i n h 7?)]g. . i.e./h). This is./i 2 ) M (19.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]. 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. achieved with the help of the Floquet solutions Me±v(z. or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions.126) For large values of the argument 2/icosh 2. 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 . / 2i<jfc + A / 8 \ ~ 2fc / drj cosh 77 1 + * _' Jo V 8fc2 c o s h 2 77 7 2fcsinh y + [ t a n . In order to be able to obtain the Smatrix.fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P . We observed that these satisfy the relation (19.125) 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 .h2). (19.h2) = OvMPfrh2).45). i. we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr. (44) of W a n n i e r .3 The Potential 1/r4 — Case of Large h2 465 Instead.fi2) can be reexpressed in terms of Hankel functions. >• * F $ 0 * e". we h a v e 77 = 0—> 6 = 0. as we saw earlier. 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 .
3.g.h) He (z.z).q. (19.130b) r_0 rll2exp\g/r (ig) ' 19.h) = e(z) exp ihez .129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H. 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. z): Re(2\z.130a). (2) (19. 3te»0.q. (19.h) Be^(z.128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.eW(z. (19. He (z.h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO.q.128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs.q.h) (3) = = = = Ke(q.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T . (19.h.4 T h e c o n n e c t i o n formulas We now return to Eq.q. 3te»0. In this way we obtain in the domain .q.q./i).q.130a) Re^(z. (19.g. HeW(z.466 CHAPTER 19.h). KeW(z./i) Be^(z.h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19.h).vr . h. exp[—ikr — in/4\ kr Re{2)(z.130b).
132) These relations are now valid over the entire range of z. / i ) + sin7r(7^)He(4)(2.h) sin irv = sin7r(7 + ^)He (3) (2.g. and He(*> exponentially decreasing.i ^ / 2 H e ( l ) (z> ?) (19.(z. (19. (19. e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e .q. these are exponentially increasing there.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19. z) and He (3) (z. h.134) in the region where the function He^(z./i). we see that the solutions (compare with Eqs.133) In a similar way one obtains the connection formulas — sin nv HeW(z. h) = Ke(q. g . t With Eq.131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z. and thus can be matched in this region. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e . . q.19.3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z. h) = Ke(q. q.g./i) — sin 7T7 Re^(z.li 2 ) Oil. (19.132) into Eqs.135) sin irv 'This means.130a).h2) a.« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e .g./i).(z.h2) Me„(z. (19. (19.129)) He (2) (z.* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z. we obtain in the domain tfcz <C 0 the relations: Me. q. — sin 777 H e ( 3 ) ( 2 . h) = sin 7r(7 + v) He (1) (z. h. h) — sin 7T7 He^(z.125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19.130b)). z) are proportional there. (19. Substituting the Eqs. q.134) Considering now the first of relations (19./i 2 ) = Me_j. (19.q.q.h). Changing the sign of z. Eqs.131) where the second relation was obtained by changing the sign of v in the first. (19.h) is asymptotically small (cf.
Singular Potentials The factor on the left.m a t r i x From Eq. Squaring Eq.(19. which in our case is the solution He^4^.119a) and (19.133)) — sin TXV He^4' (z. sin TXJ. .137) We see that this expression agrees with that of Eq. of course.e'i7ri 1. i.135) is the equation corresponding to Eq. (19.( .l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr . (19. and therefore remarks after his Eq. q. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV .119a) or (19.135). (19.119b).119b) as „4/i 1 + COS TXV ~ rrl ( g .Sin TXV. where 8i is the phase shift. (60) in the work of G.85b) obtained earlier. Thus here the S'matrix is defined by (using Eq.1 .l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19.136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtla/2) S m TXV (19. is only the dominant term for large h2} 19. Wannier did not have the relation (19. (19.5 D e r i v a t i o n of t h e ^ .135). The latter is defined by the following larger behaviour of the solution chosen at r = 0. Wannier [278].138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs.l ) (l„—ikr e From this we can deduce that Si (19. (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".468 CHAPTER 19. (19. H.e.3.133) we can now deduce the 5matrix Si — e2t51. The quantity 7 is now to be determined from Eq. h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( . cos TXV + sin TXV. (19. exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7.139) ''Equation (19.
141) COS TTU irvj — i s i n WR s i n h TTUJ. 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. we can approximate the equation for q ~ h (i.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h.138) we obtain. A{l.8h COS From Eq. (19. the real part of v must be an integer.139) could actually be neglected) implying  cos m>\ = cosh nvi.137) together with (19.e. as a comparison with Eq. 19. (19.19.§ Since the right hand side grows exponentially with increasing h the Floquet exponent must have a large 1=0 1 large h approximation small h approximation Fig.4 The absorptivity of the S wave (attractive case). Using Stirling's formula in the form z\ ~ e~zzz+1/2V2^.140) From this we obtain the absorptivity A(l. .28) shows.e. for the attractive potential. This is different from the behaviour for small values of \h2\. (19. i. i. 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. h) of the Zth partial wave.h):=l\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1 (19. 9/2 (91) with the approximation q ~ h obtainable from Eq. (19. since cos TVU is large. irrespective of what the value of I is) by cos™ + l = c o s ( T j^J y/e1.109).e.
Cvetic.^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently.142) we can see the behaviour of the S'wave absorptivity as a function of h. 11 J. and.107b) and (19. Dombey and R. "G.470 CHAPTER 19. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi . Lii. Bertocchi. Apart from the papers already referred to. In essence. B. Pao and J. such as phase shifts. H. Treiman [271]. in fact. Tiktopoulos and S. H. 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. Y. Limic [177]. N. and G. Handelsman. S. small h2) case have been considered by some other authors. 19. Eden [46]. Shabad [248]. N. Paliov and S. S . YuanBen [285]. Pope and T. VazquezPoritz [61]. Furlan [20]. A. Rosendorf [225]. Masson [191]. D. C.139) 2Tr(16h)q /32\ f e Thus with Eqs. in further contexts beyond those already mentioned. N. 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.4 (there are tiny fluctuations in the rapid approach to 1).** 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. (19. F. Fubini and G. W Some of these sources can be helpful in further investigations. A. J.P. This is sketched schematically in Fig. D. **A. Lew [128].' Rudimentary aspects of a singular potential with a general power n have also been considered previously.e. H. Lii and J. The diminishing fluctuations of A(l. Partial or other aspects of the weak coupling (i. (19. A. I r a n [61] and M. 19." as well as other aspects. E. Challifour and R. Apparently this is one of the very rare cases which permits such complete treatment. Jones [78]. L. see also M. **R.4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r . Esposito [88]. .k) in the approach to unity are too small to become evident here. Cvetic. one may expect corresponding results for other singular power potentials.4 .
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. We concentrate in this chapter on the large order behaviour of asymptotic *C. M. Bender and T.* In fact. We shall also see explicitly. Dingle [71].* However. the asymptotic solutions in different domains all have the same coefficients. e. B. T. *R. also convergent expansions are known and a lot of literature exists on the equation. Dingle and H. the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler. W. [19].Chapter 20 Large Order Behaviour of Perturbation Expansions 20. 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. Muller [74]. 471 . Wu [18]. This is indeed a remarkable connection. 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. B.g. J. how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. f R .
20. 8. Presumably this has never been done so far. as evident from Example 17.1 The function ur = pr jr\ vs.7. 19.2. 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. 10 12 14 16 18 20 Fig.1 and cases in Sec. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec. r for p = 4. 5. r for p = 4. 8.6. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec.2 The function ur = r\/pr vs. This approach leads .2. 20.6. Fig.2 and below. Since the method is also applicable to convergent expansions.8.472 CHAPTER 20.
J2 2n h E (20.e. It can be seen from these.1 and 20.e. T V lim h2N 2 n EQ?) . i. The traditional method of using an asymptotic expansion is to truncate the series at the least term. For an authoritative discussion of these aspects we refer to Dingle [70]. equivalently for g = l/h. and the integral E(g2) is called the Borel sum. even if each En is known. 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. The behaviour of such terms is illustrated in Figs.20. The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E.4) Jo Jo its Laplace transform F(z) is called the Borel transform of E.3) n=0 Thus for any given value of g . in the power expansion of the exponential function). The formal perturbation expansion of a quantity E(g ) .2) or. the function E(g2) is given only approximately by Yln=o En92n. whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power. that the maximum of the absolute value of terms in a convergent expansion and the minimum of the absolute value of terms in an asymptotic expansion are reached approximately when the value of the variable is approximately equal to the number of the term.1) is an asymptotic series if for any N lim 92^o 1 g 2N N E(g )^2En92n 71=0 2 = 0 (20. this term represents the size of the error. i. In some cases the answer is affirmative. 20. and the expansion is said to be Borel summable.2.1 Introductory Remarks 473 automatically to asymptotic series. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH). as we discussed earlier in Chapter 8. A considerable amount of work has recently been devoted to the question whether it is possible to reconstruct a function exactly from its asymptotic expansion. (20. for instance. . E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20.
. 1 we obtain oo F(z)£(*)" = — .e. In this ideal case the integral of Eq. Q. n=0 (20. . (20. In fact. (and E0 = 0). Thus if CO F(z) = aiS(z) + J2 anzn.5) then with the integral representation of the gamma function F(z) = (z — 1)!.n _ En+\ n! En+1 ^ (!)"«!.474 CHAPTER 20.4). J we obtain Now if />oo dteHz\ °° oo ' 0 dteH*\ TEn+1g^V . (20. . In general this is not so easy. .n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0. that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]). Large Order Behaviour of Perturbation Expansions Eq. (20.6)).4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation. (20.6) The power series of F(z) converges inside the unit circle about the origin at z = 0. We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0. n=0 (20. .1)? Asymptotic expansions originate from integrating a series (i. . 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. many asymptotic series do not even alternate in sign (as in the case of the series in Eq. In any case. 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. We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: . El .
Weinberg [228].3 and g < 0.9) The question is therefore: How can we determine 9 E(h).10) r as shown in Fig. (20. or can we determine Ar independently? Fig. .1 Introductory Remarks 475 except for possible subtractions... S.2.3 The potential V(r) gr». and E represents an eigenenergy. 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. ? When ?s E ^ 0. (20. we have a quantum mechanical problem where probability can leak away.. for < ^ 0. § where 7(ff) = To exp  .. N = l.10) is considered in detail in this work./ \p\dr\. e... 20. if the potential or a boundary condition is complex such that the problem is nonselfadjoint.2.N = l.g.. if the potential is V(r) = +grN.e... M.11) V. (20. 20. $s E can be calculated by the WKB procedure. Popov and V. The potential (20. This is the case if the real potential has a hump of some kind. i. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis).20.
the BreitWigner formula (10. Popov and Weinberg. (20. However. S.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. 20. (20. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available. 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 . Eletsky. The most direct way to approach the problem of determining the behaviour of the late terms of an asymptotic expansion is to consider the equation determining these coefficients. In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential.73) in the W K B or semiclassical approximation.^ The logarithmic factor in Er is a novel feature of this case.113)). the perturbation method of Sec.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. Equation (20.476 CHAPTER 20. Example 20.9) for Ar. V. Solution: The solution can be found in a paper of Eletsky. Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations.1.11) is. M. the cosine potential being t h e most completely investigated will also be seen in the following. (11. . (14. 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. of course. as we expect on the basis of Eq. Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq. An interesting application is provided by Example 20.11). (11. L.134)). as was (and still is) not the case for t h e equation 1 V .2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. Popov and V. Weinberg [83]. 8. 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.
2 A = Mx + ~M2 + 7^WM3 + •• • .14) Thus these coefficients can be obtained once the coefficients Pi(q.60).l ) ( g + 4 j .13) represents effectively a partial difference equation. Eqs.Cq. Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf. Eqs. W. or equivalently of the quantity A. Dingle and H. j . are given by i . The recurrence relation (20.j + 1). h). M2i = 2j2 J[Pi(q. (cf. (17.13) In the special case j = i the boundary conditions stated after Eq.58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g .2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential.J) = (g + 4 j .55)) X = E(q.26) and (17.M2i+1 for large i.j) +(q + 4j + l)(q + 4j + 3)i^_ x (g. (17.l(q.34)) that the coefficients Pi(q. J. (17. (17. obey the recurrence relation jPi(q.j) a n d M2i can be deduced.20. arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const. 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. i.3 ) P i _ 1 ( g .12) (20.58) and (17.l ) +2[(q + Aj)2 + l + A]Pl.h) = 2h2 + 2hq+j. Miiller [74]. Eqs. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q. B. This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods.l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf.e. (17. 3)? . % = Aq.j) of functions ipq+4j. (20.Bq. .j) are known. (20.61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q.
Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient.18.19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign. and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i . Hence exact Borel summation is not possible. In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions.l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula.91 l)]![i(g3)]!} 2 { [ .16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (20. .478 CHAPTER 20. see also below) .1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20. In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods.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.l (20. Using this relation and the duplication formula lw('\W±^.17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero.16a) (20.th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i .*.•91 {[l(? + i)]![(? + 3)]!}2 (20. also written lCj. (20.
i.* However. t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2. f H . n = 0 . but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary.4. MiillerKirsten [200]. we emphasize a few points *M. this difference was found to be real. What changes as a result of different boundary conditions is the deviation of q from the odd integer qo.20. The potential we consider here then has the shape shown in Fig. when this is g0 = 2n + l . l . .5 \ \ / \A Fig. 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. Reeve [262].3 20. . We retain the previous boundary condition at one minimum. This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions.4 The cosine potential shifted by TT/2. Stone and J. but with different boundary conditions. the difference q — q$.3 Cosine Potential: Complex Eigenvalues 479 20.e. From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g. 20. remains unchanged. . . J. 2 . W. Previously. . 20. we follow here a formulation in the context of our earlier considerations of the Mathieu equation.5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0.3. A means to achieve this was found by Stone and Reeve. \ 1 \ \ \ \ \ 0.2 solutions obtained earlier.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/?. At the risk of repeating what we explained earlier. in the calculation of the level splitting as a result of tunneling.
(1 w < 2a22) cylinder T h e solutions of this equation are the real and complex parabolic functions D ±(q~i)(±w)> In the following we find it convenient to select solutions of this type in the domains around the minima of the potential in Fig. 20.20a) As stated. Now we change this situation at the minimum on the right. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20. but t h a t with complex argument there. 20. Setting w = 2h1/2z.e. as we found these for this potential. Previously this situation was the same at the neighbouring minimum.4. 8 the equation can be approximated by d2ip(w) (20. but achieves exactly what we want.Ah2z2 • • • }^{z) = 0. a boundary condition which at the face looks somewhat abstract.T K ^ a D_i_{q+1){±iw). X + 2h2 = 2hq+—. which means replacing z by z±ir/2. We can distort the potential there or alternatively impose a different behaviour there on the wave function.4 (around z — 0) to the one on the right (around z ~ ir). and correspondingly we assume alternately even and odd oscillator eigenfunctions there. Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 . 20. (20. and allows its imaginary part to tend to infinity. i.20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. . The physical wave functions there are therefore the solutions which are real and exponentially decreasing.4. the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them.2h2 cos 2z]ip{z) = 0.480 CHAPTER 20. we shift the argument by n/2. Effectively we consider the tunneling from the left well in Fig. Thus one chooses not the solution with real argument at the minimum.21) <u> + ( i « . To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A . Here we choose t h e second way. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there. Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear).
20. 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. (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20.3 Cosine Potential: Complex Eigenvalues 481 As explained above. Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i. . . .63). Thus. the even or symmetric solution about z = 0 can be taken as *• = liDfrriVhWz) +Dh{q_qo)(2h^z)}. 2.e. . Since the large/i2 solutions valid away from the minima were the solutions of types A.4. by a different method. A derivation is given subsequently.23) .22). 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 (which we can loosely dub solutions of WKBtype). so that the deviation q — qo results from the finite height of the barrier taken into account with boundary conditions at the next minimum. Eq. considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g. i. that at z = 7r.> 0 for z > ±ioo.24) Considering t m O i n Eq. (20. we naturally constructed the even and odd solutions with these (cf. whereas here we construct these from oscillatortype solutions. This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions. 3 . We consider here explicitly only the case of the ground state around z = 0. we now choose an harmonicoscillator type of real solution around the left minimum of Fig. 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. we see that ips ~ cos \j2qhz for z ~ 0. such that this deviation becomes imaginary. whereas here we start off with a solution even or odd about a minimum. (20. 17. One should note the difference to our earlier case of the calculation of the level splitting: There we were seeking solutions which are even or odd about the central maximum of the potential. 20. however. .e.
25) qo) E2 0: (20. (20.30) 'See e. p.g. Thus the only nonvanishing contribution comes from differentiation of Y[—v).27) differ by an angle IT. (20.we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{.482 CHAPTER 20. §See e. we know from the solutions of types B and C in Sec. tp(v). and then evaluation at u = 0. W. The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function. The arguments of the functions appearing in Eq.27) requires the differentiation of this expression with respect to the index v.2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing.g. 92. Oberhettinger [181]. (20. 17.•w) + 2ir T{u) <"+!)% Dvi(iw).3. We see this as follows. p.w) = e T Dv{w) 2TT n») AvViDv^iw).29) Equation (20. Setting v = 0 in T(—v) = (—v— 1)! gives infinity. Magnus and F. Magnus and F. or logarithmic derivative of T(^) for which:§ ^(1/) r» (20.27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq. this can be handled in a few lines.(.28) Extracting Du{—w) we have A. Oberhettinger [181]. However.24) are linearly independent around z = 0 for q ^ qo. W. (20. From Tables of Special Functions'. This is the feature which gives rise to an exponentially increasing contribution in the neighbourhood of z — 0 and hence provides the dominant contribution in going away from the origin. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20.h(2h1/2 J £2=0 (20. .26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/. 2.
Thus calculations for qo ^ 1 are not given by M.29). W.e. we obtain the formulas (X) / \ 1 OO I » J ] (l+)e. This result is » not easily generalized to <jo > 1. Reeve [262]. dropping from the derivative part of Eq. those of type A.3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ 1. Magnus and F.33) = IV2~TT.)e" w /4 2 l ^Dx(iw) (20. Oberhettinger [181]. with the following expression for the largez behaviour'! (here v = E2/AK) Dv{w) ~ wue~w2/4 = evlriWe~w / 4 for  arg w\ < ^TT. For u — 0 one obtains the result (20. r(i/) We now return to Eq. (20. "See e. which are valid around z ~ 7r/2.27) exponentially decreasing contributions. i.31) dT(u) r*(v) dv 1 i/>(v) ^ o l. p. 3 .31). Stone and J.//n and ij>{y) = C . Differentiation of this equation and subsequently setting v — 0 yields. We now go to the WKBtype solutions. and "Prom W. .20. i/^0 483 T(v) It follows that d dv [T(u)\ (20. pp. including the derivative of DE2ufl(2h1/2z)).32) which we can use here since all other problems have been resolved: d D {w) dv v (20. 92. Magnus and F. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20.34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity. we obtain 4>s D0(2h^z) D0(2h1/2. Oberhettinger [181].29) v=0 (in + In w)D0(w) + v2ire (m + lnu.g.4. . . .e.+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i.. (20. (20. 5 7 7 .+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 .
(20.34) and (20.35) We construct with constants c\. (17.40). (20.36) and consider this at the other end of its domain of validity.38) in its limiting domain towards 2 = 0. that in approaching z = ir. Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left.. (20.41) 2TT . Recalling that we have shifted the potential by 7r/2. i. To this end we set z' = z — ix.34). Then V'WKB becomes (with cos z ~ — 1 + z' /2. c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20.484 CHAPTER 20.sm I We now want to match this solution to that of Eq. x . 2 c2e2hehz UJ2 .29) and (17. y COS c\ J (20. cos z/2 ~ (20.36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h. We now return to the WKBlike solution (20.— 721 . we obtain the required solutions from the earlier ones. i. smf (20.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo. (20. Eqs. 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] .e.39) Now.40) From Magnus and Oberhettinger [181] (p.cos z — 2/icos z ipA^ Tcos  and i n A —. We observe that in the matching domain (i. where we can then match these to the solution ips. near z = n the boundary condition (20. (20.e. Thus in the present case and with q ~ qo = 1 we obtain „2ft. with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ). Thus the WKBlike solution becomes p — 2/lCOS 2 V'WKB ~ D0(2h1/2z) D0{2h ' z) l2 + c2 + c2 ^Tsinf 2/i hz2 _z_ 2 (20.e.36) V'WKB cie~2hehz' 9h  .37) Comparing Eqs. by there replacing z by z + ir/2.36) cos  .
(20.38) can be written TAWKB (20. and hence (adding and subtracting equal terms) Di(i2hV2z') = ^[D^h^z') + D_1{i2h1/2z')} Z + hD_1(i2hl''1z') = \V2^D0(2h^2z') Z D_l{i2hll2z1)] D^(i2hl/2z')} + hDiWh^z1) + [D^i(i2hl/2z') Z ^D0(2hl/2z') Z . (20.e.45) Comparing this result with tps of Eq.20.44) = DQ{2h / z) ± (2h7r) / e. 1 2 4 1 2 8h ^^2 2 (20. We derive the general result in the next subsection using our own method.46) Repeating these calculations for the general case (i. (20. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial.43) Comparing now Eqs.43) we obtain. J Hence Eq.^.complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(. The general result stated by Stone and Reeve [262] is therefore presumably guessed.1/V)_le^_ Z With this result Eq.e. (20. .34) we obtain E2 = i%\ = ±ih22Qe~8h. i.39) and (20. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l .46). with c\ = exp(—2h) from above. ^'mliw <2a47) For qo = 1 we recover the result (20. (20.2/.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2. (20.3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real.
On the other side of the barrier at z = —TT/2 we consider different boundary conditions. this equation may be approximated there as d2ip ~dz^ + E + 2h2 .20a). Considering the original equation. the result is formula (334) there. harmonic osc 71/2 E>V / E<V z+ O \ E>V TC/2 \ \ .z.486 20. We have to impose two sets of boundary conditions.e.48) These conditions provide the required quantum number qo = 2n + 1. W. H.) + 7T V> = o. i. J.Ah2 ( z + . At the minimum to the right of the barrier shown in Fig. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there. (20. •+(=)* V(f) 0.5. (20.2 CHAPTER 20. and expanding cos 2z around the point z = — TT/2. Wiedemann [4]. (20. . 20.*. Thus in the immediate neighbourhood of z behaves as 2 1 2 ip ~ e ±i(E+2h ) / z ~ e 7r/2 the wave function i^{z) ±i(2hq)1/2z **We follow P.e. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq.46) we consider a half period of the cosine potential as illustrated in Fig. i. Fig.e. i.3. V(z) .5. at z = TT/2.E curre it \ j. •_(:)*<>. Eq.5 The cosine potential from — TT/2 to n/2. 20. 20.()#0. MiillerKirsten and A. Achuthan.
we demand that ^(*)L<.51b) 2i(«1)[i(9l)]'.' 4 / 1 1 / 2 COS ( Z + 7T \2 4 .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 (20.21) into the Mathieu equation we obtain.e.3 Cosine Potential: Complex Eigenvalues 487 (using Eq. three pairs of solutions of types named A. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2. (20. (20.h.49) This is a complex boundary condition which requires the coefficient of the other exponential in ip±(z) (continued to and beyond — 7r/2) to vanish. as shown earlier.r/2 ~ e i(2hq) V 2 .50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum. Recapitulating these from Chapter 17. 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.h. For the continuation we use again the WKB formalism.21)). B and C.20. (20.z)=tpA(q. for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B. i.51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}. we have IPA(Z) „2hsinz Aq{z) + O il>A(q.z). Proceeding as in our previous cases and substituting the expression (20.Hz^+Ol^ Bq[w(z)}+0 h1'2 (20.
Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function. and the second around z = 7r/2. Here the upper sign refers to the even states and the lower to the odd states.488 CHAPTER 20.52a) = Cq[w(z)}.e. the relation q . 5.=F2 2\1/2{28h2y/4e~Ah l + O (20. ^ W H . Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) . (20. We return to this condition after derivation of the . 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.^ i .48) and proceeding as in Chapter 17.54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}. (20. we obtain our first transcendental equation. we obtain 1>±W = \ a a (20.57) where go = 1) 3. (20.53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20.56) Applying the boundary conditions (20. We saw in Chapter 17 that some of these solutions can be matched in regions of common validity.55) Extending these solutions to the domain around the minimum at z = 7r/2. \vw).52b) The first solution is valid asymptotically around z = —vr/2.qo . These solutions are valid asymptotically with the first in the domain around z = ir/2 and the second in the domain around z = —ir/2. i.
58) Ignoring the correction A / 8 .Ah2 cos2 z ip = 0. . D.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 . p. F. Byrd and M. as we shall see with the result of Eq. I^A(Z) overlap parts of the domains of validity of these solutions. we see that the turning points at z± are given by 2hq . (20. C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °. 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. Friedman [40]. (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum.Ah2 cos2 z ~ 0.e.01. Thereafter we extend the latter across the turning point down to periodic WKB solutions. e x p ( . Some nontrivial manipulations are again required to uncover the generation of important factorials.3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum. i. Jz+ ' T In principle the integral could be evaluated exactly — see P.[ [Ah cos z .20. These then provide complex exponentials like (20.2hq] / dz) We know that the solutions I/JA{Z). Our procedure now is to match these to exponential WKB solutions. 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A. (20.61) [Ah2 cos2 z2hq]1/2dz. Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — . 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 (20.67).49) and we can apply our boundary condition. r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 . formula 280.
64) . 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 /•«"(.» _ i y v . we write cos2 z = l .62) as follows. using the relation 1 + sin z = 2sin2(2i/2 + 7r/4).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 (20.490 CHAPTER 20.cos2 ( z+^) = (z+ .sin2 z = l .62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20.63) We handle the logarithmic terms in Eq. We have for the following expression contained in (20.) +0 z+ IT Inserting this into the integral I. Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z. 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\. Jz++7r/2 \ 1/2 2 V Integrating this becomes z+(ir/2) 2/i h 2 Z+ V IT 2h 4/i In ^ + 1/2 zA 2h 2 2h I ln z+ i) ( I 1/2 2 + 2+ 2 1/2 2/i J (20. (20.59) 2h COS Z+  ~ 0. (20.
2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20.^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( .67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation.63) and (20.66) The WKB solution therefore becomes (approximating the denominator of (20. .3 Cosine Potential: Complex Eigenvalues Inserting the relations (20. (4^2)i/4 [ i( g _3)] ! V/ l (20. but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I . (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 . on using in the second step Eqs.+ qj V2 4 A. ./ * [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\"  .68) ' T h e use of Stirling's formula here and below — essential to obtain factorials — assumes q to be large so that corresponding corrections would have to be calculated. we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan .62).64) into Eq. 1/2 (20. We ignore this here.50b). (20. .50a) and (20.65) [Z[4h 2cos2z2hq]1/2d< e2he2hsin ^eg/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20.60) by (4h2 cos2 z)1!4).20.
68) we obtain ^±(z) = obpA{z)±t/JA{z) l/^y/4e2fe(27r)1/22^^(4/i2)1/4 1 « llfh\^e^[l(q3)]\2^ ± /(r. io(8/*)g/V^i6ft2)V4[±(g3)]! ..?/ e^2 /2(4/ l 2)i/4 [ i ±2 (^)V^ 4 1 (g _ 3)]! (/]Z+) ^WKBW Inserting here the oscillatory W K B expressions (cf.49). 1 2 1 ±2 .1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .™) . S1H (2/ KZ ) / (z f 2 (2^).2 [ ? ) Imposing the boundary condition (20.2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^^)+4 7T . we obtain l(8/i)g/ 4 e. Large Order Behaviour of Perturbation Expansions again using the Stirling formula. (14.e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20.U ^K^.67) and (20.z)}].z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i).49). (20.T 7 77 r )V2 /4(2 . we obtain T _ ( ± ) = 0. where _ 1 + [±) *) + £ (20. (14.55) and match these to the oscillatory W K B solutions V'WKE:( Z ) > V'WKB (Z) t o the left of the turning point at z+.—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { .492 CHAPTER 20.69) l(8M9/4e2^ \2 [ i ( g .+) + B^) HAh^/\ir. Eqs. i 27r(8^/ 4 2/t /4 ' r(±) . Thus with IPA(Z) and TPA(Z) taken from Eqs. i. We now return to the even and odd wave functions defined by Eq. (20.l^l !T ^^* T2 7 r ( ^^ /.50)).KW4^8 4 .i ( 2 / i g ) 1 / 2 ( z + .
20.72).h) = = E(q0.47). Eqs.26).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.21) expanded about q = qo. h) of Eq.3 ) I = }(»« !7rl/22I(.72) is our second transcendental equation. i. (20. Eqs.57) by the left hand side of Eq. for q an integer both sides are imaginary.57) and (20.72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20. (20.h)+i%Ego. (17.4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 . the result (20.3. In both cases we wrote (cf.e. who derived it for qo = 1 and guessed the form for general values of qo.h) + i l + O '4{[^ol)]!}2 E(q0. (20. Combining these equations by replacing (28/i2)<?/4 in Eq. (20. (20. (20.72) We observe that with the square root as on the right hand side both sides are complex.75) . (20. we obtain E(q.qo) El Ah' (20. Ex = 2h2 + 2hq0 + A 1 (<? .20.l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e.25)) E = A = Ei + E2. Equation (20.74) This result agrees with that of Stone and Reeve [262].h)+2h(qq0) (16h)i°+le'Sh E(qQ.73) Inserting this expression into A = E(q. We now have the two equations which have to be satisfied.3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ . in fact.h) + (qq0)( BE — 90 E{qQ.
Bogomol'nyi and V. * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles. B. MiillerKirsten and A. 7 8 ) i=l V .3. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2. we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo. Setting ng AEqo:=2iSfEqo.) {I6h)™/2e4h 1 + . J.9) the coefficients Ai of its perturbation expansion. i.e.494 CHAPTER 20. W. Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions. (20. Eq.* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state. we can recalculate with the help of Eq. Wiedemann [4] and Example 20. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj.I g  ± l ( 2 „. by others.2 below. Achuthan.W [£(901)]! < (20.80) *P. H. We had A= or _2ft2 + 2ft. (20. A = .4 R e c a l c u l a t i o n of large order behaviour Now that we have obtained QE. 20.76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level. the discontinuity across its cut from zero to infinity. The same formula can be derived in the case of the double well potential. (17. 25Eqo=8h[ 2\1/2 . . its tunneling properties and the large order behaviour.77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut. cf. Fateyev [26].+  .74b). f E . i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20. A.
J). *R.13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq.j + 1).74). and the method can be applied in many other cases. (20.13) out again: jPiiQj) = (g + 4 j . The integral is recognized to be of the type of the integral representation of the gamma function. Dingle [74]. We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results. B. For convenience we write Eq. (20. MullerKirsten [200]. Using again n = (go .81) We observe that this result agrees with the result of Eq.t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo . But as Dingle remarks. 20. j . the integration is seen to give (20. We also observe that these terms do not alternate in sign.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. i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q.l ) ( < ? + 4 j . so that an exact Borel summation is not possible. (20.. there is nothing special about this case. (20.14).83) One now sets ftH = ( 2 % P * ( 9 .3 ) P i _ 1 ( c ? .19) for the large values of i under consideration here.4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq.j) satisfy the recurrence relation (20.l ) +2[(q + 4j)2 + l + A}Pi_1(q. f H . (20.j) + {q + 4j + l)(q + 4j + 3)Pi1(q.20. a = J + \<1> (2084) .1). W. J.
Taking dominant terms. p(a. Now.85).K + ^ ] L + K^)WK. i. e. (20.e.ll{z)= L . (q + 4j) 2 ~ 4j(4j + 2g).1) + 2 (a — 1) a .86) one can convince oneself that Eq. g < 4 j . formula 13. (20. Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i.z)WaA/2(z) Setting V.h) 2+ 8h a p(a.^(Z). . which are solutions of the differential equation known as Whittaker's equation. (20. Pi(a) i. we regain Eq.27h pi(a) a = 4j(4j + 2 g .4 ) ^ .87) Taking terms of 0(h~l+1) of this equation. Whittaker functions WKjfl. (20.1/2(z) (a1)! (20.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1.h) = ^2pi(a) with p. (20. (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a .h).1) + 2 P i _i(a) + Pi_i(a + 1).89) Replacing here K by a and \x by 1/2. we obtain (2a .31.4. A.l/2l Wa. Stegun [1]. nx pi_i(a + l). 4j(4j + 2g + 4) + / . d2WKh dz2 + 1 4 K z zM2n WKtll = 0. the recurrence relation (20.a R .l.85) Considering now the generating function of the perturbation coefficients. z2 (20.85) implies p(a .e. (20.83) can be written j . 0 4j(4j + 2g) pi_i(a .h) +p(a + l. i.(a) oc .g. 1  1 / 2 (4 (20. a.e.90) See M.91) + Wa+lil/2(z) = (1 .496 CHAPTER 20. Abramowitz and I.
i/2(z) + Va+lA/2(z) = 497 2+ Va.87). . In the present case this means the replacements q — —q.4 Cosine Potential: Another Method of Calculation we obtain Vai.88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential.93) (a1)! ' where f(h).Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f .84)) {8h)q/2( Ah for j < 0.„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y. another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged. such that p(a.88) is known. h —> —h. > (20. i.h) = \~\Pi(&) with pi oc h l.g(h) are functions of h which in view of Eq. (20.20. Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz). 89. > > ^See W.(« . Oberhettinger [181]. (20.92) Comparing Eq. (20.h) ~/(fe) ' (a1)! . g(h)(20.\)V . Dingle discovered the solutions^ p(a. p. (20. ^Observe that if one solution of Whittaker's equation (20.e. Magnus and F . h) is expressed in descending powers of h.92) with Eq.(« 1) 2 ].94) Hence to insure that p(a.86) have to be chosen such that p(a.(« ..j — —j. a (20. we have to remove a factor (recalling a = j + (7/2 of Eq.i/2(z). We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20. v . h) is expressed in descending powers of z = 8/i.
j _ g / 2  1 / 2 ( .95a) Inserting (20. This expression can be obtained from the recurrence relation (20. we obtain the result of Dingle: (20. (20. ^ ...+ i f or 3 < 0 (20. Correspondingly we set for j > 0 / 1 A pla = j + q.96) and a similar expression for j < 0 with /i — —/i. n = 0. j — —j.8 / t ) for j > 0.84) 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. (20. Dingle and H. Comparing > > > this result now with Eq. B. i). Muller [74]. W. this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl. Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2.83) and is given by 1 1 .. _ i s. 2.l ) ] ! 11 R. .498 CHAPTER 20.94) here we have a series in descending powers of h.1/2(8^) ( . (20.1. h\ oc (8h)q/2e~4h ( _ .).95b) Then for j > 0 we obtain from Eq. g — —q. x 4i[2» + i(« — 1)1! il! [ ^ ( 9 . J.94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1.
i. W. p.100) **R.e. l{Q. is M2l = 2j2j[Pi(<l. Ryzhik [122]. .3)]!2 2 '+^.0) = 1 as desired (although not really enforcible for large values of i).e.{j + \<l).20.98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q.99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20. note the misprint there: (r — s) ought to be (r — s)\.97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q. 5.+i?_l]!}2 (20.j) (i + lg)[2i + i ( g .l ) ] ! 2 M [i + \q)\ i\\l{ql)]\(J + \q) [j + lq]\[ij]\ (20.97c) We assume that in a leading approximation we can cancel the factors (i + \q). formula 0.4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q. J.157(4).l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq. Thus ° 9 (i + \q) 2«(i+Igl)! p .l ] ! (20. M. (20. Dingle and H. formula (30). (20.1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g . E* J'=I i+M 2 [2i + 9 . A special case of the formula can be found in I. of Eq.14). Miiller [74]. B.l ] ! {[.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. Gradshteyn and I. S.Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (20.97d) Then the coefficient M<n of the eigenvalue expansion. (20.
.102) Thus the result is M m 1 2«Fi + g .e.2.e. . (18..3 ) ] ! _ 1 Hence M or. [19].86) and set (cf.105 ) the latter because we see from the eigenvalue expansion (18. i. i.500 CHAPTER 20.101) This formula allows us to cancel all factorials [i. i\ M i large 1 . We refer back to Eq. (18.4)) £ = ^ = ^+ w and y= ^ > ( 20 . n = 0.1. 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. since \{q — l) + \{q — 3) = \q.86)... i.' 1 i • (20. The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity.l. Eq.34) that £ is an expansion in ascending powers of h 6 /c 2 .Am i . i.i ) ] ! [ ^ +  ( g .e. [^ + i ( g .e. replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20. Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20. Eq. i.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.17). (18..81) for q — qo = 2 n + l .5 20.o i • .l l ! with the large i behaviour in agreement with the result (20.e.]\. the result of Bender and Wu [18]..5. 20.
i. h2 _ h222K+llHr+K+ll2{l)r+1 r+ K .108) Proceeding in this way. 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.110) This is the result in agreement with the large order formula (4. The result establishes the series unequivocally as an asymptotic series.107) we obtain (i)r r 71" JO yirZ£{y')dy'. (with E = h2£/2) my') = 290+1 (y')9o/2 e 2/'/3 (27r)V2[i(go_i)]!' and integrating with the help of the definition of the factorial or gamma function. in fact as one with Borel summability in view of the factor ( . in terms of K where go = 2ftT + 1. i. We leave this to the following Example.l ) r .5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20. and then the relation (20. .106) With £= E r=0 Mr (20.109) or. and hence inserting here the imaginary part from Eq.86). the discontinuity across its cut. 20. (18.77) can be verified.e.e.5. This can then be compared with the level splitting we obtained for this potential.4) of Bender and Wu [19].20.(20. (20.2 The double well It is clear that one can now proceed and apply a nonselfadjoint boundary condition also in the case of the double well potential and thus obtain the imaginary part of the eigenvalue. one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20.
J. Damburg. Cizek. The relation (20. as in later chapters. W. W. J. Cizek. Propin. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292]. Silverstone [52]. E. *R.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. J . Wu [18].77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26]. Bender and T. H. + 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. Harris. K. T.77) and its possible generality has been referred to by Brezin and ZinnJustin [36]. t The relation (20. " P . Nakai. it can be used to obtain the other. MiillerKirsten [164]. Grecchi. Solution: For the solution we refer to the literature. V. Harrell. Paldus and H. MiillerKirsten and A.Q . [19] can be traced back from articles in the Proc. in t h a t when the quantity on one side is known. Paldus. Eqs. S.77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . W. R. MiillerKirsten and A. Achuthan. Silverstone [62]. . Harrell. J. V. 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. of International Workshop on Perturbation Theory at Large Order (Florida. Graffi. J. J. t p . Its usefulness has also been demonstrated there.6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. S. J. Liang and H.^ 20. Achuthan.* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161]. Gram. Grecchi. Propin and H. (357) to (371). E. Damburg. 1982) [139]. J. M. *Some of the literature in this field following the work of C. J. K. J. R. M. M. J. Wiedemann [4]. 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.502 CHAPTER 20. R. G. 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). H. Wiedemann [4]. A relation similar t o (20. S. Large Order Behaviour of Perturbation Expansions Example 20.
'See in particular H. *R. Felsager [91]. P. The formalism is not so useful for boundstate problems (in this case the initial wave function is. However. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. R. P. Feynman and A. our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. Schulman [245].Chapter 21 The P a t h Integral Formalism 21. S. Chapters 2 and 5. ^ Our interest here focusses on quantum mechanics. which explains also difficulties.* 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. However. We illustrate the method here by rederiving the Rutherford scattering formula with this method. can be found in the book of B. Coulomb problem) we refer to existing literature. however.1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method. A further wellknown reference is the book of L. This formalism is particularly useful for evaluating scattering or transition amplitudes. 503 . in general. Probably the most readable introduction. 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. Feynman [93]. unknown). A brief introduction is also contained in R. For the application of path integrals to the hydrogen atom (i. Hibbs [95].e. P. A standard reference is the book of R. Kleinert [151]. since in such cases the probability of this system choosing a path far away may be considered to be small. 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. The path integral is.
we can compute the action S1. (21.t). ti). (21.504 CHAPTER 21.ti\tp).tf\ip) .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. x.e—»0 Sn. =J dx(xf. ti) = 5(xf . Obviously K(xf. The Feynman Path Integral Formalism 21.x fc _i) 2 2e2 V{xk) Thus: Given xk(tk). *In Dirac's notation we write this equation later (xf.2) where K is a Green's function.ti)ip(x.e.4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo.e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo. (21.5) Sn = X / fc=l mo (zfe .ti). we wish to know* ip(xf.3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e.tf. i. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf.tf\x. U.1) Let the wave function i/> at time t = ti.x). be ip{x.tf) = / dxK(xf. 2mo dx2 (21.x. */ = U + (n + l)e. the initial time. (21. The Lagrangian L of a particle of mass mo moving in a potential V(x) in onetime and onespace dimensions is given by L = TV= m0x2V(x).ti){x.
Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi. the Fresnel integral (13. of which one is illustrated in Fig.6a) . 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. Here and in the following we need one or the other of the following integrals (cf.2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e.28)): f dr\ exp ia a) ' (21. The repetition of the quadratic factors in the argument of the exponential with indices differing by 1 suggests the construction of a recurrence relation.tf).ti) to its final position at (xf. as will be shown to be possible below. 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. 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.1 A path in (x.21. The usefulness of the method therefore depends on the fact that some paths are more probable than others. 21.t).
t ~Xk+l> + xk+i 2 ) +{xk . J—oo dxneiS/h — • 0.ti) '•= f g m ) (N(€\]n+i dxL. (21. The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0. tf > U.Zfci) 2 } I +{xk+i x*. n > 0.2 + (x + x f c + 1 .* W i t h the first of these integrals we obtain (compare the result with the Green's function (7. one has j ^ dr1eifa+i^ /£ = .37)) dxk exp d(xk .506 CHAPTER 21. 2 + 2a.6c) In the evaluation of these integrals we replace a by a + ifi.7) Thus we expect t h a t — in view of the multiplicative y/e here in front of the phase factor — in the limit e — 0 > oo />oo /•oo / dxi oo / J—oo dx2.(x fc+ i . ie f me\ 2a\a~) 1/2 (21. dx exp z—{a.z f c _i) 2 } } . 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.8) Note t h a t in spite of the n integrations.ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT.6b) —c ia o drjr] e x p—V € (21. (21.xk_i) dx exp i — { 2 a . with f™ dze"™ 2 * 2 / 2 = s/2ir/w2..0 ^7 (x f c + i . One therefore removes the troublesome factor by defining (Sn being the sum in Eq.xk+i) exp .m0 . for example..xk+x oo —oo oo .5)): G(xf. dxneiSn/n.Xi.x f c _ i ) 5 \m0 J (21..tf. We then integrate and finally let /x go to zero.mQf Xk L ~2e^ ^ .s f c _i) + (xk+i .
2).tf. It is neither a variable of integration in T){x} —> Ylndxn. (21. In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path.ti) For tf > ti we now write* G(xf. The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq. (21. Felsager [91]. (21. (21. Different constructions of these families of paths usually lead to different measures. However. G(xf. as in our discretization in Eq. we arrive at a path integral without the measure factor N(e) above. Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2. (21.9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman. it may be noted that with a different philosophy concerning the summation over paths.10) can be looked at as the product of a succession of free particle Green's functions (or propagators). (21.ti) for tf > U. the Green's function K of Eq. p. .24). These integrations are indicated by the horizontal arrows in Fig. and will therefore not be attempted here.Xi.g.tf. i. i.tf. Felsager [91]. § See e.Xi.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure. We shall see later (with Eq. in fact.21.U) = / JXi=x(ti) = K(xf. A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration.e. 183.t In the following we demonstrate that the function constructed in this way is.g.Xi. and so on. (21. 11 For discussions see e. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2. 183. B. which all vary from —oo to oo.1. summation over all possible paths. See the square root in front of the exponential in Eq.5).7) for n — 1) to give a time interval.2 The Path Integral and Green's Functions 507 Eq. nor the limit of integration of some variable Xi.5). p.26)) that — ignoring the potential — the multiple integral (21. however prefixed with a new factor. 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. B. 21.e.
we obtain the same as before except that there is no factor N(e). Consider a time t = tf + e for position Xf = xn+2. One way to avoid this problem with the integration measure is to consider the ratio of two path integrals. Then the measure drops out. (21. one with the full action S.Xi. we have See. so that the piecewise linear paths fill the entire space of paths. and one is calculating the Feynman amplitude or kernel relative to that of the free theory.74)). Expanding this in a Taylor series around x.tf + e. 21.ti) = 1 j^p: f°° cte n +iexp } • fm0. pp.72) to (23. however. Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig. the discussion in B.ti. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths. Each of these linear pieces is completely characterized by its endpoints.tf + e. (23. the other with that of the action of the free particle SQ.ti). I We employ this method later in Example 23.Xi.1 and allowing n to go to infinity.3 in the evaluation of I a specific path integral (see Eqs.xi.exp 2e n ieV(xn+2) G(xn+2 + V. It is for these reasons that — unless required — the measure is frequently ignored.508 CHAPTER 21.Then we can construct a recurrence relation as follows. We have G(xn+2. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{. . Felsager [91]. we have G(xn+2.xi.11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf. 187 . 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. 16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n.188.U) imo 2 G(xn+i W)Idr.tf. Then we have on the right the factor G{x+rj).
tf.Xi. .tf.tf.ti)\ .ti) = j^rr dnexp I ^rj1 .ti) for tf > ti (21. U) + Q(e2 = (lieV(x) + 0(e2)) G{x.tf.ti) .(21.14) = G(x.ti)ip(x. Xi.2 The Path Integral and Green's Functions so that G(x. we obtain: G(x.13) We integrate with the help of the integrals (21.tf''Xi.Xi.Xi.tf. tf.xi. d and this means d i—G(x.tf)= dxG(xf.xi.ti) = 1 d2 + V(x) G(x.xi.U).Xi.x.ti) +\if(^G{x.ieV(x) + ri< —G(x.ti) N(e) or G(x.ti) 2 —l€ — ^2 + V(x)G(x.Xi.ti) +V(x)G(x. U) .6b) and (21.21.xi.ti) = G(x.Xi.Xi.U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x.tf. U) = e^G + 0(e 2 ) dtf 1 d2G(x. tf.xi.tf.6c).tf + e.16) .xi.ie 1 d2 2nV0dx^G{X. It follows that ip(xf. U) + .6a). tf + e.xi.Xi.15) We see that G(x.tf.tf + e.tf + e. 2mo dx2 (21.tf.t Thus + 0(e2 (21. Retaining only terms up to those linear in e.U) +••• \ 509 G(x.U + 0(e ).— p G ( j . (21.tf. ti) satisfies the timedependent Schrodinger equation.tf. tf.G(x. tf.Xi.xi.
10) is indeed to be identified with the timedependent Green's function.8)) Go(xf. as may be verified by inserting this into Eq.U) = / T){x} eiS\ [fdt \mx2.x.1)). (21.20) .x.510 CHAPTER 21.tf.U) = G(xf.U)6(tf . is the Fourier transform of the nonrelativistic propagator.1 Configuration space representation In the case of V = 0 we have from the previous section for the action. The function G which satisfies the timedependent Schrodinger equation (21.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0.ti) = 5(xj — x).15) is a wave function (cf. denoted by KQ. (21. as this is also called. now called So. Eq.x. The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation.U. Hence our original function K is the same as G.tf. (21.ti) = = 9{tfti)—G + G5(tfti) . X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation. 21. (21.17) Differentiating K with respect to tf we obtain —K(xf.3. 21. Eq. and the Green's function KQ\ S0= Here (cf.Xi. Thus we consider here the zeroth order of the latter case and begin with the configuration space representation of this free particle Green's function. or K(xf.U). The result.*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf.tf. (21.U)G0.tf.x. K0 = 9(tf .16). We have therefore established that the Feynman integral (21. It follows therefore that {{W ~V+2m~0~^) K{Xh t/. In a problem where V is small it is convenient to calculate K with a perturbation method.
But t = tf — ti = (n + l)e.x 0 )" n + 1 (in agreement with Eq. (21. f c .t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function.7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21.25) T h e quantity Ko(x./ .tf.oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x. Then Go(xf.tf. TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21.Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21. oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a.23) We now insert our expression for In into Eq. In Example 21. n + i .t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi.1 we show t h a t oo /. exp where Xj = xo and Xf = xn+\ and there k = 1)..ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2.ti) Ko(xf.ti)= iX (o.Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq.ti) = = G0(x.24) ( s / . exp 2 (21.Xi. . (21.Xi.tf.Xi.37) — by explicit solution of the differential equation which the Green's function satisfies.Xi.Xi.t) K0(x. (7. (21.3 The Green's Function for Potential and hence Go{xf.23) for Go and obtain Go(xf. We obtained the same expression previously — see Eq.z / c . we have Go(xf.21) fe=l and t = (n + l)e = tf — U and h = 1. (21.tf.22) • \ n/2 .ti) lim n^oo lim n>oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )(a.i ) 2 . 2t Go(xf.tf..ti)9(t).Xi.tf.21.8).
22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:.54). or else by induction. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 .x ^ 2 + y2 .x0.xo)2 + (x2 . (7. For n = 1 (see below) h This result can be verified like Eq.22). Solution: The result can be obtained by cumbersome integration.)!} J — OO . (21.f_^y e.x „ + i ) 2 = ( x n + 2 Xi + Xi .tn)••• K0(xi. (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 (21.1: Evaluation of a path integral Verify Eq.512 CHAPTER 21.^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i. (21.x „ + i ) 2 = (x„ + 2 .10) in a product form. Then oo 2 / (21.5) is carried along.ti.2y(xn+2 .e. ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l. (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.26) with n integrations and n + l factors KQ. ^ (xi . e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() .)+(1"+2. Eq.6a).7) by direct integration with ^ .25) we can rewrite the Feynman path integral or kernel (21.tn+l•xn.. i.. (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 . In our proof of induction we assume the result is correct for n. Example 21. If the potential V(x) of Eq.t0) (21. the latter being a Green's function with interaction like that for the harmonic oscillator. we have instead of 5*0 and KQ expressions S and K._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq.xi)2 = 21 xi ^2o(„. as / JXi f r D{x{t)}eiS°lh .
(21. A(n + 2) 21. iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A .30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m.i) H —3/ . This quantity is also described as the nonrelativistic free particle propagator.2y(xn+2 . (21. a dp exp [ipx + iap2] = exp — i(21.i j ) H n +— z 1 .n+2a. The second integral is an integral representation of the step function.max exp i 2t 6(t) 1/2 \2mtJ f°° 2KimQ J dp exp J—c ipx P t 2mo (21.Xi) + ( x n + 2 Xi)2. n + 2 .e. and + 2 2 l n + 2 ( x n + 2 .x2rriQ 1/2 toij_. ) H — (xn+2Xi) n+2 n+l n+2 y2 . (n+l)A n n j(a.25): Ko{xJ) = = (S) e{t)(^X'2 y J .2 y ( a .3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj). it . We pass over to momentum space with the help of the following two integrals.2 M o m e n t u m space represenation Our next task is to derive the momentum space representation of the free particle Green's function KQ.a .i) .27) we have (with a = t/2m0) . Using Eq.2.. 0(t) Ti: i r Art alr This integral can be verified by applying Cauchy's residue theorem in the plane of complex r and using one or the other of the contours shown in Fig. s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a.21.29) Hence with KQ from Eq.28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21. (21. (21. n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 .3. e > 0 .6a).27) '4a in can be verified by completing the square in the argument of the exponential in the integrand to ia(p + x/2a)2 and using Eq. i.Q . rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i . The first. 21.
Replacing 9(t) by its integral representation (21. 21. the second order two factors of V and so on. 21.28) this becomes K0(x.31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function. Thus in first . This is the type of expansion most people describe as perturbation theory.Q (21. (21.51)) is fairly selfevident.514 CHAPTER 21.4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion. The Feynman Path Integral Formalism Im t \t>0 Rex Fig.e.*' where 2mo — r. ( 3 b) ° The expression E+ V — it 2m.t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£. i. an expansion in rising powers of the coupling constant (contained in the potential).2 Contours of integration. The generalization to three space dimensions (required below in Eq. The first order correction contains one factor of the potential V.
i V e / D { a .)} (21. } e °° f l f •/ oo /"oo F ( a . ^ = k ) = lim (S^ A H*\ nKX) *—J \2irieJ r "+1 />oo X / dsi / dx2..i. .t/. (21.^ J_00 V(xk. . (21.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x..t)\. li f f 515 order (cf.21.Xi.Xi.U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2. .21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/.4 Including V to First Order: Perturbation Expansion h=l.20)) and with elS = exp i S i o / hi dtV(x.^.. dxn exp i ^ —.ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x.tf.+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq.32) Since — we recall this for convenience G(xf.tf.t) the first order correction is (G\ ~ G — Go) Gi(xf. dxr.^){iV(a..U). x (n+l)/2 \ lim .5) a n d (21.xi.t)] n+l ._i By regrouping factors this becomes Gl n+l lim ^ e n—>oo V{xhle) dxn / i ci) n+l dx z+i / * * / d / _ i . (21.(xfc .33) xG0(xi.t) JSo dtV(x. dxi ( m0 V V27rie/ m0 \^He \ exp fe=.xfc_ OO J— OO J—OO L J.i. Eqs. \2meJ n+l J. .ti.
516 CHAPTER 21.ti).t2) xKo(x2.xi. dx2Ko(xf.x2.xi. (21.U) xV(xi.e.36) {if •j. = i dt2 / dx2Ki(xf.t2)V(x2.I dtV(x. They must therefore be deleted.ti)K0(x.t) ! Jti r*f dt'V(x.t.x.tf. The factors of "V" originate from the series expansion of the exponential exp ftf . OO /"OO / oo dt / J—oo dxKo(xf.t)K0(x. (21. Jti Consider the term of second order. i. 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') + • • • . The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.tf. (21.37) Jti The last two contributions do not describe the time evolution from ti to tf. Xi.ti).t2)V(x2.x2.t) (21.t) =li dtV(x. in writing down the expansion exp i dtV(x. .xi.t2.tf. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky. / Vdt+ JIfVdt)( fti = Jti Jti ftf ) \Jti pti Jti I' Vdt' + f f Vdt' Jti rtf Vdt' J t\ j Hi Vdt rt\ Vdt' Jti Vdt Jt\ Vdt'+ Vdt+ Jti + \ S Vdt I Jti f Vdt'.t)V(x. This is cancelled by n! different time orderings.35) We also observe here that the problem of timeordering arises only in perturbation theory.i / dtV(x.ti) t2)K0(x2.34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t). i.e. t) Ju Thus the term containing n times V contains a factor 1/n!.
i ] P I dt J dxKn(xf.ti) (21.x.t)= f dpdEeipxiEtV(p. The diagrams represent mathematical quantities and hence are designed on the basis of rules.ti). called Feynman rules.t) is V(p.x.tf.Xi. tf.21.39).E).Xi.Xi.ti) i dt = K0(xf. The Fourier transform of V(x. These may be formulated in configuration space as below.t)Ko(x.xi. of course. x r l i 'V(x.4 Including V to First Order: Perturbation Expansion 517 The full perturbation expansion of the Green's function K is now seen to be given by K(xf.t) ) K K K0 'V X K o Fig.3 Feynman rules and representation of the Green's function.40) The rules for the present case are given in Fig.t. (21.t2 K o (x 2'l2 .ti).t. or in momentum space as mostly in field theory.tf. x"^ X x"2.tf.39) dxK(xf.t)V(x.tf. For various reasons — such as illustration. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory.Xi. (21. 21.38) . we are concerned with quantum mechanics.ti) = K0(xf.3 along with the diagramatic representation of Eq.t)K0(x.t)V(x.tf. Such diagrams in the context of field theory are called Feynman diagrams. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj. U) (21. Here. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms. .Xi. 21. E) which is given by (in onetime plus onespace dimensions) V(x.
The plane wave function normalized to one over a box of volume V = L3 is. mo V since ip is normalized to one particle in volume V.41) We have to pass over to three space dimensions. * k 1 • .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. (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. ^(x.t) = .ti). and dn = dnxdnydnz. with p = Kk.ti)il)i(xi. T = duration of time for which potential is switched on.518 CHAPTER 21. to one particle in all of space.ipf have to be properly normalized.e. in onespace plus onetime dimensions by A= dxf dxiil)*f{xf.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. (21. for the allowed values of k: kx = 2imx/L.tf. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions.e. There the wave functions ipi. here with E — k 2 /2mo.= l. We recall for convenience from **Note that from the wave function we obtain.Xi. i.7 = e i k ' x _ i : l i with f i/}*i/jdx. (3) The transition probability per second is = \A\2 . The Feynman Path Integral Formalism 21. dn = (L/27r) 3 dk. . 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics.tj)K{xf. (21.k = 2wn/L. i. 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).
Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p.21. Then for stationary states M^t) and itm^ = f HUt'V^t'dt'. (21. # = ^an(i)VnM) n = a^(t) + \a^(t) + . (10.4). Proceeding in this manner one obtains Fermi's "golden rule".k..47) .t) = e~^lT\ r where r = Ixl..T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential.44) = u(Ek.44) and f_oo can be replaced by fQ.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. V(x. «= — . in which T is a large but finite time. Then H'ki can be taken out of the integrand in Eq. i. Eq. and with these the following perturbation ansatz: ih^* with an(t) = HV.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. u = ^\H'kfp(k).= 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.e. 47T 1 3 7 (21.— .117) and Example 10. P(*) = j r (2145) (cf. (21. H = H0 + \H'. The transition probability per unit time is .
T a large but finite time. (21. x. we obtain A1 = clS(ti tf) / dxf(i)V(xf. also called Born approximation.tf. i.f.520 CHAPTER 21.* / .t)V(x.x. We can relate this amplitude to the amplitude ark of Eq.f. so that 0(t) can be dropped) K0(xf. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x.t) (21.i x (i)V(x.44) (here in the scattering problem with initial and final energies equal). This now has the form of the amplitude (21. (21. by threedimensional dxf and so on..tf)Ki(x. t)K0(x. (21.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P . (21. dt dxip*f(xf.52) ..49) together with the following wave functions (of asymptotically free particles) 1>i(*i.xuti) = ^ J°° d p e i p x _ i ^ * .>/>. (x*. Ai = dxf / dx.^ / .ti). (21. (21.tf.30a) K0(xf.ti)'ipi(xi.51) Inserting this into Eq.tf)ip*(xf.tf. ti) = co5(t — U)5(x — Xj) etc.tf. Xj.tf). The lowest order approximation of A.t) = ^ J*ie*<*'*><<*'*>. £.tf)K0(xf.tf)ipi(xf. Thus with the ansatz Ko(x. (21.49) i.t.39) OO /"00 / dt / OO J —OO dxK0(xf.50) Here x — xj — Xi and t = tf — ti.e.ti).). The threedimensional generalization is (taking tf » T and ti <C —T. we have to replace the onedimensional quantities dxf etc.tfx. The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas.t)K0(x. t. (21. In our previous onedimensional considerations we had with Eq. t.48) where from Eq.Xi.x.yLi.44) by contracting the propagators to instantaneous point interactions.
T) = x exp L z/ q/2 k2 521 f dx.55) We evaluate this integral with the help of the following integral in which we complete the square in the argument of the exponential and set r = t . this becomes A(p. 2 „ x v~ m o (21.f f dxi f dt f dx f dq / d q ' p V 2 (2TT)6 2 +iq' • (x — Xj) — i 2mo (t — U) + ik • x 2mo .k.58) In doing the integration over x we introduce a temporary cutoff r = 1/M by inserting the factor e~Mr in the integrand.T) = ^JdXJdt%*t2/T2exp i ( k . k2 ipx + i—t + ikx — i 1 2mo 2mo (21. (21.5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p. P2 .53) The integrations with respect to Xf and Xj (including in each case a factor (27r)3) yield delta functions 5(q — p) and <5(k — q') respectively.~ .54) Now inserting the expression for the potential V(x.2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21. £).t)exp A(p.56) (21. Then integrating over q and q' we obtain .k.).fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 .p ) . Then we require the following integral.r) = = y . At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 .59) .21. (21. (21.p 2. which is evaluated in Example 21.k.i ( k 2 .T) = JdtJdx(^\V(x.iaT2/8: L With 0 0 • .k.57) we have A(p. .
where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson).\ 2?7lo / 1 (21. (21.64) Inserting the result now into our expression for the total cross section. This expression assumes that the integration over time t is normalized to 1. ( 21 .„ 2. .a ( — ) ^ .60) Now. y 2 [ ( k . write down the probability per unit time which is ( 8 \1'2 _a2 16^ . e (21. otot contains the factor \A(p. Using the result (21.T2f 8 V/ 2 . e 8{k) = lim 7=*—^ (21. { I i _ F T l } .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." + M ] p) ' "•"• ^2 .P ) 2 + M2]2 4 VvrT2. mo m (21 . i.522 CHAPTER 21.59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P .61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2.e.k.< ^ £ \M VTTT2. k . 2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k . we obtain . 66) o . Using Eq. The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21.65) But E=*. For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T . In our case this has not been done.„. dE = PjV = ^ E ± .3.63) we have: Probability per unit time = 77777. (21.62) We want to consider the limit T — oo.T)2/T. <rtot. k2 . integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T.63) We establish this result for T — 1/e in Example 21. T ) ~ .^ .59).51 — 2 /w2 y ~ —. / / P 2 _ .
70) .21.e .M £ M * = / m0kdn = m0^2m0E (k .e. Then (x = r) foo „2jj. p 2 = k 2 . This result may also be obtained from lyukawa of Eq. (21.= e " f c a ^ o 2v^fa /4a = lim £^o —.. 2mo / dft.59) by replacing there q 2 by ( k — p ) 2 . (21.e. esfn (21. r\ poo Yukawa = = ^ Jo 2TT / r e~Mr y_i dre~Mr d(cos 9)e^r sin qr = — G / cos e = 2TT / J0 —e~Mr iqr { eiqr . with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (21.5 Rederivation so t h a t of the Rutherford Formula 523 M^f) .cos 9) = 4k 2 sin 2  .65) with the cutoff M > 0. 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. /Yukawa = ) Solution: We choose the 2axis along the vector q.4.68) into Eq. i. (21.^ .3: A representation of the delta function Determine the Fourier transformation of e~ax of the delta function <J(fe) = lim — / dxe~ax a*o 2K Joo for a > 0 and hence the following representation eikx = lim — . so t h a t (21.^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21.2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e .p ) 2 = 2 k 2 ( l .^ .j .67) T h e delta function here implies energy conservation. —Mr 4w = .j .68) Inserting (21. giving the potential a coupling constant. and inserting the expression thus obtained into the final result of Example 10.* 9 ' d r e .69) This is the wellknown Rutherford formula. setting M2 = 0.67) and (21. Example 21.
Since dxe" 1 1 1 = « / .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 .^0 a^C 2^/TTO.. 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 ) .q). however.72) co From (21.71) Solution: In general one uses for the evaluation of such an integral the method of contour integration. co oo dx co (eikx)e~ax dx J = — / ikelkxe~ax 2a 7 _ 0 0 dx = ia^ Thus g(fc) satisfies the first order differential equation g'{k) + (k/2a)g(k) yields g(k) = Ce~k2/4a with the constant C = g(0). We have x> = I J ^ l p X p l x ) .73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx '' (2L74) The normalizations (qp) = (27r)3<5(p . Va we obtain g(k) = Je~k V a = 0. In configuration space the state of the particle is written x). a simpler method suffices. (21. (21.524 The function required is CHAPTER 21. In the present case. (yx) = 6(y .71) with respect to k yields the same as °° 2 . ' i a = lim e —0 pk 2 /e2 e^TT 21. 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /.oo J / )eikx.75) . Direct integration /4a .* . '<*> = .x) (21. The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx. (21. Differentiation of (21. a. We consider again nonrelativistic quantum mechanics.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. We write the state vector of a particle of threemomentum p in the momentum space representation: p).70) and (21.
Similarly we have a position operator X with Xx) = xx). i. t\ipo)s. Thus (qPp) = p(qp) = p(27r)3<5(q . which describes the propagation of the particle from its initial position x. The time development is given by the Hamilton operator H. Then the states {x. so that \if>t=o)s = H>)H. (21. ) H = (x</>o)s.83) (21. (with ft = 1) d/dt\ip) = —iH\tp).x.6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc .82): The general Schrodinger equation has the state vector to the right.21. x. i.t).We define x. 159 .e.76) These relations imply the completeness relations 1 = J dxx><x.t'. t) = (x. Ryder [241].80) (21.79) Feynman's principle is expressed in terms of the propagator or transition function (x'. d/dt\x.160.If '/'(x.82) (21. M^ip(x. The equation 1 = J ^p)(p.t).t). H. (px) = e~ipx.e.t\.78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p). Then ( X  ^ .^ l x ) .t> =iH\x.p). as had to be shown.81) Comments on signs in Eq.e.e. (21. (21.t'x.77) (21. pp.. Hence for the infinitesimal transition (x'.0) = e""x). x' — x infinitesimal.t\ = iH{x.i) = e iH4 x. i.i) = K(x'.i'x. (21.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s.t)} constitute a "moving reference frame" in the sense that ip{x. The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H. 4) is to satisfy the Schrodinger equation. we must have d/dt{x.t) = —iHip(x. See also L. (21.. i.e.t) = +iH\x. t' — t.We now introduce the Dirac bra and ket formalism. at initial time ti to its final position xy at time tf. We consider first an infinitesimal transition. i.t).£) = ( x V ^ ' . . so thattt d — x.t) = e* H t x).
(21.85) the amplitude or expectation value or position space representation (x'#x) = . (2187) (21.* . 2mo Since (21.89) Using this and previous relations we obtain for the Hamiltonian (21.x = V(p . 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).P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«.526 CHAPTER 21.i p '..y<&(p'ix')(x'ivix)<Xp) = f dxe.q) = qq>We also have by mapping onto configuration space Pp) = = P /*dxx)(xp) = P /"dxx)e i p x = p /"dxx)e i p x f dx\x)—eipx= /"dxx)V(xp).x') + V{x)5(x' .V 2 < 5 ( x . Pq) = J 7^3Pp)(pq) = J dppp>*(p .p').91) . i.e.90) (P'IVIP) = y<&.88) Thus P has the operator representation P= /dxx)V(x.x). "  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21. (21.x y(x)e i p .
92) We now return to the expression (21.t) (cf. i.e.tj).6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p .t') \^(27rf5(p 2mo +V(pp' where we used (21. (21.p) + i{t .p') + V(p .21. the signs are reversed when these operators are applied to the ket vector \x.fx. Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj.94) In the above we have chosen the signs such that p = —iV and H = i— ot (21.p') • • • j W < 5 ( p ' . t).82)). We now consider the transition of the particle from Xj.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 . 527 (21. ti to x j . We have <x'. for the amplitude of an infinitesimal transition.p'). tf subdivided into infinitesimal transitions (xj+i. i.93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (21.j.95) when applied to the bra vector (x. But e'ipx + 0[{t . (21.84).i'x. (x'. EQ — Hi .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 '.tj+i\x.92).
i.t 0 ) + H(p2.e. .i { f l " ( p i .xn)(tn in_i)].. T h e integration over the n—1 intermediate positions. then corresponds to the sum over all paths. t n  x n _ i . We now insert for each of the infinitesimal intermediate transitions the amplitude (21. (Xf.. . x i . t n _ i ) t„2><xi.t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o .xic) by fc — Xfej H(pk..ti) rx/=x(i/) d x i dx2 .27) in which we make the replacements x —> eifc.ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa . zmo T h e n t h e integration over all p ^ can be performed by using Eq.. d x i ( x n .98) tk — *fcl # ( p f c .99) . dx.ti. d x „ _ i .x i ) H h pn • ( x n . = 1 X X<jPfc We now replace H(pfc.dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi .. The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/.97) H hH(pn.94) and thus obtain a phase space expression. x exp [ .tfei) fe=i . .2.528 CHAPTER 21. d x n . d x i . (21.a — —e/2mo and divide b o t h > sides by 2TT. .tf\Xi.. The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21. x f c ) = ^ .X n _i)] h) (21. x n which the particle passes at times tQ. .tn respectively. . .x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf.i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) . ..tj) = / ••• / d x „ _ i d x n _ 2 .tixo.e.tf\Xi.x 0 ) + p 2 • (x 2 .xk (21. x i ) ( t i .i/xj.+ V(x f c ).. i.
p)]\. > Thus the phase space representation of the pathintegral* (21.ti) = J D{x(t)}D{p(t)}expi [' dt\pi.98).tf\xi.fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk .101) Then Eq.tfci) fe=i £(**'**){ p***^} —tki m0±2k (21. i. (21. Smolyanov. to / m0 \ 3 / 2 las. . Truman [254]. A.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (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. (21.100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •.e. (21. Tokarev and A.8) as N(e) so t h a t the integral does not vanish for e — 0. (21. G.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.H(x.] ex H ^I (. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21. i.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.e. / {xf.103) is another form of the pathintegral for which the integration with the help of Eq. G..21.
We know from classical mechanics that this relation is a Legendre transform which transforms from variables x. (21. xfc). (21. t.x) (21.105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}.iJ(x fc .x) .106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t). Thus finally we have Feynman's principle: (Xf. We have now obtained the path integral representation of the matrix element ( X / .ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x.tf\xi.tf\xi. I hi dtL(x. t to x. (21. £ J ) with no operator in between.104) where for the particularly simple Lagrangian L under consideration here m0±l .8).H(xfe. nl (xf. i. £ /  X J . Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation.530 CHAPTER 21.V(xk) = m0±l .t'}. operator (21.ti) where nl D{x(£)}exp '. this agrees with the corresponding discrepancy in the onedimensional case of Eq. 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'. Note that we have n factors [iV(e)]3 but only n — 1 integrations dxk. . The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq. pk) = L(xk.e. Garrod [106]. pfc) = Pfc • xfc .8). x. p . We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C.
*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(. .0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'. With the wave functions (21. . so that we obtain a cnumber position coordinate which we can shifted across other cnumbers. .* * .6 Path Integrals in Dirac's Notation For the verification we use Eq.t'}./ ^ I < « .* * f dp = j (2^ p) ^ d .t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x".109) .77) and thus have 531 *!*') = *M = * 7 ^ I P > « .v ' x ' and thus {x". (21.107). Thus <xV'xx'. x fdxG  ^ P <p"x3)<x3e^V) x (21.i p '• x .t"\x5){x5\x'.5 l^ t 'lp / > e .21. With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'.t"\x\x'. (21. (px).108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e. ••• y (2^ x= f dp p) ^(p d We now consider an expectation value in which we replace the operator x (which is sandwiched in between exponentials containing Hamilton operators) by the expression (21.76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ).
x / / e .0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') .0 dp x„ / (27T): 1 .• dx1 (x n .t') (xV'xx'.i(t" .£ n xx n _i.iit!' 1') 2mo + vtf) + <5(x"x').tf subdivided into n transitions (xj+i. ] .£/xxj.tf — tn and x$ = XQ.x') . (21. ( x " .f ) i f e i p .x ' ) (2^ i^+vM + D «p(x"x') (21. This is the expression corresponding to that of Eq.t'){x"\H\x') = X + •••} (21.ti = to as before.i{t" .113) Inserting for each of the infinitesimal intermediate transitions the amplitude .74) of the delta function. = x"6{x" . tn2) • • • (xi.tj) at equal time intervals e.*/<x"x5)(x5tfx'> + .it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • . (21. we obtain with (21.90) we can rewrite this expression as (x".112) where x" is a cnumber.£j to Xf.94) for the case with no operator in between. Inserting here the integral representation (21.t n _i) x ( x n _ i . tixx 0 .tj+i\x\xj. .t') dp .532 CHAPTER 21. t0).110) Recalling Eq.i t " ( x "  #  x 5 ) ( x V ) +. (21.111) 1 .85) (xVxx'.£j) = / ••• / dx n _idx n _2 . We now proceed from there along parallel lines. (21.t"\x\x'. We now consider the transition of the expectation value of operator x from Xj..x') . £ n _ixx n _ 2 . we obtain (here and in the following the limit n —> oo at the end being understood) (x/. The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK.£'> = x"[«5(x" . Thus with x^ = xn. .i ( t " .
115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] .\Xi.114).t"\x'.x 0 ) + p 2 • (x 2 . .. (21.and hence may be put in front of the integrals to yield (xf. Xn _i X„ xi dx...tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{.t') sider the variation S(x".112).t"\x'.tf\x. .( x „ . dyi l xi x 2 .114) x exp [i{pi • (xi . tj  xj1 x».7 Canonical Quantization from P a t h Integrals We saw above that in the path integral method the evolution of a system is expressed by a basic functional integral like (21. we obtain* (xf.ti) 533 rxf=x(tf) = / • • • / dXidx.« • > We reconsider therefore the transition amplitude (x".2 .t') = and con 5x"Vx„{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. For simplicity we consider the case of one spatial dimension.t') = (x" + 5x"..t"\x'.{^( P i.2 .t"\x'. x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 .117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq. .xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/. tj) = xf (x/.x2)(t2 ~ h) + • • • + H(p ••• + p „ . n—>oo *•»• 21. U).t"\x'.3 " n d x f c .{x". (21.x 1 )(t 1 Z & ) +H(p2.89) is consistent with the conjugate expression dL (2L116 * ..x n _ i ) ] .. x (21. (21. tj \ xt.21.y (27F (27F •" (2^(2^)3 exp[.7 Canonical Quantization from Path Integrals (21.iyxxj...t') . We demonstrate first that the operator representation of P given by Eq. dxnn _i Xi X2 .t') (21. x n _ i x n reduces to the single factor x^.105). the string of factors xi X2 .10) or (21.x .
Hence w « = (t).118) — = (21 5x"(x". . Thus in this case of classical mechanics dL_ d fdV.x). jt/ 37 ( £T.t') =  6— / 1 V°{x}t . and we demand the validity of the equation of motion. For the purpose of enabling some formulations below. dx dt \ dx _ (21.534 CHAPTER 21. V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x.i/(x)/fi ) j ^ ) dtL(x.10) and set D{x} = D°{x}/JV(i/ .l&cdi. x) Jt Jt dL.120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t'). t". dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21.tt).t'): 1 fx" 6(x".t"\x'.t"\x'. with N = N(t" . Then.t i_(dL_ x't' h\dx J t„ (21.t'). (21. dL r /•*" . <9L' dt —dx + — 5 x ox ox f .118) JI(x)/h . (21. " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq. (21. " (t). The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0.119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'.(x".122a) = 17) 5x"Vx/. dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t. we separate the metric factors from T){x} in Eq.
t'\ = l. Streater and A. we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh.124) where with A = tf — ti.122b) By comparison with Eq.1 (A)iV.126) R. S.Xf = x(tf). F.1 (A / ).7 Canonical Quantization from Path Integrals or JL(x»y\x.t"\p(t")\x'.21.. The following properties are assumed to hold: (1) / Jti V°{x} = ^ x. .115) in three space dimensions.t').t')(x'. (21. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory.x).U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21. Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti). (21. 535 (21. / Jxi V°{x} / Jx' V°{x}.1>) = ^{x".125c) Corresponding to our relation (21. § The evolution of the system is expressed by the Feynman functional integral (xf. x' (21.125a) (2) the completeness relation holds. 6 (21.125b) and the measure property (3) N~\A + A') = A^. I(x) = / JA dtL(x. ^2\x'. Wightman [264]. (21.89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization.tf\xi.
we have the relation (xf. (21. (21. (21.t') + x"^J^{x".t') (21.t') = ih(x. (21. (21.t\x'. .ti) f ' V0{x}xlX2eu^lh.536 CHAPTER 21.t\x'.t>).. We can rewrite this expression as the commutator relation (ITU*" x"^](x".t"\x'.t"\x'.. We consider the case n = 2. We have (cf. Eq. Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21.114) (xf.t')=x"(x".133) = ih{x".tf\x(ti)\xi.x](x.t') \ i ox" i ox"J or in commutator form [P.129) = ^ y With Eq.ti) = —!— p V0{x}xx • • • xneu^h. (21.t'){x\t'\x2\xi.t"\x'.ti).t').. The Feynman Path Integral Formalism The time t is in general a time between ti and tf.122b) we can deduce the nontrivial canonical commutation relation arising here. tn lie correspondingly in consecutive order between U and tf.t"\x'.t').t"\x'.tf\xi.ti) = x' 'Y^{xf. (2) and (3) above.131) = ih(x".130) ~i5x7jX"{x".. (21.127)) (x". But for t = ti we have (and equivalently for t = tf) as we can see from Eq.127) We can now write down an expression for a timeordered product of operators. (21.t"\X'.128) where tf is later than ti and t\. Thus (xf.t').ti) = Xi(xf. Thus if T denotes such ordering.tf\T(Xl • • • xn)\xi.t"\x"\x'.132) The generalization to higher spatial dimensions proceeds along parallel lines.tf\xi\x'. Then the relation follows with the help of the properties (1).tf\T(xix2)\xi.
the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation. such as the fine structure constant of quantum electrodynamics. Besides. W = e x p ( l / z ) . The vast majority of perturbation expansions discussed e. i. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e. An essential singularity is really an ambiguity rather than an infinity. for sufficiently large r. Expansions of this type arise.e.e. It was therefore natural to search for alternative methods of expansion.e. Whittaker and G. p. i. see e. in the sense of rigorous convergence tests. In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless.g. it has some finite. E. nonzero radius of convergence in the domain of the coupling parameter. 102. 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.g.e.Chapter 22 Classical Field Configurations 22. Watson [283]. the series converges. in the context of quantum mechanics are not convergent.1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. In physical applications such cases are rare.g. Consider e. divergent. N. If the expansion is mathematically well defined. z = a + i/3 with a — oo or f3 —> oo or both. as we have seen in Chapter 8. integrated over) beyond its radius of convergence.g. so that successive contributions to the first approximation do not invalidate this approximation. T. i. Such expansions are strictly speaking. the modulus of the ratio of the (r + l)th term to the rth term is larger than one. but asymptotic. although the first few terms indicate a convergent behaviour. > 537 . Such an expansion parameter is usually known to be or assumed to be small in some sense.
There. In searching for other means of expansion. higher order corrections are of orders ±. one considered in particular an expansion which is such that the first approximation is purely classical in a certain sense and such that this ignores quantum effects. such > a series begins with a contribution proportional to (e.e. although this terminology is not precise.h.h2. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties.538 CHAPTER 22. it is an expansion in rising powers of a semiclassical expansion parameter which plays the role of Planck's constant h in the quantum mechanical WKB approximation. the latter's topological properties if any. Many of these aspects are interesting by themselves . the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27.. 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 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. and in general this change is accompanied by certain constraints.e. These methods are often described as methods of collective coordinates.e. On the contrary. The fundamental extension of the method of canonical quantization (i.. This type of expansion is decsribed as "semiclassicar. A method which achieves this is in particular Feynman's path integral procedure. A further challenging task was the development of methods of quantization of theories which incorporate classical. nature of the dominant approximation does not imply that this describes classical motion. specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative". i.g. cnumber first approximations.h3. However. The classical.. One therefore faces the problem of quantizing a system which is subject to constraints. The study of expansions of this type has turned out to be extremely fruitful. cnumber. 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. In the following we consider various typical examples. such as soliton theory. i. 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). and has led to insights which previously seemed unimaginable.) l/h2. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0)..h°.
. We shall not be concerned so much with the case of a constant first approximation. t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity. (22. d^] = d^d^ .g.V(4>). However. a static (time independent but spacedependent) solution of the classical equations of motion. or even a solution to a modified field equation (e. V{</>) = m§0V + A0(<^)2. Tong [272]. monopoles. and also to see these in a somewhat broader context (by comparison with higher dimensional cases).22.2 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. it is important for a better understanding of the considerations which follow. to keep this case in mind.* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics.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. modified by going to imaginary time). 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.2 The Constant Classical Field 539 and in any case deserve detailed study. may like to consult the article of D. 22. Thus references to field theory will be exceptional and hopefully do not irritate the reader. The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking". vortices and kinks. 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>. A main aim of this chapter is to generate some appreciation of the distinction between socalled topological and nontopological configurations (later referred to as instantons and periodic instantons and bounces respectively). This means theories with field densities which are therefore infinitedimensional. 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.
(In models of the early universe one often considers V(0) with m\ > 0 at the early stage.1 Different potentials for different signs of m§. £)) is bounded from below we must have Ao > 0.S£/<90(x. V ) .540 CHAPTER 22.2) We ask: Are there classical (i. of the equation of motion? For these we must have all derivatives of <j) zero. 22. i. To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x. For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0. as illustrated in Fig. Classical Field Configurations where d^ — (—d/d(ct). and hence here [d^ + m20}(l) = Xo(f (22. Felsager [91]. The EulerLagrange equation is d„ dC d(d. B. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0. . 433.e. An even more familiar example is the GinzburgLandau theory [111] of superconductivity § in which the phase transition to the superconducting state implies the transition to the double well potential shape). 22.1. cnumber) solutions <j)(x) = 4>c = const.e. V[<t>] m 2>o o Fig. then after cooling with a phase transition one considers expectation values 0 7^ 0 corresponding to minima at \<$>\ ^ 0. p.£). § (22.3) See e.g. m l + \04>*4> )$ = <).
TTIQ > 0.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i. 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. H = —C = V.</>2)plane. i. mg < 0. (22. But 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>. whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution. The Hamiltonian density H is defined by the Legendre transform H[<f>. 22.ir]=7r<fiC. the only such solution is the case we wish to consider.2 The spontaneously chosen phase. w A iP 3ml 5* "XT" (22.2 The Constant Classical Field For Ao > 0. we have 541 0.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig. and so ft[<M=m6:(^* 1 + Ao(0*0)2.e.22. We can convince ourselves that <fic is the field associated with a state of minimized energy. 22. The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected.e. with 0 = 0.5) . aq> For (b = const. Fig. not the solution x(t)). Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion. TT = —J.
7a) (22.3ml = 2ml = ml + TA 0 A 2 = 0. we set cf>(x) = <f)c + ri(x).Then m\ m\ = ml + A0A2 = m2.2) and separating real and imaginary parts.g. in a transverse direction.e.e. and we wish to investigate the behaviour of the field cf> in its neighbourhood. 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. (22.e. &H/d(fi = 0) if [m20 + ^>]</> = 0.2. and partly by climbing up the parabolic wall on either side. 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). >° for m 0 < °> .3). (22.7b) the constant on the left and the coefficient of A on the right of Eq.7b) In Eq.tp2. e. as indicated in Fig. V?) (22. one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 .542 CHAPTER 22. For every value of the phase (3 we have a different constant cnumber field configuration 4>c. the one for (3 = 0. i. Suppose we choose one such configuration. along the trough of V (which we can call a longitudinal direction).e. .3) obtained above.6) Inserting this into Eq. Classical Field Configurations the density Ti is minimized (i. (22. i. Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential. This is the condition (22. where (f>c = — T](x) = —= [Vl (x) + ilp2 (x)} • (22. We identify the coefficients of the linear terms on the left hand sides with those of the masses of the fields ipi. 22. ^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 .7a) vanish on account of Eq. (22. i.
7b). 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 £. ^ = h eip / \ ^2 l .3 The spontaneously chosen phase seen from above.r')S(t . has to be removed in order to allow a well defined perturbation procedure (i. t. In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar. irrespective of the question whether the perturbation series as such converges or not. This wave function is in the quantum mechanics constructed about the classical configuration (j)c at this point or collective coordinate of the classical path a vector in the Hilbert space pointing in the longitudinal direction.e. tangential to the classical path. (3=0 Fig.e. We have here an example of the Goldstone theorem which says: If the solutions of the equation of motion do not possess a continuous symmetry of the Lagrangian. The elimination of the .t').22. Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory. though not identical phenomenon. whereas ipi(x) is y ' " " " " " ^ . 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. of course. The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r. then a massless boson exists. Goldstone's theorem applies to fully relativistic field theories. r'. the existence of individual perturbation contributions).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. i. which. Thus the Green's function is similar to that in electrodynamics. We can see the problem here by looking at Eq. t') = d(r . There the wave function with an associated vanishing eigenvalue is called a "zero mode". (22. 22.
instead we shall consider static and timevarying solutions in important models of one spatial dimension. t Potentials of this type have been discussed by M. however. so that we are interested to have a parameter *For a collection of many informative papers on solitons etc. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered. F. .544 CHAPTER 22.* Potentials of higher polynomial order in $ than $ 4 can also be considered.g. 22. Behera and A. The principal aspects will always be very similar so that it really pays to study simple models in considerable detail.8) where V[<&] is a selfinteraction or potential of the field (to be specified later). *S.e. and we choose the metric ?7oo = + 1 . N.t In particular the sextic potential can be considered along parallel lines. Khare [17].* We consider the theory (later: quantum theory) of a real.3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories. 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. 7711 = — 1 ( 1 6 Minkowski manifold). i. C. MullerKirsten [291]. Lohe [179]. equates to zero the coefficient of this wouldbedivergence. To insure that the energy (see below) is positive definite we require V[$] > 0. We make one more assumption concerning V which.£) in one spatial dimension. Rebbi and G. Later we wish to develop a perturbation series about a classical configuration of $.d^} = ~d^d^V[^) = ^2^(^j V[$]. (22. see e. Zimmerschied and H. Soliani [237]. A. J. spinless field $(x. The Lagrangian density is taken to be C[^. W. In the following we will not be concerned with classical cnumber solutions of the EulerLagrange equations which are simply constants. We see therefore that the classical cnumber field configuration and the attempt to develop a perturbation series in its neighbourhood leads to intricate connections between symmetry properties. we shall need only at a later stage.
For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>.g] = \ v m For instance.l}.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series.9) with quartic potential is defined by ^V\g$\ = ^V\g*. at least in the present case.10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$. In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static.13b) We are not interested in just any solution of this equation. such that J > — — — = 0. — m$ 1 — cos Im 1 I ^ $ (22. (22. The first such condition to be imposed is that the energy of the solution of interest must be finite.12) where the prime denotes differentiation with respect to $ .«{x) = V'(<p). From Eq. (22.22. the reason for this condition being that we visualize the classical solution as representative of a lump of energy. i. The field $(x.12) we see that they satisfy the Newtonianlike equation <j. ^ ^ * + V'[*]=0. Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation. In a quantum theory the field $ is an operator defined on a space of states. 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0. (22.§ Unstable classical configurations are also important and will be considered in later chapters. i. The other condition which we impose (and study in detail) is that of stability.t) has an explicit timedependence it is a Heisenbergpicture operator.e.e. 1]. .l].13a) We write such fields <j>(x).11) (22. Since $(x. (22. but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions. (22. the socalled ^theory = ^V[g$. 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).
7r) where IT is the momentum conjugate to $ defined by dC ? r = .T = <I>. (22. The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^. the classical solution will therefore have to be such that this integral is finite.e.14) where the overdot implies differentiation with respect to time t. (22. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($. (22.2\dx 2 ' 1 fd& + V[$]. We observe here already that the integrand of the integral in Eq.T^ = 0. d.e.4 The wellknown soliton of $4theory. HI tanh m(x>k) 9 Fig.17) is reminiscent of classical mechanics if we interpret x as time and V(<f>) as the potential energy of a particle at location d>. 22.c = <£z + .15) The zerozero component is equivalent to the Hamiltonian density H. i.17) Since we have to integrate over all space. <9$ i. (22. n = & . .16) In the static limit we obtain therefore and write this as Em = H[<t>\ + E(4>) = fdx 1 2 \dx 2 + V(</>) (22.546 CHAPTER 22. dC • Too = r?oo£ + =r$ = C + vr$ = H.
See J.13b) for the specific potentials (22.5.20) (22.10) and (22. > .3 Soliton Theories in One Spatial Dimension 547 As we shall see below. 22. in the case of the first. (22. (22. (22. 4m .11). Rubinstein [240].x n ). the $*theory.e. Again XQ is an integration constant. The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq. i.1 .18) Here XQ is an integration constant. m ( x .5 The soliton of sineGordon theory.21) This sineGordon configuration is depicted in Fig. In fact.4. — XQ). it is not difficult to find classical cnumber solutions 4>c to Eq. 22. 22. "The name sineGordon is derived from the analogous KleinGordon equation or theory.777. The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n". In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 .22. around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> . .18) is depicted in Fig. one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2. O±m(xxo) (22. < = — tan [e >  ° ] 9 2 rrm/g — Fig.
. In the case of the sineGordon theory defined by the potential (22.11) we obtain r<Kx) Jd> 4>{x0) m d{g<t)/m) 9 ml g V h(lcos&) m or m(x — XQ) I.o) =m7r.e.18) and (22. and hence 4/71 9<t>{x. Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ.^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ). (22.* ( m' m 2 r)] ^2" . Classical Field Configurations The solutions (22. With we obtain g(/>(a.1 "»JtKxo) 9 1 . 4m In tan tan 5 l/e±m(*xo)tan^£^)U. <^>(x) = ± dx x=g4>/m \ / 2 ( l — COS X I Am dx lx x=g4>/m A / 4 s i n 2 ±1 In tan .10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 . or \(<P'? = V(4>)+ const.9 2 4>{x) = i. 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! . ± 1[tanh. Here the constant is zero for V(<p) and <j>' zero at x = ± o o .13b) which we can write d \W? = !"(•». Inserting the potential (22.20) are obtained from Eq.548 CHAPTER 22.
However. 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. A state of stability is therefore associated with minimized action and/or energy. i.4 Stability of Classical Configurations The concept of stability can be complex. The relation (22. (22. (22.19).21) for the energies can also be obtained from Eq. the others as antikinks. 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. Jackiw [141]. . SE((f>) 5cb(y) 0. The expressions (22. (22.17) implies dx Jdx dx n£)' + ™ ( ' • # \ dx J S(f>(y) \dx and with partial integration we obtain f dx d fd<f>\ dx\dx J  5V(</>) 5(x y) + 5(j)(x) Tx5{x~y) .22) and that its second variational derivative be positive semidefinite. in the onedimensional Ising model.22. Suppose now that we demand stability in the sense of minimized energy. 22. This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian.e. (22. the discussion of Bogomol'nyi equations) in a simpler way. instead of doing this now. the curves rising monotonically from negative to positive values are known as kinks. however.17). Then we have to demand that the functional derivative of the energy E(<p) be zero.23) See for instance R. this does not require the action to be minimized. we obtain these expressions later (cf.4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks". for instance.
Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(xy). In fact. (22. its eigenvalues w\ > 0.27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration.25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds.e.550 CHAPTER 22. in fact ip0(x) const. it is also called a "translational mode". dx (22. Classical Field Configurations Since y ^ ±00. Thus the vanishing of the first variational derivative yields again Eq. in particular the invariance under spatial translations.e. Thus we have to investigate the eigenvalue spectrum of the Schrodingerlike equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22. (22. (22. (22. A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space.13b) with classical solution 4>c. The eigenfunction 4>'c{x) is for this reason called a "zero mode". if 4>c{x) is a solution. dx2 + V"{4>) (22.24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite. a very general phenomenon which can be established as follows. The original Lagrangian was.13b) retains the invariance although the solution (f>c does not.a / 0 is not the same.e. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G. 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. namely (p'c.A. of course. the integrated contribution vanishes. i. with eigenvalue w2. i. written down in Lorentzinvariant form. i.28) .26) Comparing this equation with Eq. Since it arises from the violation of translational invariance by 4>c.13b). if we apply the generator of spatial translations d/dx to the classical equation. we see that the latter has one eigenfunction. = 0. This is. and the EulerLagrange equation retains this property. Eq. the function 4>c(x + a).25). in fact. (22. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0. The Newtonlike equation (22.
In the onedimensional case under discussion we can rewrite Eq. Then the vanishing of the first functional variation implies that 5S = 0. We can choose the potential V such that the constant is zero. in fact.20) are such configurations in two particular models. 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. it is an expression like (22. this integral has to be finite. we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC). (22. We see therefore that the occurrence of these zero modes is a very general phenomenon.22.33) x—>±oo .] = 0 or x—>±oo lim V{(/)c) = 0 for const.31) Inserting this into Eq. We return to our considerations of stability. (22. (22. Invariance of S[U] under transformations of this group implies S[U} = S[gU]. / oo Thus for E(4>c) to be finite. and that S[UC] is finite (or equivalently the energy).e.18) and (22. As stated above.e. and we have seen that Eqs.4 Stability of Classical Configurations 551 where T is a generator and A a parameter. (22. (22. (22. lim [V(<f>c) + const. (22.27). we are interested in classical cnumber field configurations of finite energy. We shall see later that these zero modes lead to undefined Green's functions for the semiclassical perturbation expansion unless one or more suitable constraints are imposed. 0 = S'[UC] = S'[gUc}.32) = V'(</>). Thus TUC is a zero mode.30) Differentiating this with respect to A (applying —id/dX) and setting A = 0.].13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const. (22.29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc. i. = 0. i.17) we obtain oo dx[V(4>c) + const.
. » In the <fr4theory defined by Eq. V{4>) = 0 at different values of </>). This number is defined like a charge in field theory by the spatial integral of the time component of a current.1). . In the sineGordon theory > • defined by Eq. Classical Field Configurations V Fig.35) and it is seen that in this case q can be ± 1 or 0. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory. in view of the antisymmetry of eM^. (22.6.34) jfc" = e^dv(l since d^k^ = 0. For a configuration to have finite energy. 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. (22.552 CHAPTER 22.6 The minima of V in $4theory at x — ±oo. (22. .10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig. Thus the classical energy or vacuum configuration is not unique. ± 2 .11) we must have correspondingly for x — ±oo: » 771 9 2vrn. . 22. the classical field (bc must approach one minimum for x — . the potentials have degenerate minima (i. In the case of the 3>4theory we can define a conserved current k^ by (22.e. n = 0 . oo J—oo y The topological quantum number q is defined by 0. .c o and another for x — oo. ± l . The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. 22.
38) 4 V(\<t>c) >m ~2[l .11) are given by ( so that for x — ±oo: > — J. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 . We can see this as follows. In the case of the sineGordon theory the minima of the potential (22. in general.A2)2 + 0 except for A2 = 1.37) for the onedimensional case here. Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real. Tfl \ (22. x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition. The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An . . Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense. First.7T (22. ± l ± 2 . . zero is not a meaningful topological number. termed topological.36) oo Consider the $ theory with 4>c — ±m/g for x — ±oo.cos(27rnA)]. . one may ask how topology comes into the picture. .22. For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22. Thus the condition of finiteness of energy does not permit A</>c deformations in this case other than those with \ = eix(e) for X (0) = O.4 Stability of Classical Configurations 553 Since we call q a topological quantum number. Again we see that E(\<f>c) is finite only for A = ± 1 .7r) does not change a boundary condition lim [\<t>c{x)\ = (/>c(±oo). and exp(ix(0))<pc reduces to 4>C when x(#) = 0. 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 . n = 0 . A property is. Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6. if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy. continuous values. and we call this "topological stability.
where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00. . the energy is minimized.554 CHAPTER 22.5 Bogomol'nyi Equations and Bounds We have seen previously (cf. MiillerKirsten [215]. 22. We observe (a) the topological current k^ of Eq. Thus with one space dimension higher there. Classical Field Configurations involves the difference An = rioo — n. 22. 00 (22.e.^ = N. (c) the time component k° is a spatial divergence. i.13b) or ^ ' ( * ) ] 2 = V(</>).428. W. pp.40) (22. its integral is nonzero only on account of nontrivial boundary conditions. if oo / dxy/2V{<l>)<f>'{x).42) Thus the energy has a lower bound given by the right hand side of this inequality.g. (22.41) dxy/2V{$)4>'{x). (22. Eq. (22. J.** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents). The inequality is saturated. (For a broader view we consider briefly in Sec. the factor A = exp[ix(#)] varies over a complete circle). (22. 00 (22. H. 425 .17)) f 00 > 0.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion).9 the foregoing arguments in a higher dimensional context. (d) the topological quantum number iV arises in a completely classical context.31) to (22. Eqs.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22.43) **See e.33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22. From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf.
. In the sineGordon theory we can consider. (22. (22. 555 (22.19).21). the set of configurations (fic which interpolate between 0(—oo) = 0 and <f>(oo) = 2nnm/g. Bogomol'nyi [27]. 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— .22.n > 1.5 Bogomol'nyi Equations and Bounds We observe that (cf.40)) this occurs when 4>'(x) T V2V(4>) = 0.^ Clearly it solves the second order equation or (22. Eq. B.39). 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. We discuss this in the following Example. n E . Here (V(c/)) > 0. 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. 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.40) is called the "Bogomol'nyi equation" since it was first introduced by Bogomol'nyi. for instance. (22.44) This first order equation which saturates the inequality (22.
The kinetic energy T of the bob at position xn at time t is (recall " m r 2 r / 2 " ) 1 —mr 2(d<j>{xn.e.t) 4>{xn.1: Classical analogy to sine—Gordon theory Consider a string along the xaxis.556 CHAPTER 22. z)plane and only such that their elastic strings remain taut. } n= — oo roc —> / L J . Solution: Consider a string along the xaxis as described. Assuming that the pendulums can move (i. 22. Initially the pendulums are suspended freely in the gravitational field.g.7 The pendulum analogy. t)).ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji. At equidistant points xn = na along this string we attach strings with pendulum bobs.g. from the yaxis as shown in Fig. Construct an expression for the energy of such a classical system and its Lagrangian. Fig. We can now write down an expression for the energy of such a system of (say) n pendulums.t) measured e. . H.46) **See e. ka —> K.cos</>(x„.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 .e. * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22.2 V dt + k[(f>(xn+i. rotate) only in the (y.t)\2 +mgr{\ . 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. Assume that the pendulums can move (i. Chapter XI. V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22. xn+\ — xn = a) 00 a —> dx.7. Classical Field Configurations Example 22. 22.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na. rotate) only in the (y. Goldstein [114]. total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) . so that oo > u.t) <j>(xn. all being identical. with mass m and connected with short strings to neighbouring bobs.t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i.
which represents an expansion in terms of normal modes. (22.25) which we obtained as the condition of classical stability for eigenvalues w\ > 0.47) where n(x. This number. ' Equating coefficients of exp(—iw k t).7 linking the pendulum bobs).25) obtained earlier. (22.13a) and (22. t) of Eq. of the continuous curve in Fig.47) into (22. Eq. (22.6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22. the topological quantum number. We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc. (22. The equation is called stability equation or equation of small fluctuations.t) = <l>c(x) + r}(x. Hence we set ${x. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x). is therefore also called "winding numbed.48) becomes ( d2 \ X I [Q^2 + wl)Tpk(x) e*v(iwkt) k ^ X = k ^2v"{4>c{x))iljk(x)exp(iwkt). We can understand what stability means in the present case.t): {w With the ansatz r)(x. each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i.49) exp(iwkt)ipk(x).22. t) is the fluctuation (field).t) is a small .t) = ° (22'48) (22. 22.50) This equation is identical with Eq. It is instructive to obtain this equation by yet another method. If some w\ were negative.e. Inserting (22.49) would imply an exponential growth in the future t > 0 or in the past t < 0. (22.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs.13b) we have for small fluctuations rj(x.6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards. wk would be imaginary and so the factor exp(—iwkt) in rj(x.t). 22. (22. and so would invalidate the procedure which assumes that rj(x. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x.
V"(<b) = 2m2 + 6g2(f>2. Thus for stability w\ must not be negative.20). since we shall need the explicit form of V"{<f>c(x)) we calculate this for the two cases: (a) ^theory: Here "<*> = ^  i n*) = 2m 2 (l£^)*.558 CHAPTER 22. V"((/)c(x)). If that is necessary. In fact.50) has the general form of a timeindependent onedimensional Schrodinger equation.A \ Inserting <f)c of Eq. i.18). The equation of small fluctuations (22. i.51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ . Inserting (j)c of (22.x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22. t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists. we obtain V"(<t>c) . and need 9 = tan" 1 j e±m^Xo) 1. cos46» = 2cos2(2(9) . 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. we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22. This means that the stability is not universal but only local in a certain sense. we shall see later that the fluctuations n(x. In fact. (22. . the zero mode.e.52) tan 6 := e±m^xo).2 m 2 + 6m 2 tanh 2 m(x . Classical Field Configurations fluctuation.e. For x — ±oo the quan> tity acting as "potential".1.
(22.22. Those of the other case are complex and periodic at infinity.* For such a potential the spectrum can be b o t h discrete and continuous.2 (22. . We can therefore construct solutions of definite parity. — XQ)] cosh[±m(x — XQ)]: 1 . i. We can investigate the spectrum as follows. The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable.ijj^ respectively. ip+. Consider the equation fd2 dz2 . we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo. ^+(0) d 0.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 .53) = m(x — (22. and as a consequence a limiting form of the small fluctuation equation of all basic potentials. even and odd solutions.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. (22.e. A  *(* + !) cosh 2 z tp(z) = 0.56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation.55) The differential operator is even in z.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 .tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a. which satisfy the boundary conditions V>(o) = o.
when in the even case n ( n .I) .Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22. A ^ An = .e.l2 = .( 2 n + l .Z)2.4 n ( n .( 2 n .57) Then the equation for x{z) 1S (22.60) terminate after a finite number of terms.e.e.61) Discrete eigenvalues are obtained when the series (22.Z ) + ^(A + Z2) = 0.60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n. 2n < Z. . and in the odd case n+\)U+\l)+\(\ i. A * An = .59) Setting (22. i.58) = (1 + sinh2 z)~l/2x(z). (22. Since for £ — oo: * the function ^>+ is normalizable provided Z > n. (22. Classical Field Configurations We now set (with ip either ip+ or ip) tp(z) = (cosh z)~lx(z) and change to the variable £ = sinh2 z. i. + \{\ ) 2 ~Q"m 1/2 oo «n+l (n + i ) ( n + l ) ( n + ^ ) ( n + ± .e.Z ) 2 . i. + l2) = 0.560 CHAPTER 22.
m i. w2 = Am2.is normalizable provided 2(n + . (22. the continuum is w > Am2. (22. for N = 0.6 The Small Fluctuation Equation 561 and the function ip.e..2.e. N = 0. i. (22.64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz.2Z = .j < I . We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 .54) by setting X = k2 Then 1 .22. (22.Ny. d w = tanhz.62) where even (odd) iV are associated with even (odd) eigenfunctions.1: w2 = 0.63) we see that the continuum starts at A= 2 2^ = 0. m N<2. From (22.65a) _dz2 cosh 2 z l=2.N<1. (a) ^theory: the equation Inserting (22.62) we see that the equation possesses discrete eigenvalues w —2 .11).. We can see the continuous spectrum of Eq.50) we obtain with z = m(x — xo) tp(z) = 0. (22.1.63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0.51) into (22. We can now derive the spectra of the stability equation for the potentials (22. 2 .10) and (22. 3m 2 .e.\=Kl2 From (22...( 2 .u2)—^V 2 dio and ip(z) = elkzV{to). °2 i.
It is now particularly interesting to look at their wave functions.53) into (22.2 The discrete eigenvalues are now given by w ^l m^ = (lN)2. Thus 4>'c(x) has no node. .50) we obtain with z = m(x — XQ) the equation (22.. .8 The zero mode as typical ground state. also be deduced from d(j)c/dx.562 CHAPTER 22. This property can.e.=1A=^_..8 and we see it has the shape of a typical ground state wave function. Thus it has no node and so represents the eigenfunction of the lowest eigenvalue. of course. (22. We know that 4>c is a monotonic function. . Thus either spectrum contains the expected eigenvalue zero. i. and — as expected — this eigenfunction is even. 22. d . 22. Eq. m d . . so that its derivative is nowhere zero (except at ±oo). %(x) Fig.27)) .66) We see that this is a nonvanishing function for any finite value of x. Classical Field Configurations (b) SineGordon theory: Inserting (22. .65b) dz2 cosh^J. the zero modes. In the $ 4 theory we saw that (cf. for N = 0: to2 = 0 and the continuum starts at w2 = m2. m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22. N<1. The zero mode is depicted in Fig.
t a n h x . (22.oc (22.6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to . 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 . .= E(<f>c).3 . Finally we consider the normalization of the zero modes. . (22.22.68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i. — XQ) OC Hence the conclusion is similar to that in the previous case.67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ). d .z 0 )]oo = 8 . (22. 1 o rrr tanh x — . .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.70) .e. ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(xXQ) —tan e g dx m D±m(za. . i. Imposing the condition F we have in the <I>4theory: /•oo dx[^o(x)}2 = 1. without quantum corrections).o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a.
e. I.* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem.> 4>a{x) = 4>(ax). The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here. This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces. Derrick [68]. i. This question has important consequences. Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0).564 CHAPTER 22.72) with T{$) = I' ddx]{V(j))2 > 0.4) <j>(x) —• a<f>(ax). H. *The considerations of this section extend in part beyond quantum mechanics and into field theory. for if such configurations exist in a realistic theory with three spatial dimensions.2d. A.g. however. Kastrup [145].g.41) for the energy of the classical configuration but now with respect to a ddimensional position space.71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . we consider E(<P) which we also write as (22.g.T(4>).7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions. . Classical Field Configurations 22. Olive [113]. They may help. H. I. Goddard and D. it is important to understand their physical implications. ^For basic aspects of scale transformations see e. *See e. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension. to underline the fundamentally different nature of topological and nontopological finite energy configurations. J As a first example we consider the expression (22. (22. P. P(</>) = f ddxV{(f)) > 0. Affleck [5] who uses in his Eq. One can have different types of scale transformations — see e. f G .73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax . ddx {V4>f + v{4>) (22. (2.
22. the second derivative is positive) only for d = 1. a=l (22.e.V(<P) (22.78) For the energy of a static configuration.. since d2E(<pa) da2 2(2d)T(<f>).e.e. Next we consider more complicated cases. In fact. i. to be finite the field <P ~^ 4>oo £ Mo for x — oo. d = 1 minimizes the energy. (22. (22..74) From (22. Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a.xd). 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}.\ eMo(22. we write • = lim J r . 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.7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1.x2.79) In order to emphasize that (f> — (p^ in any direction. Thus this scaling argument excludes the existence of static finiteenergy configurations of the given theory in more than one spatial dimension. (22.76) we see that E(<f>) can be minimized (i..80) .77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}.75) Since both T and P are > 0 and d is integral. (22..73) we see that this implies (2 .71).d)T(<f>) = dP{4>). It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>). i. Consider first the Lagrangian density £[&M = \{d^)(d^) . x = (x1. we set 0 dE(<pa) da J a = i (22.
1 manifold Sd~1\d=i <• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (nn'/ 0) const. If we go to two spatial dimensions (d = 2). i. as is the case for the constant solutions.e. —00. = —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 .38). 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). i.81) and (since d = 1) S = ± 1 .e. is a connected region.and sineGordon theories we assumed $ to be real. — 1). therefore. are the same. (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a.e. Table 22. the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00.1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = . (ptx = <f>oo.1. We see. ( with — = lim <pc(x) 9J *>±oo TYl \ (22. one set can be mapped into the other. 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 . and. of • course. Then each point at infinity is characterized by a different value of the polar angle 9. in fact. S1. —00 or equivalently S = + 1 . . 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00. 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. —1.566 CHAPTER 22. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22. These observations are summarized in Table 22. i. that we have a set of disconnected classical vacua (</>oo><^oo. Thus in this case S^1 is a disconnected set consisting of two points.
g.e.e. i. assuming invariance of C under the transformations of the group SO(3) in internal space. for instance. cj)(r. (22. On the other hand. by allowing <fi to have three components in some "internal" (isospin) space. that if the problem is to be topologically nontrivial. if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}.e. —00 or S = 1 . Similarly. VMx)) with (0(oo) = linvxx) (p(r. Thus. given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$. $(r) : Sl . phase transformations 6Q ^> 9Q + 59.e. i. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}. i. we have w e ca {Mx)}> and 1 3 1. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial. Thus. 9) would depend only on r = x and hence the problem would be effectively onedimensional.e. the one point S = + 1 . the field must map position space into field space at infinity.83) .e. therefore. i.1 . to have topologically distinct classical configurations. .82) Now Mo := \ 4>(oc) ^24>2 = a2\. This will be "nontrivial" in the above sense only if 0oo is independent of 9. map into. if cj> were real and position space twodimensional. we would have only two points at infinity. We see. i.7 Existence of Finite Energy Solutions Suppose.2. If position space were onedimensional. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to.• M0.3 (^real) v{4>) = / (22.22. * n reach any other <^(oo) by simple phase shifts. 00. This can be achieved e. i. if <p is real.
Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line. L. a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved.e. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3). Georgi and S. B.e.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) £(</>. Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space.568 CHAPTER 22.Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3.e. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories.</>] = ~G^Gai. the appropriate rotation.V{4>a].e. 11 H. the winding number is n.2. $ — $exp(ia) in a U(l) symmetric theory. i. Summarizing we see that </>(x) is a mapping from S^T1 to <S^1. it is effectively a surface (and so a continuous set) S2. through each point at infinity). the subsequent discussion requires only the general form of the Lagrangians. + i ( D " f t W ) a . invariance under replacements $ — —$ in $ 4 theory. going once around S%~1 (i.2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A". circle. we cover <S^_1 n times. $ — $ + 27 > • T in sineGordon theory. In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo. The symmetry of the potential (and Lagrangian) — i.„ where G r = 9 M £ ..3. i. . The theories are: (a) The Nielsen—Olesen (vortex) model (also called Abelian Higgs model) " with Lagrangian density V(4>) = Ug2 2A 9 2 FM„ = du. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit. % . Example 22. . A) = T(4>.a M J . (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM. Olesen [220].. a = 1. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. A) + T{A) + P{<j>). $ — > > $ exp{i Xlo=i Pa. If. sphere. so that details of these models are irrelevant. Nielsen and P. Glashow [108].
d T ( A ) + AdP(</>).Ax) dX Since T((f>. >• A„ x ( x ) = AAM(Ax). Thus we set dE(<fix.d)T(4>.dP(4>) = 0.T{A). = 0. P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution.<^[x) transforms like a vector.i T(AX) = X4~dT(A). (i)^ 2d T Considering the contributions of (22. A). l j ddxGijGai] > 0.84) and investigate the existence of finite energy classical field configurations.22. some contribution must be .. (22.d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ). (22.84) in this way one finds that the quantities involved transform T ( ^ .89) P{4>) are positive semidefinite and not all zero. (22. and so B ( ^ A .e. (22.d)T(A) .A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0. We now consider scale transformations with A £ R. J 4 ) + A 4 .87a) C?M„(x) — \2G^{\x). > Since T>IJ. A) + (4 .d T ( ^ . Solution: Each contribution in (22. T(A) = f ddxV(4>) > 0. i. (22. A A ) = A2dT(^. negative. it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also.86) (22. we have {^/»)A^ (22. ^ A ) = A 2 .85) Here the difference in the transformations results from their difference as scalar and vector quantities which are defined by the transformations ^V) = 0(x'>.e. <0 >• <£A(X) = <KAx).A). 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd . (2 . Mx') = E with a scale transformation x — x' = Ax. A.84) is positive semidefinite and so must be finite by itself.87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d.7 Existence of Finite Energy Solutions where T{<j>. i.
exemplified by soliton and sine—Gordon theories. Landau [111].92) The expression ^ ( x ) ! is known as the "order parameter.570 CHAPTER 22. One can thus define a scalar field or wave function ty(x.g. (22.** e. A) or P{4>) for d = 3. In the following we consider a related theory. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models.89). This case is. 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. J.8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory. D. We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory. Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x). (22. ft V . L. (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example. MiillerKirsten [268]. a > 0.89). R. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D. Ginzburg and L. a = a(T) = a T —T c . if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives. is temperature dependent and changes sign at the critical temperature Tc. H. 22. of course. Tchrakian and H. (e) For d > 5: In this case it is not possible to stabilize Eq.e.90) and we would have to add to Eq. H. (d) For d = 4: This is possible only if the sign of V is reversed — this is exemplified by a socalled pure gauge theory with socalled instanton solution (which we shall encounter in a simpler context later). Thus such a contribution could counterbalance T{4>. 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. Skyrme [253]. (c) For d = 3: All three terms must be present as in the GeorgiGlashow model. (22. (22.89) the contribution (4 — d)T(4>). timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity. 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. We then have a situation of the potential like that described in the context **T. tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory.t) such that  ^  2 describes the density of Cooper pairs. . T^{<t>x) = \4dT(</>). We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i. W.
95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory. ^(00) = a (22. but for T < Tc it has a double well shape.e. Sec. i. i. Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material. (22. Then the static energy density becomes (cf.94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ .e.8 GinzburgLandau Vortices 571 of the complex Higgs model (cf. setting S£. assuming there is no superconducting material in this domain).94) to check the dimension) 1 = I—a. The parameter 7 serves to put the minima of the double well at V = 0.2) £(^. 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.A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4. 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. Example 22.22. which we recognize as one half of the kink solution we obtained previously as the topological instanton solution of the Schrodinger equation with double well potential (in the present case ip(x) = 0 for x < 0.2): For T > Tc the overall potential has a single well (minimum).93) J' (22. Variation as before. (22. is a measure of the distance from the surface of the superconductor to where . from x < 0) means — as is familiar from electrodynamics — replacing V by V .(tp) = 0. Clearly the length (see Eq.
yJ—a/[3.e. inside the superconductor it falls off exponentially away from the surface (this is the MeissnerOchsenfeld effect generally described as the expulsion of lines of force from the superconductor below the critical temperature T c ). i. V x M = j s . 2mi Setting < / > \P\e vp B = ViA. i. (22. 22. W. i. h [^DV'V'DV].96) ipH.e. H. . 22. the order parameter \ip\ attains (approximately) its asymptotic value. MiillerKirsten [215]. The magnetic field on the other hand. We can see this as follows.9 Behaviour of ip and B in the superconductor. Classical Field Configurations superconductor (x>0) Fig. J.g. The density of the Cooper or super current is given by the typical expression of a current. 450. In the Maxwell equation See e. p.9. (22. exhibits a completely different behaviour — in fact.M.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 .572 no superconductor CHAPTER 22.98) where M is the magnetization resulting from atomic currents.e. (22. as indicated in Fig.
97) this becomes (since curl grad = 0) AB = Mo—Vx m A = /x 0 —B. m Here A^ is known as the (London) penetration depth (or length) (in practice this is of the order of 1 0 . This is zero in the present case.M i.L B = O. Using the relation "curl curl = grad div . and V x (V x B) = / J 0 V x j s . V x B = [i0V x M = /lois. These vortices therefore carry magnetic flux which is necessarily quantized . In the onedimensional case we have B{x) = B(0)ex/XL. Also dD/dt = 0 in a static situation like the one here.e. this becomes A B = no V x j 5 . those of the socalled type I and those of type II. Approximating p by its equilibrium value. K i ™*.100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter.e.99) Inserting (22. (22.div grad" and V • B = 0. Hence 0 = V x H = V x ( — B .8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside. 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. There are two types of superconductors. More and more vortices are formed as the magnetic field is increased further and further until only small superconducting domains remain which then disappear completely beyond a second critical value of the magnetic field.22. Superconductors of type II are characterized by the fact that when the magnetic field is again increased and beyond a first critical value. i. m xi A B . (22..5 to 10~6 cm).
Finiteness of E requires \<t>\ = \<fioo\ on this circle. Schaposnik [69] after their equation (3. Classical Field Configurations (see below). where exp(ix(0)) is an arbitrary phase factor.9). 22.37).9 Introduction to Homotopy Classes In this section we consider topological properties of classical finite energy configurations and introduce the concept of homotopy classes. De Gennes [67]. The existence of these vortices has been confirmed experimentally.9) ^e^ '>\<poo\. in other words. We can think of (e. The existence of these vortices with quantized magnetic flux was predicted by Abrikosov* and are therefore frequently referred to as Abrikosov vortices (the idea was originally rejected by Landau). See also P. In the NielsenOlesen model defined in Example 22.. i. Abrikosov [2]. In the two models of Example 22. J. 6')r_>00 represents a mapping from the circle at spatial infinity to the circle defined by x(#) (or. (22.574 CHAPTER 22. the space consists of two diametrically opposite points on the circle.2 we saw that finiteness of P(V) puts no restriction on the phase of (f> so that r e (22. de Vega and F. 'See the remark in H.g) twodimensional space as being bounded by a circle at r = oo. x—>oo Thus at spatial infinity </» has the value of a zero of the potential. A.36). In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM).101) (p(r. . 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. Thus the asymptotic field </>(r. A. of two points on a line. *A. by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI. we must have that lim V ( H ) = 0. G. i.2 — identified as above with the GinzburgLandau theory — type II superconductivity occurs^ for A > 1/4 and type I superconductivity with the complete MeissnerOchsenfeld effect for A < 1/4.e. (22.e. in fact a periodic function of the polar angle with period 27r.102) In the onedimensional soliton case of Eqs. (22.
The two theories we considered above in Example 22. W. Suppose <j>. States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig. 2n] remains continuous if the function assumes the same value at the two endpoints. Sl := {zeC\\z\ = 1}.1).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". commutative). defines the orientation of this line element. 0 < 0 < 2TT. Misner [96].* 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. (22. X is called the domain of / and y its range.22.e. (22. Finkelstein and C.1.102) is called the exponential map of S1 onto S1. Fig. Thus smooth deformations *See D. The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism. topological group with the usual multiplication as group operation.10 States of winding number 0.103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq. We may define this as « = ^[X(0 = 2 T T ) .10. Then a continuous function of (j> on [0. abelian (i. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure. The map A : S1 > S1 defined on S1 C R2 by (22.2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n. i. 22.X ( 0 = O)]. . The 1sphere is the unit circle in the space of all complex numbers. 22. Naturally an integer cannot change continuously.e.118)). In mathematical language S1 is therefore a compact.
Thus two homeomorphic spaces have equivalent topological structures. we required that lim r—>oo w o. if it has one twist as above or a hole. Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below). the winding number must be a constant of the motion.. r^oo 1 dx(9) n r o0 (22..g.i'ldx(0) 9A Thus for (22. A field configuration with n ^ 0 therefore cannot be deformed continuously into one with winding number zero.e.g. For that reason n is called a "topological invariant?.576 CHAPTER 22. In our consideration of the NielsenOlesen theory we also required the gauge field A^ to behave such that the energy is finite. We observe that since time evolution is continuous. it retains this twist or hole under the continuous distortion) and is described as a homeomorphism. connectivity.) remain the same under homeomorphic transformations and are therefore called topological invariants. Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n. The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e.105) is described as "pure gauge" since it is determined entirely by the phase x(#). since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr. one hole.e. finite. The winding numbers n are such topological invariants. (22. no hole. elements in the same class have the same topological characteristic. i.104) so that d2x hJ Since A has direction e#..105) the corrections falling off faster than 1/r. i. The gauge field required in the asymptotic limit (22. we can write 1 d h J ee rde w At h r^oo . it cannot be deformed continuously into the socalled "classical vacuum" (the latter is a constant configuration in our soliton example).e. i.104) to be finite we must demand that g . The different kinds of connectivity of these topological spaces (e. In other .
Therefore. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6). In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22.l]. i.105) insure that B ^ O .108) map the spatial circle S^. the vector potential A. it is determined by a gradient (l/r)d/dd. (22.9):=t(f>£\9) + (lt)(f)W(9) with te[0.<b rd9—4^ = n.110) .22.e. we can consider transformations which describe continuous transformations of a configuration with fixed n. i. (22. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed. the group space of (the phase transformations of) U(l) continuously.109) A W(0) = e^ W.106) J g j r d d g Thus the magnetic flux is quantized. It follows that the field Fij. the gauge field. 9){^ and hence the functions (22. the higher order contributions in (22. is also zero at r = oo. we can go n times around S^. However. Of course. cannot be pure gauge everywhere.107) The fields <f>(r. going once around S^. the number of flux quanta 2irh/g being the winding number n.9 Introduction to Homotopy Classes words. we have B = V x A = curlgradx = 0. An example is the mapping H(t.e.105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22. with n = 1. i. into S i . Ffiu. A field configuration with one loop (or twist). However. We can calculate the magnetic flux $ through the region with $ = * rd9A0 = . if A = V x . for B / 0 . (22. cannot be deformed continuously into one with two loops (n = 2). 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.
111) as a map from the space described by 9. n' ^ n. i. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G. (22.578 CHAPTER 22. the unit interval I = [0. 8) = cfinl\9).9) = <fiW(9) and H(l. A brief introduction can be found in the very readable book of W.lQSl^S^ with H(0. (b) symmetry and (c) transitivity.112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9).T. and the space of t. time cannot make a field jump from one homotopy class (see below) to another. i. Since the passage of time is continuous. Solution: As stated.3: Homotopy an equivalence relation Verify that the homotopy map H(t. to the internal group space of £7(1). We verify these in the following example.0) = $\6). (22. Thomas [270]. like <pn (0).1]. But this type of combination is excluded sind H(t. Steenrod [261] and S. H. it is clear that this is also a homotopy. The map H(t. We can consider (22.e. Classical Field Configurations which is continuous in the parameter t and is such that H(0. Reuter [77]. i. i. are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9). f T h e standard texts on homotopy theory are N. (fin (9) in the above example. Hu [134]. S%. The relation (22. we have to show that the map satisfies the three properties of (a) identity. Dittrich and M. t The homotopy relates a representative function or curve (fin ' (0) to another function or curve 4>n' (9) by giving a precise meaning to the idea that the one can be deformed continuously into the other. e.g. 9) is an equivalence relation. i. Example 22. Two curves which can be related to each other in this way. H.9) = ^°\0).113) . the phase factors of (f)(9). (b) symmetry and (c) transitivity. these two maps (fin from S^.e. A superposition of classical configurations with different winding numbers does not minimize the energy.111) In principle an expression like (22.112) is an equivalence relation — and as such it is defined by the three properties of (a) identity. Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8).e.e. In fact.8) itself must be a configuration with definite winding number. n.110) could be written down for configurations with different winding numbers. H(1.e.9) is called a homotopy (emphasis on the second syllable). — S\ > (22.
9) = <f>W (0). 0) = 4>W (0). \ M)\ H!(2t. we see that of the set of these maps some are equivalent and others are not. 9) = H(l. 9) := H(l . 0). Hence the set of all such maps breaks up into equivalence classes or homotopy classes. at t = 1/2 both i?i and H 2 give c/^1'. i.t. Moreover. We now define the function H ( f 0 ff . Hence <f>(0^ ~ <p 2 ). these groups are then called homotopy groups (cf.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. On the other hand H(0. 0) = tf2(l.0) = </>(0) is continuous there. so that H is a homotopy connecting <jyl>{0) with <f>^>(9).0). is isomorphic to) the set of integers Z (positive. so that H(t. being the continuous map connecting </>(°) with <j>(1'>. . (b) Next we check that if 0<°) ~ <p^.* 22. which therefore demonstrates that the homotopy relationship is transitive.e) tf2(2il.e. as they are called.9) = . <f>W .10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t. 0). ' .0) for for 0<t<. (c) We can demonstrate the property of transitivity as follows.9) H(0. a class being a set of equivalent maps. i? is a homotopy connecting </>(°) with <^>(2'. Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. negative or zero). i<t<l. Returning to the maps \(x) : Si ^ S1^ (i. Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i.ffi(O. 0) = ^ (0) and 6(1. since and tf (1. U(l)). Homotopy classes are sometimes also called Chern—Pontryagin classes. is such that H(0. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent. (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. .22. <^(2). 9) = H(0.e. then </>W ~ <t>{0)• We define an H(t. In some cases the set of homotopy classes possesses group structure (satisfying the group properties). 9) connects <pn (0) with itself continuously. Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'.e. 0) = <£{2). Then H(t. In the above case of 7ri[[/(l)] we know from our earlier considerations that every equivalence class is characterized by a specific value of the winding number.9) = <f>(0\6) and H(l. below).
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. n=0: t=1: t=1/2: i. but TTI[S2] = 0: S 1 is a circle. to make the result plausible.11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T. How can we see this? We first sketch the idea. and then introduce some of the related mathematics.11 The trivial n = 0 map allowing shrinkage to a point.11 the circle drawn with the continuous line is the circle S%. S 2 the surface of a sphere in a threedimensional Euclidean space.) why 7i"i[S2] = 0.1] 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. p. 22. Appropriate maps.e.115) Varying t € [0. Fig. Rajaraman [235]. Sketch of idea demonstrating that nilS1] = Z. 22.g. which is to be mapped into another circle Si indicated by dashed lines. In order to really appreciate how topology comes into the picture here one should understand (e. . T (22.580 CHAPTER 22. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors. This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig. 22. In the diagrams^ of Fig. 52. T < 9 < 2TT.
0 < X(P) < 2vr.h.12. 22. 2 the maps are wound around the dashed circle.103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x). .h. is invariant under infinitesimal (continuous) deformations. d (22.e. phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S. The maps for these cases are illustrated in Fig. Coleman [55]. and hence the winding number. (22.s. the transformations^ (the result may be looked at as 8n of expression (22. of (22. i.I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l. For winding number n we have the mapping (i.117) in ±.12 The n = 1 and n = 2 maps not allowing shrinkage to a point. i.22.117) 2^ d6X(x)\1(x).e. One can prove that the expression (22.116) In fact from this we obtain (see below) 2  n and this means r. In the cases n = 1. 22.s.e. (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. to the case of winding number zero.117).
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. the cyclic group with two elements (since 50(3) is doubly connected). Classical Field Configurations Under this transformation n changes by 5n Now. Thus there is only the trivial map into a point. whereas 7ri(50(3)) = Z2 (i.e. only one homotopy class. 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. il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W). i. there is a theorem which says thatH TTg(Sn) = 0 for q < n. 1 1 See N. . Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations. the identity class (the group space S2 is simply connected.8.582 CHAPTER 22. It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point.e. In fact. Steenrod [261]. 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. an integer modulo 2) i.e. 15. Sec.
In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation.Chapter 23 P a t h Integrals and Instantons 23.e. This is an important standard example of the path integral method which serves as a prototype for numerous other applications and hence will be treated in detail and. Together with the infinite Euclidean time limit this requires for the calculation of the tunneling contribution to the ground state energy the FaddeevPopov method for the elimination of the zero mode. 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. Since we have not yet introduced periodic (i. finite Euclidean time) classical configurations (which are nontopological). 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!). different from literature.1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies. 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. and in particular the exponentially small level splittings of the double well potential. 23. in such a way that the factors appearing in the result are exhibited in a transparent way.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem. We consider such a theory now with action S((p) depending on the real scalar field 583 .
1 The double well'potential. Patrascioiu [110] and Chapter 18.m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf.v{4>) <?VV m4 __ 1 .2) Fig. (23. n = 0. Gildener and A. Path Integrals and Instantons S(<f>) / dt \i>2 . ! " ^ j = ^ 2 . i. and ^2(0) : m* .2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq. 23. 23.1)) and hence E0 = mh/y/2. note that here the mass of the equivalent classical particle is taken to be mo = 1.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O .* /•tf CHAPTER 23. in particular E. "For comparison with literature.584 <)>{t). Fig.2. Chapter 6) 1 2 . (23. The Hamiltonian of the simple harmonic oscillator (cf. 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a. 2 2 4 ™ = ~.e.1.
Fig. W(0) = 0. 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).2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability.e. i. e. as we did in Chapter 18 — but as we saw. The states of the system here must be even or odd under x — ±x.x) = x2 V(x). we have the Lagrangian L(x.In mechanics of a particle of mass 1 with ~ position coordinate x(t).2 The wave functions of the lowest states. 23.2 — and we can deduce from this the boundary conditions. this alone does not yield the level splitting. We can develop a perturbation theory around either of the minima. </._(0) = 0. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator . This tunneling affects the eigenvalues. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig.g.23. 23. must have definite parity as in a physical > situation. but there is tunneling between the two minima if the central hump is not infinitely high. and so the question is: What sort of boundary conditions do we have to impose on the wave function other than its exponential falloff at 4> — ±oo? We can decide this if we observe that S = J dtL is invariant • under the exchanges 4> ~ * ±</>.
tf\ . we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time. dr m+™ (23.3). i. i.7) so that (23. that given by Eq. Then (0o. r .7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo. For small h we can attempt a stationary phase method. SE.  .7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time.4) J <bi =—(An rTf (ITLE = I JT. n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23.e.e.0o.e. d(f> d ' dr2 dr It follows that the equation of motion is now given by 55. Equation (23.U) = J'D{<f>}exp[iS(<l>)/h] (23. But quantum mechanically it can tunnel through the hump from one well to the other.6) d*l dr 2 .5) and iS = dLE d dr[d(d(f)/dT)} i.9) . We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o.<h. dr We observe that Eq. (23.2 m 2 ( l m z (23. is finite if we go to imaginary. Path Integrals and Instantons approximations into an even ground state and an odd first excited state.3) for tf — oo. (23. dLE = 0.2 V'(<l>) = .£'(0) = 0.586 CHAPTER 23.e. The transition amplitude for this.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21).8) V(4>) + const.7. classically forbidden domains then become "classically" accessible. Euclidean time T = it. A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g. i. In the semiclassical path integral procedure that we want to use here. (23. (23.2.
Fig.e. — (23. and hence kinetic energy gained is equal to potential energy lost. Thus we know the solution from there which is the kink solution (22.3 The instanton path. we see that the equation is the same as the classical equation for a static soliton (i. 23. in the present case (T — To) = m tanh m(r — TO). is depicted in Fig. i. i. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p).2 Instantons and AntiInstantons 587 In fact.18).e.3. Its trajectory. 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. from fa — —rn/g.23. . 23. timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously.e. If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below). and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest. this means the total energy of the particle in Euclidean motion is zero.10) where TO is an integration constant. the classical path. In the present context of Euclidean time the configuration (f)c is generally described as an "instanton".
The classical equation (23. Inserting the result (23.10) which rises monotonically from left to right is the pseudoparticle configuration called kink in the context of soliton theory and instanton in onedimensional field theory or quantum mechanics with topological charge 1 and Euclidean action given by (23.13) 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. Using Eq. the constant must be zero.8) we obtain T f SE dr[2V((j)c) + const. we obtain lim f+ OO.4).5).7"/! ).^ .4. For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26.13) into Eq. (23.10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1.12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23. 23. thus indicating the nonperturbative nature of the expression. The configuration (23.13).588 CHAPTER 23. We also observe that the expression is singular for g2 —> 0. Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = . 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression. but with the same Euclidean action (23. Path Integrals and Instantons The Euclidean action of the instanton solution (23. Tf = 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. (23.TJ T fro.10) is given by (23. (23.7) admits another configuration which is the negative of (23.].11) For this quantity to be finite for TJ = —oo.r 0 ) ' (23.13).14) But this is not yet enough. This antiinstanton configuration rises monotonically from right to left as shown in Fig.
/ > Looking at the instanton (23. Questions which arise at this point are: Is the nonmonotonic configuration (23. 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. b u t if TQ. 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. The sum (23.r0).TO far apart and T'Q 3> TQ.g. 23.15b) a solution of the classical equation? Is it a topological configuration — if so.g. what is its associated topological charge? W h a t is its Euclidean action? The sum of 4>c and (j)c = —4>c defined by (23.Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 .15b) a mending together of an instanton and an antiinstanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ. and an antiinstanton localized at (say) TQ.5.TO) + ^ c ( r . 0c(r) = &(T).15b) In the case of (23.15a) we have a simple superposition. as we can see e.2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig.15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r). (23.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig. (23. so correspondingly one defines the instanton or kink with topological charge + 1 and the antiinstanton or antikink with topological charge — 1.5. . #(r) = </>C(T . 23. in the case of (23.15b) since we require configurations from vacuum to vacuum.15b) has the shape shown in Fig.15b) is not exactly a solution.4 The antiinstanton.23.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity. e. 23.
In fact. 5^ In order to answer the second question we observe that (23.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which. #(r) dr 1 1 m 9 cosh m(r — ro) cosh m{r — T'Q)_ cosh2 m(r — TQ m cosh2 m(r — TQ) g cosh m{r — ro) m 1 [1 .T 0 ) } ] . of course. at oo the variation is <<> = 0).15a) is not exactly a classical solution. T\] with LECIT = 0. g cosh m(r — TQ) (23. Path Integrals and Instantons '.5 Superposition of an instanton and a widely separated antiinstanton. m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig. Finally. vanishes as T\ —> ±oo (of course. it also does not exactly stationarize the Euclidean action.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 . 23. since (23. Ignoring exponentially small contributions the Euclidean action of the configuration (23.16) . we see that in the domain [—oo. Thus it belongs to the socalled vacuum sector of solutions and its topological charge (determined by the boundary conditions at plus and minus infinity) is zero.15a) can be argued to be zero.15a) is a configuration which starts from 4> = —i^/g and ends there.CHAPTER 23.2 m ( T o .
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).
whereas the determinant arises only once in the original determinant Dn.exp • /_:T JV(T')" dr' = e x p [ .T')dT' (23. Our final step is to relate the VolterraFredholm determinant to the required functional determinant.. Cameron and W.j. 2! . Thus in our case one of the kernels is zero and we have to decide what we do when.lnJV(T)}] N(T) N(T) (23. = 1 52 K(n. (23.67) i=l v ) = lim exp V In ( 1 + ' L K(T. as in Eq. r ' ) is to be evaluated at r = T ' .n) o K(TJ.62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: .TI) K(T2.Ti) i=l : exp \ £ K(T.. Thus the factor a inserted above has to be taken as 1/2.e.n) i=l N lim exp £*(T.63a).T1)dT1 + J dn J dr2.66) •Dr) .n) z=l x (23.64) D (23. .64).68) D above. so that (with T *T) 1/2 •Dz N(T) (23.n) K(Tj. n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1. where a f 1 See R.4 Field Fluctuations 607 This equation represents a set of n linear equations which has a unique solution provided the determinant Dn of the coefficient matrix on the right does not vanish. the kernel K(r. .7 Y. ^This is the method explained by B. X(TI. 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. (23.64) Inserting the explicit expression for the kernel given in Eq. 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.N{T) Finally we add the following remarks concerning the discretization (23. The Fredholm kernel is in general — as in our illustration above with a Green's function — a quantity with a discontinuity between the two constant integration limits.TI) 0 K(T2. Thus D„ = l + 5 ^ K ( r i . H. p.65) The quantity D is really the Fredholm determinant which is obtained as above by solving the Fredholm integral equation with constant integration limits.a { l n i V ( r ) . the value of the determinant is half (or arithmetic mean value) along the diagonal. Martin [43].. In the limit S —* 0 . we obtain exactly the nonvanishing terms on the right of Eq. (23.63b).64) as above and consider the determinant of the expression. n )d? (23. the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — T ) . (23. We observe that if we discretize Eq. Felsager [91]. [44].TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i. 192.23. are summed over all values from 1 to n. This determinant can be expanded in powers of K (more precisely in powers of a parameter A which we attach to K). i.T2) I exp I K(n. T.. r i ) + .
3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23.Z ( T ). Eq. In = dzi dz2 • • • dzn e x p . (23.71) Hence (with h = 1) l = j .Z i . 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. (23.N(T) .69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf. f(T) = N{T) J^ dr'^^z(r') (23.59)) + f(T)] = — f dz(T) f°° dae^W+f™.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. I2 and using induction one obtains (cf.0.2o) 2 rc^O. Path Integrals and Instantons Example 23. Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T). (21.73) .57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T . Evaluating Ii. T + N(T) / T' = T JT dr' N(T') . and hence may be ignored.70) and this now requires also integration over the endpoint coordinate z(T). With partial integration (in the second step) we obtain (since (cf. (23. t h e range of T is subdivided into n + 1 equal elements of length e with (n + l)e = 2T and TO = —T.59) the endpoint conditions »)(±T) = 0 and evaluate the integral.l + £7t) «i (23. n J=.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). Solution: From Eq. In every integration / dzi the contribution €7. Eq. so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo. .22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l . Briefly.72) The endpoint integration dz(T) = dz(rn+i) can be combined with this.( .J dz(T) J T>{z} J da•Dz hJ dz{T) v{z} I f_A\^+W)i{T)} J H § I"exp [ . rn+i = T. (23.— . represents a constant translation.e—*0 In. (23.(T) + /_ dr — .^ . N(T') .
N. (21. (23. . ^E.12).3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function. 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.76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE. J co § dr Recall that / f ^ doce^ = yfVfp.4 Field Fluctuations 609 In the required limit the argument of the exponential is proportional to 1/T and thus vanishes in the limit T = (n + l)e — oo. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const. a = const.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. Eq..11). D. This result verifies Eq. Gildener and A. / — / / We started off with V(T) = ^CnVniT). (23. £n = ~2 dTT](T)7]n(l (23. i.59) and agrees with that in the literature. Patrascioiu [110]. 1/2 T W 2 (r)J (23.4.74) where we reinserted h. Faddeev and V.e.£Q) — aTo]dTod£o. Popov [89]. 11 L. Ignoring the remaining metric factor with the reasoning explained > between Eqs. (3.10) and (21. j / dTT]n(T)r]m(l r .^ 23.23.
^O) with dependence on ro which the replacement (23.J drm(r)Mr + T0) (23. e. Thus. / dTm{r)n{T). (23. in the direction of the zero mode. The delta function 8[F(TQ. i. We consider the following integral / ( r 0 ) := .80) i r or •i /*oo dr7?o(r)0(r) = 0.75) requires.£o)/a.75) enforces the collective coordinate ro to be given by F(ro. in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate. (23. We now impose (i. . The number £o is the component of the fluctuation rj(r).e. (23. demand validity of) the constraint £o = 0.81) the rdependence is integrated out and hence the integral does not provide the function F(TO. In the integral of Eq. Then cn = ~2 / dr(pc(T)rjn(T). if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) .76) we set 4>c{r) = ^CnVnir). in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r).81) 2 / dr»7o(r)0c(r) = .£Q) — aro] contained in Eq. at r = ro. i /*oo (23. In analogy to (23. i.e.610 this means that CHAPTER 23.82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T). We establish the function ^(To>£o) with the following reasoning. this constraint implies that we go to the subspace which is orthogonal to the zero mode.78) <t>ij) = Y. Path Integrals and Instantons _^ ^ w .77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration. (23.e.79) (23. where L = in + cn. (23. (23.83) **The collective coordinate in the present context is also called "instanton time". A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T).^n{r).g.
jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0.87) l/M) (23.86) = Jdrrjoir^Mr F(TO.(23./ drri0(T)<pc(T) + PJ (23.90) Now tt 0^C(T + TQ) ro=0 <9T0 dr (23.88) ro0 Then in Eq. (23.y/SET0 = 0.Zo)y/SETQ We now define A FP '• =0 • (23.23. (23.<t>c{r)} .£O) OTQ = = £ " .. Felsager [91]. We have AFP = /F(T0.84) (23.82) we can replace here 7?O(T)0C(T) by 77O(T)T7(T).£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o.y/S^ro] (23. We evaluate this quantity as follows in the leading approximation. so t h a t with F(TQ.85) PJ W i t h Eq. (23.4 Field Fluctuations 611 for an infinitesimal translational shift TO of the position of 4>c(r). .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 ) .75) 1 = AFP f dT06[{F{rQ.91) " F o r more discussion of this point and related aspects see also B. p. 158.89) Here the factor App is called the FaddeevPopov determinant. . Expanding /(TO) about TQ = 0 we obtain (with (J>'C{T) = T)Q(T)\J13E/'p) /(TO) = = Hence / dTT]Q(T)[(j)c(T + To) .^) .
using the relation (23. 7 Thus there is no fluctuation along the direction of the zero mode in the sum.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. 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.e. iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ).77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction.75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 . the "static energy" invariant. 0 d r (23. 770(7"). in d(j)c(To) + dr/fa). ~ There is a quick way to obtain the result (23.e.0AFP8 4o H P . if the coefficient of 770(70) vanishes. Inserting (23.93). (23.612 CHAPTER 23.e.90).93) where AT = TT — 00.94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals. 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./ dT0AFP. Consider. P the relative sign being of no significance. i. if ^ + d£o = 0. Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i. we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe. tangential to (J)C(T)) leaves this and hence the square of this quantity. i.91) into Eq. /(TO) ^/SE~T0} V~SET0 c d£. =~ (23. . P J AT (23. fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To).
4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23.96) lim .97) is a constant still to be evaluated.96) does not yet allow a comparison with (23.93).23.35a) since the dependence on Tf — Ti = 2T of both expressions differs significantly.60a) in not involving the zero mode contributions. (23.55b). We can now perform the Gaussian integrations and write the result lim T^oo p 4 m3 exp 3^J (23. (23. (23. We determine I'Q by observing t h a t this differs from IQ of Eq.4. we can set I 2TT ^o = l'o\ wlp2'' which with Eq. (23. Since w\ is for finite T given by WQ of Eq.4 Field 23.95) P Here £o 2 in the exponential is multiplied by WQ which is very small for large T and approaches zero in the limit of T —>• oo. (23. so t h a t h = det r)/(n\ppATexp 4 m3' ATAFPdet(^ \ a£ jexp (23. 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 (23.43)). 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.52a) and replace in this J d£o by the expression (23.A T A F P i " o e x p T—>oo p where I'o n^O.60a) for I0 and Eq.55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23. We observe t h a t the result (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.
Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{. ~ rKo')e{Tx . (23. Patrascioiu [110] make the replacements \QP <> 2g 2 and p?GP <» 2m 2 . This means we are concerned with configurations like* 2n+l d(r .TiMiy+'Mr «>.100) in such a way that it exhibits the crucial factors of which it is composed. we drag the factor p2 along for ease of comparison with literature. *For the validity of the calculation the instantons or kinks have to be far apart. the result becomes h lim 2 m r ^ ^ 1 e .614 CHAPTER 23. Path Integrals and Instantons It follows that the kernel becomes* h = Urn ATAFPI0( T^oo p W^TT ^L exp 4 m3"1 3^h 4 m6 .a n t i . . T0 < 4l) <C Ti « 42) < • • • «C T 2 n + 1 = T — oo (23.5 I n s t a n t o n . (23.* Finally we rewrite the expression (23. Patrascioiu [110]. in the present case we also have to include multiinstanton configurations in the form of one instanton plus any number of instantonantiinstanton pairs which describe back and forth tunneling between the two wells.2 m T e x p 2 irgh ' T»oo (23.99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App. Gildener and A. i.e. their Eq.r). In view of their normalization of the normal modes.55b)) w0 = v0e VQ 23.101) 2T > oo (cf. (3. 0c M. Gildener and A. ^See E.32).i n s t a n t o n contributions As already mentioned.20). the approximation is then known as the "dilute gas approximation". Although we choose p2 = 1. they have to introduce the inverse of that factor on the far right of their expression (3.4.4>C exp h (23.100) We observe that the dependence on the normalization constant p cancels out.102) *For comparison with the parameters of E.
this means we have to consider the amplitude (23.4>I) ld2SE SE(c/>i. 7 i = . .H)2 + . (23.2 T / 3 .<Pi)2 + • • • .104) Considering the case n — 1 of a kink with one additional kinkantikink pair.Tl}{4>l.36) (/.4>l)eSE{4>l.r / 3 . Consider (with T 0 = T. . (3) (23.Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23. are varied. Proceeding as before we now consider first in analogy with Eq.23.l ) i + 1 in the sum of Eq.17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed. 1 0 6 a ) and SE{(J>I.106b) SE{<t>l.Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f.Tl\(j)i.<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h . T3 = T) d(f>l((pf.4 Field Fluctuations 615 for n = 1.105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations. (n = 0 implies the single instanton solution).<t>ff + h . (23.Tf\(p2. ( 2 3 .102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left). we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f.T2\(f)l.fo) = Id2 St SE(<Pf. the total Feynman amplitude is given by (^/.<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 .r 0 ( 1 ) = . T2 = T/3. The factor ( . Taking these configurations into account.(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0. r 0 ( 2 ) = 0.r/l&.(l>i) (23.103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig. (23. Since these endpoint configurations.<t>2)eSE{<t>2. here (f>2. 23. r0(3) = 2T/3. .4>i.6.T2}((t)2. 2 .<j>i) .
of course.106a).4>f) d4>\ 4>s72 (23.109) The result of these integrations is that the amplitude l2n+i involves 2n intermediate integrations.12) we have JM~°°) i.(01 .106b) is to be substituted into the second relation of (23. that each applies only to a fraction of the overall Euclidean time interval of r = 2T.4>i) / d(f>2 e x p {<h . rn ^2 J2H 9 1m' ±2gcj). each of which implies a multiplicative factor 7. It follows that — always within our approximations — c (^TflfaTjW SE{<t>f.e.e. It has to be remembered.. With our remark after Eq. d2SE{4>) d(j)2 2m.616 CHAPTER 23. the dilute gas approximation.2d2SE ~c 4>i d(f)i e x p SE^f.