•
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
With various techniques and deeper studies. screened Coulomb potentials and maybe singular potentials. 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. However. whereas for the calculation of discrete eigenvalues the Schrodinger equation. 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. in fact.e. again point the way: For scattering problems the path integral seems particularly convenient. 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. its mathematical foundations. The introduction to quantum mechanics we attempt here could be subdivided into essentially four consecutive parts. Thus this first part . in Feynman diagrams in quantum electrodynamics). Excluding spin. but arise also in recently studied models of large spins. numerous problems could. any problem more complicated is frequently dispensed with by referring to cumbersome perturbation methods. Chapters 1 to 14. we recapitulate the origin of quantum mechanics. and it will be seen in the final chapter that potentials with degenerate vacua are not exclusively of general interest. basic postulates and standard applications. but also about complex energies that he encounters in a parallel course on nuclear physics. Thus important level splitting formulas for periodic and anharmonic oscillator potentials (i. be treated to a considerable degree of satisfaction perturbatively. with degenerate vacua) were first and more easily derived from the Schrodinger equation. like periodic potentials. even though the expansions are mostly asymptotic.g. With the growing importance of models in statistical mechanics and in field theory. To what extent the two methods are actually equivalent. has not always been understood well. In the first part. that is the Coulomb potential and the harmonic oscillator. These basic cases will be dealt with in detail by both methods in this text.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. one problem being that there are few nontrivial models which permit a deeper insight into their connection. the aforementioned exactly solvable cases. 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.
which is presumably the only such singular case permitting complete solution and was achieved only recently. The most prominent examples here are the double well potential and its inverted form. After a treatment of power potentials.XVII deals mainly with standard quantum mechanics although we do not dwell here on a large number of other aspects which are treated in detail in the longestablished and wellknown textbooks. we deal mostly with applications depending on perturbation theory. In the final chapter of this part we discuss the large order behaviour of the perturbation expansion with particular reference to the cosine and double well potentials. Chapters 15 to 20. i. the method of matched asymptotic expansions with boundary conditions (the latter providing the socalled nonperturbative effects). In the majority of the cases that we treat we do not use the standard RayleighSchrodinger perturbation method but the systematic perturbation procedure of Dingle and Muller which is introduced in Chapter 8. The following Chapter then deals with Schrodinger potentials which represent essentially anharmonic oscillators. and their eigenvalues. 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. The potentials with degenerate minima will be seen to reappear throughout the text. periodic instantons. we derive respectively the levelsplitting formula and the imaginary energy part for these cases for arbitrary states. Using perturbation theory. In part three the path integral method is introduced and its use is illustrated by application to the Coulomb potential and to the derivation of the Rutherford scattering formula. the chapter thereafter deals with Yukawa potentials. and the elliptic or Lame potential — here introduced earlier as a generaliza . 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. Thereafter the concepts of instantons.e. In the second part. We also consider inverted double wells and calculate with the path integral the imaginary part of the energy (or decay width). and the behaviour of the eigenvalues is discussed in both weak and strong coupling domains with formation of bands and their asymptotic limits. 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 earlier Chapter 17 also contains a brief description of a similar treatment of the elliptic or Lame potential. This is followed by the important case of the cosine or Mathieu potential for which the perturbation method was originally developed.
This puts Schrodinger equations with e.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. the Mathieu equation. the Schrodinger equation can be solved for practically any potential in complete analogy to wellknown differential equations of mathematical physics. cubic). and whenever available also with the results of WKB calculations. we can make the following observations. it is. for instance. The application of path integrals to the same problems with the same aims is seen to involve a number of subtle steps. except that these are no longer of hypergeometric type. quartic. 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. Employing anharmonic oscillator and periodic potentials and reobtaining these in the context of a simple spin model. All results are compared with those obtained by perturbation theory. however. from our precise reference to unavoidable elliptic integrals taken from Tables). An expected observation is that — ignoring a minor deficiency — the WKB approximation is and remains the most immediate way to obtain the dominant contribution of an eigenenergy. Comparing the Schrodinger equation method with that of the path integral as applied to identical or similar problems. We then illustrate the relevance of this in the method of collective coordinates. anharmonic oscillator potentials on a comparable level with. 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. With a fully systematic perturbation method and with applied boundary conditions. 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 particular solutions and eigenvalues of interest in physics are — as a rule — those which are asymptotic expansions.g. we consider the topic of transitions between the quantum and thermal regimes at the top of the barrier and show that these may be classified in analogy to phase transitions in statistical mechanics. The introduction of collective coordinates of classical configurations and the fluctuations about these leads to constraints. such as limiting procedures. Our fourth and final part therefore deals with elementary aspects of the quantization of systems with constraints as introduced by Dirac. The physical behaviour there (in the transition region between quantum and thermal physics) is no longer controlled by the Schrodinger equation. for instance. This method is therefore more complicated. this comparison on a transparent level being one of the main aims of this text. an approximation whose higher .
Their deep involvement in the attempt described here is evident from the cited bibliography. formulas taken from Tables or elsewhere are referred to by number and/or page number in the source. so that the reader is spared difficult and considerably timeconsuming searches in a source (and besides. The author is deeply indebted to his onetime supervisor Professor R. Dingle (then University of Western Australia. B. B. Throughout the text some calculations which require special attention. Liang (Taiyuan).g. which is particularly important in the case of elliptic integrals which require a relative ordering of integration limits and parameter domains. in particular Professors J. Dingle for paving him the way into this field which — though not always at the forefront of current research (including the author's) — repeatedly triggered recurring interest to return to it. Whittaker and Watson [283]. Nonetheless. . shows him that each such formula here has been properly looked up).XIX order contributions are difficult to obtain. D. In writing this text the author considered it of interest to demonstrate the parallel application of both the Schrodinger equation and the path integral to a selection of basic problems.* H. Watson [283]. H. thereafter University of St. we also consider at various points of the text comparisons with WKB approximations. D. whose research into asymptotic expansions laid the ground for detailed explorations into perturbation theory and large order behaviour. e. as well as applications and illustrations. MiillerKirsten *In the running text references are cited like e. also for the verification of results. As a rule.g. Tchrakian (Dublin) and Jianzu Zhang (Shanghai). Instead a glance at a nearby footnote provides the reader immediately the names of authors. J. the references referred to are never cited by mere numbers which have to be identified e. 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. not the least for permitting a more detailed and hopefully comprehensible presentation here. an additional motivation was that a sufficient understanding of the more complicated of these problems had been achieved only in recent years. The line of thinking underlying this text grew out of the author's association with Professor R. N. For ease of reading.Q. Whittaker and G.g. other topics have been left out which are usually found in books on quantum mechanics (and can be looked up there). Since this comparison was the guideline in writing the text. are relegated to separate subsections which — lacking a better name — we refer to as Examples. at the end of a chapter (after troublesome turning of pages). The author has to thank several of his colleagues for their highly devoted collaboration in this latter part of the work over many years. This endeavour developed into an unforeseen task leading to periodic instantons and the exploration of quantumclassical transitions. Andrews). with the source given in the bibliography at the end. K. T. W. Park (Masan). like E.
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. 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. 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.Chapter 1 Introduction 1. Thermal radiation (with wavelengths A ~ 10~ 5 to 10 . that the formula he had established for the energy distribution of electromagnetic black body radiation was in agreement with the experimentally confirmed Wien. which involved also thermodynamics and statistical mechanics (in the sense of Boltzmann's statistical interpretation of entropy).1 Origin and Discovery of Quantum Mechanics The observation made by Planck towards the end of 1900. 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. Although practically every book on quantum mechanics refers at the beginning to Planck's discovery. Instead. the amount of radiation absorbed by a body is equal to the amount the body 1 . Kirchhoff's law in thermodynamics says that in the case of equilibrium. Planck had arrived at his formula with the assumption of a distribution of a countable number of infinitely many oscillators.
1 Absorption in a cavity. i. (1. v + dv in equilibrium with the black body at temperature T.e. y J c 3 \ex .l ) where x = ^ = ^ . Let us look at the final result of Planck. The parameters k and h are the constants of Boltzmann and Planck: k = 1.e.1) ' y Here u(v..2) and the . In Eq. 1.e.3 4 J s. kT kXT (1. the formula (to be explained) u(u. h = 6.1) 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. radiators. Black bodies as good absorbers are therefore also good emitters. Introduction emits. How did Planck arrive at the expression (1.2 CHAPTER 1.1) c is the velocity of light with c = u\.T) = 2*?£(?)kT.38 x 1(T 23 J K'1. u(u. energy per unit volume) of the radiation (i. It was found that one expression agreed well with observations in the region of small A (or AT).e. T) were known and tested experimentally. i. (1.T) = dv3eC2U/T. and then measuring the increase in temperature of the heat bath. and the other in the region of large A (or AT).626 x 10 . These expressions are: (1) Wien's law. of the photons or photon gas) in the cavity with both possible directions of polarization (hence the factor "2") in the frequency domain v. A being the wavelength of the radiation. Fig. T)du is the mean energy density (i. The (equilibrium) radiation of the black body can be determined experimentally by sending radiation into a cavity surrounded by a heat bath at temperature T.
(1. Considering Eq.3) Ci.2 Distributing quanta (dots) among oscillators (boxes). so that W represents the number of possible ways of distributing the number P := NU/e of energyquanta ("photons".e.T) = av e6"/Ti' where a and b are constants.1.1) in regions of a. exp(x) ~ 1+x) and "large" (exp(—x) < 1). Moreover Planck assumed that these oscillators do not absorb or emit energy continuously. Indeed.T) = 2^C3T. 3 (1. To this end he considered Boltzmann's formula S — klnW for the entropy S.1) as approximations. discretization begins to enter. "small" (i.2) and (1. every oscillator corresponding to an eigenmode or eigenvibration or standing wave in the cavity and with mean energy U.1 Origin and Discovery of Quantum Mechanics (2) RayleighJeans law: u(i>. Planck now imagined a number T of oscillators V or iV oscillating degrees of freedom. but — here the discreteness appears properly — only in elements (quanta) e. that the formulas (1. C3 being constants. When Planck had found this expression. where the Fig.47TZ/ 2 {x small). 1. in the first place Planck had tried to find an expression linking both. (1. and he had succeeded in finding such an expression of the form u(v. We see. (x large). This is the point. . which are indistinguishable) among the N indistinguishable oscillators at . C2. T) u(i/.3) are contained in Eq. he searched for a derivation. we obtain: u(i/.T) 2^^kT. e xhv. Here W is a number which determines the distribution of the energy among a discrete number of objects. and thus over a discrete number of admissible states.
2) requires that e ex is. h = const.343(2). Agreement with Eq. Introduction temperature T. p.1)!P! (1. as for small values of e).g.2. e = his. formula 8.3 Comparing the polarization modes with those of a 2dimensional oscillator. N * oo. I. i. with the quanta represented schematically by dots as indicated in Fig. = 1/T). Magnus and F. . W.e. We now obtain the energy density of the radiation. i. (1. 1. (1.6) Fig. e. 940. We visualize the iV oscillators as boxes separated by N — 1 walls. by multiplying U with the number nvdv of modes or oscillators per unit volume with frequency v in the interval v.e. with riydu — 2 x —w—dv.7) *See e. . u(i>.g.T)dv. Gradshteyn and I. one obtains (cf. as in most other Tables.3. p. The Stirling formula or approximation will appear frequently in later chapters. Then W is given by w = (N With the help of Stirling's formula* {N + piy. Oberhettinger [181]. U{T) being the average energy emitted by one oscillator. S. M.4) IniV! ~ JVlniViV + O(0). (1. there not called Stirling's formula. 1.1) u = vmrri (L5) as the mean energy emitted or absorbed by an oscillator (corresponding to the classical expression of 2 x kT/2. Ryzhik [122].4 CHAPTER 1. and the second law of thermodynamics ((dS/dU)v Example 1. v + dv.
where^ 2 [2KUY . The number of possible modes (states) is equal to the volume of the spherical octant (where n^ > 0) in the space of n^. as in electrodynamics..2. so that .1.7) for instance. as indicated in Fig.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. dj\l dM .3. v + dv..e. d ™4*±\IL> — —dv = nvdv dv dv _8 3 \ c / 14 8 2 4TTV2 = 3 dv 83 ^ ^ = ^ ^ ' as claimed in Eq. (1. We observe that u(v. We obtain therefore u^T) = Unv = 2^fJ^—i.UJ = 2KV. is given by .T)dX = ^ehc/*kT_idX. '''From the equation I \ JW .3.2.1)) The solutions of this equation are ^max = 4. .1. rii = 1. L for i = 1. KT = I J .8) This is Planck's formula (1.1).. 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.V 2 ) E = 0. In terms of A we have u(X.7). i.2.T) has a maximum which follows from du/dX = 0 (with c = vX).i = 1..3 (as for ideal conductors). We obtain the expression (1. (1. (1.UJ2/C2 + K? = 0.965 and xmin = 0.. 2 2 2 r2 L K — 7T n . Then L^j = nrii.. nvdv per unit volume. so that (lvL\A 0 I I = rr. . The number with frequency v in the interval v. so that the derivative of u implies (x as in Eq.
the behaviour of the system at zero absolute temperature. i. One might suppose now. 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 .965K = Const.—v <L9) We observe that for T — 0 (i. of course. . Since temperature originates through contact with other oscillators. which can not be eliminated without a different approach. We are not dealing with the linear harmonic oscillator familiar from mechanics here.l. We therefore examine such contradictions next. and from which the constant h can be determined from the known value of k. . A. . Later it was realized by H. ra = 0.2 (1. but one can expect an analogy. we then have at T = 0 independent oscillators.8) could be derived much more easily in the context of statistical mechanics. 2 . This is Wien's displacement law. . 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=—. such a procedure leads to contradictions.10) Thus here the socalled zero point energy appears. which can assume the discrete energies en — nhv. which had also been known before Planck's discovery. One expects. n = 0. Introduction he 4.e. However. that we arrive at quantum mechanics simply by discretizing the energy and thus by postulating — following Planck — for the harmonic oscillator the expression (1. which did not arise in Planck's consideration of 1900. x — oo) the mean energy vanishes (0 < U < * > oo). that it is easier to consider first the case of T = 0.e. Lorentz and Planck that Eq. Thus we have a rather complicated system here.6 The first value yields AmaxT = CHAPTER 1.10). If an oscillator with thermal weight or occupation probability exp(—nx) can assume only discrete energies en = nhu. . (1. that of an oscillation system at absolute temperature T ^ 0.1.
In . for 2 degrees of freedom.1)! .1. .2 Contradicting Discretization: Uncertainties 7 Example 1.1: 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. kT/2. 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".kT This means U is then the classical expression resulting from the mean kinetic energy per degree of freedom.ln(TV . so that 1 f ds\ T — \dUjy . 1. {NW •• 1)! (TV1)!P! and P = UN Show with the help of Stirling's formula that the mean energy U of an oscillator is given by U •• exp(e/fcT) . d ..= .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.e.1)! .+ 1 e \U e u = exp(e/fcT) which for e/kT — 0 becomes > .k dU 1+U\ f ln(l T + U\  U U k ( e .1 ' u~ e . i. around 1926. with Heisenberg's discovery of the uncertainty relation.1 Solution: Inserting W into Boltzmann's formula and using In TV! ~ A In T — TV. 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. In the following we attempt to incorporate the above discretizations into classical considerations* and consider for this reason socalled thought experiments (from German "Gedankenexperimente").i n .2 Contradicting Discretization: Uncertainties The farreaching consequences of Planck's quantization hypothesis were recognized only later.
8
CHAPTER 1. Introduction
shall see that we arrive at contradictions. As an example^ we consider the linear harmonic oscillator with energy E = mx2 +  w V .
ZJ
(1.11)
Zi
The classical equation of motion dE n — = x(mx + mco x) = 0 permits solutions x = Acos(u>t + S), so that E = mco2 A2, where A is the maximum displacement of the oscillation, i.e. at x — 0. We consider first this case of velocity and hence momentum precisely zero, and investigate the possibility to fix the amplitude. If we replace E by the discretized expression (1.10), i.e. by En — (n + 1/2)HUJ, we obtain for the amplitude A
A A
^ =\[Ef+l
(i i2)

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

(1.17a)
In the first of the examples discussed above the contradiction arose as a consequence of our assumption that we could put the oscillator initially at
1.2 Contradicting Discretization:
Uncertainties
11
any point in phase space, i.e. at some point which does not belong to one of the allowed ellipses. In the second example we chose n = 0 and thus restricted ourselves to the innermost orbit. However, we also assumed we would know at which point of the orbit the pendulum could be found. Thus in attempting to incorporate the discrete quantization condition into the context of classical mechanics we see, that a system cannot be localized with arbitrary precision in phase space, in other words the area AA, in which a system can be localized, is not nought. We can write this area AA > An+1 Any'=
(1.17a) 2TT/L
since the system cannot be "between" An+i and An. Since A A represents an element of area of the (q, p)plane, we can write more precisely ApAq > 2irh. (1.18)
This relation, called the Heisenberg uncertainty relation, implies that if we wish to make q very precise by arranging Aq to be very small, the complementary uncertainty in momentum, Ap, becomes correspondingly large and extends over a large number of quantum states, as — for instance — in the second example considered above and illustrated in Fig. 1.5.
Fig. 1.5 Precise q implying large uncertainty in p. Thus we face the problem of formulating classical mechanics in such a way that by some kind of extension or generalization we can find a way to quantum mechanics. Instead of the deterministic Newtonian mechanics — which for a given precise initial position and initial momentum of a system yields the precise values of these for any later time — we require a formulation answering the question: If the system is at time t = 0 in the area defined by
12 the limits 0 < q < q + Aq, 0<p<p
CHAPTER 1. Introduction
+ Ap,
what can be said about its position at some later time t = T? The appropriate formulation does not yet have anything to do with quantum mechanics; however, it permits the transition to quantum mechanics, as we shall see. Before we continue in this direction, we return once again briefly to the historical development, and there to the ideas leading to particlewave duality.^
1.3
ParticleWave Dualism
The wave nature of light can be deduced from the phenomenon of interference, as in a doubleslit experiment, as illustrated in Fig. 1.6.
Fig. 1.6 Schematic arrangement of the doubleslit experiment. Light of wavelength A from a source point 0 can reach point P on the observation screen C either through slit A or through slit B in the diaphragm placed somewhere in between. If the difference of the path lengths OBP, OAP is n\,n € Z, the wave at P is reinforced by superposition and one observes a bright spot; if the difference is n\/2, the waves annul each other and one observes a dark spot. Both observations can be understood by a wave propagation of light. The photoelectric effect, however, seems to suggest a corpuscular nature of light. In this effect* light of frequency v is sent onto a metal plate in a vacuum, and the electrons ejected by the light from the plate are observed by applying a potential difference between this plate and another one. The energy of the observed electrons depends only on v and
"See also M.C. Combourieu and H. Rauch [58]. "This is explained in experimental physics; we therefore do not enter into a deeper explanation here.
1.3 ParticleWave Dualism
13
the number of such photoelectrons on the intensity of the incoming light. This is true even for very weak light. Einstein concluded from this effect, that the energy in a light ray is transported in the form of localized packets, called wave packets, which are also described as photons or quanta. Indeed the Compton effect, i.e. the elastic scattering of light, demonstrates that photons can be scattered off electrons like particles. Thus whereas Planck postulated that an oscillator emits or absorbs radiation in units of hv = hu>, Einstein went further and postulated that radiation consists of discrete quanta. Thus light can be attributed a wave nature but also a corpuscular, i.e. particlelike, nature. In the interference experiment light behaves like a wave, but in the photoelectric effect like a stream of particles. One could try to play a trick, and use radiation which is so weak that it can transport only very few photons. What does the interference pattern then look like? Instead of bands one observes a few pointlike spots. With an increasing number of photons these spots become denser and produce bands. Thus the interference experiment is always indicative of the wave nature of light, whereas the photoelectric effect is indicative of its particlelike nature. Without going into further historical details we add here, that it was Einstein in 1905 who attributed a momentum p to the light quantum with energy E = hv, and both he and Planck attributed to this the momentum
The hypothesis that every freely moving nonrelativistic microscopic particle with energy E and momentum p can be attributed a plane harmonic matter wave ip(r,t) was put forward much later, i.e. in 1924, by de Broglie.t This wave can be written as a complex function ij)(T,t) =Aeikriut,
where r is the position vector, and to and k are given by E — hio, p = /ik. Thus particles also possess a wavelike nature. It is wellknown that this was experimentally verified by Davisson and Germer [64], who demonstrated the existence of electron waves by the observation of diffraction fringes instead of intensity distributions in appropriate experiments.
f
L. de Broglie [39].
14
CHAPTER
1.
Introduction
1.4
ParticleWave Dualism and Uncertainties
We saw above t h a t we can observe the wave nature of light in one type of experiment, and its particlelike nature in another. We cannot observe both types simultaneously, i.e. the wavelike nature together with the particlelike nature. Thus these wave and particle aspects are complementary, and show up only under specific experimental situations. In fact, they exclude each other. Every a t t e m p t to single out either of these aspects, requires a modification of the experiment which rules out every possibility to observe the other aspect.* This becomes particularly clear, if in a doubleslit experiment the detectors which register outcoming photons are placed immediately behind the diaphragm with the two slits: A photon is registered only in one detector, not in b o t h — hence it cannot split itself. Applying the above uncertainty principle to this situation, we identify the attempt to determine which slit the photon passes through with the observation of its position coordinate q. On the other hand the observation of the interference fringes corresponds to the observation of its momentum p.§ Since the reader will ask himself what happens in the case of a single slit, we consider this case in Example 1.2. Example 1.2: The SingleSlit Experiment
Discuss the uncertainties of the canonical variables in relation to the diffraction fringes observed in a singleslit experiment. Solution: Let light of wavelength A fall vertically on a diaphragm Si with slit AB as shown schematicaly in Fig. 1.7.
^y
Ax
Fig. 1.7 Schematic arrangement of the singleslit experiment.
On the screen S2 one then observes a diffraction pattern of alternately bright and dark fringes, in the See, for instance, the discussion in A. Messiah [195], Vol. I, Sec. 4.4.4. Considerable discussion can be found in A. Rae [234].
1.4 ParticleWave Dualism and Uncertainties
15
figure indicated by maxima and minima of the light intensity I. As remarked earlier, the fringes are formed by interference of rays traversing different paths from the source to the observation screen. Before we enter into a discussion of uncertainties, we derive an expression for the intensity I. Since the derivation is not of primary interest here, we resort to a (still somewhat cunbersome) trick justification, which however can also be obtained in a rigorous way." We subdivide the distance AB = Ax into N equal pieces AP\, P1P2,..., as indicated in Fig. 1.8.
Si
t
Ax
\
»v
"^^J
B ^ ^
*
w
p3
Q
Fig. 1.8 The wavefront
WW.
We consider rays deflected by an angle 9 with wavefront WW' and bundled by a lense L and focussed at a point Q on the screen S2. Since WW1 is a wavefront, all points on it have the same phase, so that light sent out from a source at Q reaches every point on WW1 at the same time and across equal distances. Hence a phase difference at Q can be attributed to different path lengths from Pi,P2,... to WW'. Considering two paths from neighbouring points Pi,Pj along AB, the difference in their lengths is Axsva6/N. In the case of a wave having the shape of the function 2?r sin kr = sin this implies a phase difference given by < ;v = 5 2n Ax sin 6 \ N (1.20)
Just as we can represent an amplitude r having phase 6 by a vector r, i.e. r — rexp(iS), we > can similarly imagine the wave at Q, and this means its amplitude and phase, as represented by a vector, and similarly the wave of any component of the ray passing through AP\, P1P2, • • •• If we represent their effects at Q by vectors of equal moduli but different directions, their sum is the resultant OPN as indicated in Fig. 1.9. In the limit N — 00 the N vectors produce the arc of a > circle. The angle 5 between the tangents at the two ends is the phase difference of the rays from the edges of the slit:
27T
5 = 2a =
lim NSN = — A a ; s i n 0 .
(1.21)
If all rays were in phase, the amplitude, given by the length of the arc OQ, would be given by the chord OQ. Hence we obtain for the amplitude A at Q if AQ is the amplitude of the beam at the slit: . length of chord OQ , 2a sin a , sin a A0 , ,,—; =A0 . (1.22) 7^=A0 length of arc OQ a2a "S. G. Starling and A. J. Woodall [260], p. 664. For other derivations see e.g. A. Brachner and R. Fichtner [32], p. 52.
16
The intensity at the point Q is therefore
CHAPTER
1. Introduction
h = h
where from Eq. (1.21)
•K ,
.
a = flisinB = Aisint A 2
.
k
Fig. 1.9 The resultant OPM of N equal vectors with varying inclination.
Thus the intensity at the point Q is
Ie=Io
sin 2 (fcAx sin6(/2) (fcAx sin 0/2) 2
(1.23)
The maxima of this distribution are obtained for fcAxsinfl = (2n + 1 )  , i.e. for A x sin0 = (2n + 1 )  = (2n + 1) and minima for 1 fcAx sin # = 7171", i.e. for A x i
: TlA.
A
(1.24a)
(1.24b)
The maxima are not exactly where only the numerator assumes extremal values, since the variable also occurs in the denominator, but nearby. We return to the singleslit experiment. Let the light incident on the diaphragm S i have a sharp momentum p = h/\. When the ray passes through the slit the position of the photon is fixed by the width of the slit A x , and afterwards the photon's position is even less precisely known. We have a situation which — for the observation on the screen S2 is a past (the uncertainty relation does not refer to this past with px = 0, rather to the position and momentum later; for the situation of the past A x A p is less than h). The above formula (1.23) gives the probability that after passing through the slit the photon appears at some point on the screen 52. This probability says, that the photon's momentum component px after passing through the slit is no longer zero, but indeterminate. It is not possible to predict at which point on S2 the photon will appear (if we knew this, we could derive px from this). The momentum uncertainty in the direction x can be estimated from the geometry of Fig. 1.10, where 6 is the angle in the direction to the first minimum: Apx = 2px =2psin6 = — sing.
A
(125)
From Eq. (1.24b) we obtain for the angle 9 in the direction of the first minimum Ax sin 6 = A,
1.4 ParticleWave Dualism and Uncertainties
17
Fig. 1.10 The components of momentum p.
so that Ax Apx = 2h. If we take the higher order minima into account, we obtain AxApx A x Apx > h. We see that as a consequence of the indeterminacy of position and momentum, one has to introduce probability considerations. The limiting value of the uncertainty relation does not depend on how we try to measure position and momentum. It does also not depend on the type of particle (what applies to electromagnetic waves, applies also to particle waves). = 2nh, or
1.4.1
Further thought experiments
Another experiment very similar to that described above is the attempt to localize a particle by means of an idealized microscope consisting of a single lense. This is depicted schematically in Fig. 1.11. light
Fig. 1.11 Light incident as shown. The resolving power of a lense L is determined by the separation Aa; of the first two neighbouring interference fringes, i.e. the position of a particle is at best determinable only up to an uncertainty Ax. Let 9 be one half of the angle as shown in Fig. 1.11, where P is the particle. We allow light to fall in the direction of —x on the particle, from which it is scattered. We assume a quantum of light is scattered from P through the lense L to S where it
18
CHAPTER 1. Introduction
is focussed and registered on a photographic plate. For the resolving power Ax of the lense one can derive a formula like Eqs. (1.24a), (1.24b) . This is derived in books on optics, and hence will not be verified here, i.eJ Ax~±. (1.26a) 2 sm 0 The precise direction in which the photon with momentum p = h/X is scattered is not known. However, after scattering of the photon, for instance along PA in Fig. 1.11, the uncertainty in its xcomponent is 1h Apx = 2psin0 = — sine A (1.26b)
(prior to scattering the xcomponents of the momenta of the particle and the photon may be known precisely). From Eqs. (1.26a), (1.26b) we obtain again Ax Apx ~ h. The above considerations lead to the question of what kind of physical quantities obey an uncertainty relation. For instance, how about momentum and kinetic energy T? Apparently there are "compatible!'1 and "incompatible" quantities, the latter being those subjected to an uncertainty relation. If the momentump x is "sharp", meaning Apx = 0, then also T = px2/2m is sharp, i.e. T and px are compatible. In the case of angular momentum L = r x p, we have L = rp' = rp', where p' = p sin 0. As one can see, r and p' are perpendicular to each other and thus can be sharp simultaneously. If p' lies in the direction of x, we have Ax Ap' > h, where now Ax = rAip, ip being the azimuthal angle, i.e. rAipAp'>h, i.e. ALA<p>h.
Thus the angular momentum L is not simultaneously exactly determinable with the angle </?. This means, when L is known exactly, the position of the object in the plane perpendicular to L is totally indeterminate. Finally we mention an uncertainty relation which has a meaning different from that of the relations considered thus far. In the relation Ax Apx > 0 the
"See, for instance, N. F. Mott, [199], p. 111. In some books the factor of "2" is missing; see, for instance, S. Simons [251], p. 12.
1.5 The Complementarity
Primciple
19
quantities Ax, Apx are uncertainties at one and the same instant of time, and x and px cannot assume simultaneously precisely determined values. If, however, we consider a wave packet, such as we consider later, which spreads over a distance Ax and has group velocity VQ = p/m, the situation is different. The energy E of this wave packet (as also its momentum) has an uncertainty given by
AE « T^Ap = vGAp. op
The instant of time t at which the wave packet passes a certain point x is not unique in view of the wave packet's spread Ax. Thus this time t is uncertain by an amount
At w Ax
vG
.
It follows that AtAE^AxAp>h. (1.27) Thus if a particle does not remain in some state of a number of states for a period of time longer than At, the energy values in this state have an indeterminacy of Ai£.
1.5
Bohr's Complementarity Principle
Vaguely expressed the complementarity principle says that two canonically conjugate variables like position coordinate x and the the associated canonical momentum p of a particle are related in such a way that the measurement of one (with uncertainty Ax) has consequences for the measurement of the other. But this is essentially what the uncertainty relation expresses. Bohr's complementarity principle goes further. Every measurement we are interested in is performed with a macroscopic apparatus at a microscopic object. In the course of the measurement the apparatus interferes with the state of the microscopic object. Thus really one has to consider the combined system of both, not a selected part alone. The uncertainty relation shows: If we try to determine the position coordinate with utmost precision all information about the object's momentum is lost — precisely as a consequence of the disturbance of the microscopic system by the measuring instrument. The socalled Kopenhagen view, i.e. that of Bohr, is expressed in the thesis that the microscopic object together with the apparatus determine the result of a measurement. This implies that if a beam of light or electrons is passed through a doubleslit (this being the apparatus in this case) the photons or
20
CHAPTER 1. Introduction
electrons behave like waves precisely because under these observation conditions they are waves, and that on the other hand, when observed in a counter, they behave like a stream of particles because under these conditions they are particles. In fact, without performance of some measurement (e.g. at some electron) we cannot say anything about the object's existence. The Kopenhagen view can also be expressed by saying that a quantity is real, i.e. physical, only when it is measured, or — put differently — the properties of a quantum system (e.g. whether wavelike or corpuscular) depend on the method of observation. This is the domain of conceptual difficulties which we do not enter into in more detail here.*
1.6
Further Examples
Example 1.3: The oscillator with variable frequency
Consider an harmonic oscillator (i.e. simple pendulum) with timedependent frequency w(t). (a) Considering the case of a monotonically increasing frequency w(t), i.e. dui/dt > 0, from LUQ to u>', show that the energy E' satisfies the following inequality Eo < E' < —yEo, w o (1.28)
where Eo is its energy at deflection angle 6 = 0Q. Compare the inequality with the quantum mechanical zero point energy of an oscillator. (b) Considering the energy of the oscillator averaged over one period of oscillation (for slow, i.e. adiabatic, variation of the frequency) show that the energy becomes proportional to ur. What is the quantum mechanical interpretation of the result? Solution: (a) The equation of motion of the oscillator of mass m and with variable frequency co(t) is mx + mui (t)x = 0, where, according to the given conditions, — > 0, dt
dui
u> = u>o a,t t = 0, w = ui at t = T,
.
_
i.e. io{t) grows monotonically. Multiplying the equation of motion by x we can rewrite it as
1 , 1 w mx • 2\—mui 2 (t)x 2 W 2 2
2 1 n —mx 2 ^ = 0. 2 dt
dt The energy of the oscillator is l „ E — mx1 2
l 0 , z l 29 + mu} (t)x , y 2 '
so that
dE 1 — = mxz dt 2
9
dJ1 > 0, dt ~
(1.29) v '
where we used the given conditions in the last step. On the other hand, dividing the equation of motion by UJ2 and proceeding as before, we obtain  [mx + mur (t)x\ = 0, i.e.
1 1 1 — —mx 2 H—mx 2 dt u22 2
1 mx 2
2d
— . 2 dt\u)
"See e.g. A. Rae [234]; P. C. W. Davies and J. R. Brown [65].
1.6 Farther
Examples
21
d ( E\ 1 , d / 1 \ — — ) = mx2 — ( — = dt\uJ2J 2 dt\u>2)
mx2 dw < 0, UJ3 dt ~
(1.30) v
where the inequality again follows as before. We deduce from the last relation that 1 dE —2 u} dt Integrating we obtain fE' dE ^ f"'2 / < /
2 dw2 . _ E , _, n • n 2 ,a,' —7T, ie. [[InE f„ [ lna; 2J u 22 , i.e. 2 E < E
E dw2 < 0, UJ4 dt ~
i.e.
1 dE 1 dw2 < —2 . E dt ~ u) dt
(1.31)
JEo
or
E  Jui
u, '
> °
"o
E0 ~ UJ22
' ^ u'2 — < —
E'
<
^EQ.
Next we consider the case of the harmonic oscillator as a simple pendulum in the gravitational field of the Earth with
e + wgeo, ^o = f.
and we assume that — as explained in the foregoing — the length of the pendulum is reduced by one half so that J2 = 2  =2u;2. Then the preceding inequality becomes E' < 2E0. In shortening the length of the pendulum we apply energy (work against the tension in the string), maximally however EQ . Only in the case of the instantaneous reduction of the length at 6 = 0 (the pivot does not touch the string!) no energy is added, so that in this case E' = EQ, i.e. E0 < E' < 2E0. We can therefore rewrite the earlier inequality as , u'2 < 5Bo.
E0<E'
Just as the equality on the left applies in the case of an instantaneous increase of the frequency (shortening of pendulum string), so the equality on the right applies to d = # m a x . In classical physics we have 1 2 1 —mx H—? 2 2 If no energy is added, but u> is replaced by 2a; 2 , then x changes, and also x, i.e. x becomes shorter and x becomes faster. The quantum mechanical expression for the energy of the oscillator in its ground state is the zero point energy E = Hu>/2. Here in quantum physics we cannot change UJ without changing E. This means if we double tj instantaneously (i.e. in a time interval A t — 0) > without addition of energy (to fojj/2), then the result E' = Tiw is incorrect by A E = HUJ/2. We cannot have simultaneously A t — 0 and error A E = 0. > (b) The classical expression for E contains u> quadratically, the quantum mechanical expression is linear in OJ. We argue now that we can obtain an expression for E c i a s s i c a l by assuming that w(t) varies very little (i.e. "adiabatically") within a period of oscillation of the oscillator, T. Classical mechanics is deterministic (i.e. the behaviour at time t follows from the equation of motion and
22
CHAPTER
1.
Introduction
the initial conditions); hence for the consideration of a single mass point there is no reason for an averaging over a period, unless we are not interested in an exact value but, e.g. in the average
(lmX/ = ^I0 \mx2{P>dtx2(t) = UJ2X2 and hence
(L32)
E i.e. in
If u> is the frequency of x(t), i.e. x(t) oc cosujt or sinu>t depending on the initial condition, then lmw2x2\ (as follows also from the virial theorem). Eq. (1.29), for mx2 / 2 the mean value = (mx2\ =
If we now insert in the equation for dE/dt,
/I 1£ 2\ (  mx ) = , \ 2 / 2u2' we obtain dE_/l
~dt ~ \2mX
and hence
2\dw
2
_Edw2
dE _ 1 dw2 _ du
/ ~dT ~ 2w2~dT'
E — = const. w
°r
~E ~ ~iU> ~~ ~u7'
In quantum mechanics with E = hw{n + 1/2) this implies H(n + 1/2) = const., i.e. n = const. This means, with slow variation of the frequency the system remains in state n. This is an example of the socalled adiabatic theorem of Ehrenfest, which formulates this in a general formJ
Example 1.4: Angular spread of a beam
A dishlike aerial of radius R is to be designed which can send a microwave beam of wavelength A = 2irh/p from the Earth to a satellite. Estimate the angular spread 6 of the beam. Solution: Initially the photons are restricted to a transverse spread of length A x = 2R. From the uncertainty relation we obtain the uncertainty /\px of the transverse momentum px as Apx ^ h/2R. Hence the angle 0 is given by
~~ p
2R\2nh)
~ AKR'
See e.g. L. Schiff [243], pp. 25  27.
Chapter 2
Hamiltonian Mechanics
2.1 Introductory Remarks
In this chapter we first recapitulate significant aspects of the Hamiltonian formulation of classical mechanics. In particular we recapitulate the concept of Poisson brackets and reexpress Hamilton's equations of motion in terms of these. We shall then make the extremely important observation that these equations can be solved on the basis of very general properties of the Poisson bracket, i.e. without reference to the original definition of the latter. This observation reveals that classical mechanics can be formulated in a framework which permits a generalization by replacing the cnumber valued functions appearing in the Poisson brackets by a larger class of quantities, such as matrices and operators. Thus in this chapter we attempt to approach quantum mechanics as far as possible within the framework of classical mechanics. We shall see that we can even define such concepts as Schrodinger and Heisenberg pictures in the purely classical context.
2.2
The Hamilton Formalism
In courses on classical mechanics it is shown that Hamilton's equations can be derived in a number of ways, e.g. from the Lagrangian with a Legendre transform or with a variational principle from the Hamiltonian H(qi,Pi), i.e.
rt2 r
6
/
^2PiQiH(qi,Pi) dt = 0,
where now (different from the derivation of the EulerLagrange equations) the momenta pi and coordinates qi are treated as independent variables. As
23
Pi. qi(t + 6t) .1). whereas the velocity requires observations of space coordinates at different times. the standard reference for the application of Poisson brackets is the book of P.t). chapter VIII.1). this expression is simply a functional determinant. in analogy with Eqs. 6t>0 5t Real quantities which are directly observable are called observables. All functions u(qi. i. (2. It was only with the development of quantum mechanics by Heisenberg and Dirac that Poisson brackets gained widespread interest in modern physics. du \ x^/dudH du dH\ .24 CHAPTER 2.4) contains as special cases the Hamilton Eqs. (2..4) as the generalization of Eqs.1). since .1) as d . ^As H. It suggests itself therefore to consider more closely the properties of the symbols (2. The following properties can be verified: *See e. (2.Pi) is contained implicitly in the canonical variables q^ and pi.e. the entire timedependence of observables u(qi. H. One now defines as (nonrelativistic) Poisson bracket the expression^ With this definition we can rewrite Eq. Compared with an arbitrary function f(qi. (2.P. A system consisting of several mass points is therefore described by a number of such variables. Goldstein [114] remarks at the end of his chapter VIII. chapter VII.qi(t) qi = hm f . S »(«. dH In this Hamilton formalism it is wrong to consider the momentum pi as mqi.^ ^ j .2) If we have only one degree of freedom (i = 1). the equation of motion of the observable u. as mass times velocity. which all together describe the state of the system. .3). Goldstein [114]. (2. which can be observed directly at time t. The total time derivative of u can therefore be rewritten with the help of Eqs. Dirac [75]. (2. one obtains the Hamilton equations* • OH . A. M. We can therefore consider Eq. . Rather.2) as This equation is.g. (2. x^fdu. One can verify readily that Eq.p% are therefore again observables. (2.) = £ [wm + WiK) = £ [WiWi .pi) of qi.n n. Hamiltonian Mechanics is wellknown. Pi has to be considered as an independent quantity.
it is irrelevant whether we write {A.BC} (5) Jacobi identity: {A.B}* (4) product formation: {A. we expand A and B in powers of qi and pi and apply the above rules until only the fundamental brackets remain.6) We can now show.A}. a i S i + a2B2} = ax{A. we wish to evaluate {A.5b) (3) complex conjugation (note: observables are real.5d) is useful in calculations.B} = {B.e. which combines the Hamilton equations.5e) = {A. If.Pk} = 0. B} is completely evaluated. the Poisson bracket {A.5d) = {A*.4). B}} = 0.2 The Hamilton Formalism (1) Antisymmetry: {A. C}} + {B. As an example we consider a case we shall encounter again and again. but could be multiplied by a complex number): {A. for example. Bx} + a2{A. As long as we are concerned with commuting quantities. Later we shall consider noncommuting quantities. in other words without any reference to the original definition (2. If we evaluate the Poisson brackets for qi. (2. Since Eqs. like here.C}.B}C or C{A.5c) The first three properties are readily seen to hold.6). (2. These are {Qi. {qi. then the ordering is taken as in (2.Pi.Qk} = 0. (2. we obtain the fundamental Poisson brackets. B2}.2. B}. A}} + {C.B}.5a) (2. Property (2. {A. (2. where A and B are arbitrary observables. that of the linear harmonic oscillator.B}C + B{A. i.6) give the values of these. (2. 25 (2.B*}. can be solved solely with the help of the properties of Poisson brackets and the fundamental Poisson brackets (2. (2) linearity: {A. {C.Pk} = 5ik. {B. The original definition of the Poisson bracket will not .5d) above. {pi. (2. that the very general Eq.3) of the Poisson bracket.
or q(t) = qo+ Pot . (2.\(p2 + q2)} = l({q. and P={p.. These are transformations qi—>Qi = Qi(q. q = {q. (2.8a) (2.p) = ±(p2 + q2). Then we obtain: (2. (2.p\) P. Since constants are also irrelevant in this context.8b) p=q.8a) and use the properties of the Poisson bracket and Eqs.6).12) .9) and (2. and so q = q.9) Similarly we obtain from Eq.p.7) into (2. we consider as Hamiltonian the function H(q.p) is an ordinary function is also irrelevant.t).7) q = [q. (2. (2.q2}) = = 2i{q.t). (2.P2} + {q.10) we deduce q = p = q.. From Eqs.yPot3 + •••• (2.H}.4) we have for u = q. q + q = o.qot2 . According to Eq. In the evaluation one should also note that the fact that g..p.p}p + p{q. Hamiltonian Mechanics be used at all.H}. "4' = q.p..11b) (2.8b) We insert (2. and Pi are ordinary real number variables and that H(q.26 CHAPTER 2. (2.11a) 'q' = q. from which we infer that q(t) — qocost + posint. Pi—> Pi = Pi(q. (2.10) In classical mechanics one studies also canonical transformations.
Pi back to their constant initial values. (2. {PhPk} = 0.Qk} = 0. The proof requires the invariance of the fundamental Poisson brackets under canonical transformations.p <> Q.2.e.P provided the transformation q.P One can then show that {A. this transformation is described precisely by the equations of motion but we shall not consider this in more detail here.2 below contains a further illustration of the use of Poisson brackets.3) we can now express the Poisson bracket {^4.p = {A.* Hence in Example 2.pk} = 5ik. for which Hamilton's equations hold. which means that a Hamilton function K(Q. {qi. Eq. as {A.t). {qi. .15a) °raS {AiB}Q. and Example 2.2 The Hamilton Formalism 27 for which the new coordinates are also canonical. B} of two observables A and B in terms of either set of canonical variables. (2.P is canonical in the sense defined above.P.e.e. i. (2.1 we verify only Eq.t).15a).15a). i. *H.Pk} = 0. Example 2. {Qi. chapter VIII.qk} = 0.1: Canonical invariance of Poisson bracket Assuming the invariance of the fundamental Poisson brackets under canonical transformations Qj = Qj(Q'P)>Pj = Pj{Q>P).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.14) With the help of the definition (2.e.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. Pi^Pi=Pi(Q. i. Goldstein [114].13) We write the reversal of the transformation (2. verify that the Poisson bracket of two observables A and B is invariant.B}q. those at a time t — 0.e. _ax p__dK_ (2. P) exists. which transforms qi. . {QhPk} = Sik.B}q. (2. Of course.12) Qi—>qi = qi{Q. i. i.B}Q>P. (2. Example 2. that (dropping the subscripts therefore) {Pi.P.
15a) r„ ™ v . .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).p^v dP.q/c) and (E.p^ "Pk.A}q.p^— + {A.F}r ~di d~E ~ d~E~8t Consider F = H(q.+ dQ. Qk}q.p = dB dA dB \ rA 1 {A'B^ E b r s r . Mittelstaedt [197].( 9A = {Pk. (2. Replacing here A by Q and B by A.Qj}q. Thus whenever q and p are multiplied.pt to fourvectors (in a (1 + 3)dimensional space) define relativistic Poisson brackets. 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 „ „•.p v ^ (9A dB ^  _ _ *? V dqj dpj dA 8B —— dpj dqj 8B_dI\\ dPk dpj J _ 8A_f^B_dQk dpj \ 8Qk dqj dB 8Pk dPk dqj = E E k dAL/'^B_dQk dqj \ dQk dpj {A.p) .3: Relativistic Poisson brackets By extending qi. Solution: Relativistically we have to treat space and time on an equal footing. (2.Pj}q. 1 „ H = p + m0gq and solve the canonical equations with Poisson brackets for initial conditions q(0) = qo. V V °Qk dPk dPk dQk J E x a m p l e 2.A}q. dA {Qk. 236.E(t).B}QIP. The relativistic Poisson bracket (subscript r) therefore becomes du dF dq dp du dF dp dq du &F _ du dF ' {u. {A.P^pr.28 CHAPTER 2. § E x a m p l e 2. we obtain as claimed by Eq. OQk Pk}q. p. we have —Et. § See P.pc).B}q. we obtain dA {Qk.p(0) = poSolution: The solution can be looked up in the literature. their product Et — qp being relativistically invariant. Thus we extend q and p to spacetime vectors (t.s F 5 7 r ={A.p = J2 d~Fk Replacing in the above A by P and B by A. we obtain analogously dA dQk' Inserting both of these results into the first equation.15b) dA {Qk.
and E as a function of t). it is the equations of motion which lead from Go(q.t). (1.y. Since Hamilton's equations give a continuous map of one domain onto another.po is a point on Go. where dr is the difference of proper time given by dudH 1 du dE(t) du — = 1 du .17a). dE .p) to G\{q.t).1 Liouville Equation. Then . so that Go(qo. p= p(qo. But now we consider a system whose phase space coordinates are not known precisely. if qo. since H is expressed as a function of q and p. but partial derivatives of F do not vanish. i. This probability is evidently proportional to AqAp.14) and area A of the phase space ellipse given by Eq. of course..3 Liouville Equation.E(t)}r = Hence at Relativistically we really should have clu/dr..dt y/l  qZ/c2' 2. Let us assume that at some initial time to the system may be found in a domain Go(q. in the case of the linear oscillator with energy E given by Eq. (1.po.p).' ±=j1V c \dtj' u^.. rrr . I . and at time t > to in a domain G\(q.t0. Probabilities 29 (This is. g oc AqAp. . to . also boundary points of one domain are mapped into boundary domains of the other.Pi are the coordinates of some mass point m.3 2. (2.po) = Gi(q. We consider first the a priori weighting or a priori probability. one obtains Gi with q = q((lo. We distinguish in the following between two kinds of probabilities. du ..to. c dt 2 2 w^ du du dr .e.2.Po.. we have A = (p apaq — J .p).)•'=(*)**£. dudH s l u .p) . g._.3. Of course.e. and the space spanned by the entire set of canonical coordinates is described as its phase space. i. Instead we assume a case in which we know only that the system is located in some particular domain of phase space. Probabilities Single particle consideration We continue to consider classical mechanics in which the canonical coordinates qi.p). H(q. (*.16) For example.p) around some point qo. 2TTE UJ and hence g oc — . du q\ P = —. 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.Po. numerically zero. .
15) — an ellipse of area § dpgdp. 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. Je=0 .30 CHAPTER 2.p^.14) t o (1.5: A priori weighting of a molecule If the rotational energy of a diatomic molecule with moment of inertia / is 1 / 2 .. Thus g must be independent of t. Hamiltonian Mechanics If g depended on time t it would be dynamical and would involve known information about the particle. as is demonstrated by Liouville's theorem in Example 2. dt dq and with with Hamilton's equations (2.5 thereafter provides an illustration of the a priori weighting expressed in terms of energy E./.1): d . the (pg. — dp A A — In(AqAp)s = — + — = Example 2. this has the same value at a time to. dq Aq dq 2 dq Aq dq 2 qH to the right and q to the left. ? « + •sin 2 6>/' 21E 2IEsin2e' in spherical polar coordinates.— = — Aq. Pi 2 / \. = 2TrIEsm6. Hence from the difference: — . dpdq and similarly dt = — Ap. J0=0 Hence g oc 8n2IdE.)curve for constant E and $ is — as may be seen by comparison with Eqs. as at a time t'0 ^ to • Solution: We consider dln(A9AP)rf(A<Z) ' . Show that the total volume of phase space covered for constant E is 8n2IE. (1. Example 2. Solution: Integrating over the angles we have =2TT fe=7r 2frEsin. Example 2. in view of this independence it can be expressed in terms of the conserved energy E. dt dq dq dp dp d2H dqdp d2H = 0. and hence g oc %ir2IdE.4.4: Liouville's theorem Show that A q A p is independent of time i.eded<t> = 8TT2IE. which means.
20) .e. Thus dn is that number of systems which at time t are contained in the domain q.t)dqdp= 1. (2.*).3 Liouville Equation. i.19) Thus W is the probability to find the system at time t at q. d d 1 P = JJdqidpi.1 The system moving from domain Go to domain G\.p. whose positions in phase space are characterized by points. 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. 2. (2.2. The total number of systems N is obtained by integrating over the whole of phase space.Thus we assume a large number of identical sytems.p.e.Po.t)dqdp = N.3.2 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. Probabilities 2. leaving the basis of classical mechanics! Hence we set / «w)^ =. W=p(q. dn — p(q. it is suggestive to introduce a factor 2KK with every pair dpdq without.p + dp.18) With a suitable normalization we can write this / W(q. by jr = P(9»P.p.q + dq.t). (2.p. (2. i. Since W has the dimension of a reciprocal action.p.17) F ^GT^^ 0 Fig. however.
e.p) denote the velocities in directions q and p — p{q. 2. — if vq(q.e. p(q. 2.t)dp{ jt Q. i. The number of points at time t + dt in domain G. that in our consideration no additional points are created or destroyed. 2.3 and establish an equation for the change of the number of points or systems in G in the time interval dt. In doing this. p + dp G P O q qndq Fig. t)dqdp p(q.a t dt + q+dq.p.3 The region G. i.p.P dt p(q + dq. The equation of motion for n or W is the socalled Liouville equation. * q Fig. i. t + dt)dqdp — p(q.p) and vp(q.e.p / . In order to derive this equation. Hamiltonian Mechanics We can consider 2irh as a unit of area in (here the (1 + l)dimensional) phase space. we take into account. is equal to the number in G at time t plus the number that went into G in the time interval dt minus the number that left G in the time interval dt.32 CHAPTER 2. t + dt)dqdp.p.t)dp( — I . We are now interested in how n ov W changes in time.2 The ensemble in phase space.p.p. we consider the domain G in Fig. how the system moves about in phase space.
p(q + dq. With Eqs.p) = q = g^.P.21) we can also write dW(q.p.21 ) This is the Liouville equation which describes the motion of the ensemble or.p + dp.3 Liouville Equation.t)vq(q + dq.p(q.p + dp.t)vq(q.t + dt) .t)vp(q.p) .P)]K(Q.p)dtdp +p(q.p. (2. Probabilities and thus p(q.p(q.P *' % = iH>P} ( 2 .t) dt {H(q. . so that dt ~~dq\P Hence dp) + dp\P ~dq)~ ~~dq~~dp~ + dp~~dq~ ~ { . p.t)vp(q.p.22) The generalization to n degrees of freedom is evident: The volume element of phase space is . = K(9.p. . t)dqdp = p(q.p(q + dq.p) dq p(q.p + dp) dp = or  However.t)vp(q.P)].p)dtdq . the probable motion of the system under consideration.p. (2. put differently.t)^ = 1. p. t)vq(q + dq. .2.t)} with JW(q.W(q.t)vp(q.19).p) .p + dp)dtdq. dH Mq. (2. .p.p.t)vq{q.p)dtdp .p)=p = 3H —.4) shows that p and u satisfy very similar equations.t) dt p(q.p).20) and (2. Comparison of Eq.p.p(q.p. (2. vp(q. 33 Dividing both sides by dqdpdt this becomes p(q.21) with Eq. (2.p.t + dt)dqdp — p(q.
24) since the total derivative is made up of precisely the partial derivatives contained in Eq. With Eq.p. that dtj\y v if qo. i. 2.4 Expectation Values of Observables Let u = u(q.34 CHAPTER 2. We define as expectation value of u(q.1 the area Go is equal to the area G\. Example 2. We deduce from the Liouville equation the important consequence that ^ M = 0. p. for i<«>!/«<*p>"w>(^)". (2. Equation (2.p (cf.e.p. and hence that equal phase space volumes contain the same number of systems. q+dq. Thus in Fig. (2.that the density or probability W(q. Hamiltonian Mechanics where is the probability for the system to be at time t in the volume q. 2.24) implies that p is a constant in time.e. The first and most immediate possibility is — as indicated . i.p).t)(^J. (2.4).24).po are the initial values of q.p) be an observable. We now inquire about the time variation of the expectation value (it) of u.p)W(q. (. We shall see that we have two possibilities for this. t) depends explicitly on time t (if determined at a fixed . since no systems are created or destroyed. and this means — since these systems are contained in a finite part V of phase space — that dt We have in particular.26) we described the time variation of the observable u(q. p+dp. (2.4).p) the following expression: (u)=Ju(q.
^ Solving the equations of motion for q.p){H(q.t).p) assumes certain values) that depends explicitly on time. W is the density in the neighbourhood of a given point in phase space and has an implicit dependence on time t.t).0) =u(g(qo.29).8.t). (2.po. (2.po. (2. .28) = Ju(q.i).29) The distribution of the canonical variables is given by W(q.2. However.p.Po.22). at t = 0.f(qo.p) = uo{qo. We expect.0) = u0{qo. (2.Po. Sec. since W oc p is constant in time according to Eq.t). that (u) = (u)0. (2.Po.t) = = W(g(q0.34) S e e also H.p.t).W(q.e.t) W(q0. Then Eq. (2. (2.27) becomes !<»>  /«(*P>!"W>(^)" (2.p.po. (2. We verify this claim as follows. ie. Reversing Eq.24): W(q.po.31) i.32) In this expression the time t is contained explicitly in the observable u(q.p.t)W0(q0. we can express these in terms of their initial values qo.33) po = f(q. With these expressions we obtain for the expectation value (U)Q: (u)o = Ju0(qo.p0.30) p = f(qo.t).4 Expectation Values. of course.p.Po) at time t = 0. where we used Eq. Goldstein [114]. and the time variation d(u)/dt is attributed to the fact that it is this probability (that u(q.0) = W0(q0.p.Po.t). and hence u(q.t)}(^y.t). 8. so that Q = g(qo. we have Qo = g(q. (2. Observables 35 point in phase space).p.p0)(^^J. we can also employ a more complicated consideration.po.f(q0.p) = u(q. Thus we can write.p).Po. 1 (2.t).t).
[ du0{q0.31)) W0(q0.t). (2.po. (2. (2.P.Po.t) dB__ (du\ °P \°PS P=f(qo.po.30) Moreover.e.p.32).t).36 CHAPTER 2.t). (2.t).f({H(q.37) as had to be shown.M 27Th i .) Wo(qo.p.p.u{q.o.t) d t _ (du\ \ d<i dq / P=f(qo.t).32). Eq.Po.36) and (2.t).p0) = W(q. Inserting Eqs. (2. (240) .p.(dq0dp0\n We deal with the partial derivative with the help of Eq.(2.Po.p.t). we obtain an expression which is different from that in Eq.P.p).t).t) . (cf.P)}).t).=9(.f(g(q.t).35).t).28).p.t). (2.p.po)(~ .t) = (2 30) (2.35) Inserting these expressions into no we obtain uo(qo.t) (2.36) = (2." V " " ™0' > • • » « » 0 .p)W{q.Po)(^^)n.p)})q=g(QO.0)=u(q.t) dp M du\ ° q 'P=f(qo.t)\—^\ = (u).f(q.p. . Eq.\„.p.t)J(q. (2.t) v{g(<j(q.P0.39) Substituting this into Eq.t).f(q.30): duo(qQ.POit)i p=f(qo.PO.f(q. (2.37) and (2. (2.u(q.t)W "(q0.p.p). i. / ^fdq0dpc .0) u(q.P0.t) M f 9u\ V °P / P=f(q0.f(q.29) q = g(g(q. d dt {U)0 = dt.25) into Eq.t).t) 0H_ 0 q =  {{H(q.p.po. Taking now the total time derivative of (it)o. (2. p = f(<j(q.p).) f / u (q0.30) u0(g(q. (2.38) we obtain £<«)„ = .Po. Hamiltonian Mechanics so that on the other hand with Eq. we obtain («)o = / u{q.
.. lim ——uW — 0. and the time variation of (u(q.W(q. the observable u is treated as a function u(q.u(q.W} in Eq. from the properties of the Poisson bracket.2.p.44) iji^ioo dpi (which is reasonable since the density vanishes at infinity). dt <«)o v f u(q. (2.t)}.p). we obtain {H.p. W}.(2.W(q.p.45) in agreement with Eq.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. u}W + u{H.t) of u assuming certain values q and p.42) the relation d .41) This expression contains {H. ^ „.p). (2. (2.p). described as "Schrodinger picture".p). 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. Alternatively we could deduce from Eqs.p..4 Expectation Values.28) and (2.p){H(q..45) the Liouville equation. However.43) ——uW = 0. u}W instead of u{H. Observables 37 Here we perform the transformation (2.t)} .28). we obtain zero after partial integration of I and hence from Eqs.t) dt {H(q. x r „.p.41) and (2. uW} = {H. so that ~d~t <«>0 = J{H(q.t) dqdp\n 2irhJ (2.25) and (2. In the first case.34) and use (2.31). / d q d p \ n 2irh J (2. This timedependence is described by the Liouville equation dW(q.p)) is attributed to the probability W(q.28). . (2.42) wxtf)" ~dHd(uW) dqi dpi vanishes under certain conditions. Consider {H. The phasespace integral of a Poisson bracket like I (2.p)}W(q.
4)) ^ = {«(. the probability of the initial values Wo(qo. . The time dependence of the expectation values can be dealt with in two different ways.p)}.po are constant initial values — we have du0(q0.t).38 CHAPTER 2. " T h e author learned this approach from lectures of H.p0.t) dt __ ~ (2J30) ( 4) du0(qo.e.H(q. the reason being that — since qo.5 Extension beyond Classical Mechanics With the above considerations we achieved a general formulation of classical mechanics.F(«. " This means: Only in the Heisenberg picture dW/dt = 0.po) is assumed.p) It = {u(q.rt. i.p)}. for which the axioms of a commutative group apply. (2.p). and probabilities W. 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.** 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. but does satisfy the usual associative and distributive laws. 2. as we observed. (2.po. as the equation of motion of an ensemble or of a probability distribution on the one hand. does not have to be commutative. Hamiltonian Mechanics In the other case.46) We thus also recognize the connection between the Liouville equation. and the equation of motion (2. (c) A multiplication of the quantities among themselves is defined. Koppe [152] at the university of Munich around 1964. (b) As usual a muliplication by a complex number is defined. which are described as "Schrodinger picture" and "Heisenberg picture"— all this on a purely classical level but with the use of canonical transformations..t) dt du(q.4) of an observable on the other. Eq. which." and the explicit timedependence is transferred into the correspondingly transformed observables Uo(qo. The equation of motion is then that of an observable. (cf. This formulation deals with observables u representing physical quantities. called u Heisenberg picture".po.
Thus we can write the expectation value of an observable u (cf. (b) TV(cm + j3u) = oTV u + /3Tr v.q}=0.6) a socalled "commutator''' [A. for a phasespace integral we define the word or symbol "trace". (2.) = TV (vu).p] = l.5 Extension beyond Classical Mechanics 39 Quantities satisfying these properties define a linear algebra. ~[q. In Chapter 9 we attempt a corresponding approach for classical systems with . which all apply to (2. px > in—. (2.p]=0(247) One verifies readily that the commutator relations are satisfied by the differential operator representations qx > x. by Traceu:=Ju(q. (c) TV (u*u) > 0. y:\p. so that we consider this next. (d) Tr(ut. therefore. B] := AB — BA in the following way: 'i {q. That this is the case.p} = 0 • — ^ ~[q. also called "density matrix".2. ox In view of our later considerations. Moreover.q} = 0 {p.p)(^y. Introducing it here assumes. The quantity corresponding to W in quantum mechanics is the socalled "statistical operator'''. Eq.49) With these considerations we have reviewed aspects of classical particle mechanics in as close an approach to quantum mechanics as seems possible.48): (a) TV u* = (Tr «)* = TV^.26)) (u) = Tr(wW).p} = l {q.48) In matrix theory the symbol "trace" has a welldefined meaning. if u ^ 0. (2. also written "TV". can be deduced from the following characteristic properties of a trace. that its use here implies the essential properties it has in matrix theory. We shall then interpret as "canonical quantization" the procedure which allocates to each of the fundamental Poisson brackets (2. it is helpful to introduce already at this stage some additional terminology. and to be able to correlate these with the above classical considerations.
. i. electrodynamics. a constraint) has to be taken into account. and.e. Thus we can now proceed to prepare the ground for the extension of classical mechanics into an operator formulation. These aspects will be considered in Chapter 27. the Poisson brackets require modification to Dirac brackets. However. since gauge fixing (i. excepting the Poisson brackets.40 CHAPTER 2. Hamiltonian Mechanics a wavelike nature. obtain corresponding results — as one would envisage in view of the expected particlewave duality in quantum mechanics. it will be shown that in electrodynamics.e.
of measurable quantities.e. 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. A set M. which turn out to be those of the theory today known as quantum mechanics.2 Hilbert Spaces We first recapitulate some fundamental concepts of linear algebra and begin with the axioms defining a linear vector space.Chapter 3 Mathematical Foundations of Q u a n t u m Mechanics 3. i. = {ipi} is called a linear vector space on the set of numbers I 6 {C}. We found that the Poisson algebra permits extensions to noncnumber formulations. These somewhat abstract considerations — although later in many cases not referred back to — are a necessary prerequisite for the formulation of a mechanics which is not of the cnumber type as classical mechanics. quantum mechanics: The Hilbert space as the space of state vectors representing the states of a physical system. if the elements ipi of M satisfy the usual axioms of addition and K 41 . with the canonical commutation relations or Heisenberg algebra defining the basic product relations.1 Introductory Remarks In Chapter 2 we investigated the algebraic structure of classical Hamiltonian mechanics. i. In this chapter we therefore introduce important basic mathematical concepts of this noncnumber mechanics. we can define the Hilbert space as the space of states of a physical system. 3. Building upon this. and selfadjoint operators in this space as representatives of observables.e.
^) called inner product. ip2 of this space can be associated with a complex number (V'l.4) The vector space . . A + tpj = tpj + i'i.~ip2) '• M. E (3.42 CHAPTER 3. i=l (3. . i = 1. (ipi + ipj) + ijjk = ipi + (ipj + ipk). (a.5a) (3. In each case n linearly independent vectors are said to form a basis. .^i) . n. i.1) aipi + aipj. . ../?eK).M is said to be a metric vector space or a preHilbert space. • • • . and n is the smallest such number.i = 1. . (ipi.5b) . 2 .2) Vectors ip\. if numbers eti. are linearly dependent.ip2).aiV>i + a 2 ^ 2 ) = ai(ip. (aptyi. n exist (not all zero). —»• IK. . (V>. the vectors il)i.e.ip2.tp2. (3.  .ipn are said to be linearly dependent. .M. x M. (3. such that n ^ = YJCi^i. If n + 1 elements ^ € M. if any two elements ip\. . 2 . n is called the dimension of .02)* (hermiticity). so that n 5 > ^ = 0. (0 : null element and with complex numbers a and j3: a(tpi + tpj) = a{Wi) lipi = = < M). if every vector ip E M can be associated with numbers Q .^1) = (01.ipn are said to be linearly independent. (3 5c) + a2{ip. fa. ( n (3. with the properties (a* G IK): (^2. . f > 0 if V T ^ O .3) If all on = 0. Mathematical Foundations of Quantum Mechanics multiplication by complex numbers. ^ i = 0 if ^ = 0.
The distance between two vectors if>i. ^ i ) + A * ( ^ .is defined by d(ipi. tp2 £ M: 1(^1.)> (36) i. The first two properties imply ( a i ^ i + a2ip2..2.. »2 2 . ^ 2 )  . (3.9a) we start from if) = ipi + \if>2 6 M. linearity in the second component.i>2\\2 = 2'i/'i2 + 2'i/'22 (parallelogram equation).9c) HV'i + tp2\\2 + \\1p1 .9d)   V i   = sup  ( ^ i . ^ ) + A 2 (V2. In order to verify Eq. antilinearity in the first component (also described as sesquilinearity.7) In addition the following relations hold in a metric space M.9b) = 0 (Pythagoras theorem).9a) (3.rh) (3. for arbitrary A and ip2 ^ 0: (V>i + AV>2. meaning oneandahalffold linearity). (3.tp2 £ M.V') +"2*(V. WA+Ml2 = ll^il 2 + ll^2 2 .2 Hilbert Spaces 43 where the asterix * means complex conjugation.^) = ai*(V'i.V.V>i + A</>2) > 0 . The norm of the vector ip (preHilbert space norm) is defined as H:=(^)1/2. ^ i ) + A ( ^ i .^ 2 ) =   ^   2 + 2 ^ i . \\1p1 + ^211 < HV'ill + IIV^II (triangle inequality).*h)=\\AH\(38) (3.^2)1 < H^ill • 11^211 (Schwarz inequality). A V 2 ) + A2V22 2 \m\\ llwl For if)2 7^ 0 we set A so that (^2^l)2 2 IWI + ll^i. (310) (^2. for ip\.3.e. (3.^1) 2 h 11 0 <%«e + M . (3. if Wi.9e) 11^11=1 We restrict ourselves here to some remarks on the verification of these wellknown relations. which we can write 0 < ( ^ i .
(3.* Two vectors i f i . Examples of metric vector spaces: (1) Let M.44 or CHAPTER 3. ^ e M are said to be orthogonal if (</>!. a £ K : ll^ill HVill 11^1 +^211 M If for a vector tp e M: > = < = 0. for which CO 2 := V^l^l 2 < 00. i=l * Not all the wave functions we consider in the following and in later chapters are automatically normalized to 1.  ^ l l = (HlMI +   « 2 .\\ih\\. HIHI (3.9c). (3. hence verification in each case is necessary. We also omit the verification of the following properties of the norm of a vector tpi € M with ^ € X . V 2 ) = 0 . In verifying the triangle inequality we use this result (3. (3.11). Mathematical Foundations of Quantum Mechanics 1(^2.ii) thus verifying Eq.9d). so that   ^ l + ^ 2   <   ^ l   + ^2.10): ll^+^H2 = < <   ^  2 + V22 + 2K(Vl. (3.^1)1 Vi 2 + ^2 2 + 2   ^   . (3.9e). beginning with A = 1 in the second line of Eq.12) We omit here the verification of the remaining relations (3. (3.^)1 <IIV>il.9a). o ^ V i = o.V2) IIV'l2 + V'22 + 21(^2. be the set of all column vectors v\ V = (Vi) = I 2 V with complex numbers Vi. ll^ill + llVdl.13) the vector is said to be normalized. .
= 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. etc. are combined to an equivalence class [/] (with space L2).e. But this applies also in the case of any function which is nonzero only on a set of measure zero.3.w) :=J2v*Wi. is defined by ([/]. 2 and so on. and one defines addition and multiplication by complex numbers with respect to these classes.e. with inner product oo 45 (v. / ) = 0. (3. all squareintegrable functions / which are "almost everywhere equal".5c). i. In order to avoid this difficulty.2 Hilbert Spaces Then we define v + w := (v{) + (wi) := (vi + Wi). The Schwarz inequality is then oo oo \M\<* ^2\Vi\ ^2\Wj\2. (3. which differ solely on a set of measure zero. Js the space C? is not yet a metric vector space although for (cf.5b). (3. i.[$]):= / JM /(x)* 5 (x)d 3 *. Elements of the classes are then called representatives of these classes. (2) Let M. which satisfies relations (3. With the scalar product (f.g)= [ /(x)^(x)d3x.5a). . •'^ \ 0 otherwise. Then L2 is the space of all these equivalence classes. */ x } = / /o for x = 0. Eq. for which the scalar product.5c)) / ( x ) = 0 = > ( / .
46 and
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
ll[/]=0=»[/] = [0],
where [0] is defined as the class of all functions which are almost everywhere zero. This means that functions that differ only on a pointset of Lebesgue measure zero are looked at as identical. Unless necessary, we ignore in the following mostly for simplicity the distinction between C2 and L2. Convergence of sequences in Hilbert space is then called convergence "almost everywhere". With the help of the concept of a norm we can introduce the concepts of convergence and point of accumulation. Definition: A sequence {tpn} € M. is said to converge (strongly) towards tp E M., if the distance \\tp — ipn\\ tends towards zero, i.e. lim   ^  V n   = 0.
n—•oo
(3.14)
The vector ip is then called point of accumulation. The point of accumulation does not have to be an element of A4. If M. contains all of its points of accumulation, the set M is said to be closed. A normalized vector space M. which with every convergent sequence contains a vector towards which the sequence converges, is said to be complete, i.e. if ipn e M. with lim \\ipn  i/jm\\ = 0
m,n—>oo
(called Cauchy sequence), there is a ip 6 M. with ip = lim ipn,
n—>oo
i.e.
lim \\ip — ipn\\ = 0.
n—>oo
(3.15)
Every finitedimensional vector space (on IK) is complete in the sense of the concept of convergence defined above (so that completeness does not have to be demanded separately). In order to see this, we consider the convergent sequence
n
<Pa = ^2CmiPi,
where i/ji,...,ipn€.A>i ras)
CaieK,
(3.16)
constitute a basis in M Then (according to Pythagon
nv«  M = i sec™  c0i)A\\ =YI i c  c^2'
2
2
n
(3i?)
3.2 Hilbert Spaces
47
a relation also known as Parseval equation. The convergence of the sequence i/ja implies the convergence of the sequence {Cai} towards a number Cj. Then for the vector
i=l
we have
ll^aV'II^ElC^Cil2,
i=i
(3.18)
i.e. that the sequence of the vectors tpa converges towards tp. We thus arrive at the definition of a Hilbert space. Definition: An infinitely dimensional, metric vector space, which is also complete with regard to (strong) convergence, is called a Hilbert space "K. The given definition of a Hilbert space is that usually given in mathematics.t In physics this is generally supplemented by the requirement that the space be separable, i.e. of a countably infinite dimensionality. Naturally Hilbert spaces with a countable basis are the simplest. We supplement the above by referring to the concept of a dense set or subset M of "K. A subset M of "K is said to be dense in "K, if to every / 6 "K there exists a sequence of vectors fn, fn < M, so that fn — f, S > i.e. fn converges strongly to / , implying that every vector / e Ji can be approximated arbitrarily precisely. We consider next some examples. Examples of Hilbert spaces: (1) The hyperspherical functions Yitm(6,(p) define a complete set of basis functions on the unit sphere. Any function f(9,(p) with
[\f(e,tp)\2dn<™
can be written as a convergent series
oo 1=0 /
f{9, ip) = Y, E
m=l
C
l,mYl,m(0, ip).
(3.19)
For completeness we recall here the definition
(o<p)twi r(2*+iwm)!i x v„ sin ™ 0 (±y +m (cos2e _ lV Yl
*i,mW<P)2in
[
47r(Z + m ) !
J
e
sm
\
d 9
)
^cos V
^'
See e.g. N. I. Achieser and L. M. Glasman [3].
48 so that
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
Y0fl = —=,
J Air
1
Yito = \ —cosO,
V 47T
/ 3
Yit±i =
I o
^\—e±tvswt
•
A Hilbert space contains a complete set of orthonormal basis vectors or a corresponding sequence precisely then if it is separable. (2) On the space L2(0,2n), i.e. on the space of squareintegrable functions on the interval [0, 2TT], an orthonormal system, called the trigonometric system, is defined by the functions
1
einx,
±n = 0 , 1 , 2 , 3 , . . . .
(3.20)
/2TT
(3) On the space L2(a,b), where (a, b) is an arbitrary but finite interval, a complete but not orthonormal system is given by
In order to obtain the orthonormalized system, one employs the orthogonalization procedure of E. Schmidt.* The sequence of polynomials thus obtained consists of the Legendre polynomials which are defined on the interval — 1 < x < 1. These polynomials are defined as follows: Po(*) = l, *>„(*) = _ 1 dn —(x2ir, n = l,2,.... (3.21)
These polynomials satisfy the following normalization conditions, i.e. are orthogonal but are not normalized to 1:
l
/_
Pm(x)Pn{x)dx
l
= 0, (m ^ n),
(3.22a)
and
J' [Pn{x)\2dx = 1 ^
l
.
(3.22b)
(4) By orthogonalization of the following functions
x2/2
xe~x 2
/2
x2px
2
/2
one obtains the following functions defined on the space L2{—oo,oo): ^[x) = (  l ) n e * 2 / 2 f ^ e  * 2 = Hn{x)ex2'2 dxn (3.23a)
''This procedure is wellknown in analysis, and hence will not be elaborated on here.
3.3 Operators in Hilbert Space with
/*C5tJ
49
j —(
bn(x)<t>m(x) = 2nn\yft6nm,
(3.23b)
where Hn (x) is the Hermite polynomial of the nth degree (which we do not go into in more detail at this stage).§
3.3
Operators in Hilbert Space
Having defined the admissible states of a physical system as the vectors which span a Hilbert space, the next step is to introduce quantities representing operations in this space. This is our objective in this section. We begin with a number of definitions. Definition: Let T>A, the domain of definition, be a (dense) subspace of the Hilbert space "K. Then one defines as a linear operator A on "K the mapping A : VA > H with (a,/3 G C, ^ , ^ e ^ c M ) A(cnl)i + 0rfo) = a(A^i) + /3(At/>2). (3.24b) (3.24a)
Definition: One defines as norm (i.e. operator norm) of the operator A the quantity sup l i M . (3.25) ipevA\{o} WW Definition: An operator A is said to be bounded, if its norm is finite, i.e. \\A\\ < oo. Definition: Two operators A : T>A — "K and B : T>B —>"Kare said to be > equal if and only if Atp = Btp, for every ip G VA and Z>A = T>BExample: An example of a linear operator is given by the differential operator D:= —
d
A:=
_.:.LU ^ f./. _ T1 ^ rwith VD = iil>eL*,jeL*>.
T2
(3.26)
Definition: We define the operations of addition and multiplication of operators by the relations (A + B)tp:=Aip
s
+ Bip, \/i/j<EVA+B = VAnVB,
(3.27a)
Hermite polynomials are dealt with in detail in Chapter 6.
50
CHAPTER 3. Mathematical Foundations of Quantum Mechanics (AB)ip := A(Bip),
V^GPB,
B^eVA.
(3.27b)
Definition: We define operators called commutators as follows: Let A : T>A —> "K, and B : T>B — "K be linear operators in the Hilbert space "K. Then we > define as commutator the expression [A, B] := AB  BA with V[A,B\ = D A n £>B » ft. (3.28)
Definition: If .A : X^ — IK for i/> 6 T>A\{0}, then ^ is called eigenvector with • respect to the eigenvalue A € C if and only if AV> = \i/>. (3.29)
Very important for our purposes are the concepts of adjoint and selfadjoint operators. Definition: Let A : T>A — IK and ip € P A C IK. Then A^ is called adjoint » operator of A if for A^4> := <jj, <f> € T>Aj, the following relation holds: {A* 4,il>) = (<!>, Ax/,). (3.30)
Definition: The operator A : X>A — ft, is said to be symmetric if and only if >
0 M ^ ) = (MV), V ^ i e P
A
c%
(3.31)
Definition: The operator A, A : X^ — IK, is said to be hermitian or selfad> joint if and only if A = Af, One can verify that: (A*)* = A, (A + S j ^ A t + (AA)+ = A*A , (AB)t = S U * , (A" 1 )* = (A*)" 1 .
f
i.e. D A = D A t
and A ^ = A</>.
(3.32)
(3.33a) fit, (3.33b) (3.33c) (3.33d) (3.33e)
We are now in a position to construct relations between operators A and B in K, which correspond to the relations (2.5a) to (2.5e) of Poisson brackets, however, we omit their verification here: [A,B] = [B,A], (3.34a)
3.3 Operators in Hilbert Space [A, aiBi + a2B2] = ax[A, B{\ + a2[A, B2], [A,B]* = [A\B\ [A,BC] = [A,B}C + B[A,C], [A, [5, C]] + [B, [C, A}} + [C, [A, B}} = 0.
51 (3.34b) (3.34c) (3.34d) (3.34e)
The last relation is again called a Jacobi identity. Comparison of Eq. (3.34c) with Eq. (2.5c) requires here in the definition of the corresponding quantity the introduction of a factor "i" (see Eq. (2.47)). As important examples we consider the following operators. (1) q3 : Vqj  L 2 (R 3 ). We write sometimes the application of the operator qj to <f> G L2(IR3): (qj4>){x), and we define M)(x) :=arj0(x). (3.35) We can read this equation as an eigenvalue equation: Operator qj applied to the vector <f> yields the eigenvalue Xj multiplied by <> in the present case on /, 3 R . Since Vqj = {<£ G L 2 (R 3 ) : qj* G L 2 (R 3 )}, we have
^ = ^ t 
Furthermore, for instance for ip, <p G L2(IR1), we have (ip,q<p) = / ip*(x)(q(f))(x)dx = / ip* (x)x<p(x)dx
=
(#,«•
(3.36)
Since for the adjoint operator A^ of A:
it follows (with Pg = V t from above) that qj — q3^. (2) P j : VPj > L 2 (R 3 ) defined by (p 3 »(x) := ih^<t>{y). In this case we have VPj = <<t>€ L 2 (R 3 ) : c continuous, ~ G L 2 (R 3 ) I = / > Vpj]. (3.37)
52
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
Here for ip,(f> e L2(Ul): (V,W>) = /V(aO(p0)dx
(3
=7)ifi
f^*(x)^(f>(x)dx
= inr(x)<i>(x)\f00J^r(x)<i>(x)dx
— — I [ ih—i/}(x) \ (f)(x)dx = J (pip)*(x)<f>(x)dx = (p^,0) = (pty,0), (3.38)
so that p* = p. Something similar applies in the case of the following operator which represents classically the kinetic energy T: (3) T:VT^ L 2 (R 3 ) and
™ ( x ) := "£ A<Kx) = ( ^ & 2 V ( x ) '
As a further example we consider the commutator. (4) Let the commutator be the mapping
(339)
Then for <p E L 2 (R 3 ): \Pj,Qk]<l>(x) = i.e. formally The following commutators which define the Heisenberg algebra \Pj, Qk] = ihSjk, \pj ,Pk] = 0, [qj ,qk] = 0 = (pjqk ~ 9fePj)^(x) = (pjqk(t>)(x)  (qkPj(t>)(x)  i ^ f ^ X f c 0 ( x )  xfe—</>(x) J = ih6jk(j)(x.),
are called canonical quantization conditions with respect to a theory whose classical version possesses the fundamental Poisson brackets {Pi, ?*} = Sjk, {Pj,Pk} = 0, {qj,qk} = 0.
The simplest example to consider is the harmonic oscillator. We postpone this till later (Chapter 6). We add, that the quantization must always be
3.4 Linear Functionals and Distributions
53
carried out on Cartesian coordinates. Moreover, the above relations assume that the three degrees of freedom are independent, i.e. there are no constraints linking them. Systems with constraints can be handled, but require a separate treatment.^
3.4
Linear Functionals and Distributions
We now introduce the concept of a continuous linear functional on a socalled test function spaced Our aim is here, to provide a frame in which the formal Dirac bra and ketformalism to be developed later finds its mathematical justification. We require in particular the delta distribution and Fourier transformations. A subset of a Hilbert space "K is called a linear manifold D, if along with any two elements (f>i,4>2 G ^ C 'K this also contains the linear combination of these, i.e. « i ^ i + a2<^2 £ ^ , ai,a2€C. (341) A linear functional [/] in the Hilbert space "K is a mapping of the manifold T> into the set of complex numbers, i.e. [/] : V  C and is written [/]<<£> = / < 0 ) : = / with the property of linearity, i.e. f(<l>i + <h) = f(<l>i) + f{<h), (i.e. the expression (3.42) is antilinear in the first component). If / ( x ) is for instance a locally integrable function (i.e. for a compact set e R n ) like exp(ik • x), then it is clear that the function </>(x) has to decrease sufficiently fast at infinity, so that the expression in Eq. (3.42) exists. Functions which provide this are called test functions (see also later). Instead of the Hilbert space we therefore consider now a space of test functions (vector space of test functions), and on this space linear functionals. Definition: The compact support of a continuous function <fi : Rn — C is > defined to be the compact (i.e. closed and bounded) set of points outside
f
/(x)0(r)dx,
(V<P€V)
(3.42)
S e e Chapter 27. "We follow here to some extent W. Giittinger [127].
54
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
that of (ft = 0. Test functions cf> with compact support are exactly zero outside their support; they define the space D(Rn). A different class of test functions <j> consists of those which together with all of their derivatives \Dn(f>\ fall off at infinity faster than any inverse power of x. These test functions are called "rapidly decreasing test functions" and constitute the space S(Rn): D(Rn) S(Rn) := {(f> G C°°(DRn > C) : support of </> compact}, := {4> £ C°°(Rn > C) : \x\m\Dn(j>\ bounded, m,n,... e I > 0}. N (3.43) Definition: Distributions f{<f>) are defined to be the linear functionals on D(Rn) and tempered distributions the linear functionals on S'(IRn). A subset of distributions can obviously be identified with ordinary functions, which is the reason why distributions are also called "generalized functions".
3.4.1
Interpretation of distributions in physics
It is possible to attribute a physical meaning to a functional of the form
f (</>):= J dxf(x)<j>(x).
(3.44)
In order to perform a measurement at some object, one observes and hence measures the reaction of this object to some tests. If we describe the object by its density distribution f(x), like e.g. mass, and that of its testing object by <t>(x), then the product f(x)(p(x) describes the result of the testing procedure at a point x, since f(x)4>(x) = 0, provided f(x) ^ 0 and <p(x) ^ 0, i.e. if object and testing object do not meet. The expression then describes the result of the testing procedure in the entire space. If we perform the testing procedure with different testing objects and hence with different test functions <pi(x),i = 1,2,..., we obtain as a result for the entire space a set of different numbers which correspond to the individual 4>i{x). These ^dependent numbers are written as in Eq. (3.44):
m = j f{x)<t>{x)dx.
If f(<j>) = 0 for every continuously differentiable function <f>{x), then f{x) — 0. In general one expects that a knowledge of the numbers f{<j>) and the test
3.4 Linear Functionals and Distributions
55
functions <p(x) permits one to characterize the function f(x) itself, provided the set of test functions is complete. In this way one arrives at a new concept of functions: Instead of its values y = f{x) the function / is now determined by its action on all the test functions <f>(x). One refers to this as a functional: The functional / associates a number /(</>) with every test function (f>. Thus the functional is the mapping of a space of functions into the space of numbers. f{4>) is the value of the functional at the "point" <f>. With this concept of a functional we can define quantities which are not functions in the sense of classical analysis. As an example we consider in the following the socalled "delta distribution".
3.4.2
Properties of functionals and the delta distribution
The delta distribution is defined as the functional 5(<j)) which associates with every test function <fi(x) a number, in this case the value of the test function at x = 0, i.e. 6(4>) = </>(0), (3.45a) where according to Eq. (3.44) 6((j)) = f 8(x)(/)(x)dx. (3.45b)
The result of the action of the "delta function" 5(x) on the test function 4>(x) is the number 0(0). The notation J 5{x)(j>(x)dx is to be understood only symbolically. The example of the delta function shows that a function does not have to be given in order to allow the definition of a functional. In order to insure that in the transition from a function f(x) defined in the classical sense to its corresponding functional /(</>) no information about / is lost, i.e. to insure that f(4>) is equivalent to f(x), the class of test functions must be sufficiently large. Thus, if the integral J f(x)4>(x)dx is to exist also for a function f(x) which grows with x beyond all bounds, the test functions must decrease to zero sufficiently fast for large x, exactly how fast depending on the given physical situation. In any case the space of test functions must contain those functions <f>(x) which vanish outside a closed and bounded domain, since these correspond to the possibility to measure mass distributions which are necessarily restricted to a finite domain. Furthermore, for the integral (3.44) to exist, the test functions must also possess a sufficiently regular behaviour. For these reasons one demands continuous differentiability of any arbitrary order as in the case of 5(0?") above. Certain continuity properties of the function f(x) should also be reflected in the associated functional. The reaction of a mass distribution f(x) on a test
56
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
object <p(x) is the weaker, the weaker cp(x) is. Thus it makes sense to demand that if a sequence {(pi(x)} of test functions and the sequences resulting from the latter's derivatives of arbitrary order, (p^(x), converge uniformly towards zero, that then also the sequence of numbers ((pi) converges towards zero. We refrain, however, from entering here into a deeper discussion of convergence properties in the space of test functions. The derivative of a distribution f((p), indicated by a prime, is defined by the following equation f((P):=f(cP>). (3.46) This definition suggests itself for the following reason. If we associate with the function f(x) the distribution
oo
dxf(x)</>(x),
/
oo
then the derivative f'(x) becomes the functional
oo
/
dxf(x)cP(x)
oo
= /($'),
(3.47)
as in Eq. (3.44). Partial integration of this integral then yields, in view of the conditions <p(±oo) = 0,
oo
dxf(x)<P'(x) = /(</>'),
/
oo
as in Eq. (3.47). Equation (3.46) defines the derivative of the functional f(<p) even if there is no function f(x) which defines the functional. For instance in the case of the delta distribution we have 6'(cP) = S((P') = <j>'(0) according to Eq. (3.45a). Formally one writes, of course,
oo
(3.48)
/•oo
/
oo
dx5'(x)<p(x) =
[<J(a;)0(a;)]?foo  /
J—oo
dx5(x)(P'(x) (3.49) (350) (3.51)
= 5(<P>) = <P'(Q). For an infinitely often differentiable function g(x) one has apparently (5 •/)<</>> = / ( # ) , so that (x6)((P) = S(x(P) = [x(P}x=0 = 0 x (P(0) = 0,
3.4 Linear Functional and Distributions or ' x6(x)</>(x)dx = 0,
/ '
57
x5{x) = 0.
(3.52)
Thus formally one can operate with the delta distribution or delta function in much the same way as with a function of classical analysis. As a further example we consider the relation f(x)S{x) = f(0)S(x). According to Eq. (3.50) we have f{x)5{x)4>{x)dx = «J(/0) = [f(x)4>(x)]x=0 = f(0)<f>(0) (3.53)
/
=
f(0)5(<f>) = J
f(0)6(x)<j>{x)dx,
as claimed in Eq. (3.53). Formal differentiation of the relation (3.53) yields f(x)6\x) = f(0)S'(x)  f'(x)S(x). (3.54)
One can convince oneself that this formal relation follows also from the defining equation (3.46). In particular for f(x) = x we obtain the useful relation x5'{x) = S(x). (3.55)
A very important relation for applications is the Heaviside or step function 9{x) which is defined as follows:
From Eq. (3.56) we can deduce the relation 6'{x) = 6{x). (3.57)
For the verification we associate with the step function the following functional
oo
roo
/
oo
0(x)<f>(x)dx = / <f>(x)dx.
JO
/•oo
For the derivative we have according to Eq. (3.46) e'(x)(<p) = 8(x){4f) = dxc/)'(x) = (f)(0)  0(oo) = 0(0)
= 8{x)(<i>),
58
CHAPTER 3. Mathematical Foundations of Quantum Mechanics
or symbolically 0'(x) = 5(x), i.e. Eq. (3.57). After the introduction of the delta function it is customary to consider briefly Fourier transforms, i.e. in the case of one dimension the integral relations f{x) g(k) = =  =
V Alt
I
1
\ /
f°°
dkg(k)eikx dxf(x)eikx
=
F~lg{k), (3.58)
Joo
roo
 =
= Ff(x).
\Z2TT JOO
We assume here some familiarity with these integral relations. It is clear that the existence of these integrals assumes significant restrictions on the functions f(x),g(k). As a formal relation in the sense of the theory of distributions we deduce from Eq. (3.58) the important formal integral representation of the delta function, i.e. the relation
I
/•OO
S(x) = — /
dkeikx = S(x).
(3.59)
One can see this as follows. According to Eq. (3.58) /(0) =   L fdkg(k) yzir J and g(k) =  L y lis J fdxf(x)eikx.
Inserting the second relation into the first, we obtain
/(0) =
Jdxf(x)^jdke^,
and comparison with Eqs. (3.45a) and (3.45b) yields (3.59). We close this topic with a comment. The singular delta distribution was introduced by Dirac in 1930. The rigorous mathematical theory which justifies the formal use of Dirac's delta function was only later developed by mathematicians, in particular L. Schwartz. Thus today the singular delta distribution is written as a regular distribution, i.e. as an integral operator, by writing, for instance, 5a= f <5(x  a)^(x)dx = ^(a) V 0 e S ( [ R n ) , (3.60)
and one derives from this that the delta function <J(x) has the (impossible) properties of being (1) zero everywhere except at x = 0 where it increases so enormously that (2) the integral over <5(x) is unity:
f S(x)dx = 1.
Chapter 4
Dirac's Ket and BraFormalism
4.1 Introductory Remarks
In this chapter we introduce the (initially position or momentum space representationindependent) notation of Dirac* which is of considerable practicability. We also introduce those properties of the notation which make the calculations with states and operators amenable to simple manipulations. The notation will be used extensively in later chapters. The integral representation of the delta function discussed in Chapter 3 can be considered as a formal orthogonality relation for harmonic waves exp(ikx) which finds its rigorous justification in the context of distribution theory. Thus we have — again for simplicity here for the onedimensional case —
1
/"OO
S(x x') or 6(k k')
= — /
27T
i
dkeikxe~ikx'
(4.1a)
J_00
/oo
= — /
27T J_00
dxeikxeik'x.
(4.1b)
Since k and similarly x here assume a continuum of values, Eqs.(4.1a) and (4.1b) are described as normalization conditions of the continuum functions exp(ikx) as distinct from orthogonality conditions for functions depending on integers m,n as, for instance, the trigonometric functions um(x) = cos(mx), vm{x) = sin(mx),
'See P. A. M. Dirac [75].
59
60 for which
CHAPTER 4. Dirac's Ket and BraFormalism
1 fn  / dxum(x)un(x)
^
= 6mn,
(4.2a)
JK
i r
 /
^ J7T
dxum(a;)i;ri(x) = Smn,
(4.2b) (4.2c)
i
 \
^
r
dxum(x)vn(x) = 0.
JK
In the following we shall therefore "orthogonalize" continuum states represented by vectors of a Hilbert space which is no longer separable (i.e. which has at least a subset whose vectors are characterized by a continuous parameter) in the sense of the relation (4.1a) to a delta function instead to a Kronecker delta as in Eq. (4.2a). With this formal use of the delta function we can manipulate continuum states easily in much the same way as discrete states which implies an enormous simplification of numerous calculations, t The Fourier transforms introduced in Chapter 3 permit an additional important observation: We have several possibilities to represent vectors in Hilbert space, since the Fourier transform describes the transformation from one representation to another (" configuration" or "position space representation" <> "momentum space representation"). A position space Schrodinger wave function ?/>(x), which we shall consider in detail in numerous examples later, corresponds to the representation of the vector ip of a vector space (as representative of a state of the system under consideration) in the position space representation, i.e. as a function of coordinates x. In the momentum space represenation the vector tp is the Fouriertransformed V>(k) of ip(x); this representation will be used in particular in Chapter 13.
4.2
Ket and Bra States
In the following we introduce the notation of Dirac.* We define ketvectors as elements of a linear vector space and bravectors as those of an associated dual vector space. The syllables "bra" and "kef are those of the word "brackef. The spaces are linear vector spaces, but not necessarily separable Hilbert spaces, unless we are dealing with an entirely discrete system. More as an excercise than as a matter of necessity we recall in the following some considerations of Sec. 3.1, expressed, however, in the notation of Dirac. Hopefully this partial overlap is instructive.
'Formally this means that we have to go to an extended Hilbert space, which contains also states with infinite norm, see e.g. A. Messiah [195], Vol. I, Sec. 7.1.3. *P. A. M. Dirac [75].
4.2 Ket and Bra States
61
In order to achieve a representationindependent formulation we assign to every state of the system under consideration a vector, called ketvector and designated ), in a vector space V. In the symbol \u), for example, u is an index of the spectrum, i.e. a discrete index in the case of the discrete spectrum, or a continuous index in the case of a continuous spectrum. The linearity of the vector space implies, that if £ is a continuous index, and A(£) an arbitrary complex function, then with £),
\w) = J\(0\Odt
(43)
is also an element of V. As mentioned, the space can be finitely or infinitely dimensional. In the vector space V of ketvectors a set of basis vectors can be defined, so that every vector can be expressed as a linear combination of these basis vectors. With the vector space V of ketvectors we can associate a dual space V of bravectors, which are written (x. The bravector (% is defined by the linear function <X{«» of ketvectors \u). For a certain ketvector \u) the quantity x has the value (xit), which is, in general, a complex number. For a better understanding of the concept of the dual space, we recall first the difference between a linear operator and a linear functional. According to definition, the linear operator which acts on a ketvector \u) G V yields again a ketvector \u') G V. On the other hand, a linear functional x i s a n operation, which assigns every ketvector \u) G V linearly a complex number {x\u}, i.e. X= with x ( A i K ) + A 2 « 2 )) = A i x O i ) ) + A 2 x ( M ) The functionals defined on the ketvectors \u) G V define the vector space V called dual space of V. An element of this vector space, i.e. a functional x is symbolized by the bravector (x.§ Then (x\u) is the number that results by allowing the linear functional (x G V to act on the ketvector \u), i.e. X(\u}) = <xl«>One should compare this with our earlier definition of the functional / on the space of test functions <> We wrote this functional /. «)eV>x(l«»
m = w».
By construction every ketvector is assigned an appropriate bravector. The reverse is not true in general. See e.g. C. CohenTannoudji, B. Diu and F. Laloa [53], p. 112.
that the vector \u) E V is associated with a vector in V. (2) (u\u) > 0.4a) \v) = J\(m<% and (4.5c)).5) 4.(t. 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. or (4. Two ketvectors \u).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). if x\u) = 0 V kets \u) E V.e.\u) = (u\0*. Hermitian Operators A linear operator A in the space of ketvectors V acts such that A\u) = \v) EV . Dirac's Ket. (3) (u\u) = 0. It follows that we can now write the Schwarzian inequality (3. i. Because of the antilinearity the conjugate bravector associated with the ketvector u) = Ail> + A22> is (v\ = Xl(l\ + X*2(2\. The dimension of V is the same as the dimension of V. the norm squared. if \u) = 0 (see (3.62 CHAPTER 4.and BraFormalism Furthermore we have: ( x  = 0. \v) E V are said to be orthogonal. Also: (xi = (X2I) if (xi\u) = (X2\u) for all \u) E V.4b) (4. if (u\v) — 0. This expression is linear with respect to \u) and antilinear with respect to \v). V \u) (in the case of £) with infinite norm!).9a): \{u\v)\2 < (u\u)(v\v). As in our earlier considerations we demand for (v\u) the following properties: (1) (u\v) = (v\u)*. which we write (u\. (4.
. ltt) = \u). the space Vi <S> V2. Let \u)(l> be vectors of the space Vi and similarly ?j)(2) the vectors of the space V2. Then (x(^4it)) is a linear functional of \u) £ V (since A is linear).4. This does not always exist. as we can see as follows.\u) = ((X\A)\u) = (X\(A\u)) = (x\A\u). It follows that (r. This product is commutative. denned as a linear operator on V. if \v) — A\u) i. Furthermore. The inverse of A is written A .e. defines the unit or identity operator. If Vi has the dimension A/i.e. The inverse of a product is (AB)'1 =B1A~1. Then one says: (771 results from the action of A on (x. on bravectors are similar to those on ketvectors. In this way the operations of operators A... 1 and \u) = B\v). The operator A. Hermitian Operators 63 with A = 0 if A\u) — 0 for every vector \u) £ V. B. we have \u^u^) = X1\v^u^) + \2\w^u^) and so on. if AB = 1.or direct product of vector spaces. Multiplication by unity. AB^BA.e. The product vectors l?^ 1 )^ 2 )) span a new vector space. Furthermore linearity applies. The product of such vectors is written: \u^u^) = \u)^\u)^. Let the following bravector be given: (x\ £ V. Then the following rules apply: A + B = B + A. Finally we define the tensor. i. Also A = B it for every vector \u) £ V : (u\A\u) = (u\B\u). and one writes: (V\ = (X\A. let (771 6 V be the bravector defined by this functional.3 Linear Operators. Two operators A and B are said to be inverse to each other.or Kronecker. BA = 1. i. also acts on the dual space V. the space Vi <S> V2 has the dimension . with n)(1) = Aiw)( 1 )+A 2 k) ( 1 ) .
Examples of operator relations: (1) (AB\u)(v\Cy = C^\v)(u\B^Al (4.A^] = 0. Let A^> be an operator in the space Vi and A^ space V2. Dirac's Ket. in the product space.e. [A^. Assuming now that \v) G V and (u\A G V are conjugate vectors as explained above.6) for all \u). and (v\ = (u\A. As a consequence we arrive at the following properties (e. If (v\ = (u\A. AW\U)W = \V)W.7b) (4.7d) .3). we write \v) = A*\u). \t) G V. Since we demanded the scalar products to have the property (t\v) = (v\ty. Every such operator is then also associated with an operator.64 CHAPTER 4. 3. the operator A^ is called hermitian conjugate or adjoint operator of A (cf.and BraFormalism an operator in the Mi A/2. i. since AWAW\UW)\UW) = \vWuW) = AWAW\UW)\UW). AW\uW)\uW) = \vWUW). by replacing in the above the bravector (t by <tflt) (AB)* = B*A\ (cA)* = c*A\ (A + 5) t = A* + B\ (4. here designated with the same symbol. Every operator A^> commutes with every operator A(2\ i. we obtain the conjugate relation <tAt«> = (u\A\t)* (4.e.e. then \v) is a linear function of \u).g. also Sec. i.7C) («)(u)t = \v){u\. i. Next we construct the following scalar products: (t\v) = (t\Ai\u).e.7a) (4. (v\t) = (u\A\t).
and (b) that the eigenvalues with respect to \u) are equal to those with respect to (u\ and vice versa. K] = 0. [H.e. for we had (\u){v\)i = \v){u\. An hermitian operator A has the properties (a) that all its eigenvalues are real. as is also (u\u). Obviously the quantity \a)(a\ is an hermitian operator. The operator H is said to be positive definite. i. Verification: Since A — A^ and A\u) = a\u). i. Every operator A can be written as the sum of two such parts: A=\(A + A*) + \(Atf) anti—hermitian . It follows that a is real. or (u\A\u)* — {u\A^\u) = (u\A\u).e. if H = H' and K = K\ then (HK)* = HK only if tfrf = KH. Analogous considerations apply in the case of bravectors (u. it follows that (u\A\u) = a(u\u). if H — H' and antihermitian. but only if these commutators commute.K}.K]^ = = [HKKH]* KHHK = = K^H^ [H. so that (a)(a)t = a)(a. hermitian The product of two hermitian operators is not necessarily hermitian.4. so that (n^4n) is real.K\. The commutator of two hermitian operators is antihermitian: [H. An important theorem is the following. H^K] The separation of HK into hermitian and antihermitian parts is therefore HK=±{HK + KH)+l[H. . 65 The linear operator H is called hermitian.3 Linear Operators. Hermitian Operators (2) The conjugate bravector of AB\u)(v\C\w) is {w\C^\v){u\B]A]. if H is hermitian and (u\H\u) > 0 for all ketvectors \u).
i. Assuming A\u)=a\u). The operator Ps which projects onto S is defined by the properties: Ps\u) = \us) = Ps\us). we have 0 = (uArt) — (u. then the hermitian operator defined on this space is an observable. {us\us*) = 0.66 CHAPTER 4. Let S be a subspace of the Hilbert space !K. since a ^ b.e. so that (v\u) = 0.9a) . (n\n')=5nnl. The entire set of ketvectors spans the extended Hilbert space. A continuous spectrum can immediately be included by passing to the extended Hilbert space.8) The projection operator has the following important properties: Ps is hermitian. and let S* be its complement. a ^ b. it follows from A\u) — a\u) that (u\A = a(u\ or the other way round.e. since a is real.u'). i. Ps2 = Ps(4. (u\Ps = {us\. i. (u\u') = 5{v . Then every vector \u) £ "K can be written \u) — its) + \us*) with \us) e S. Dirac's Ket.and BraFormalism Moreover. \u8*) E S*.Au) = a(v\u) — b(v\u) — (a — b)(v\u). The above theorem remains unchanged. Next we introduce the very useful concept of projection operators. If these form a complete system. (v\A = b(v\. ( n  z / ) = 0 .) = 0. which includes vectors of infinite norm. i. as we can see as follows. but the continuum vectors are normalized to a delta function. (4.9b) (4. Ps is idempotent. This follows from observing that (u\Ps\v) = (u\vs) = (us\v3) = (us\v).e. Orthogonality: Two eigenvectors of some hermitian operator A belonging to different eigenvalues are orthogonal. Ps\ua.e. so that (u\Ps = (us\.
2).1. P\p) =p\p). Example: The vector \a) normalized to 1 with (a\a) — 1 spans a 1dimensional subspace. Ps is projector onto S. so that (Ps2 ~ Ps)\u) = 0.4. Then (a\u) = (au a ) + (a\ua*) = (a\ua) and so {a\u) — {a\ua) and hence {a\u} = c(a\a) = c.e. iV) is a set of orthonormalized states. where £ is a continuous index. if the set of vectors 1). Obviously N PN = y]ln)(re n=l is the projection operator onto the A^dimensional subspace SN. (1 — P)\u) are orthogonal. (4. i. c = (a\u). The quantity \a)(a\ is called elementary projector. i. Hermitian Operators This property follows from the observation that Ps2\u) = Ps\us) = \us) = Ps\u). . .9c) This property can be seen as follows. p 2 . Let a continuum state be written £). 67 (4. i.9d) Thus the vectors P\u). The projection \ua) of an arbitrary vector \u) onto this subspace is ita) = a)(aw). Very similarly we can construct operators which project onto the subspace spanned by the continuum states. Ps2 = PsThe only eigenvalues of Ps are : 0 and 1.. (1 — Ps) is projector onto S*.p = 0. = (p2p)\p). so that \ua) = (a\u)\a) = \a)(a\u). Then 0 = (P2P)\p) and hence p = 0. Let p be an eigenvalue of the projection operator P.3 Linear Operators. Set \u) = \ua) + \ua*) with (a\ua*) = 0.. i. \ua) = c\a).e. If for these (by construction) the projector P2 onto a subspace S2 is P2= J*I f2\odm..e. as can be seen as follows. \u) arbitrary.e.
one says the degree of degeneracy is r .r\ "An example is provided by the case of the hydrogen atom. . see Eq. ' The projector onto the subspace with eigenvalue Aj can then be written Pi = ^2\ui.1 and 11. In the case of the infinitedimensional Hilbert space with a countable number of orthonormalized basis vectors.e. Let \u\). Coulomb potential.and BraFormalism and hence '6 Numerous properties of the projection operators Pi are selfevident. be a system of basis vectors in this space. In general it is possible. u2). 4. i. the operator P is correspondingly P2\u)= f2\0m\u). in this case the spectrum also has a continuous part. If there are r such vectors.6.114b). Observables are representatives of measurable quantities. 8.g. which we introduced earlier. The eigenvalues Aj then form a discrete sequence with associated eigenvectors \ui) € !K. that one and the same eigenvalue Aj is associated with several eigenfunctions.r)(ui.. (11. However... Examples 8. which are orthogonal to each other (i. (11. and Eq. See e. Degeneracy will be discussed at various points in this text. P = £n)(n. Sec.6.e.68 CHAPTER 4. Let us assume that A is an hermitian operator with a completely discrete spectrum (as for instance in the case of the harmonic oscillator). Dime's Ket. 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.114c). are linearly independent).4 Observables Operators which play a particular role in quantum mechanics are those called observables.
i.11) i. i. or also as subdivision of unity or of the unit operator.10) follows from the fact that i Applying the operator A to the projector (4.r.10) expresses the completeness of the orthonormal system. i.\i)Pi = 0. If Aj ^ Aj'. the projection of this operator is the entire space.r (4.r \ui. the relation (4. and APi = XiPi.e.12) .r\u)\2.r>(ui. The operators Pi are linearly independent.r\ = l. It follows that the norm squared is given by (u\u) = {u\PA\u) = ^2(u\ui.r =^ i. PA = y£pi i = Yt\ui.r)(ui. we obtain i i i i i. Applied to an arbitrary vector \u) G "K Eq. (4.r) (the convergence. then PiPi' = 0. in the case of degeneracy with (ui.e.r'} = Sii>6rr>. (4. i i (4.4 Observables The dimension of this subspace is that of the degree of degeneracy. Together with the orthogonality condition.10) This expression is known as completeness relation or closure relation.r\u) i.r\ui/.r)(ui.e. is always implied). The uniqueness of the expression (4.10) gives the linear combination it) = ^ i.10).r\u). and the eigenvectors \ui. Let us set {A . the operator A is completely determined by specification of the eigenvalues A. 69 i If A is an observable and if the spectrum is purely discrete.r)(ui. which we do not enter into here.4.r \(ui. ^ A i P i = A = ^Aiui.e.
(3. we have the corresponding generalizations.e. (note: {uv\A\uvi) is the matrix representation of A) A\uv) = \(v)\uv). (4. the eigenvector \ui): f(A)\ui) = f{\i)\ui). i. as explained earlier.e. for instance. Let v be the continuous parameter which characterizes the continuum.and BraFormalism This is the expression called Parseval equation which we encountered with Eq.14) . (u„zv) = 8{v . If the spectrum of an observable A consists of discrete as well as continuous parts. 1/2)) PA = 22 \ui)(u*\ + / dv\uv){uv\ = 1.v'). (4. i. now however. One defines as action of f(A) on. Then f(A) = f(A)PA =^/(AOk)^! + / f(\{v))\uv)(uv\dv. Then we denote by \uv) the eigenvector of A whose eigenvalue is \{v). 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 a similar way we can handle functions f(A) of an observable A.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). > rV2 + / dis\(uu\u)f \uj)(ui\Xi + / \(v)\uv){uv\dv. written in Dirac's notation. Dime's Ket.70 CHAPTER 4. The ketvectors \uv) are orthonormalized to a delta function.17) previously. for instance exponential functions.
In the onedimensional case we write {x\x') = S(xx'). and (b) if they possess a uniquely determined system of basis vectors. as a formal orthonormality condition. here presented without proof.17) The Fourier transform provides the transition from one representation and basis to the other.58) in terms of ket. We shall return to this theorem. 4.C.. \k)€Fk.e.15) and (4.B. In the following we consider more generally ketvectors \ip) as representatives of the physical states. Eq.16) represent subdi . with energy E. i.16) Correspondingly we also have a complete set of basis vectors {\k)} of an associated vector space F^.e. if (a) they all commute pairwise. / dx\x){x\ = 1. Therefore we want to reexpress the Fourier transform (3. Extending this we can say: A sequence of observables A. their commutator vanishes.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.59). (4.and bravectors. 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 We began by considering ketvectors \u) in which u is a parameter of the (energy) spectrum. [A. (3.4. x) € Fx. Since both expressions (4. (4. if and only if they possess at least one common system of basis vectors.. for which the completeness relation is f dk\k)(k\=l. i. (4. where the symbol ip is already indicative of the Schrodinger wave function ij. form a complete set of commuting observables.. which is discrete in the case of discrete energies and continuous in the case of scattering states (with continuous energies). First of all we can rewrite the integral representation of the delta function. which all commute with one another. For many practical purposes the use of the Fourier transform is unavoidable.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. Finaly we recapitulate the following theorem from linear algebra: Two observables A and B commute.. again later.
(3. The representation of a state vector \ip) G 'K in position space Fx is the mapping of the vector \ip) into the complex numbers (x\ip).22) All of these expressions can readily be transcribed into cases with a higher dimensional position space.20) The representation of the corresponding bravector (^1 G IK in the position space Fx is correspondingly written (il>\x) = {x\ipy. .72 CHAPTER 4.1a).e. called wave function.21) provides the ketvector \ij)) in the A. (4.Fk.L e " * * ' = {x'\k)\ V27T (4. V 27T (k\x') = . i.15): 5(x . shows that these expressions are the corresponding continuum functions (the continuous parameter k replaces the discrete index n). <a#) = [ dk{x\k)(k\ip). i.e. ${k) := (k\if>).19) Comparison with the orthonormalized system of trigonometric functions (4. (4. Dime's Ket.and BraFormalism visions of the unit operator. (4.18) According to Eq. we can rewrite Eq. this expression has to be identified with — / dkeikxeikx\ i. Rather \x) and \k) serve as basis vectors in the representation spaces Fx. The Fourier representation ij){x) = )= V 27T J Ieikx^{k)dk (4. The vectors \x) and \k) G F are not to be confused with the vectors u) or \ijj) G !K. Obviously we obtain this by inserting a complete system of basis vectors of Fk.x') = (x\x') = (x\t\x') = (x\ f dk\k)(k\x').space representation. which are representatives of the states of our physical system. (x\k) = ^=eikx.2a) etc. il>{x) := (x\i/}) : "K * C. (4.59) or Eq. (4.e.
i.t). We postulate the (timedependent) Schrodinger equation* (with the Hamiltonian as the timedevelopment operator) and obtain the quantum mechanical analogue of the Liouville equation. with the complementary system or *E.Chapter 5 Schrodinger Equation and Liouville Equation 5. 5. 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. Schrodinger [244]. 73 . if we do not take into account the complementary part of the universe. the calculation of which for specific cases is the subject of Chapter 7. 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).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. which is the quantum mechanical analogue of the classical probability density p(q. This will then also clarify the role of p as an operator in the quantum mechanical analogy with the Liouville equation.1 Introductory Remarks In this chapter we remind ourselves of the measuring process briefly alluded to in Chapter 1 and of the necessity to recognize the abstraction of a physical situation that we perform.p. We mentioned earlier (in Sec.2 The Density Matrix We shall establish a matrix p.e.
. i eH ®^' (5J) We assume that the states are orthonormal and in their respective subspaces also complete. (52) (5. \a). the matrix can be diagonalized by a transition to a new set of states ^The hermiticity can be demonstrated as follows. The actual and real state of a system is then a superposition of two types of states. but also of the states of the complementary set of the universe. is then given by (A) := {MA\^) = Yl U\{b\A\a)\i)CaiCbj* a. i. i.. called the density matrix..e. are called "pure states".j = where J2 (J\A\i)5baCatCbj* = X>'l4*>/0« a. as well as 2~] «)(* = 1 in the subspaceof pure states i (i\j) = 5ij.. These are the quantum mechanical states which are also referred to as microstates.^ Since p is hermitian.i. Schrodinger Equation and Liouville Equation the measuring instruments. The states \a) £"K'on the other hand are the corresponding states of the complementary part of the universe. p" ( C C * ) t = CTC* = p. we have (a\b) = 5ab. \a). where i represents collectively all quantum numbers of the state of the system under consideration.b. of an operator that acts only on the states of the system we are interested in. The difficulty we have to face is the impossibility to specify precisely the stationary state of our limited system. The states \i) £ "K.j ij (54) Plj = (i\p\j)^YCiCJ*a (55) Here p is an hermitian matrix.3b) The quantum mechanical expectation value of an observable A. we have to deal with a socalled "mixed state".3a) and N J  a ) ( o  = l in the complementary subspace. that is.b.e.. (a\i)=0 = {j\b).74 CHAPTER 5. which in Dirac's notation we can write as = J]C a »i) a.. a (5. T . 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). where T means transpose: p = CTC*. \i).i..
i. (5..e.7) p = ^2\ii)(i. This ensemble is interpreted as a large set of identical systems.3 with Pi^YtCaiCaj*.g. as we saw earlier. The ensemble can be represented as a set of points in phase space.* We can specify the state of our system. This transition to a new set of basis vectors in which the matrix becomes diagonal.. a complete orthonormal system with properties (5. which today is a fundamental term in statistical mechanics. This projection operator represents a quantum mechanical probability as compared with the numbers ui{ which represent classical probabilities.4) we had the expression (A) = Y/Pij(j\A\i) i. In the following we write simply \i) instead of \i'). (5. in a specific representation. The operator Pi = \i)(i\ projects a state \<p) G "K onto \i). seen macroscopically of Z copies of the system).) = YJc t . In this sense we have p(x'. If the ensemble corresponding to the system under consideration consists of Z elements (i. which are all subjected to the same macroscopic boundary and subsidiary conditions.ji(x / )i*(x). quantum states.2) and (5. P ^ ) = i)(#) = (#)i) (5. or (5. e. a (512) *In his reformulation of the statistical mechanics developed by Boltzmann.\p\i")(i"\ = Y. i i (511) In Eq.e. . Thus the real system is in the pure state \i) with probability uii/ J2i Ui. then (cui/ '^Zi^i)Z is the number of ensemble elements in the pure state i).j(x'i)(ia. in the position state representation or xrepresentation.e. can be achieved with the help of the completeness relation of the vectors \i): \i') = J2\i)(i\i') i.10) and is therefore. (5. but occupy in general different microscopic. Thus the pure states \i) form in the subspace of the space of states which is of interest to us. where x represents collectively the entire set of position coordinates of the (particle) system. Gibbs introduced the concept of ensemble..6) (i'\p\i") = Wi'Si'vi = real.5. i. a projection operator. as we discussed previously.e.9) is diagonal.^'\i'^i'\> (58) since then (f\p\j") = ^Ui'ij'li^ii'lj") v = ^Ui'Si'fSi'j" v = UjiSj/j.3a). (see below) eM.x) := (x'\p\x) = yju.2 The Density Matrix 75 \i') 6 "K.
Schrodinger Equation and Liouville Equation We now define the operator p by the relation (cf.4)) (5.3 = pvv = uv. (5. (5. Then the expectation value of the observable A is (cf.76 CHAPTER 5.13) {A) = ^(j\A\i)(i\p\j) ij = 52{j\Ap\j). Hence Wj/ > 0.) := (mm = (#') W> = l(#')2 > 0. i t (5. On the other hand.18) Substituting this into Eq.18). In particular for A = 1. 3 (5.4). (A.5) above) (i\p\j)=Pij. Thus Tr(j>ii)<i)=l. we have Tr(p) = l. Eq. according to Eq. We recall that the elements of p originate from the coefficients Cai in w = Y^cai\a)\i) = Yl ( E ^ M V ^519) . To this end we set A —>• Ai' := \i')(i'\ (no summation over i'). (5.14) which we can write (A) = Tr(pA).17) Hence the sum of the classical probabilities is 1.16) (5.14) we obtain (Ai') = ^2(j\i')(i'\i)(i\p\j) ij = ^Sji'Su'Pij i. (5. (5. (5.15) 5^0'l5^w«l*X*b') j i = 1 » or J^wi(it> = ^ W i = l. as claimed by Eq. We can also show that Ui > 0. (5. Eq.
We observed above. but we see here. by representing the vector \i) by the wave function i(x. We now consider the latter expression. thus enters these states. in all other cases the state of the system is a mixed state. Equation (5. and in the other states with probability zero. t) p= i The expression (5.x'). For the case n we have TJ=J (520) (5.20) expresses that the system can be found with probability 1 in the state \i). i. We see this more clearly by going to the position space representation. represents a quantum mechanical probability. Thus we have a system in a pure state. so that p{x. .22) In the following (x\i) is always to be understood as (x\i(t)}. In the following we are mostly concerned with considerations of systems in pure states. the system is in a pure state.24) s (x\x') = 5{x . since they do not take into account the (interaction with the) rest of the universe.e.t) we now write tp(x.e.8). i.t)\2.5. Then {x\Pi\x) = (x\i)(i\x) = \{x\i)\2. If all ui but one are nought.t) = {x\i). ^2ui\i){i\. defines the statistical operator. (5. not in a mixture of states.t) and instead of (x\Pi\x) we now write p(x.e. The effect of the interaction of the system under consideration with the surroundings. in which the numbers Wj represent classical probabilities and the operator \i)(i\ quantum mechanical probabilities or weights.t). (Question: Is the state of the universe a pure state?) 5.t) = \^{x. the only remaining one is 1. one says. (5.§ {\x)} is a complete system of basis vectors in the position representation space Fx with dx\x){x\=l. i.3 The Probability Density p(x. (5.23) Instead of i(x.3 The Probability Density 77 We see therefore that p defines the mixed states.21) p=\i){i\=Pi. that — different from u>i — P. that these are actually not sufficiently general. which is contained in the coefficients Cai.
t) = / dx{i\x)(x\i) = (i\i) = 1.27) is known as the Schrodinger equation. Schrodinger Equation and Liouville Equation The relation (5. can be found with probability 1 in the space R1.28) Differentiating Eq. (5.25) J Thus the particle whose state is described by \i).3.t) = \rp(x. This differentiation requires the time derivative of a state vector \i) E "K. the Schrodinger equation is always postulated in some way.e.22) can also be written (xi) 2 = (i\x)(x\i). We now want to obtain the quantum mechanical analogue. (5. the expression (5. t). 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.^ The equation which is the conjugate of Eq. 2.27) where H is the Hamilton operator of the system. Equation (5. and hence the integral over all space dxp(x. (5. t h e equation of motion with states represented in a system for which the q's or here x's are diagonal).27) is iHJ\ jt= m . or the wave function tp(x. To this end we have to differentiate the density matrix p. (5. with respect to time. The generalization to a threedimensional position space is selfevident: JK 1 / p(x. but the linear operator states are moving — even if the system is observably stationary).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. which is chosen in such a way that the desired analogy is obtained. Hence we postulate: The time development of a state \j) E "K is given by the equation ih^\j)=H\j). it is not "derived"). i.8) and multiplying by ih. and the timeindependent equation should correspondingly be called Schrodinger wave equation (i. (5. We shall convince ourselves later that the postulated equation is sensible (in fact.e. (5. .t)\2. We therefore postulate first an equation for the time dependence of a state vector.26) 5.78 CHAPTER 5.8).
M\/n\ (530) where in statistical equilibrium dt £^">01=03 (5.Xi)\j). (5. We can now consider the Schrodinger equation (5.p}.5. (3. (5.34) . such a formulation is described as Schrodinger picture. i. we observe that here on the left hand side we have the total time derivative. dp/dt ^ 0.28) the states are considered as timedependent.32) with its classical counterpart (2. the relation ih^ = [H. (5.p} + ihJ2^\J)(J\. (5. p also contains "quantum mechanical probabilities" (i.4 Schrodinger Equation and Liouville Equation 79 Inserting here Eqs.27) and (5. Eq. (5. both uji and i)(i).21). With Eqs. Eq. not the partial derivative.32) Comparing Eq. (2.37)) (5.27) in the position space representation: ih(x\\j) = (x\H\j) = {x\H(pi.e. We thus obtain the operator form of the Liouville equation.28) we obtain (see also Eq.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). i. The Schrodinger picture and the alternative Heisenberg picture will later be considered separately. The correspondence to the classical case is obtained with the substitution h J< > <5'33> and dp/dt = 0 (cf. The reason for this is that in addition to the "classical probabilities".e. (5.27).31) below) 3 j 3 = y £u>J{HPjpjH)+thyE^m\ dt 3 = [H. (cf.24)).e.
.36) however. to) = exp[—iH(t — to)/H\. we can replace these states by corresponding timeindependent states.. For a mixture of states \i) of the system under consideration caused by the rest of the universe we have Ui ^ 1. i.38) (x\j)t=o = Yt(x\E^(En\J)t=o. since the solution of Eq.x 1 ^ ( x . it is plausible to refer to this equation at this stage. 10.Xi){x\En) = En(x\En). we write V>(x.35) Here we can look at the Hamilton operator as a "time generator3''. appearing in the Boltzmann distribution. in Sec. (5.36) in in The initial wave function or timeindependent wave function ^(x.and temperaturedependent Green's functions. T meaning temperature. however. (5.t): > ih—</>(x.80 CHAPTER 5.t) = Hld .t) = exp ip(x. In expression (5.e.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.35) can be written'! ip(x.. 0) or \il))t = \il>)t=o exp (5. 5.** In view of the close connection between time. we shall describe as "time development ator given by U(t. since the timedependence cancels out (the exponential functions involve the same operator H but with signs reversed.8) for p.4. or H\En) = En\En). Schrodinger Equation and Liouville Equation or with (x\j) — ip(x. which we shall need later. in p = ^2i^i\i}(i\. the state \i) is still timedependent. „ / ith—. . P. with respect to time t.e. operator" the exponentiated oper . this is an application of the BakerCampbellHausdorff formula which we deal with in "Later. 0) can be expanded in terms of a complete set of eigenvectors \En) of the Hamilton operator H. **R.4. t ) . but with respect to the parameter /3 = 1/kT. (5. i. n (537) This equation is described as the timeindependent Schrodinger equation. namely that the density matrix satisfies an equation analogous to the timedependent Schrodinger equation — not.0): where H(ih—. Feynman [94]. With the help of Eq. a (5.
the expression _Zie^\Ei)(E>\ also as (see below) e0H (5.39) therefore (x\p\x') = J2"i(x\^)(Ei\x'). x') = ^ i LUiMxWix'). i.41 ) or with 4>i(x) := (x\Ei) this is p(x.45) since on the one hand •PEi i i i and on the other hand .9) with En > nhu) L0icxePE\ so that Yliui becomes = f3 = l/kT.42) Inserting here Eq. (5. Hence we have (with (Ei\Ej) = 5^) in what may be called the energy representation p = YJ"i\Ei)(Ei\.x'(3) = Si e ~^W*V). (543) Since H(f>i(x) = Ei<f>i(x).e. i ( 5 .^ n tne position space representation Eq.4 Schrodinger Equation and Liouville Equation 81 Example 5.5. we can rewrite p. (5. (5.x')^p(x. Eq.39) Without proof we recall here that in the socalled canonical distribution the weight factors uji (similar to those of the Boltzmann distribution) are such that (cf.40) 1. i (5. (5.44) r (TrePH)' (5. we obtain p(x. or ^ = ^ e 0Ei _m.1).40). (1.
x.tt tt For a more extensive discussion see R.45) without normalization as PN(J3) := e~pH. Wilcox [284].47) (5.1: Baker—Campbell—Hausdorff formula Verify that if A and B are operators. (5. B}} . we obtain = HxPN{x. (3).35).x'.82 CHAPTER 5. [A.X'.55)).L [B. the following relation holds: exp(A) exp(S) = exp ( A + B + i [A.47) with respect to (3.51) is seen to be very similar to the Schrodinger equation (5. (5. [B./3). (5. (5.The quantity pjv is obtained from the solution of the Schrodingerlike equation (5. This solution is the Green's function or kernel K(x. Equation (5./?).(3). We now rewrite the factor exp(—(3H) in Eq.49) In the position or configuration space representation Eq.46) is pN(x. B] + L [A. Schrodinger Equation and Liouville Equation so that p is given by Eq.51). (7.45) and (5. PN(0) := (EilpNiP^Ej) = (Ei\ePH\Ej) =e"^^ (5.4> = ^ g ^ .(3) := (x\pN(P)\x') = (x\e~0H\x'). M. Differentiating Eq. A}} + • Solution: The relation can be verified by expansion of the left hand side. Tr(pN) = / dxpN(x.x'.45).46) we can write the expectation value of an observable <^Tr<„. (5. (5.x'. .52) Hence this expectation value can now be evaluated with a knowledge of pN.50) Differentiating this equation with respect to j3.46) In the energy representation this expression is pNij(P) with PNij(P) = 6ij. Example 5. whose calculation for specific cases is a topic of Chapter 7 (see Eq. (5. (5. (5.51) where the subscript x indicates that H acts on x of PN(X.48) = 1. With Eqs. (5. we obtain 9pN gpP) = SijEie^ = EiPNij(J3).
* Comment of Y. (8. E." Quite apart from this basic significance of the harmonic oscillator. its associated space of state vectors also provides the best illustration of a properly separable Hilbert space.Chapter 6 Q u a n t u m Mechanics of t h e Harmonic Oscillator 6.the present form of quantum mechanics is largely a physics of harmonic oscillators. The importance of the harmonic oscillator can also be seen from comments in the literature such as:* ".51) to (8.1 Introductory Remarks In the following we consider in detail the quantization of the linear harmonic oscillator. which is often described as "second quantization" as distinct from the quantization of quantum mechanics which is correspondingly described as "first quantization" (thus one could visualize the free electromagnetic field as consisting of two mutually orthogonal oscillators at every point in space). S. like the Coulomb potential. Kim and M.. we shall see on the way. Noz [149]. 83 . can be treated exactly and therefore plays an important role in numerous quantum mechanical problems which can be solved only with the help of perturbation theory. 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. Since the spectrum of the harmonic oscillator is entirely discrete. 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. However.53)). that the oscillator.. can — in fact — be quantized in quite a few different ways. This is a fundamental topic since the harmonic oscillator.
of which the only nontrivial one here is [q. p^ih—. hence we do not use a distinguishing notation. However. . the boundary conditions on the solutions ip(x) would be square integrability over the entire domain of x from —oo to oo. To this end we introduce first of all the dimensionless quantities rriQUi fl q. When such a term arises one has to resort to some definition.84 CHAPTER 6. which is — in the first place — representationindependent.to = 1 reduces to the parabolic cylinder equation.2) which for mo = 1/2. 1 2 2 The quantization of this oscillator is achieved with the three postulated canonical quantization conditions. Also observe that H does not contain terms like pq or qp.p]=ih (6. This can be done in a number of ways.^ The first problem is to determine the eigenvalues and eigenfunctions of H. ox and then to the timeindependent Schrodinger equation Hip(x) = Eip(x). This equation would then have to be solved as a second order differential equation.p. so that there is no ambiguity due to commutation.2 The OneDimensional Linear Oscillator In the case of the onedimensional. we can also proceed in a different way. like taking half and half. which is called Weyl ordering. 2mo dx \ 2 2 (6.1) for the hermitian operators q.MvT« 7=r)^(VTvir>' + («> 'Note that we assume it is obvious from the context whether q a n d p are operators or cnumbers. in the present case + [EmooJ2x2)ip = 0. h = l. For instance we can go to the special configuration or position space representation given by q^x.' ^mgLOh by setting and *~. linear harmonic oscillator the Hamilton operator is given by the expression XJ 1 2 . Quantum Mechanics of the Harmonic Oscillator 6.
2 The OneDimensional Linear Oscillator 85 (p. A\ we obtain h A + A^ rriQLU y^ (6. C}B. for ^ A  a ) = a\a). i.B}C + B[A. We now use the relation (3.BC] = [A.14b) [A* A. C] + [A.7) (6. if A*A\a) = a\a).6) and p = y/motoh AAi =^. + 1).12) in order to obtain the following expressions: [tfA.9) We observe first that if \a) is a normalized eigenvector of N with eigenvalue a. Prom these relations we obtain ( ^ = ^ . C] = A[B. (6.8) The eigenstates of H are therefore essentially those of N := ^ f A (6.10) then a = (a\A^A\a) = \\A\a)\\2>0.1) we obtain immediately the commutation relation [A.14a) we deduce for an eigenvector \a) of A^A.34d): [A.A^] = 1.e.11) Thus the eigenvalues are real and nonnegative. A*} = A^A. i. With the help of Eq. . (6.1 ) .e. (6.5) Reexpressing q and p in terms of A. (6.14a) (6. (6. q are hermitian). (6.6.13) ( ^ A ) i t = A\A^A From Eq. (6.C]. and [AB. A] = [A\ A]A = A. (6. i\/2 Inserting these expressions for p and q into H we obtain H = hiu{A^A + AA*) = hi* (A^A + ^ . A^] = A\ (6.
unless A 2 a) = 0. (6.86 the relation CHAPTER 6.21) . (6. Thus for a certain value of n > 0.17) (6. or Ata) = v ^ T T . (6.l)A\a). we must have (a) Let \an):= Ana)^0. but (b) An\n>\ llA An+1a)=0.11) this is not possible. unless A^\a) = 0.l)\a) = (a .l)\a) = A(a .19) i.A)\a) A{AAU . unless A\a) = 0. However.18) = '= a ) A(A^A .. A 2 a) is eigenvector of A^A with eigenvalue (a — 2). In this case we have (A*A)A2\a) (6 (6. (6. (6. The norm of A la) is Aa) 2 = {a\A^A\a) = a ( a  a ) = a. (6. Quantum Mechanics of the Harmonic Oscillator (A*A)A\a) = A{A*A . since this equation implies that the eigenvalues cannot be negative.e. If we continue like this and consider the vectors An\a) ^ 0 for all n.16) This means.l)A\a) = A(A^AA . Similarly we obtain from Eq.2A)\a) = A(Aa . Next we consider the vector A 2 a). we find that A n a) is eigenvector of A^A with eigenvalue (ct — n).20) .14b) : (AiA)A*\a) = A\A^A + l)a) = A\a + 1) a) = (a + l)A^\a). in view of Eq.15) Thus A\a) is eigenvector of A^A with eigenvalue (a — 1).2A)\a) 2 (a .  a) is eigenvector of A^A with eigenvalue (a + 1). or \\A\a)\\ = Va. This would mean that for sufficiently large values of n the eigenvalue would be negative. (6. Similarly \\A^\a)\\2 = (aAAta) = (al + A^A\a) = (a + l)(aa) = a + 1.2)A a).
2 The OneDimensional Linear Oscillator be a normalized eigenvector of A^A with eigenvalue (a — n).26) <0A4 t AA t + A4 f 0) = (Q\2(A]A + 1)0) = 2.27) According to (b) of (6. v.. (6. (6. (6. \ 9 \ I / II 4 T ) . (6. so that A^\n)^0 In particular we have A^\0) ^ 0 and II^IO))!2 = (OIAA+IO) = (0\l + AU\0) = (00) = 1.n ) = " A»a>=» ' = A«a)F = 1 . In the following manipulations we always use the commutator (6. (6. Moreover.. HAU+IO)!!2 = = (O\AAAWI\O) = {O\A{A^A + I)A^\O) for all n. The state 0) is called ground state or vacuum state.23) Hence the eigenvalues of the operator iV := A^A are nonnegative integers.18) we obtain for a = n: p t  n )   2 = n + l.24) This is a very important relation which we can use as definition of the state vector 0). \ I 2 2 x 87 Replacing in Eq.20) we obtain from this that the right hand side vanishes. or a = n > 0.' " /. I „ . For a — n = 0 we deduce from (b) of (6.n  a .20) that A0) = 0.17) a by (a — n).25) (6.I n U .e.6. i. we obtain an = \\A\an)\\2(6=) An+l\a) \\An\a) (6. (6.20) we have for a = n = 1: A2\1) = 0. From relation (6.28) . that an = 0.22) With relation (b) of Eq (6.5) to shift operators A to the right which then acting on 0) give zero. But A 2 A t 0) = A(AA^)\0) = A(A^A + l)0) = 0. so that (a(A n )tA n \a) __ \\An\a)\\2 ( a .
Quantum Mechanics of the Harmonic Oscillator A2AtA*\0) = A(A^A + 1)4*10) = (AA^AA^ + AA*)\0) = = = {AA\A* A+1) + A* A+ 1}\0) (AA^+ 1)\0) = (A^ A+ 2)\0) 20) ^ 0. A^} = 2A\ and [A.32) we have (nm) = . we have 2> = ^=dAi\0). (6. According to Eq. and in general [A.29). (6. In general we have \n) = ^(A^n\0) (6.33) . Similarly we find 2)oc^Ut0).32) (6. i.^ ( O l A ^ n o ) . (6. and in view of Eq.11) we have [A. (6. But using Eqs. (0^ 2 (At) 2 0) = 2.26) the equality l) = A t 0). (6.e.Al] = 2A^A^ + (A^)2 x 1 = 3(A^)2. The states \n) thus defined are orthonormal. (A*)2} = [A.5) and again (6. (A^)3] = [A.11).(A*)n] = n(Ai)n\ (6.(6.30) (arbitrary phase factors which are not excluded by the normalization have been put equal to 1).31) (6.29) Hence we obtain and in view of Eq. A^A* + A*[A. (At)2]A+ + (A*)2[A. as we can see as follows.88 and CHAPTER 6. l>ocAt0).34) (6.
'n\ and = v n ![ ( A t ) M i= + (6.1 0).1 ) .39) A^A\n) = y/nA^\n . •(» (6. in quantum mechanics. . With Eq. .8) we obtain therefore H\n) = hu (A^A + ]. . 2 . the eigenvalues of the Hamiltonian H are En = hw\n+\+ i ) .35) Inserting this result into Eq.1) (6.33). however. In view of the same properties A is also called annihilation operator and A^ creation operator of socalled u quasiparticlesv.34) A\n) = ±=A{Al)n\0) .40) (6. Here. 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) = n\n).38) (6. (6. .35) the operators A^ and A are called respectively raising and lowering operators or also shift operators. chosen in close analogy to the postulated "second quantization relations" of field theory. n = 0 . The terminology is.e.34) and (6.J \n) = hw (n + i J \n). we do not have creation or annihilation of any real particles as in field theory (hence the word "quasiparticle"). . . (6. whose number is given by the integer eigenvalue of the number operator N = A^A.1 (A + ) m . l .32): rf\n) = .L ( A t ) n + 1  0 ) = VnTT\n Vn! and with Eq.36) + 1) (6. (6.41) The contribution hu/2 is called zero point energy.37) n^)711}^) = v n  n . i. (6. we obtain (n\m) = 5nm We also deduce from Eq. ldnm = n\5nm.2 The OneDimensional Linear Oscillator so that {0\An(A^)m\0) = = {0\An1A^mA\0) + (0\An1m{A^)ml\0} 89 0 + m(0A n .6.
3 The Energy Representation of the Oscillator The energy representation.e. 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 (6.1... is defined by projection of operators on eigenfunctions of energy. A^ from the relations (6.38) and (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. Quantum Mechanics of the Harmonic Oscillator 6. i. Eq.7) the energy representation of the operators q and p.. (5.'\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 . The representation of the Hamiltonian in this energy representation is {n\H\n') = En5nn. n = 0. other elements zero. (6.46) v^ .39)). other elements zero. Vn + 1..g.43) = = y/n(n'\n — 1) = y/n5n^ni. (6.39): {n\A.2. and — in fact — we have used this already previously (see e.45) Correspondingly we obtain with Eqs.44) yfn. (6.6) and (6. (6.90 CHAPTER 6.42) We can deduce the energy representation of the operators A.. En = hu(n + ). also called Heisenberg representation.
2 . The ground state wave function (/>o(x) is defined by Eq.6. Eq. (6.4 The Position Space Representation 91 It is an instructive excercise to check by direct calculation that Eqs.4 The Configuration Space Representation We saw that the eigenstates are given by Eq. A\Q) = 0.32).3)) ^ (q 2n \ + (6. by (cf.5) are also satisfied as matrix equations.e. Correspondingly the position space representation is given by the wave function <t>n(x) := (x\n). 6.1) and (6. This is a simple differential equation of the first order with solution (x0) = The normalization constant C is determined by the condition* OO /C 1 = (00) = / / dx(0\x)(x\0) = \C\2 I dx(0\x)(x\Q) = \C\2 / OO J —< emou}x2/hdx = \C\2 mow' so that We choose the arbitrary phase 6 to be zero. ™ou(x 2h \ + J _ d ] m = 0 (649) mQUJ dx CemouJx2l2h. (6.47) LP 0) = 0 . mow / (6.48) Applying from the left the bravector (x\ and remembering that (x\p\<f>) = we obtain ih—(x\(j)}. (6.24). i. (6. Hence ( * Recall J™ dxe™ * ? 2 2 2 1/4 W \ i/4 \ em^x*/2h_ (g 5 Q ) = yfln/w .
e. they are called Hermite polynomials.54) (6 55) i/4 rzri7r 1 / 4 vo. Quantum Mechanics of the Harmonic Oscillator This is therefore the ground state wave function of the onedimensional harmonic oscillator. i. Vn! Now {XlA (6.0 = ( .L ( x  ( A t ) »  0 ) .92 CHAPTER 6. it suffices according to Eq. In order to obtain the wave functions of higher states.l ) n f f „ ( 0 .56) < ^ = We have as an operator identity the relation . (6. (6.51) \l 2h {X q \ m0uP)\l 2h \X m0u. c/>n(x) := (x\n) = . (6.32) to apply the appropriate number of creation operators A^ to the vacuum state 0). so that <t>n(x) = ( . ) e"^2.l ) > n ( x ) .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 . These polynomials are real for real arguments and have the symmetry property # n ( .2 n n! The functions Hn(£) are obviously polynomials of the nth degree.
en_ dt. gain as an operator relation we have de = een± = een da ee/2±_^ee/2 d£ 1 u ^ d? dt.It follows that 0en(Pe/2^Lez 2 n\P?i2 — c2 d d£ ? = 2f.4 The Position Space Representation i. Magnus and F.6. Oberhettinger [181]. p. (6.2)m = (±t)m for some function VKO.58a) ' T h i s definition — apart from the usual one — is also cited e. 81. . ( 2 0 4 . e.54) we find that (observe that there is no factor 2 in the arguments of the exponentials!)§ Hn{i) = ( . (2036(20. 93 .12(20 .e. (202"2. (6.57b) • Generalizing this result and inserting it into Eq. d£" (6.58b) by W. 2£.l ) n e ^ In particular we have #o(0 #i(0 #2(0 #3(0 tf4(0 = = = = = 1.g.12. — = n+ d ~^ i 2 + d 2 de) ' 1+e ~ (6.
60a) and I d£ Hen{Z)Hem{Z)e?l2 = n! V & „ m  (6.^ + n) He n(0 = 0 (6. £. Oberhettinger [181].59d) The normalization of Hermite polynomials is given by the relations: oo / d£ Hn(OHm(Oe~e oo = 2 n n\^5nm. Quantum Mechanics of the Harmonic Oscillator The differential equation of the Hermite polynomials Hn(£) defined in this way is ( ^ 2 ^ + 2n )ff"(O=0. £ 4 . Magnus and F. (6. (6. Mathematicians often define Hermite polynomials Hen(^) as^ Hen(0 Then He0(O Hex{0 He2(0 He3(H) fle4(£) = = = = = 1. 81. (6.58c) Here special care is advisable since the Hermite polynomials defined above were motivated by the harmonic oscillator and hence are those frequently used by physicists.59c) The recurrence relations satisfied by these Hermite polynomials (the second will be derived later in Example 6.He'n{£). = nHe^iO. (6. £33£.(e^/2) or Hn(0 = 2n/2Hen(V2ti).59a) The differential equation obeyed by the Hermite polynomials Hen(£) defined in this way is (^2 . = ffen+itO+ntfeniCO.60b) J—< "See for instance W.59b) = (l)^2/2^.94 CHAPTER 6. £2l.6 £ 2 + 3.3) are (as given in the cited literature and with a prime meaning derivative) Hen+1(0 Hen(0 He'n(0 = ZHenifi) . . (6. pp. 80.
p.. 2(£  s)£2& s=0 s=0 ~ ds = m sf2]e?= •««) : S=0 2(2e2l) = (2e)22 = ^2(0.61) e?l*2nl2Hn{Z).. Their normalization is therefore given by OO /"OO / OO J —OO z (6.s)=F(£.64) n\ See e. s) as a Taylor series. 93.63) The meaning is that if we expand F(£. W.6. Oberhettinger [181]. Thus the generating function can be written F(Z. Magnus and F.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. and so on.g. .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£) = ( .l ) ^ e^HeniO 2 / 4  .s) = J2 n=0 (6.^ = e^/42~n/2Hn(aV2).s) = eM?(Zs)2}(6. F(S. (6.62) Joo The following function is generally described as generating function of Hermite polynomials Hn{£): F(S.
i. In classical mechanics the energy of the oscillator can be zero.e. 6. 2 . E0 ~ —{Apx)2 + IrriQ l mQuj2{Ax)2 moLoh/2 2TUQ 1 moU!2h/2 2 TTIQU = Ku). We see t h a t the lowest state is symmetric. We also see from the eigenvalue t h a t the energy of the lowest (i. ground) state is fko/2.96 CHAPTER 6. mow' Apx = ^m0Luh/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.e. and we see t h a t even and odd states alternate. In fact we can use the uncertainty relation AxApx ~ H/2. T h e comparative behaviour of t h e wave functions of the three lowest states of the harmonic oscillator is shown in Fig.e.1 The comparative behaviour of the lowest three wave functions of the harmonic oscillator. 6. i. in order to estimate the lowest energy of the harmonic oscillator. has parity + 1 . W i t h Apx = moAvx we have = moLoAx Ax and thus h/2 Apx h/2 mQuAx Ax I h/2 . Furthermore we observe t h a t the wave function of the nth state has n zeros in finite domains of the variable x. Quantum Mechanics of the Harmonic Oscillator Fig. for x(t) = 0 and p(t) = mox(t) — 0.1.
4) to obtain by contour integration the eigenvalues of the harmonic oscillator. and is directly observable. and with potential V(q) = 2K2mov2q2 and total energy E = p2/2mo + V(q).plane Imz z .4 The Position Space Representation 97 The zero point energy plays an important role in atomic oscillators. This observability can be taken as direct proof of the uncertainty relation. mass mo. Solution: Consider the simple harmonic oscillator of natural frequency of vibration u. Thus. back to 0.1/a O l/a Rez all at infinity Fig. original path q . 14. According to "old quantum theory" with integer n. 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)). Example 6. in corrected version instead with n* = n + 1/2.: 27 T mov2a2. /classical orbit Here q goes from 0 to a.2 contains only the answers since the evaluation of the integrals proceeds as in Example 6. pdq. In Example 6.1. The analogous Example 6. Finally we point out that the Hamiltonian of the harmonic oscillator can be diagonalized in many different ways. In the plane of complex q the integral has no poles for finite values of q. 27rmoi/ <f> V'a 2 — q2dq = n*h.plane pole at z=0 ^ . But it has a pole at infinity. Sec. to —a. which is the complete orbit.2 The original path and the transformed path at infinity. n*h. classical orbit i. setting q = 1/z. E .e. to 0. 6. The kinetic energy is always positive or zero.6. like twoatomic molecules.. we obtain the integral .1 we demonstrate how the eigenvalues may be obtained by the method of "poles at infinity".1: Eigenvalues obtained by contour integration Use the corrected BohrSommerfeldWilson quantization condition (cf.
harmonic oscillator and Coulomb interaction. In the following we investigate in more detail the important differential equation of the harmonic oscillator. 6. i. are the most basic and exactly solvable cases in quantum mechanics. We can evaluate the integral with the help of the Cauchy formula (the superscript meaning derivative) Jc Hence we obtain (z . 2 . With a2 = E. After removal of a Gaussian or W K B exponential.e. . (m .c2 = (I + 1/2) 2 (natural units.5 The Harmonic Oscillator Equation In the above we considered mainly the energy representation of the harmonic oscillator and encountered the Hermite polynomials. and obtain with this all properties of the solution. 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). (11. verify that II = d> dq yi92 aq + ft 2TTI 9 /y/F^I 9z V a + bz z=o .a) m = 2 .e.e. i. 6. since these two cases.2: Contour integration along a classical orbit Setting q = 1/z. / " ^ . £ = n*/w = ( n + .5 . for instance — without deriving all properties of the solutions.( 6 . « qdq + I2 = f dq . . in q at infinity).3.It follows that 1 = s2 i ^V»a*ai ~2f z=0 2wi r m '„2 = 7ria 2 La*Va 2 2 2 . . 1 .1 D e r i v a t i o n of t h e g e n e r a t i n g function From the above we know t h a t the energy is quantized.2b = Ze2.1. Example 6.65) f ^Va2z2 .1. E = Z2e4/4(l + n+ l ) 2 (cf.2TTI a ^ = . we start with a Laplace transform ansatz for the remaining part of the solution. but here we leave this .5.. z = 0 Vi 2Kmov(Ka2) = n*h = E/v. In Chapter 11 we perform similar calculations for the radial equation of the Coulomb potential — as in Example 11. hence we consider the following differential equation in which n = 0 .1)! (6. mo = 1/2) the integral I2 can be checked to give the eigenvalues for the Coulomb potential Ze2/q (2 turning points). a J \d Solution: The solution proceeds as in Example 6.) hv. Eq.V62 a2).98 CHAPTER 6. Quantum Mechanics of the Harmonic Oscillator This integral has a triple pole at z = 0 (i.114a)).
(this determines the limits).p t dp = \ps(p)ePt} .66) First we have to remove the £2term. Thus — ignoring details — / dpf(p)erpx is the Fourier transform of / ( p ) . i.68) § = J p2s(p)e*dp. (cf. (11. / dpf(p)e~px the Laplace transform of f(p). (6.6.J ±(ps(p))e*dp.i ^  (ps(p)) P n P Integrating this expression.5 The Harmonic Oscillator Equation latter property open for determination later: 99 ~dP + 1 1 2 V> = 0 . fp{ps{p)) = {v2 + n)s{p). (6. Hence we set (with the chosen sign in the exponential) m=y{t)et2l* with *0. E.t*ij& + ny{t) = O. (6. This has to be true for all values of t.g.67) the latter being the differential equation of Hermite polynomials Hen{t) for integral values of n.67) we obtain \ps(p)e pt] + I P2s(p) + —(ps(p)) + ns{p) e~ptdp = 0. since this can give rise to exponentially increasing behaviour (at most a factor t can be tolerated). Substituting these expressions into Eq. (6.** We make the Laplace transform ansatz^ (with limits to be decided later) y(t) = [ s(p)e~ptdp. Oberhettinger [181]. Magnus and F. p. W. . . We demand that C(p) := — \ps(p)e~pt] = 0.69) **See e.e. and / dpf(p)pn the Mellin transform of / ( p ) .g. Then (with partial integration) *dt = ~*/Ps(p)e. Eq. we obtain P2/2 ln(ps(p)) = — p — nlnp or s(p) <x Tl+l P (6. so that the integrand of the remaining integral vanishes. 80.111)) the Mellin transform of e~p is n!.
as indicated in Fig.e. This can be attained if p is allowed to be large (t can be of either sign). we can choose a path from zero to infinity (the condition will vanish at p = 0). 6. 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..67)) M) * e"* /4y(t) ~ et +2/ 2 /4 But this is not permissible. when p = —t (which is possible. then for large t we have y(t) ~ exp(i 2 /2).3. Quantum Mechanics of the Harmonic Oscillator ptp2/2 » ( « ) ptp2/2 dp with C{p) = (6. Thus the only solution left which does not involve large values of \p\ is that with § Q for n an integer. of course). since otherwise (with p = \p\ exp(i8)) we will get part of the solution from 0 to infinity. so that y(t) ~ exp(i 2 /2). (1) This will be satisfied if \p\ — ±oo. The value of the exponential at this point is ~ exp(£ 2 /2). 6. Hence a possible path of integration > is from — oo to oo. The quantum mechanical wave function is then (cf. To see this. = / • Im p Rep n not an integer Fig. n must then be an integer.100 and hence CHAPTER 6. This exponential assumes a maximum value when pt \—p2 = minimal.70) = 0 P.71) . consider the exponential in the integrand. (6. (2) If n < 0. Eq. y(t) oc 2vri J pn x dp for n integral and ept p 2 = coefficient of pn in ~^ (6.3 The path when n is not an integer. For singlevaluedness. Therefore we take ptp2/2 y(t) . i. so that p cannot go to infinity.n+l between the two limits of integration. We can show that if \p\ is allowed to be large.
„y2 v l { ' ^ r (^) \ 2 ' where v = 0. .6. pt p2/2 (6. 4 .5 The Harmonic Oscillator Equation 101 with Cauchy's residue theorem.3. In y(t) above. if n is even and v = 1.l ) n n ! x coefficient of pn in = = (~l)nn\ e~ ~ x coefficient of pn in V l z ^ e . We have 27T! Setting q — p + t. The Hermite polynomial Hen (t) is now defined such that the coefficient of tn is unity. We can also obtain the derivative form of this Hermite polynomial. n—v even ^ 2 ' Hence we obtain the polynomial of degree n Hen(t) = nl £ see below ( _ 2 ) ( n .if n is odd. Therefore the coefficient of (pt)n in exp(—pt — p2/2) is (—l) n /n\. The exponential function here is described as the generating function of the Hermite polynomials Hen(t). The Hermite polynomial is therefore given by H ^{t){^—f—^r~dp. 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.l ) n n ! x coefficient of pn~v in V LJ_L e p 2 /2 *^ v\ u=0. . .P 2 / 2 00 C—tY ( .. the term of highest degree in t in the exponential corresponds to taking all p's from exp(— pt)..72) We now obtain the polynomial form with Cauchy's residue theorem as Hen(t) = ( . . this becomes jfj. (676) .5.
Solution: We have to evaluate °° 2 .g.n + l ) f f e n ^ i ( £ ) + (n. and elsewhere. (6. we obtain tHe„{t) (l)"n! ~ i. for the evaluation of expectation values.1 '2dt using Hen(t) = (l)n27 r . 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+<.)} 2 /2dt = ePde(P 2 W)/2^^} J —CO . one can construct the tower of polynomials Example 6.it) _ (n + l)Hen+1{t) (l)""1^1)! (—l)"+ 1 (n + 1)! ' tHe„(t)=nHen1{t) + Hen+1(t).59a). Hei(t) Henj. Starting from Heo(t) = 1. Example 6. (n. n + l) = l. 8.e. in radiation problems.n— 1) = n.4: Orthonormality of Hermite polynomials Establish the orthogonality and normalization of Hermite polynomials. 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). We leave the consideration of the recurrence relation of Hermite polynomials (of paramount importance in the perturbation method of Sec. Hen(t). as in the normalization of asymptotic expansions of Mathieu functions). ( n .7 that we use throughout this text) and their orthogonality and normalization to Examples 6.4. Example 6.3 and 6.3: Recurrence relation of Hermite polynomials Show that tHen(t) = (n.102 CHAPTER 6. Reinterpreting the remaining integrals with Eq.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. (6. (6. Pptp 2 / 00 /2 dp. Quantum Mechanics of the Harmonic Oscillator which we recognize as agreeing with Eq.70).5 is a further application of the method for the evaluation of integrals that occur frequently in practice (e. n — l)Heni(t).77) = t.
82) ^ . (6.6.q) = = e . 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. g) is the product V^Tre" ^oM!(n/i)!(n + 2rM)! (6. so that /2 Q2(n+rrf in / ( p .5 nm .jQ) n n!I ^—' ^Q fi\(n .q). zero otherwise (6.n)\ i s an+2rij.^ + 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 .78) \/27rn!. the coefficient of qn+2rn the coefficient of p"qn+2r m exp(.(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: Generalized power integrals with Hermite polynomials Establish the following general formula for Hermite polynomials: / Jo t2sHen(t)Hen+2r{t)et ir n\(2sy./(n __ 2r — /i)!. Hence the normalized wave function of Eq.66) is \Zn\V2w Example 6.5 The Harmonic and hence Oscillator Equation 103 / = (!)" But (2iri)2 { J Jp=0 J q0„ pn+ i0m+l P I e d d P 1 1 / epq p" <k rrdq = coefficient of qm in epq = — 27ri / qm+1 m Therefore (IT ^n+m ^ n l I Jp=o ' v d P ' 27U pnm+1 =2ni if Ti=?7i. (6. g) is n! Next. and 0 otherwise \/2n(—l)n+mn! if n = TO.80) f(p.
( n + Jr).r . Hence t4 He„ (t) = = Hen+i Hen+4{t) (t) + [(n + 3) + (n + 2) + (n + 1) + n] H e n + 2 (t) + • • • + 2(2n + 3)Hen+2(t) + . n + l ) ( n + 1. n ) ( n ." ) a2^) r coefficient of (2s)! in f^ (s . A « 2 ( r '+ I/ ) 2 coefficient of (2s)! in (n + 2r)\e ' ^ .l ) ( n . n + l ) ( n + 1.t • 2l/2dt _ J : Q2 » a2(n+r^) ~ f ° ° f f „ t2/2 „ !ZoHen(t)e2 Wdt .i / ) ! ( 2 r + i/)! > .r . i.n + 2) + (n. K is the coefficient of a / ( 2 s ) ! in the expression J ..° i i f e T 1 ( t ) i f e „ + 2 r ( t ) 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.e.e.n+ l)(n + 1.— — ^ . n + l)(n + 1.v)\(s .104 CHAPTER 6.e. n + 2) + (n.^ / 2 * . /i = (n — v) _ / r o o e .' '2dt= / Jo t4 Dn(t)Dn+2(t)dt = V2^(2n + 3)(n + 2)!.u)\ 2s<—" v\(n . so that now with integration limits from 0 to oo K = coefficient of a2s (2s)! in J a = 2.v)\(2r + v)\ > Q (2s)!(n + 2r)! AT 2~ s r 2V v\(2r + u)\{n . n + l)(n + l . i / ! ( n . As an example we obtain for the integral from 0 to oo (also reexpressing the integral in terms of parabolic cylinder functions Dn (t)): / Jo t 4 f l e „ ( t ) H e n + 2 ( t ) e . i.' . n)(n.78).^ ) ! ( n + 2 r . Quantum Mechanics of the Harmonic Oscillator It now follows t h a t — in later equations with v := (n — /i). 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. the relation (n.l . i. n + 2) + ( n . n . Prom Example 6.r+v) a2s coefficient of in (n + 2r)\ V (2s)! h . n + 3)(n + 3. n.(t)e«2/2dt Now. n + 4)_ffe n +4(t) + [(ra.e '' ^ o^M ! ( n .77).2s _oo ( a 2 / 2 ) i . n — 1) are given by the first index. n + 2){n + 2. . n + 2)(n + 2.A o?(.Y^ i[ ^0vKnv)\C2r + ») cc2s A (n + 2r)! a 2 ^ .79). by n!(7r/2) 1 / 2 .83) = (n + 2r)!e a / 2 V „ „. The result then follows with half the normalization and orthogonality integrals (6. the coefficient of T 2s »{sr u)\ As we observed.ra + 3)(n + 3. n + l ) ( n + 1. n + l ) ( n + 1. n + 2)]ife„ + 2 (t) + ••• . n + 2)(n + 2.v)\ ^ Multiplying both sides by the value of the normalization integral. we obtain the result (6.3 we know that upgoing coefficients are 1 and downgoing coefficients (n. .A i ) ! (6.
which we write G{x.x'. which we write here K(x. and those which are timeindependent. We 105 . of a differential equation of second order. 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. With reference to this case. Let H^ = E^> (7. This example also offers a convenient context to introduce the inverted oscillator potential.1) be the timeindependent Schrodinger equation which has to be solved. more generally. These are the Green's functions of the timeindependent Schrodinger equation or. and is thus an important point to be noted here. in Chapter 21 (with a different calculation).2 Timedependent and Timeindependent Cases In the case of Green's functions we distinguish between those which are timedependent. which is a point of unstable equilibrium for a classical particle.1 Introductory Remarks In this chapter we consider Green's functions of Schrodinger equations in both timedependent and timeindependent forms.t). The first of these cases will reappear later. in Feynman's path integral as the kernel or free particle propagator. 7.Chapter 7 Green's Functions 7.x'). which is normally not discussed. we consider the sojourn time of a quantum mechanical particle at a point. We consider first timeindependent Green's functions. The other example of the oscillator enables us to evaluate corresponding expectation values of observables in the canonical distribution introduced in Chapter 5.
2) where Ho consists. of the kinetic part p2/2mo and some part of the potential.e. cf. although the explicit calculations can be much more involved. (7.g. x') has to satisfy correspond to those of * (to insure e.g.EW)xG{x. e. Green's Functions H = H0 + Hi.4) where the functions ipi(x) are solutions of HQij)i{x) = E^iix). The boundary conditions which G(x.x') can be written t G (7.x').13). (7.x') or ^2^(x)^*(x') i = S(x . Eq.x') by recalling the completeness relation. (7. square integrability. aq2.x') is defined by the equation* (Ho . We can readily see that G(x. (7. 'The case of £(°) equal to some Ei is considered later. i. i.5) M = Z*W=EW i l for Ei0) E * (7 6)  This holds since *For simplicity we consider the onedimensional case. and Hi is some other contribution. the relation X') = 5{x . . for instance. e. since the generalization to higher dimensions is selfevident. where £ ( 0 ) = lim E. a perturbation part like (3q4. appropriately decreasing behaviour at infinity). The timeindependent Green's function G(x.g.106 assume a Hamiltonian of the form CHAPTER 7.3) i which has the specific representation ^2{x\i)(i\x') i = 5(x .e.x'). We can obtain an expression for G(x.
E^ of the eigenvalue.2 Green's Functions: Timedependent and Timeindependent 107 We note that the Green's function possesses simple poles in the plane of complex E(°\ and that the residue at a pole is determined by the wave function belonging to the eigenvalues Ei.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. x'){E .x'.E® (EE{0) Hi{x))m{x).EM)X*(X) ( .9) = i H0YieE^i%(x)^(x') = H0K(x.0) = S(xx').x'. (7. (7. For the solution V we can immediately write down the homogeneous integral equation (again to be verified below) *(a. x'\ t) (7.7.11) = = 3) = fdx'(H0 .E^ We can check this by considering (H0 .#. The timedependent Green's function K(x. We rewrite the complete timeindependent Schrodinger equation H^ = E^ as (#0£(0))* = (££(0) #/)*.x')(E .x'..) = f dx'G(x.E<®)xG(x. (JL x'\ t) = H0K(x. We see that the initial condition (7.£(0) . (7.7) with the initial condition K(x.x'){E .x'.10) The contribution on the right hand side is the socalled perturbation contribution. . i (7.x. .(*'))*fr') Hi{x'))^(x') f dx'5{x .t) (K for "kernel") is defined as solution of the equation ih—K(x. We can see the significance of the timeindependent Green's function for the complete problem as follows.t) since ih^K(x.t) = ^Eie^ihM^i(x') i = Y^eE^ihUx)^(x').Hi{x'))^(x').8) is satisfied as a consequence of the completeness relation of the wave functions ij)i{x). which consists of the perturbation part Hi of the Hamiltonian and the corresponding correction E .t).
. Merzbacher [194]. All such methods of solution are based on the analogy of Eq.108 CHAPTER 7. Green's Functions Can we add on the right hand side of Eq. D. The timedependent Schrodinger equation ih\><l>)t = H\il>)t (7.t) for the complete problem as follows. But then the Green's function (7.. Booss and D. see B. H. This will be different. Maharana. the function ipi(x) would not be a solution of and it is not possible to add an inhomogeneous contribution. Instead of Eq. (7. This difficulty* can be circumvented by demanding from the beginning that 0= f dx'tf(x'){EEjHi(x?))V{x'). In this case the perturbation is restricted to that subspace of the Hilbert space which is orthogonal to ipj. 17. it is possible to solve the integral equation by an iterative perturbation procedure (see later: Born approximation). integral equations are more difficult to solve than the corresponding differential equations. Sec. that (E .15) ''This problem and its circumvention can be formulated as a theorem. however.2 (not contained in the first edition!).x'. see E. Bleecker [33].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).6) is not defined. 2nd ed. But if we assume that E^ ^ Ei for all i.Ej ffj(x'))*(a:') is a vector in TL which is orthogonal to ipj.11) with a system of linear equations of the form Vi = Mijyj. (7. Wiedemann [183]. (7. Examples in various contexts have been investigated by L.12) Equations of the form of Eq. The theorem is also known as Fredholm alternative. (7. In general.6) we would have (7.e. (7. W.11) are called (homogeneous or inhomogeneous) Fredholm integral equations. i. In cases where an inhomogeneous contribution is given. MiillerKirsten and A. then at best such a contribution would be something like cipi(x).14) We can now see the significance of the timedependent Green's function K(x. if we assume that E^ = Ej. J.
. for the oscillator potential.51) shows that we obtain a very analogous expression for the density matrix PN(P) a s for the Green's function K(x. (7. t) with the Note that when K(x. We can write this expression also as§ ip(x.The solution of Eq.0).17) is the initial condition of \4>)t. this relation provides the probability density \ip(x.g. 2nd ed.20) the Green's function K(x. (5.17) as initial condition is obviously W)t = Y.Q) = 6(xx'). (7.20) is the timedependent Green's function.158. \En)(En\ip)t=oeE^\ n (7. (7.x'. E.5. (7.t) = ^ e £ " f / « V n W < ( ^ ) .7. 1 See e.x'. in different formulations. 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.t) = where if(x^'.. (7.22) According to Eq.x'.15) which contains (7.2 Green's Functions: Timedependent and Timeindependent permits stationary states \En) defined by \1>)t = \En)eE^ih. Eq. 7. p. x'.t) describes the evolution of the wave function from its initial value ^(a^O). We use this in Sec. (6. Merzbacher [194].0)eEnt/in.t)il>(x'. x'. <a#) t = ]T n J [dx'(x\En)(En\x'){x'\^)t=0eE^\ (7. 109 (7. H\En) = En\En). *>0. e.2 for the computation of the sojourn time. t)\2.16) As usual. Comparison of Eq. (7. (7.42)) Wi=0 = E l ^ ) ^ l ^ = o n (717) Thus at time t = 0 the state \ip)t is a linear superposition of the vectors \En) with coefficients (En\i/})t=o.t) = J2 fdx/^n(x)^*n(x')^(x'.' which obviously satisfies the initial condition K(x.18) or.e. i.t) is known.g.7) with Eq.19) i/>(x.21) dx'K(x. t > 0.
7.E . between (with E^0' = E) G(x. i. we obtain / W : = . Inserting for GE+ie the expression above.x') = GE(x. 7.1.25) in agreement with the timedependent Green's function .23) along the contour C in the plane of complex E as shown in Fig. we obtain then also the corresponding density matrix.ie (7 .110 CHAPTER 7. t > 0. Green's Functions difference that (5 = 1/kT plays the role of it. t) = J2 e £ " f M i ( x ) < ( x ' ) . As a consequence of the above considerations one wants to know the connection between the timedependent and the timeindependent Green's functions. Jc 27T „ hn . (7.e.«^«2 W .x') and K(x. We therefore consider the following integral with e > 0: I(t) := i J ^eEt/*hGE^e(x.1 The contour of integration. With Cauchy's residue theorem we obtain I(t) := J2eEnt/ihMxWn(x'Mt) n = K(x. 24 )  ReE Fig.t).i /f e «/.x'. In the following we shall derive the Green's function for the case of the harmonic oscillator.x'. x'. = J2 ^„(s)C(*') En — E We see that G = GE depends on E.t) K(x. x')6(t) (7.
(7.0) = 5(xx').x>. (7.7).30) into Eq. A and B being constants.e.7.t) = ^eB^X In this case we have dK ~dt and dK dx 2 8K dx2 A A 2t3/2 A tV2 '^'K (7.28) B(xx')2 t2 DB(xx') 2 /t (7. i.e. K(x. which is moving in one space dimension.3 The Green's Function of a Free Particle 111 7.30) Inserting Eqs.X>. we obtain *I£I = h2 —(2AB). It is clear that it is nontrivial to solve an equation like Eq.29) and (7.x1) t B(xx')2/t B(xx')2/t 2AB 4AB2(xx')21 ~¥l 2+ W 2 (7.29) tJJ 2 2B(x . and identifying coefficients of the same powers of t on both sides.31) The constant A has to be chosen such that K(x.27).— d2 —2K(x.x'. We therefore consider first the simplest case with Hn = P 2mo h2 a2 2mo dx2 (7. In this case the Green's function is the solution of the equation d hK(x.t) t h2 = . B m0 n _ 2ih 2iK (7.26) This is the case of a free particle with mass mo.t). / dxK(x.x'.27) An equation of this type — called of the type of a diffusion equation — can be solved with an ansatz. v ' 2m 0 v h i. Thus we try the ansatz. (7.0) = 1.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. . 2mQK h h2 ih(AB) = —{4B2A).
(7. from K{x. (7.33). Green's Functions For parameter values such that the following integral exists. n For a free particle moving in the onedimensional domain \x\ < L — oo.(7. Mx)^M^) „ n so that J =^ . Merzbacher [194].zEnt. 2m0 (7.x'."737) \27rii/i in agreement with Eq.33) V Lirvnt Can we demonstrate that this expression can also be obtained from Eq.x't) = Y. (7. j3 = i(xx'). we > have to make the replacements V ^ fdk.ihMxWn{x').38) .112 CHAPTER 7. we have This is 1 provided A=JB= ^ (7 . p.. 158.0)oce"Qx2+ifcox. 32) It follows that x'. t) = /jJo_ e mo(««') 2 /««.36) 2mo Then — provided that the parameters assume values such that the intergral exists — K{x. i.x'.t)= f^em2t/2m0iheik(x~X')_ J 27T (735b) We set a = i—. v27r En^^.36) ^ W Q ^ J emo(ii') 2 = —eP2lia 27T dke~^k^2^2 J„00 /2«tS . (7.t) into Eq.35a) K(xy.20) and can then obtain ip(x. 1st ed.21). (7.H We can insert the expression for K(x.t) — for instance for a wave packet given at time t = 0 of the form ^(x.oo 27T (7. "See the excercise in E.t) = — / dkeak2+Pk 2vr J. (7. *>0.34) K(x.e.
.——K(x.7.x'.t) 1 + m0u2x2K(x. 1 2mo u>n ox n with the initial condition (7. i.40) becomes 2 . in this domain *Recall that S(x) = ± f eikxdk. i.33)./ ) + fK(x.e.x'.t).39) In this case the timedependent Green's function K of Eq.. .37) will later be obtained by a different method — see Eq. . §jK(x. the relation* 6(x) = o<y(os).e.40) Then Eq. .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. K(x. x'. —ih—K(x..t) We now set h2 d2 = .x'. (7.t) + ~x K(x. (7.0) = 6(xx') at / = 0. x'. We consider the onedimensional harmonic oscillator with Hamilton operator HQ.e..25) — in the context of Feynman's path integral method. (7. (7.3 The Green's Function of the Harmonic Oscillator 113 The result (7. i. mQLu . (21. / ) = ~^K(x. z .7) is the solution of d ih—K(x. .d r.4 Green's Function of the Harmonic Oscillator The next most obvious case to consider is that of a particle subjected to the harmonic oscillator potential. .8).42) .t). (7. We rewrite this initial condition in terms of £ and use for a = const.X . 7.t) = h2 2 d2 TW .x'. 5{ax) = ± J eikaxdk = ArS(x). / ) (7. x'. I (7. —KK(X.x'.e. HQ = ^—p2 ZrriQ + \m0uj2q2.x usn ot i.
50a) we obtain (7. a = . we must have /o = 0.c o t h 2 / . we obttain a' = l .51) See R. 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. b' = Aab.47) We insert this ansatz for K into Eq.) a'£2 + b'i + c' = (1 . that is. . as the limiting case / — 0. p. P. W ) .b2.33).4a 2 )£ 2 . f i.50b) (7.114 CHAPTER 7. (7.48) ocexp[{a(/)C 2 + 6(/)e + c(/)}] (7.t) become similar to expression (7.Z'. i.47) becomes (7. (/ = 2a. and we expect K(x.f) with a(0)»y. it • is suggestive to attempt for K the following ansatz^ m. Green's Functions to the particle behaves almost like a free particle.50c) (7. (7.50a) (7.^ . c(0)^.x'.49) a=icoth2(//J0 ) w 2 "' 2tanh2(//0) * To ensure that the expression (7. (7.. (7.46) If we interpret Eq. (7. (7. Integrating Eq.b2.4a&£ + 2a . Feynman [94].48). 50. so that for / — 0: > a^—. Identifying coefficients on both sides.e.45) in this sense.e.45) in accordance with (7.4 a 2 .42) and obtain the equation (with a1 = da/df etc.
i) (7. / —————rexp y 27rmsin(u.55) 'For an alternative derivation and further discussion see also B. Finally Eq.X'. and to ensure that we obtain the prefactor of Eq. in particular for the derivation of the sojourn time in Sec.54) For t —• 0 this expression goes over into the expression (7. c(0) = £' / 4 / . (7.27). 2TT^ Inserting a(f). as one can verify.50c) yields for c(/). c(/) = i ln(sinh 2/) + ^ 2 c o t h 2 / .48) 6(0) = — £'/2f. t. if we return to x.3 The Green's Function of the Harmonic Oscillator Correspondingly we obtain from integration of Eq. 115 To ensure that in accordance with Eq. x'. we can use K(x. we must have A = ('.0). (7. In order to satisfy Eq. p.£) _  {(x2 + x /z ) cos tot — 2xx'} 2msm(o.t) to obtain this element (x.In B.53) or. x .e. B independent of / .e. i./3) = J IJIQUI 2irhsmh(hLu/kT) 2hSinh(hw/kT) \{X +X j C x exp ° S n kT lXX J (7. i. (7. (7.48).2.5.* With this result we have another important quantity at our disposal.45).b(f). 7. (7. 174.47) we obtain ^ c o t h 2 / + ^ + ^ c ° t h 2 / (7. (7.51) for the density matrix PN(X.33) for the Green's > function of a free particle. as we shall see in the following.7.t) = . Felsager [91].c(f) K = 5 =exp Vsinh 2 / into Eq.x'.50b) Kf) = —r^77) smh 2 / A independent of / . K(x. we must have (besides A = —£') = /m 0 u. (7. (5. x') of the density matrix p^ (with respect to the canonical distribution with (5 = 1/kT): pN(x.x'. with A. Comparing the Eqs. . (7.40) of the timedependent Green's function with Eq.
.» l. eP E i = i + y. 52. Eq. ^—• Irpiv J pN(x. (5. we obtain § < Z> = ^ c o t h ^ ^ ° — .52): Tr (PNq2) Jx2pN(x. 11 For T ~+ 0: u>0 . = jdxp(x.56a) {q2) = = Y. )x Inserting into Eq. (7. we skip the algebra here.eft >E*) E (7 57) ' of the number of systems of the ensemble occupies the quantum mechanical state i.x. Feynman [94].56b) w v ' 2mQu} 2kT 2m 0 u ' What is the meaning of this expression? At temperature T the fraction (cf.uij>0 >• 0.116 CHAPTER 7. For instance we have with Eq. R. Thus the system is in a mixed state and the expectation value "Cf. <2 (7. Eq.e. at temperature T): {A) = Tl{pA) = ^ > .p)dx Thus for (q2) = Tr(pq2) = ^2i(i\pq2\i) we obtain: 2 (7. (5.p)dx W ) = —^r = —p .x.x.P) 2. p. Green's Functions With this expression we can evaluate (cf..55). [ [ fdxdx'dx"(i\x}(x\p\x')(x'\q2\x")(x"\i) Idxdx'dx"{x"\i){i\x){x\p\x'){x'\q2\x") J2 [ [ i = fjf^»(x>)(xW)(xW) dxdx'dx"5{x" i. P.56a) the expression (7.52)) the expectation value of an observable A in the canonical distribution (i.40)) 1 epEi 1 Ui = y. 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"}. (5.e. .
21). We return to the Green's function (7. 1 mo^ exp lmow / 2 2 ft X (7.QUJ (7.t "With the normalized ground state wave function of the harmonic oscillator given by Eq. 2 + x'2)le^2xx' to—>w—it t—>co 2 .50) we obtain. e > 0. i. x'.x'. We assume t > 0 and t —> oo.56b) is that with respect to this mixed state (whose cause is the finite temperature T). x'.t) w—>UJ—ie to—n t—>oo moco/irh ^ ' piuit g—iwt exp m ^\(x 0 ^ . <*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. 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. (7. If we consider the system in the pure state \i).54).7.54) is K(x. which means in the oscillator state \i) with eigenenergy fku(i + 1/2). /2N 2~7T: = eE°/iht(t>Q(x)(t)0(x') for £ > 0. (6. This is the first and hence dominant term of the expression (7. In this case the Green's function K(x.J (w — ie)< e (n+l/2)Et e i(n+l/2)u.e. and we replace to by LO — ie. setting a = mooj/h.58) Next we explore the connection between the explicit form of K and the latter's expansion in terms of a complete set of states.59a) E 0 = ^fiw.3 The Green's Function of the Harmonic Oscillator 117 (7. t) of Eq. of K(x.59b) n+ l \ ftwt here 2hf exp _ i ( n + .
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)]. k = k.(3) = J2^EnMx)K(x') n „ ^ ° e^MxWo^') (760) 7. which are planes.62) .. and with this we estimate the sojourn time T.5 The Inverted Harmonic Oscillator We encountered the inverted harmonic oscillator already in some examples. to infinity) the contribution with n — 0 dominates. the velocity of planes of equal phase. We first introduce the concept of a wave packet and then use the particular form of a wave packet in order to describe the state of the particle at time t = 0. the particle will stay there only for a finite length of time T. Considered classically.1 we estimate T semiclassically. is defined as vv = . Green's Functions For t large (i. of dpN as pN(x. 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. lie on surfaces.5.x'.59a).e.118 CHAPTER 7.e.61) The word "plane" implies that the points of constant phase <p := k • r — uit at t = const. (7. i. (7. The wavelength A is given by The phase velocity v<p. However quantum mechanically in view of the uncertainties in position and momentum. 7.e. a particle placed at the maximum of the inverted oscillator potential (which is classically a position of unstable equilibrium) will stay there indefinitely. i. In a very analogous manner we obtain the solution of the equation for the density matrix. as in Eq. (7. In Example 7.
a > 0 .4 The Inverted Harmonic Oscillator 119 Every frequency u belongs to a definite (particle) energy E (cf.Q ( k . u = w(k).64) where / ( k ' ) differs substantially from zero only near k' = k. i ) = /*/(k')e i(k '" r '' . for which \ip\ assumes its largest value: OO . We now ask: How and under what conditions does the time variation of the function ^(r. One defines as centre of mass of the wave packet that value of x for which d(p — = 0.k ') 2 . / oo J—oo . oo (7.e.e. i.65) Let f(k') = \f(k')\eia and <p := k'x . the fundamental postulate on matter waves in Chapter 2): E = hu. The wave packet describes a wave of limited extent.63) The relation u = w(k) is known as dispersion or dispersion law. ' t) dk'.e.Jt + a. i. or dco' da The centre of mass determines the particular phase. to the function ip given by oo / fik'y^'x^dk'. i.)¥'(*:")) +ism(<p(k') . /OO / / / oo /*oo \f{k')\\f(k")\ei^W^k"»dk!dk" l/(fe. ^ ( r . t) describe the motion of a classical particle? For reasons of simplicity we restrict ourselves here to the onedimensional case.7. essentially again a Gauss curve. duo' da . t) is the spatial Fourier transform of this Gauss distribution — as we know. e.)ll/(fc")l[cos(¥'(fe. e . xt— + — = 0. (7. then ^(r. (7. A wave packet is defined as a superposition of plane waves with almost equal wave vectors k.<p(k"))}dk'dk". If we assume for / ( k ' ) a Gauss distribution.
Green's Functions This expression assumes its maximal real value when <p(k') = ip(k") = const. Example 7. For E = hio. we can use a simpler method. p = hk the group velocity is equal to the particle velocity v.^ g=r a rad. a > 0. we obtain the deBroglie relation p — hk.. as claimed for =0. Thus the expression g{k) is solution of the following first order differential equation g'(k) + Aff(fc) = 0.120 CHAPTER 7.fc w> vv = ~a^ = S r a d P E «i7 Sd u .— g(k). . 9 cku dk dhw dhk dE dp dE dp' We can also argue the other way round and say: By identifying vg = v. In the present case.69) In general one uses the theory of functions for the evaluation of this integral. .1: Fourier transform of a Gauss function Calculate the Fourier transform of the spatial Gauss function e . Solution: The function to be calculated is the integral dxeax OO ezkx. The threedimensional generalization is evidently du .e. i. (7. (7. however. = — (768) It is instructive to consider at this point the following examples.67) The centre of mass moves with uniform velocity vg called group velocity of the waves exp[i(kx — cut)]. Differentiation of g(k) yields g'{k) = / dxe~ax 2a y_oo 7 f°° 2 {2ax)eikx = 7 2a / f°° Jx dx With partial integration we obtain from this 7 f°° 2 k g'{k) = — / ikeikxea* dx = . dE v 9 = .
—±=ek*/ia2 / 4 " = lim 1 — — lim ^e* = The second important example can be verified by immediate integration. We have — 1 / * ° ° dke~eWeikx =  I / " 0 0 dke~tk cos kx =  1 . with partial fraction decomposition.43) one can verify the following important relation* 5[{x a)(xb)] = T—!— [S(x .1 verify the uncertainty relation Aa.72) Solution: From Eqs. Example 7. (7.73) we obtain the requested representation of the delta distribution with 1 f°° 1 € Six) = lim — / dke~e^eikx = lim . MiillerKirsten [215].69) and (7.70) to verify the following representations of the delta distribution: X2/€2 5(x) = lim ?—=e ^ o eV7r and 5{x) = \\m^r^.69) and (7. (7.Afc = 81n2. J.4 The Inverted Harmonic Oscillator Simple integration yields g(k) = c e " f c 2 / 4 a . Since /•c 121 9(0) = J—00 dxe J_ we obtain S(fc) = ^ e " f c 2 / 4 a ( 7 ..70) we obtain 6(h) = lim — f°° dxeax2eikx = .74) Example 7.0 0 7T 7 0 7T e From Eq. 2 f + .7.a) + 6{x . or see H.3: The uncertainty relation for Gaussian wave packets For the specific Gauss wave packet of Example 7.. (7.. (7. (7. (7.2: Representations of the delta distribution Use Eqs. .71) (7.73) 27T 7 . \a — b\ (7. W. g .75) ' E .r „• With the help of Eq...7 °) With the help of this example we can obtain some useful representations of the delta function or distribution as in the next example.&)] for a + b. (7. XZ (e > 0). where c = 9(0) is a constant. Appendix A.
76) 2ira The uncertainty Ax is defined to be the width of the curve at half the height of the maximum.r)1/2a Re f(x) Ax Fig.CHAPTER 7.2 we sketch the behaviour of the Gaussian function p—x 11a ~—ikox (7. 7. i. where g(k) = max<?(fc)/2. Green's Functions 1/(27i)1/2a — 1/2(2.77b) (7.e. i.t) is a measure of the probability to find the particle at time t at the position x.77a) g(k) = e . requires according to Eq. (7. According to Eq. the width Ax is a measure of the uncertainty of the probability. Physically .e. Thus a slim maximum of the curve of /(a. Thus a sharp maximum of the function f{x). In quantum mechanics the square of the modulus of the wave function ip(x. The breadth A/c of the curve g(k) around k = kg.g. The wave function corresponds to the function f(x) in the above considerations (e. 7.2 The Gaussian curve.77a) small values of a. where  / ( x )  = m a x  / ( x )  / 2 .70) the Fourier transform of f(x) is (7. and hence implies large values of Ak. at time t = 0). is Afe = 2 V 2 1 n 2  (7.* o ) 2 / 2 .) leads to a very broad maximum of the Fourier transform \g(k)\. (7.« 2 ( * . Solution: In Fig.78) It follows that the product of the uncertainties is AxAk = 8 In 2. A simple calculation yields A i = 2v/2~In2a. when Ax is very small.
e. .5. i. nr = S(x)..".2 ) and U)t.81) *G. This means the particle is localized at x = 0.54) with the substitution ui — iu. Barton recalls that W.x'. (7. Barton [15]..4 The Inverted Harmonic Oscillator 123 p = hk is the canonical momentum associated with x.54) these parameters appear in the combinations moui/h (dimension: l e n g t h . ^In Eq.76) ex 2 /2a2 ^2na a>0 um 1/0*01 = lim a^O . The result (7.20) 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. E.t) = V 27Ti s i n h t exp 2 sinh t {(x2 + x'2)cosht2xx'} . (7.75) implies therefore: The more precise the coordinate of a microscopic particle is determined. 7. As G.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.79) According to Eq.71) > and (7. For reasons > of simplicity in the following we set in addition t mo = l.«> = / ^ C ' W . <7*» We assume that initially. Barton mentions at the beginning of his paper. In the same limit the momentum uncertainty Afc grows beyond all bounds. The function g(k) is therefore described as momentum space representation of the wave function f(x).* We obtain the Green's function K{0 for the inverted harmonic oscillator from Eq. In the reverse case a sharp localization of the particle in momentum space implies a correspondingly large uncertainty of its spatial coordinate.7. so that Ki0(x. 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. Thus G. Drastic idealizations are then required in order to reformulate the classical situation into that of a quantum system.u> = l. O ) . In the limit a — 0 we obtain from Eqs.h = 1. (7. (7. Estimate the maximum length of time compatible with quantum limitations before the pencil falls over. the less precise is the determination of its associated momentum. which means all values of the momentum are equally probable. (7. at time t = 0. many supposedly elementary problems request the calculation of the sojourn time of a quantum system near a classically unstable equilibrium configuration.
88) .79) into (7. = .87) <9 2 „9 2 sinh21 i? = A + B f / ^ d x e " ^ / 2 " 2 = v^F<*.83) ip(x.84) we obtain* ip(x.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. . ' 2\Aswht % cosh t (7. 9 (7. + cosh 2 1.(i/2) sinh 2t 2 2 2C 2 (i) + *c^. 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.cosh t sinh t' Wisll2b ^ (7.85) = —r= o2 7 2 sinh 2i.(Ax' + Bxf \/27risinh + (B2 + ^cothi)a.124 CHAPTER 7.2 (7. (7.t) = J oo dx' Setting A2:=±we can rewrite ip as ex b2 and AB sinh t' (7.84) dx' oo tvirl/2b With some algebra one can show that (B2 2y and A2 sinh 2 t f7 s ^ sinn / + icotht) = .83) sinh 2 t + (i/2b2) sinh 2i 2 sinh £ 2A 2 sinh 2 1 (7..81) and (7. Green's Functions Inserting (7.80) we obtain „/2 oo ex P 2 s i n h t { ( : C + x')^ht2xx'} \/27usmh .86) Evaluating the Gaussian integral of Eq..t) = J oo P . A . (7.
95) . •&(t)=//c(t) = / or..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. 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.94) / dxf(x)=g'(y)f(g(y)). C2(t) (7.90) with the expected limiting behaviour \im \ilj(x.93) where —{dQ/dt)5t is the probability that the particle leaves the vicinity of the origin in the interval St around t (partial integration leads from the second expression back to the first).89) = exp[x2/C2(t)} 7rV2C(t) (7. t) the phase ip drops out and one has (x. In taking the modulus of ip(x.t)\' 2 125 (7.4 The Inverted Harmonic Oscillator The real part of Eq. We have therefore .£ 2 (7.7.91) C(t)' (7.t)\' 2 exp[x2/b2] rrVaft In the following we choose 6 = 1 . since dy Jo the derivative dT_ ~dT For b = 1 we have f°° dt= / (7.t)\2 Ji == 71 / V " Jo dO ? 6' (7. (7.86) has C2(t) := b2 cosh 2 t + 72 sinh 2 1. _2 1 exp C 2 (i) = l + 2sinh 2 i.
=o V(i v W + v2) oV "dT J L 1 f A x 2 __H_I r'=s/21 Adx'..From this we obtain C(t) C'(t) Z2 ^(r?2_. (7.(4 2 . B.126 We set I C(t)' so that dT ~d~l Now.99) .97) and obtain dT ~dl or with £ = ?y/7: 2 2 d^e.2)(p_7?2)(7.98) We insert this into Eq.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. 136.= \ / l + 2sinh 2 t. C(t) = . we obtain dT _ 1 ~dJ~l 'See for instance H.96) °° l Jr. Dwight [81].=0 V^ C'(t) I Jt=0 V^ (7..1 ) 2 \n J ^"•""ff^^V^K?. + 0(e'2) (7. £ 1 dr\ 2Z %_ rl 2 2 yfx Jr. p. V and (cosh t = y 1 + sinh2 t) 1 r CHAPTER 7 Green's Functions dr} dt C'lt) ' C(t)' (7.97) sinh 2 1 = .
(7. p.7. is instructive in revealing the basic quantum mechanics involved in the finiteness of the sojourn time. this equation is the equation of motion of this scalar field with negative masssquared. also motivated by this paper. 238. Explain the parameters entering the calculation.100) Reintroducing the dimensional parameters we had set equal to 1. The equation there describes the classical "rolling down" of this tachyon. 2 2 2mo mow x The classical equation of motion is therefore' it — uj x = 0 (7.4: The sojourn time calculated semiclassically A tiny ball of mass mo is placed at the apex of an upright egg.3 How stable is the particle on top of the eggshell in quantum mechanics? E x a m p l e 7.g. Fig. T h e following Example 7. B.102) 'As a matter of interest we add that in the theory of a free scalar field x. See e. this is T ~ ]n(ly/mou>/h).4. Zwiebach [294]. It follows t h a t T~lnZ.4 The Inverted Harmonic Oscillator 127 for I sufficiently large. . Such a field arises in the spectrum of states in string field theories and is there called a "tachyon". Solution: In the case of the inverted oscillator (representing the egg) we have in the usual notation the classical Hamiltonian Hwith time derivatives dH dp P mo ' max = p dH dx = mow x. 7. Calculate with the help of a semiclassical consideration the (quantum mechanically limited) maximal length of time which passes before the ball is observed to roll down. U) (7.101) A more detailed evaluation of the constant can be found in the paper of Barton cited above.
p(0) cannot be determined with arbitrary precision. 2 x(0) ^ where A i .5)) Ax = <J(x2){x)2. (x 2 ) h 2mou 2mou> and so l^J^ULl).e.102)) p(0) = ^/(p 2 >. For h —> 0 the time T —• oo in agreement with our classical expectation. (p2):=mW(x2). For a symmetric state like the usual ground state of the harmonic oscillator we have (since the wave function is an even function. ^(x2)(p2). We ask: At what time T > 0 does the ball reach the point with horizontal coordinate x = I. Ap= v /(p 2 )<p)2. Green's Functions x = Aco sinh cot + Bco cosh cot. .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. 2V ' Let A and Bio be the values of x and x at time t = 0. and solution x = A cosh Lot + B sinh cot. Eq. x(0) = A. but instead are subject to an uncertainty relation of the form AxAp > h. Thus we set I = A cosh LOT + B sinh coT ~ (A + B)e"T.128 with x2 = LO2X2 + const. 2 10 x(0) = B u .e. the expectation value of x is the integral with an odd integrand) <x) = 0 = <p). so that x(0) + il^L mow It follows that For a minimal uncertainty (hence the factor of 2 in the following) we then have h= 2 Hence uT P ( ° ) _ O . Thus quantum mechanically x(0). i. Quantum mechanically x(0). (7. so that e. (7. the smallest macroscopic length. so that AxAp= In our semiclassical consideration we set therefore x(0) = y/{x*). v A x A p = \ (x2)(p2) v = m0u(x2).g. A p are defined by the mean square deviations given by (see Eq. (10. and for simplicity we take (cf. CHAPTER 7. i. pifi) have to be replaced by the positions of the spatial and momentum maxima of a wave packet. so that raoto _ We assume now that at time t = 0 the ball is placed at the point x(fi) with momentum p(0). / 7 ^ 2 =2J(x }.
1 Introductory Remarks The Schrodinger equation can be solved exactly (i.2h2 cos 2 a # = 0. 129 . in closed form) only for very few potentials.e. It follows that in general one depends on some approximation procedure which is usually described as a perturbation method. In the case of the harmonic oscillator potential ax2 the Schrodinger equation can be solved exactly with ease. but this is no longer the case for an anharmonic oscillator potential like ax2+(3x4.e.Chapter 8 TimeIndependent Perturbation Theory 8. i.E^ is already a bigg problem. In general perturbation series do not converge. An example permitting convergent perturbation series is provided by the trigonometric potential cos 2x with onedimensional Schrodinger equation given by ip" + [E. 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. For mathematical purists the question of convergence of the series is an even bigger challenge. one would set tt E = = ^(°)+/ty. EW+(3EW+(32EW +  ((3 can also be thought of as a kind of "bookkeeping" parameter in retaining corresponding powers of some kind of expansion).(i)+/^(2) + ". Frequently even the calculation of the next to leading contributions ip^.
.g. ^For further details. Meixner and F . 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. W. the expansions i> = v (0) + /*V (1) + /*V 2 ) + • • •. (8. . at the end of Chapter 26) — and by comparing the results of WKB. T. (8.2 However.1) and its terms with the behaviour of the following series and its terms: f(x) = 1 . h hl are in the strict mathematical sense divergent. Cll E = a_ 2 n + a_i/i + a0 + — + —r \ .2) n—»oo n It is interesting to compare the behaviour of the series (8. The series n=0 of the exponential function is well known to converge absolutely for all real and complex values of x. the expansions in descending powers of h2. 581. It will be seen later (e. Watson [283].3) *See e. the expansions in ascending powers of the parameter h2 can indeed be shown to have a definite radius of convergence.e.g. ..g.• + (ir~ X X* 1' 2' r?l Xn + Rn(x). 0. N. which is known as the Mathieu equation. perturbation theory and path integral methods — that the lightminded way in which perturbation theory is sometimes discarded is not justified. E = E^ + h2E^+hAE^ 71 72 7 + . In the following we explain the difference between convergent and asymptotic series.* i. see e. This convergence can be shown with D'Alembert's ratio test since''' lim ..g. concerning also the uniform convergence of the exponential series. p. J. TimeIndependent Perturbation Theory In the case of this equation. 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. Whittaker and G. E. explore the latter in somewhat more detail and finally consider methods for deriving perturbation solutions of the Schrodinger equation. Schafke [193]. e.+ ..= 0 < 1.130 CHAPTER 8.
we have ^ ' ~ = 3 + 9 ~ 2 7 + 8 1 ~ 2 4 3 + 7 2 9 ~ 2187 + " ' ' 1 . + 0. Comparing the asymptotic series (8.e. and in fact increase indefinitely. . We observe in the first place that since lim n—>oo co (8. the series (8.. i.1).g.000 000 06 + • • • .4) for every value of x..33333 . begin to increase again.0. 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. It is evident that the larger the value of \x\.l n ( l + ^ ) n=l L ^ ' (8..222. for larger values of x. / ( I ) = 1 — 1! + 2! — 3! H • However.g. 235. E...9876. the better the approximation obtained.0. after reaching 6/27 = 0.0. for x = 3. 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.001 + 0..0.. This type of behaviour is characteristic of the terms of an asymptotic expansion as we shall see in more detail in Chapter 20. N.4938. + •••. /(1000) = 1 .5) Cf.0. and we see that the moduli of successive terms first decrease and then.. However.29629. The theory of asymptotic expansions claims that if the expansion is truncated at the least term.3) with the convergent series (8. Whittaker and G. .. + 0.22222 . Normally a divergent series is characterized by an ever increasing behaviour of its terms as in the case of x = 1.000 002 00 . T.2 Comparison of Asymptotic Series with Convergent Series 131 where Rn(x) is the remainder sum. As a second example we consider two series expansions of the gamma function or factorial T(z) — {z — 1)!. Thus e. +0.3) can still be used to obtain almost correct values of f(x) for x sufficiently large. ..2. Watson [283]. ..8. e. p. . From the Weierstrass product which defines this function* one obtains the series _ M*l)! = 7(*l) + E [ ^ .376 .3) diverges for every value of x..22222 . in spite of this the series (8. . We can see this as follows.
J In2. p. Merzbacher [194]. 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. . Messiah [195]. ^E. This is not appreciated by mathematical purists. 16. at a later point in his treatise Messiah admits that the expansions are mostly § Cf.132 CHAPTER 8. for ln(z — 1)! one can also obtain the Stirling series ln(zl)! = ln(2vr)5 z+ (z. On the other hand Schiff" says:" We assume that these two series (for \I/ and E) are analytic for e between zero and one. Sec. Schiff [243]. In fact. 11 L. 152.5772157— Here the series on the right is an absolutely and uniformly convergent series of the analytic function. TimeIndependent Perturbation Theory where 7 = 0. 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. (16. Thus one can say that the vast majority of expansions of this type in physics is asymptotic and not convergent. In many respects Messiah aims at more rigour in his arguments.1." These "rapidly converging power series" in physical contexts are most likely very rare cases. that E and \ijj) (i.1 between Eqs. it is sensible to make the assumption. **A." This is a very clear statement which does not try to pretend that a perturbation expansion would have to be convergent.41) and (16. p. It is interesting to observe the comments of various authors on this point. § However. It seems that Schiff tries to cling to the idea that a proper series has to be analytic. It is inherent in the nature of an approximation in a physical problem that in deciding between dominant or primary effects and those of secondary importance. 371. p. ip) can be represented by rapidly converging power series in e. in fact. Watson [283]. E. It is therefore not surprising that he** says: "7/ the perturbation eV is sufficiently small. whereas applications of the convergent series (8. Whittaker and G. Applications of Stirling's series can be found in all areas of the physical sciences (particularly in statistical problems). I.5). Vol. the leading term of an asymptotic expansion.e. meaning convergent. This latter observation hints already at the importance of asymptotic series in applications. the dominant approximation is. 236. N. although this has not been investigated except for a few simple problem^.5) are practically unknown. T.
Kato: "Indeed the perturbation expansion is in most cases an asymptotic expansion . 198 (German edition). we can rewrite Eq. Replacing t h e exponential in Eq. Merzbacher.2.9) 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. there is no need for such bias. Messiah [195]. which have been studied in great detail. B.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. . It is therefore suggestive to write iy poo <t>(x) = 1 . ft Jx du = 2tdt. (8..8. (8. 8. 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.ft somewhat afraid t o say so himself a n d therefore with reference t o investigations of T .. We know from books on Special Functions t h a t all of these functions.11) A .= / e~t2dt.10) Vn Changing t h e variable of integration to u = t2x2. (8.2 Comparison of Asymptotic Series with Convergent Series 133 asymptotic.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.^=e~x2 / e^^dt. Vol. solutions of second order differential equations. (8. In fact.7) This function has been considered in detail in t h e book of Dingle. p. possess asymptotic expansions which have for a long time been important a n d accepted standard results of mathematics. 2 (8.* and we follow some of t h e considerations given there.7) as / e'* dte* dt = 1 . *R. (8." Of these three authors t h e first. He says..
« \~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. (8. 7T (ni)!sin{7r(n+^)} ( „ . we obtain 1+ i.^ X e U ~ ( 1 + * \ 2 .13) presupposes t h a t < 1.15) < 1 we can insert this expansion into Eq.^e~x \Ar i. 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. TimeIndependent Perturbation Theory <P(x) = 1 .134 we obtain CHAPTER 8.e.i)[(_!)«' Then. du. since (—4)! = y/n.' we have 2 nH 7T 7T (8. Using the reflection formula (*!)!(*)! = for z = n + i .12) VK Jo n V oo 7 n=o x) 2 2*Qn\(±n)\\x2) (8.14) sin7T2.e. (8.12).16) . 2 9 f00 / Jo du e" 2(u + x2)^ u <£(*) = 1 T h e binomial expansion .
—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^. much more useful than the convergent expansion. In an asymptotic expansion the first few terms successively decrease in magnitude. In the foregoing discussion of the error function we assumed that x was real and positive. and in fact so slowly. .e. /•oo T(n) = (n .19) implies either that we ignore the remainder or the correction term given by the integral in Eq. the asymptotic expansion of the error function is. i. i. We can actually understand the deeper reason for this in practice. argx = 0. We now see how the asymptotic expansion (8.r<7r.18) or that we insert the binomial expansion (8.2 Comparison of Asymptotic Series with Convergent Series 135 We can evaluate the first integral with the help of the integral representation of the gamma function. whereas in a convergent expansion the terms first increase in magnitude. The expansion (8. the fact that we effectively use a binomial expansion beyond its circle of convergence implies that the resulting series is divergent so that even if the first few terms decrease in magnitude the later terms will increase eventually as a reflection of this procedure. We now want to relax this condition and allow for phases argx ^ 0. (8.8. As in the case of the gamma function.17) 0(X) = i. Whichever way we look at the result.19) originates.e. Frequently "=" is written instead of "~".12) ignoring the fact that the latter is valid (i.19) approximates <f){x) the better the larger x is. ' arg.e. in general. that a large number of terms has to be summed in order to obtain a reasonable approximation of the quantity concerned. (8. It is fairly clear that the above considerations remain unaffected for argx < 7T.19) where "~" means asymptotically equal to which in turn means that the right hand side of Eq. (8.15) into (8. (8.l)! = / Jo Then 2 cHnxdt. convergent) only in the restricted domain u < x2.
20) where we set t = is.vds. when argx = 7r/2. thus with \v/y2\ < 1.^ f V e ^ .e. (8.24) We can read off the binomial expansion of the factor in the integrand from Eq. however. we can try to proceed with Eq. We therefore consider this case in detail. Proceeding along lines similar to those above we write <f>{iy) = ^=ey2 I* e^^ds. 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). we set x = iy in Eq. TimeIndependent Perturbation Theory since for this range of phases the decreasing nature of the exponential is maintained. (8. =dv. (8.21) = y2s2. dv = 2sds = .136 CHAPTER 8.2 y V .^ = .7) and obtain 4>(iy) = —= j e~l dt = = / es2ds.15). (8. VK Jo V^ Jo (8. (8.23) Throughout the range of integration 0 < v < y2 the integrand is real. The situation becomes critical. V7T Jo Changing the variable of integration to v (8. (8. We can therefore rewrite the integral as (j)(iy) i 2 fy2 y —=e / e~v 1 . i.22) we obtain 2 4>{iy) = .26) or else return .
i.24). Eq.30) yvn f Jo V y J Cf. 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\. ey is much larger than something of order 1/y.. and the larger y. we obtain 4>(iy) = ^ey2TZ f°° e~v (1 .28) because then the correction term oc f°£ becomes smaller and smaller. (8.e. (8.24). We know that for v < y2 the integrand is real. for large values of y. If we proceed with Eq. (8. and so the integral is real. However.^] * dv. .27) Then .28) ignores the integral contribution of Eq.19).27) and (8. the better the approximation expressed by Eq.2 Comparison of Asymptotic Series with Convergent Series to Eq. 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. i (8. — (8.8. „2 \v J v^(ni)! y(nj)l ^—' y2n ie"2 /00 ^ ieir /•" W7r /„2 y ^—' _v^{n\)\(v n! . e ^ ( i y ) ^ ^ . We have therefore <j>(iy) = ^e^TL fV e~v ( l . (8. Returning to Eqs. (8.^ ^ p .29) where 5t means "real parf.26) we can write* 137 v* Jv2 ie « „2 OO / 1 n=0 M ~k n! .28) n=0 We observe that this expansion differs significantly from (8. For v > y2 the second integral is seen to be f imaginary and thus drops out in taking the real part.28) we observe that the expansion (8.27). Suppose now we consider Eq. (8. (8. which will be studied in more detail below. y y 2 ( n _i)! y arg(* = iy) = i * . (8. i.e.2 yir (8.
(8. (8.138 CHAPTER 8. y. (8. Since (cf.e.8)) 4>(x) = <f>(x) we see that _x2 OO / ] \  n=0 for 7r/2 < I arg x\ < 3TT/2.e. This result is identical with that obtained previously.31) and (8. we use the expansion outside its circle of convergence and hence obtain a divergent expansion.* We observe that this Stokes discontinuity is a property of the asymptotic expansion but not of the function itself. e xZ ^ (n — £)! „ ^ ) . 7r = . With a similar type of reasoning one can show that the same expansion. (8.25) and integrate from 0 to oo. TimeIndependent Perturbation Theory If we insert here the expansion (8. Looking at Eqs. Dingle [70]. .30) can be written </>(iy) = ^ey2 fV ev(l^) yVn Jo V y J "dv = ^!rV"E^(4)V (8.3!) which is (8." — E ^ ^ forarg. since (8. For the sake of completeness of the above example we continue the phase to  arg x\ > IT/2. (8.. Eq. is obtained for <f>(x) with arg x = —ir/2.2 OO rz=0 V . R. . We have therefore found that XTT n=0 ^ ' and .26). i. and • higher order terms of the associated series all have the same sign.33) we see that the Stokes discontinuities occur at those phases for which these expansions (i. B.33)) would have • a maximally increasing exponential e+x . Chapter 1. *Cf.28).31). (8.33) The sudden disappearance of "1" at argx = 7r/2 hints at something like a discontinuity of (j>{x) at arg x = n/2 which was discovered by Stokes and is therefore known as a Stokes discontinuity. The corresponding phase lines are called Stokes rays.
1/1 ipv{x) v1 vl (±ix).8. (6. Chapter 1. 9 . e. (8. Eq. which can be written (for v not restricted to integral values) D^(x) DM(x) f xu(f>u(x). 5. and 4>(x) argx=7r/2 = 2 r2 ^ + <f>(x) a. though. 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.36) has an exponentially decreasing solution and an exponentially increasing solution for zero phase of x. Later we shall make extensive use of parabolic cylinder functions Du(x) which are solutions of the equation dx2 + v+ :X y =o .13. in the sum of the right hand sides of Eqs. pp. These rules. Thus if we set x = XR + ixj. p.2).2. i.g. for arg x = 7r/2. 91. apply predominantly to asymptotic series of the solutions of second order differential equations. Equation (8. Dingle [70].32) and (8.37) Cf.rgx<n/2 argx>7r/2 _ X1T •<—' v — ( n=0 £ (ni)! X2)n ' oo (8. . Oberhettinger [181]. i. R. and so may not be universally applicable. See also Tables of Special Functions. —1 cancel out.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.e. Magnus and F.e. B. W.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. we have e z2/2 _ 2 e(x Rxj+2ixRxI)/2_ The exponential is maximally increasing for XR = 0.36) 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.35) 8. '"Cf.34) the contributions 1.
= 7r/2. 1 1 Cf. v — — f — 1. 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. Thus D{J\x) (i.140 where^ CHAPTER 8. > > The equation (8. Chapter 17. 8. We consider Dv (x). (8. as can most easily be seen in the case of the Mathieu potential.7. Hence Dil){x) = xv(j>v(x).e. R.40) § Cf. 0 < argz <  (8.ei<v+VxvHv{x\ argx = £ . i.36) is invariant under these replacements. 9) and XXI. ' S u c h symmetries are extensively exploited in the perturbation method of Sec.e. x~l/~1ipu(x). (838) We observe that one solution follows from the other by making the replacements^ x — ±ix. Chapters I (in particular p. Chapter 1. ti asymptotic series (8. Eq.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 2 oo «*)  — E^fi=0 v . .37). We also observe that the late (large i) terms of the series in (8.39) (as in the example of the error function).e.39) onto the Stokes ray and from there into the neighbouring domain is determined by the following rules which will not be established here:" • Dingle's rule (1): On reaching arg a. R. (14).38)) has a Stokes ray at Dl '(X) the arg x = 7r/2. Dingle [70]. i. At arg x = 7r/2 the exponential factor becomes an increasing exponential and late terms of the series have the same sign. B. TimeIndependent Perturbation Theory i. B. The continuation of (8. with a real proportionality factor. (8. D^(x) = = xv4>v{x)Jr\^l'1^'Kl'1\A~v~X^^) xv4>v{x) + \a. Dingle [70].
8. The value of Du (x) on the other side of the Stokes ray.e. = ir. (8.2 Comparison of Asymptotic Series with Convergent Series 141 for argz = 7r/2. for xj = 0) and therefore possesses a Stokes discontinuity there. Applying rule (2) we obtain D$\x) = U™ + ^iap) (xyMx) + a{x)vlMx) (843) for IT < arg a. i.e. i. Since Dv (x) is real when x is real. In (8. where in the first line on the right hand side the second term contains the real proportionality constant a/2. < 7T. another half of the discontinuity appears on leaving the Stokes ray on the other side.44) + ^iap) = 0 . the extra phase factor u ei7r/2 a n c j t n e p h a s e ew7r/2 0 f x <j>u(x) (so that as the rule requires the added contribution is 7r/2 out of phase with the first). e +fx2 _ 2 2 e±(x Rx I+2ixRxI) is maximally exponentially increasing (i.41) xuct)v{x) + a{x)~u~lipu{x).42) el™ + ±iafi\ {x)v<t>v{x) + a{x)vl^u{x). On reaching ir the part containing ^ ( x ) .41) to arg a. = ir. is given by • Dingle's rule (2): One half of the discontinuity appears on reaching the Stokes ray. beyond arg x — 7r/2.40). we must have at arg x = 7 in the dominant factor on the right of Eq.e. We therefore proceed to continue (8. sm(iri/) = —a/3/2 or a/? = 2sm(irv). as yet an unknown real constant. It is determined by continuing the asymptotic series to argx = n and demanding that the result be real since Do (x) is real when x is real. < 37r/2. (8. Applying Dingle's rule (1) (to the Stokes discontinuity of ^v{x)) we obtain on the ray D<P(x) = x^M^ + aUxy^M^ + l^e^e^lxrMx)] (8. Thus for arg x > ir/2 we have D$\x) = xp(t>u{x) + = 2]iaei<v+^x1/l^v{x) ^ < arga.e. (8. (8.42) T ^(eiwu i. with \x\ = (—x) for arg a.41) the Stokes multiplier a is.
45) that af3 = 2sin(7r^)) D^\x) = (8.1) (8.1). Eq. (8. a{y)P{y) = a(u . 8. Dl \X) is (cf.1) = 2sin7ri/. i. (8. a(u)P(v) = 2sin(7r^) = 2sin7r(i/ . (8.47) Equation (8. or sin7r(2: + 1) sin irz' W ( = 7 ^ ) ! = 2sil"r2Comparing this with Eq. .50a ) (aMb) »W = f "M=(=^)i In order to decide which case is relevant we observe from Eq.51) .e.1)! TT.e. (8.42) that a multiplies ipv(x).45) Thus the right hand side of this equation remains unchanged if v is replaced by —v — 1.14). i.1) = ±a(y).v .48) We compare this with the reflection formula (8.e.42)) for argx = ir (in the second line of Eq. (z)\(ziy.46) For a given function a(y) this equation determines /3(u).42) we insert from Eq.38) = cos7ru(x)u(f)1/(x) + (—x)ve~*x COS7T^— —r a(u)(x)l'1^u(x) (2i . argx = vr. TimeIndependent Perturbation Theory Now a and (3 can still be functions of v. (8. (3{v .e.142 CHAPTER 8.48) we can set <8 49) ' ^ ) = {_V\)V PM = ^f> or ( 8 . V i=0 \U0 v 2w . We can see that the equation is satisfied by (3{u) = ±a(u . i.l)0(v . (8. Hence the left hand side must have the same property. (8.1) (8. i.45) is therefore given by a{v)a(v .—^ ON • — ^ > + ^ ^ L .
(2inl)! (n1)! n! {n2i)\ So far we have been considering the asymptotic series expansion of Di.e. ' ^ (i/l)!z! ^ . (8.1 ) ! 4~L i\{2x2Y i=0 V2^(x).50a) because this enables us to define normalizable parabolic cylinder functions. It should be noted that in Eq. —f.36).53) We have therefore obtained in a natural way the quantization of the harmonic oscillator.) (8. However. Then (in the second step using (—u — l)\v\ = — ir/sm(Trv)) DW(x) ~ c o s ^ ^ * f ^ .35). Eq. the second contribution vanishes and we obtain £>«(*) . (8.36)) in the domain 0 < arg x < IT..e. the error .52) (argx = 7r). which vanish at x = ±oo. 8.2. Proceeding along similar lines we can continue the expansion into the domain (2) ir < arg x < 2ir. ' (X) (cf.7).. (2x2)1 = COSTTU(X) v . (8.)(.2^1 v y 7T i=0 ' (2i + v)\ z!(2x ) v (8.e.8..e^ .A / ± sin^Cx)"M* 2 V ^ .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. (8. Chapter I. solutions of Eq.1.1 ) ( .I / .53) the factor (—n — 1)! is to be understood in association with (2i — n — 1)! in the numerator..*"ei°> f ( ^ _  _ L _ = (!)««(.„ _ix2^(2ivl)l e 2 > —. i..7r). For details we refer again to the book of Dingle. u 1 y> {2i + v)\ 1 n 0. i.3 Derivation of Asymptotic Series from Differential Equations 143 We choose (8. We observe that for v — n = 0.. i. (8. In a similar way we can examine Dv (x) and its Stokes discontinuities (at argx = 0.
e. x—>±oo (8. (8.36). Eq. • • • can be determined.54) we consider again the important equation of the harmonic oscillator. x.36)) 0 where + x2Q(x)y = 0. 5„=4 Sl=o. S\. \ + U (8. (8. how series of the type (8.55) The large x asymptotic expansion can also be obtained from the differential equation. previously we had ^. Thus 2S 0 Si 5o + 25 0 5 2 + ^ + ^ and so on.144 CHAPTER 8. + — + ^ + • • • .. Instead of dealing with Eq (8.54) (8. Solving these equations we obtain = = 0. ^ = 0 dxA dx with the boundary conditions lim 4>(x) = ± 1 . ^ + 2 . and demonstrate.58) Since x2 is the highest power of x appearing in x2Q(x) we set S(x) = ]S0x2 + SlX + S2 In a.38) can be obtained. S2..e.37). (8. cf.56) ~l^(8 57) Q{x) =  We then set y = es^ so that the equation becomes (8. 0. i. (8.. we obtain equations from which the coefficients So.59) and equating to zero the coefficients of powers of x2.^ + 1 ) . % = {.36) in the form (coefficient of — x2 here chosen to be ~.60) 2 x x l Inserting (8. We write Eq. i. x°.60) into (8. TimeIndependent Perturbation Theory function can also be obtained as the solution of a second order differential equation. (8. Eq.
etc.e.3 Derivation Then of Asymptotic Series from Differential Equations 145 eS(x) = 2 e±^ xS2 1+ 0 (8. i.19). 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+ .37) with (8. therefore.35) and (8.91.8. 2 X 2Wi.38). In particular we shall need the asymptotic expansions of parabolic cylinder functions in various domains. In the following we are mostly concerned with asymptotic series in some parameter. do not enter here into their derivation and simply quote the result: W i t h \x\ .63) **W. (8. (8.37). The derivation of the asymptotic series from integral representations (cf.38) in a variable for the solutions of approximated differential equations. T h e expansions (8.b. Magnus and F. n 22e"T JziL (8. 4~l~2 ' i i _i. / i N . function.+ X M W = Q.61) (the dominant terms of (8.36) have correspondingly e±x / 2 ) . are asymptotic series in a variable x. Nonetheless we shall need asymptotic expansions like (8. and it is clear t h a t higher order terms follow accordingly. » 1. Oberhettinger [181]. . p. We. 2 ) + V2(l)! lt1! 2 1 — v 3x\ 1 2 2' 2 where iF\(a. Whittaker and Watson [283]) is now somewhat elaborate (although in principle the same as before).z) is a confluent hypergeometric T.62) 4 \ 2 XiFi 2'2 . \x\ ^> \v\ and (a) for  a r g x  < D„{x) 3/4TT: e 4 x x i/(i/l) 2x2  v(y\){V2){yS) 2Ax2  . a z a a ( + 1) z2 which has the unit circle as its circle of convergence.
67) Cf. Thus there is no contradiction^ between (a) and (b).4x 2 27F (*/!)!  e^e^x""1 1 + (v + l)(i/ + 2) 2x 2  (i/ + !)(»/ + 2)(i/ + 3)(»/ + 4) 2. Whittaker and G. T. 8.0. TimeIndependent Perturbation Theory (b) for 5/47T > arga? > n/4: Dv{x) ~ e 4 v x v{y\) 2x 2 V{y\){y1){yS) 2. whereas e~x '4 increases exponentially. Watson [283]. 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 3.64) (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 " ' (8.66) 7 =e(^) 2 i(3x)2ei(^)(?Jx) decreases 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.146 CHAPTER 8. 349. E. 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. top of p.4 Formal Definition of Asymptotic Expansions Finally we introduce the formal definition of an asymptotic expansion. If Sn(x) is the sum of the first n + 1 terms of the divergent series expansion .4x2 (8.1 a2 an a0 + — + ^ + .+ — + ••• of a function f(x) for a given range of a r g x . 4 (8. N. .
5 RayleighSchrodinger Perturbation although Theory 147 lim \xn [f(x) . (8. we set E^En * > ^ n = En°1+eEnV = + e2EnV + .Sn{x))  = oo n—>oo (8.73) Vn + q/#> + e2T/i2) + • • • .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).e.68) is (effectively) simply a statement of the observations made at the beginning. lim xneW x—>oo = 0 (8. first published in 1902. (8. .68) The definition(8. Chapter I. It may be noted that Whittaker and Watson's internationally aclaimed text on "Modern Analysis". A critical discussion of the definition can be found in the book by Dingle [70].69) and the consequent nonunique definition of the expansion. which is assumed to be small. The definition is due to Poincare (1866).72) For the eigenvalue E near En and the eigenfunction \l/ near ipn we assume power expansions in terms of the parameter e. thereby (unknowingly) discarding the vast majority of expansions which have been used in applications to physics. 8. In fact. 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. The equation to be solved is the equation H<S> = EI/J with H = H0 + eV. possesses a whole chapter on asymptotic expansions. (8.g. the divergent nature of asymptotic expansions together with a vagueness of definition (which is avoidable as explained by Dingle) frequently tempt mathematically prejudiced purists to turn away from asymptotic expansions.8. in particular with regard to possible exponentially small contributions to the expansion for which e.
J f°° dxrn E(^ 0) . 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.n.e.76) The first equation is that of the unperturbed problem. so that we choose / • dxrpM^O.4 0 ) M = / ^C(4 1} . (8. i.w » . (879) We multiply this equation from the left by ip^ and integrate over x.77) and obtain {Ho ~ 4 0 ) ) E "i^i = (Ein] ~ W n . (8.77) The states \tpn) of the unperturbed problem span the Hilbert space"K.70) and equate on both sides the coefficients of the same powers of e.n = 40)V>n.V)i. With the help of the orthonormality condition (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. We obtain then the following set of equations: Hrj.As a consequence the state vl/) is a vector in this space. (8.75) HoW + vW = E^+E^+E^. TimeIndependent Perturbation Theory We insert these expansions into Eq. We insert the ansatz (8.e. ( ^ .e. (8. The second equation can be rewritten as {Ho . i. .72) we obtain oo roo dxifc{H0 OO / ~4 0 ) ) E • / °i ^ = / J— OO d x < i E n ] ~ V )^n.i4 0) )V4 1} = {EM . i.148 CHAPTER 8. (8.74) (8. ^ ) = 0. °° i J or 0 = EWJdx{rnVil>n).78) into Eq. Ho^+V^n = EW^+E^n.
i.Viprt e (8.a\^(E:S0) „• 4 0 ) )^ = / J—oo dxrl?m(E$V)1>n. (8.V)^ Setting + 42tyn.i ^ 0.^(EW ~ V ) ^ + E& / J—oo dx^miln . (883) Thus the procedure forbids the equality of any E\ '. we return to Eq.76) which we rewrite as (H0 .e.4 0 ) M 2 ) = ( 4 1 } . the procedure forbids degeneracy and we can consider here only nondegenerate eigenstates.79). (8. (8. (8.5 RayleighSchrodinger Perturbation and so Theory 149 E& = {n\V\n) = Wn.Vil>n) (880) In order to obtain ipn . We therefore consider Eq. (il)i. ^ n ) + 0(e 2 ). multplying Eq. S ' ti^+Q^)• We observe that Eq.83) makes sense only if Ef] ^ EnV for all i + n.e.82) *> = ^ + E .84) from the left by ip^ and integrating over the entire domain of x we obtain coo J —( oo />oo / oo dxi. i.8. the coefficients a\ . (8. and hence It follows that to the first order in the parameter e: En = 4 ° ) + e ( ^ . In order to appreciate the structure of these expansions it is instructive to go one step further and to derive the next order contribution. It is clear that we can proceed similarly in the calculation of the higher order contributions to the expansions of En and \I/n.m ^ n and integrate: oo /•oo / oo dxrmY. with En . We multiply the equation by ipm.84) (885) ^2) = £ a S 2 ) ^ (excluding i = n for reasons discussed above).
90) + £ ( 4 0) 4 0) )(4 0) ^ 0) ) ^ +' .88) (Q) Inserting Eq. ipn ) = 0 i.86) E^ iy^n JJn J J i ' i^n For m ^ n w e obtain from Eq. TimeIndependent Perturbation Theory Setting m = n we obtain with (^„.88) into xjjT we obtain similarly * r i + f ^ i^n (lpi.79) •/—oo oo / Using (8.Vlpn)(lpn.89) • • • where we guessed the term of 0(e 3 ).150 CHAPTER 8.£i°») 2 Hence +^£o (£«? . FVn) + e 2 ^ ( V > n .^ ) + $& w £ (^fc. En0) (8.85) together with (8.^(8.Vlpn)E.^w . ~Ej E E.Vlpn) E^ .V'>pTl (E^Eny V'fc (8. n) £f) (8. Inserting the expression (8.4 0) ^+ e 2 E (lpk.VV .^j)(V .V^j)(^j.B i 0 ) ) ( £ f . (8.V^n) .V4>n){lpn.86) into the expression for En we obtain En = 4 ° ) + e ( ^ n . j.e.Vlpn] M _ P (0K 2 ^r . ^ n ) j&iift + ^n .2 (V> m . (8. V ^ i ) X (Q) (Vi.87) (i>k. (8.(1) (i&0) .81) we obtain (2) _ _ {lt>m.
e. One would prefer to lump it together with the other contribution.g.(0 4>k The sum J2j^n includes a term with j = k which is exactly cancelled by the other contribution.n+i = Ckn(2) We can now rewrite ipn ' as / (2) _ V"^ c ^ n knc0 ~ 2j ~ TJfi)0 ) k^n (4 ^) 2 n(0). The functions ipn(x) are eigenfunctions of some secondorder differential equation.90) is the term with only one summation J2 m the contribution of 0(e 2 ). i.. (8.5 RayleighSchrodinger Perturbation Theory 151 The expressions (8.Vlpn) = y^Ci(lPk. 8.. Merzbacher[194]).7 is based on this procedure.90) can be found in just about any book on quantum mechanics. in the form i Adopting this procedure for the perturbing potential V(x) we can set V(x)i. These functions generally obey one or more recurrence relations which are effectively equivalent to the differential equation in the sense of difference equations. We can therefore rewrite tpn ': The perturbation theory described in Sec.lpn+i) = ^ Cj5k. + l^t /„(0) V~^ c kjCjn &(Efi»Ei )(EfE&»)\ 0) „(0)w. We then obtain (lpk. These recurrence relations allow one to reexpress expressions like Xm^n as linear contributions of ipi with constant coefficients.91) The coefficients Cj follow from the recurrence relations.n(x) = J2crtn+ii (8.8.89). sometimes expressed in terms of projection operators which one can construct from the states \ipm) (see e. An ugly feature of the expansion (8. .
93b) The lowest order eigenfunctions belonging to En. are allowed to be equal since otherwise the expansion becomes undefined. however.94a) into Eq. i. in a central force problem all have the same energy unless the system is placed in a magnetic field.. i. Equations (8.e. TimeIndependent Perturbation Theory 8.94a) * m = c 2 i ^ n + c22tpm + e ^ + e ^ + ••• (8. (8.97b) with ipn y£ ipm.93a). that of double degeneracy En = Em . degeneracy.96) are effectively the known equations of ipnitpm and so yield no new information. is not uncommon. so that En = 4 ° ) + eE™ + e2EP + ••• .97b) are . Equations (8.97a) (8.. i. Em can now be linear combinations of ipni'ipm.ipm) = 0 = (^m.6 Degenerate Perturbation Theory It is clear from the expressions obtained above for En and tyn that no two eigenvalues En '. (H0 + eV)^n and (HQ + eV)Vm = Emym.4 0) )</4 1} = (En1] ~ V)(cu^n (Ho . . (8. (8.93b).96) (Ho . Proceeding as before we obtain Ho(cniln + Ci21pm) = En0) (cU1pn + C121pm) = EnVn.We now insert Eqs.4 0 ) ) ^ } = (Em] . . all n.95b) H0(c2l^n and + c22ipm) = En°\c21^n + c22^m) (8. Em = 4 ° ) +eE$ + e2E$ + .ipn).94a).152 CHAPTER 8.e. .95a) (8. (8.V)(c2l^n + ci 2 ^ TO ). I — 1 . For example the magnetic states enumerated by the magnetic quantum number m. (8.94b) with (ipn.Hence we write * n = CllV'n + CisV'm + € ^ + d 2 ^ 2 +••• .97a) and (8. —I.e. (8. .70). In order to deal with the problem of degeneracy we consider the simplest case. + c22^m). . m = I. Equality of eigenvalues.93a) (8. (8. (8. (8.
H0lP$) = C22[E$ ~ (Vm. and ( ^ m .ci 2 (V„.^ ( ^ .(lPn. we obtain with (ipn. ^ ) ) = CU[EW . Vl/. H0lpV) . Vi/>n)] . Vlf>m)] .H0xl>$) .C 21 (^ m . Similarly Wn.ipm ) the equations (iPn. c2iV^n + c22V7pm).^ ) ( ^ m . Wm.V^n). (^n.m.99) can now be rewritten as (frvvA \1"'Vr\ )(cn)=w(cn) (siooa) .H0^) .Vil>m)] V^m). . ^ o ^ ) ) = 0.98b) + Culpm) + cuV^m). Then (^.£ < « = ag)^) . (^m.EJ®(lPn.(^04°))41)) = EM^CuTpn (ipl. We now multiply these equations from the left by ip* and integrate. (8. ^ ) = ( V .98a) Setting I = n. Equations (8.99) and ( ^ ^ O ^ ) " ^ ^ .ipn n) = 0 = (ipm.H0il>W) = cn[EW ./>£>) = C2l[E£> . But (8.J#»a&> = 0j since ii4i = En (degeneracy).m). V^n)\ .C22(Vn. (8. V^n). t f ) = a k 1 } 4 0 ) " Eg>aU = 0.(Vn.6 Degenerate Perturbation Theory 153 inhomogeneous equations. £ aW^°Vi) . cn(ipm.cnVipn and tyi.y>m.8.
Similarly Eqs. 11. (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.Vrl>n) . However. the degenerate perturbation theory must be used. 9} ip. Nonetheless we shall see that the Schrodinger equation of the hydrogenlike problem can also be separated in some other orthogonal coordinates.154 and V (^m.6. Then / > i>" 2— = V ip Hence 0 = V ' V . \nip = ln(j> + c o n s t . Example 11. An example which emphasizes the significance of the spatial dimensionality in this context is treated in Example 8.Vl/>m) J \ C22 J m \ C22 J V ' These matrix equations have nontrivial solutions if and only if (lf>n. Solution: We consider the Schrodinger equation with potential V in the simplified form i>" + (V .E™ (Vn. (8. if)' (/>' — = —. Example 8.5). If one wants to perform a perturbation calculation in such a case.Vlpn) CHAPTER 8. .100a) determine the coefficients Cn. there is degeneracy in the case of the hydrogen atom with Coulomb potential (Chapter 11) which is based on threedimensional spherical polar coordinates r. like that providing the Stark effect (proportional to rcosO).^ ' V = — (•>!>'<t><t>'i>). i. d>" E=^.. dx i.C22. It follows that ip and <j> are linearly dependent. ip = c<p.E)tp = 0.e.e. as we saw. cannot be degenerate because there is only one quantum number corresponding to one definite energy and hence cannot be degenerate. It is then possible that a special perturbation. Let ip and < be two eigenfunctions belonging to the same eigenvalue E. just as there is no degeneracy in the case of the onedimensional harmonic oscillator (Chapter 6).This completes the derivation of the first order perturbation corrections if an unperturbed eigenvalue is doubly degenerate. 4.e.1. (V>V</>V>) = const. splits the problem into two independent onedimensional systems. i> 4> Hence there is no degeneracy. i.3).100b) determine C2i.ci2. in particular in parabolic coordinates (see Sec. TimeIndependent Perturbation Theory (l/>m.^Vm) 0.101) This secular equation determines the first order correction En • The equations (8. the secular determinant. A onedimensional system. At x = ±00 the constant is zero. .
in particular in applications to periodic potentials. Eq. In the following we describe such a procedure as first applied to the strong coupling case of the cosine potential or Mathieu equation. See also R.105) (8. one quickly arrives at clumsy expressions which do not reveal much about the general structure of a higher order perturbation term.106) (8. Muller [73]. Dingle [72] and H. the considerations given below would be valid even if D contained only the derivative of the first order. H. Applications of the method are. W. J. W.V0)<f> = 0 (8. i.104) are chosen such that the equation Dcf) + (E0 .103). anharmonic oscillator potentials and screened Coulomb potentials.7 DingleMuller Perturbation Method 155 8. The subdivisions (8. Muller [216].t Since in general a solution is valid only in a limited region of the variable. Muller [210]. and the solutions then have to be matched in their regions of overlap.e. W. B. We consider an equation of the form Dcj) + (E . cf. .103) such that if E is written correspondingly E = E0 + e.8. and such that one can even obtain a recurrence relation for the coefficients of the expansion. with some restrictions on the function V. J. T For instance spheroidal functions can be studied with this method. B. Dingle and H. J.* In various versions this method is used throughout this book.102) where D is a second order differential operator and E a parameter.104) "This method was developed in R. (8. (8. also possible in areas not of immediate relevance here. It is immaterial here whether D does or does not contain a first order derivative — in fact. a potential.e)<t>.V)(p = 0. of course. V = V0 + v. It is therefore natural to inquire about a procedure which permits one to generate successive orders of a perturbation expansion in a systematic way. the method has to be applied in each domain separately. (8. (8. We assume that V can be expressed as the sum of two terms.7 DingleMuller Perturbation Method We have seen above that although the procedure of calculating higher order contributions of perturbation expansions is in principle straightforward.102) can be written D<f> + (EQ .V0)<t> = (v. Thus the method is a systematic method of matched asymptotic expansions.
VQ) ^ s c s4>m+s = 22 s C s(Em ~ Em+s)<fim+s = ( u .Vo)(j)m+s = (Em .113) .) This result is obtained as follows. 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.V0)4>m = 0.m + s)4>m+s s (8. TimeIndependent Perturbation Theory is exactly solvable and has the eigenvalues Em and eigenfunctions (f>m.102) with E ~ Em. (8.V0)<f>m+a = 0. since D<f>m + (Em .109) D(j)m+s + (Em .m + s)" in order to relate them to "steps" from m to m + s.E0\ (the latter in a restricted domain).Em+s)(/)m+s. v = {D+EmVo)<t>m. m + s)0m+s. and so v (8. the equation Dcf>m+S + (Em+S .108) with coefficients (m. Then if ]e < \E\ and \v . the function 4>(0) = <t>m (8.m + .V0)<f>^ .+s ' (8.e ) ^ = 2 ( m . Considering (frm+s we have. s an integer: (ve)(pm = 'Y^{fn.111) s then (see below) 0(1 = E (m. (8. where m is an additional integral or nearintegral parameter (depending on boundary conditions). m + s) which are determined by the recursion formulae.107) represents a first approximation to the solution (f> of Eq. We write the coefficients "(m. Consider (D + Em= (D + E m V0)(4>W + ^(1)) = (D + Em .156 CHAPTER 8.e\ < \V0 .110) If we write the next contribution <f>W to <p(°> in the form m+st (8.
T? S The first expansion determines the solution 4> (apart from an overall normalization constant).m) 1^1^ IW .m) — —.102) to order (1) of the perturbation (v — e).8.7 DingleMiiller Perturbation Method and so (m. V ^ V ^ {rn. We observe that the coeffcients of expansions (8.114) To insure that 0 = ^(°) + f^ ) H is a solution of Eq. (8.m + s) 157 (B\IA\ s^O: 1 cs = J m _ ^m+s .117) „ . We then find altogether ^  l^m  Em+S) + y^y^ SJ.m + r) m ^ (8 118) r^O ' ~~ Em+s)\Em ~ Em+r) together with n _ / x y^(m.110) and are therefore related to the inverse of a . Eq. (8. . (8.119) follow a definite pattern and can therefore be constructed in a systematic way..^ (Em . /01_.119) is effectively the secular equation). . .118).116) m ^ v (m.m)4>m in (8. The factors (Em — Em+S) in the denominators result from the right hand side of Eq.m + s)(m + s.m + s)(m + s. the coefficient of the term (m.113) must vanish.m) + > ^ .TP \(T? .m + s)(m + s.m) — 0 (first approximation).e.Em+S) = 0 (second approximation).m + r)(rn + r. £ ^ T^o (^m ~ Em+s){Em — Em+r) ( m ' m + s ) ( m + g ' m + r ) $m+r (8. and the second expansion is an equation from which e and hence the eigenvalue E is determined (i. (8. . (8. Thus (m. (8.115) Repeating this procedure with cf)^1' instead of 4>^ we obtain the next contribution 0(2) =YY together with (m.Q (m.m) E J^ ( m ~ Em+S) ' • fa.n\ ^119) + .m + s)(m + s. .„N This procedure can be repeated indefinitely. (8.
I2q + 64) + 256Z 3 (41 g . 2 k2 + g2 = ga(2l + q)^[3(q2 „3 + l) + 4(3q~l)l + 8l2] 32ag r . (8. n = 0 . i. 1 .[ 4 ( 8 5 g 4 + 2q2 . S. as will be seen later.6q + 1)1 + 24(5? .9) + 4096J4] + 0 ( l / g 3 ) . .120) n4 . This means the solution is not normalized. The eigenvalue expansion obtained is an asymptotic expansion. Kaushal [146]. this is — so to speak — the price paid for obtaining a clear systematics which is essential. and when the solution is multiplied by this factor one obtains the contributions involving </>m in the higher order terms (for an explicit demonstration of normalization see Example 17.2 therefore illustrates the use of this perturbation method in the derivation of the eigenvalues of the Gauss potential.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.* Example 8. and is typical of a large number of cases. An additional aspect of the method is the full exploitation of the symmetries of the differential equation. The following Example 8. One should note that in this perturbation theory every order in the perturbation parameter also contains contributions of higher orders.[q(llq2 + 1) + 2(33g 2 . J.1)( 2 + 64J3] — . ^The potential A r 2 / ( 1 + gr2). An explicit application of the method (not including matching) seems suitable at this point.e. The result is the following expansion. for instance. 2 . i. Imposing normalization later one obtains the normalization constant as an asymptotic expansion. does not occur in any of the later contributions. A further conspicuous difference between this method and the usual RayleighSchrodinger perturbation method is that the first approximation of the expansion. for instance.38). as already mentioned after Eq.158 CHAPTER 8. . for the investigation of the behaviour of the late terms of the perturbation series. Other applications can be found in the literature.e.7 1 9 2 + 32g + 2976) 3•215g2 +32Z 2 (252g 2 . has been treated by R. since it will be used extensively in later chapters. TimeIndependent Perturbation Theory Green's function. .2).423) + Z(2720 9 3 . The systematics of the coefficients suggests that one can construct the coefficients of an arbitrary order of this perturbation theory. Miiller [211]. § H. . (f>m in the above. This can in fact be done and will be dealt with later. W. in which q = in + 3. (8. as will be seen in later chapters.
122) where A is of order 1/g. The solution i.2a2 A.b. Therefore in the general case we may set k2 + g2 = ga(2l + q) . multiply the equation by onehalf and set h = —a/g. (8.2 '(l + l) .• 0 ( ° ) = ipq(z) = zl+1ez2/4<S>(a. . 1 1(1 + 1) 1 .a + l)if>(a + 1) + (a. 1 . Then a?ip dz2 fk2+g2 2ga l(l + l) 1 z*)i> ±(T'(^*and S=z2.123) the contribution fl<°> Ah + .123) The first approximation (note the last expression as convenient notation) ^ . Setting q = in + 3 implies k2 + g2 ga(2l + q). We now insert this equation into Eq. b. a .8. . The recurrence relation for the functions ip(a) follows from that for confluent hypergeometric functions: z2f(a) = (a. 2 .2) where 1/ .z2 is a normalizable function if a = —n for n = 0 . 2 The solution XQ(S) = $ ( a . <«»> In the limit g —• oo we can neglect the right hand side and write the solution ijj —* I/IQ.121).1).l)ip(a . 11. i/> with Dq = d2 .^z2) = i>(a) leaves unaccounted for on the right hand side of Eq. Sees. „2 J„2JL. S) is a confluent hypergeometric function..b.V ! 4 I z S *! V2 ipq(z). . .7 DingleMuller Perturbation Method Solution: We expand the exponential and rewrite the Schrodinger equation in the form rfV (ir 2 . (8.0(z) = zl+1ez2^fa. Then the equation is Dqi> = Ah+y 2 4 ^ 1 1 » fc i\ . (8. . (. 159 •g'g'a'V U E Here we change the independent variable to z = y/2gar. (fc2 + <?2) 4ga and b = I \—._ „!2 ^ ( " « 2 r r 2 > . 3\ 2 V ' 2. a)ip(a) + (a.6 and 16. We then set V>o(2)=2i+1ez2/4X0(2) Now the function XQ{Z) satisfies the confluent hypergeometric equation (cf. . \2 .
the latter determining to that order the eigenvalue.a + 2]i P . a + i) = 0 for i ^ 0. one can write down recursion relations for the coeffcients of powers of hl as will be shown later in a simpler context.S m (a. a + r) = 0 for r > m. „ . fa. a + r) = i 2 \ m ?n z J 0(a) = J^ j= — m S m ( a . fa. a]i + h fa.160 where (a.3). So (a. _ J [a . a + r) + S m _ i ( a . a + r — l)(a + r — l .a + j}i+1^(a + j). The above remainder i j ' 0 ' can now be rewritten as oo i 1 R(0)=J2h + i+2 i+2 ]T (i+2) [a. a + r) with the boundary conditions Sn(a. a) = 1. a) = b . of course.1) = a . . a ] i + L ' ^ J 1 [ a + l .b = (g + 3) . The coefficients S m ( a . a + r) satisfy the following recurrence relation: Smi(a. As a check on the usefulness of the perturbation method employed here.j^0 ' + jU+1 J ^(a + j) with h[a. a + 1) = a = In general CHAPTER 8. If we now evaluate the various coefficients in the last expansion we obtain the result (8.123) and so in R(°> can be taken care of by adding to the previous approximation the contribution fitp{a + j)/j except.Z.2. 0 = h[a. i. a + j)i/>(a + m). when j = 0. .a]1=0. Now we make the following important observation that Dqip(a + j) = jip{a + j).120). (a.a]2+L ^ J1[a + 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.^ 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. o + r) + S m _ i ( a . TimeIndependent Perturbation Theory (g . (8. a . a + 111 r [a.i ^(i)=£V+i i=0 J2 j = (i+2). a + i)(a + r. In fact. one could now use the RayleighSchrodinger method and rederive the result (8. where [a.e. Hence the next order contribution to tp^ becomes oo i+2 r [a a .120).a]i = . This equation shows that a term fiip(a + j) on the right hand side of Eq.a—lli.a]i It is clear that the various contributions can be obtained from a consideration of "allowed steps". a ] i + 0(/i3) [a. a ] x + L _ 2 J 1 [ « .l .a — 211 .2a = I + q. a + r + l)(a + r + l . ( 5 m ( a . (a.
Regrettably this has to be left out here in view of lack of space. although not exactly straightforward. except for the canonical algebra which has to be formulated with Dirac brackets replacing Poisson brackets. i. We therefore begin with the appropriate classical considerations with a view to generalization. 161 . and we shall then see that these lead to analogous results as in Chapter 2.Chapter 9 The Density Matrix and Polarization Phenomena 9.e. as we shall see in Chapter 27. proceeds along similar lines.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. 9. light. and this means electrodynamics. In continuation of our earlier search for an approach to quantum mechanics as close as possible to classical mechanics by looking for the generalizability of classical mechanics — as already attempted in Chapter 2 — it is reasonable to consider also classical systems with a wavelike nature. We know from classical electrodynamics that a planar light wave with propagation or wave vector k and frequency u> is represented by a vector *So far our considerations were restricted to oneparticle systems.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. the generalization to many particles.
(9.5) which is described as right polarization as compared with left polarization for which (A0R x A 0 /) • k < 0. It follows therefore that A(r.1. When AQR is parallel to Ao/ the ellipse becomes a straight line and the wave is said to be linearly polarized.e. i. 9.162 CHAPTER 9. (9. In the latter case one has the possibility of {A0R x A 0 /) • k > 0. (9. This is the equation of an ellipse as indicated in Fig. t) given by A(r.i). t) = A 0J R cos(cji) + A 0 / sin(u.1) Fig. In general the constant Ao is complex.3) (9.oei(krwt> which satisfies the transversality or Lorentz condition k • A = 0. 9.1 The vector potential ellipse. A 0 = A0R + iA0I.4) (9.t) = 3ftA. The Density Matrix and Polarization Phenomena potential A(r.2) (9. When Ao/? is perpendicular to Ao/ and Ao# = Ao/ the ellipse becomes a circle and the field is said to be circularly polarized. t) = A(XR cos(k • r — cut) — Ao/ sin(k • r — cot) and at r = 0: A(0.6) .
however. Here we are interested in observable properties of light waves — like. the intensity of a wave. 17. A2. From the vectors A(°). p.A« we can construct only the following scalar products with this invariance: (A(°\A(°)). as e.A«).2) the vector potential A is completely determined by its two (say (x.11) 'This means we consider the polarization. (A(°). represented by a filter "F".e. I=<l2x2>:=E40)*l*40)i.e.7) We consider now measurements with polarized light.9. . i. The intensity 7 of a light wave with amplitude coefficients A\' .8). 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. i.A 2 .9) The intensity is characterized by the fact that it can be looked at as the "expectation value" (1) of the unit matrix 12x2. we recall.y)) components A i . Observable quantities can only be those which are invariant under translations and rotations.t in which k and ui remain unchanged. (9.AW). with k . (9. for instance.2 Reconsideration of Electrodynamics 163 As a consequence of the Lorentz condition (9. the fields E and B are observables. determined with the help of a polarizer. not. the connection must be linear. (AW. a calcite crystal or a film of nitrocellulose. (9. Then their connection is given by Eq. Rae [234]. i.2 *i*4 0) (98) with coefficients F^. the vector potential A. ie. / = 4 0 ) * 4 0 ) + 4 0 ) * 4 0 ) = (A(°\ A<°>) = A(°)2. which leads to an outgoing wave which we give the label "1". We set therefore A 0 = Ai + A 2 .e.k (910) We give the initial wave the label "0" and subject this to an experiment.g. (AW. AW = £ fc=l.A i = 0 = k .A(°)).A2 is the sum of the moduli. See for instance A. In classical electrodynamics. (9. the direction of polarization.
(9.164 CHAPTER 9. i.16) i. the connection between the initial polarization and the final polarization. if (F) is an observable quantity. We use definition (9. it must be real and hence (F) = <F>*. To this end we observe that as a consequence of Eq.12) i. Then according to Eq.A( 1 )) = ^ ^ F i.k (9.13). Thus (F) = J2FikPki = ^(Fp)a.15) This matrix is hermitian since (Pife)t = (AiAtf = (A*Akf = (AkA*f = {pkif = (Pik). so that (F) can be looked at as the expectation value of an observable F.e.17) . (F)=Tr(Fp). (9. (9. (9.14) We now introduce a matrix p — called density matrix — which is defined by the relation pik := AiA%.e. But we can convince ourselves that we achieve exactly the same as with our earlier consideration of observables.13) the quantity F must be an hermitian matrix and thus representative of an observable.A*kF£A* = E and hence F = Fl A p i tkAk. (9.A*FikAk. i. pki = AkA*. (A(°).13): j2^FikAk = Y^AiF*kAi = Y.k l f c 4 ° \ (9. We write the observable quantity as (F):=Y. i.13) This consideration of translation and rotation invariant quantities is somewhat outside the framework of our earlier arguments. Thus.k i (9.15) in order to rewrite Eq. (9. The Density Matrix and Polarization Phenomena Only the second last of these combinations describes the experiment.e.
But light. "incoherently.d "»=(o ! ) " In the case of an admixture with equal portions of 50%.e. it is a statistical admixture of light rays whose polarization vectors are uniformly distributed over all directions. <»^> (9. i. A2 = b as </*) .9.15) with A\ = a. The socalled pure case. mutually orthogonal polarization directions x and y described by the matrices (see also below) * .21a) follows from Eq. like that of a bulb. 165 (9. b = 1 we obtain the expressions (9. . i.e. 1 The classical polarization vector or wave vector A corresponds in the quantum mechanical case to the wave function ip.( J o ) .* For a lightwave travelling in the direction of z. we have (9 19) ' In this case there is no preferential direction. is given by a single polarized wave. (9. b = 0 and o = 0.(AAt) = ( £ £ ) . i. 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. that of a pure state.2 Reconsideration of Electrodynamics In particular we have / = Tr(p). is in general unpolarized.18) The vector A describes a particular ray of light. In the present case of a polarized beam of light we have two possible. (9.19).22) For a — 1. The density matrix for the pure state (9.e. This admixture is described by introducing matrices p.21b) This state is still a pure state. For a 45° polarized wave we have 1 .
(9.Z^ 3. We set in analogy to Eq.166 and CHAPTER 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". (9.15) Pik — A i A k . l l\ " i °\ ) ' ^925^ 9. We shall see that the matrices appearing in the description of polarization properties of waves possess the same generalizability as the particle considerations of classical mechanics. b = —=.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. (9. Then according to Eq. in which the intensity I = Tr(p) is not conserved (e. Pi35° = \ \ 2 ? )• (9.8) describes the relation between the initial and final polarization vectors A^°) and A^1) respectively in a measurement of the observable F. as a consequence of absorption). We now inquire about the the way p changes in the process of the transmission of the lightwave through an apparatus. The Density Matrix and Polarization Phenomena P45° = I 1 For a 135° polarized wave we have a = and l M .23) 1 .3 Schrodinger and Heisenberg Pictures As in Chapter 2 we now consider on the purely classical basis the (analogues of the) Schrodinger and Heisenberg picture descriptions. Equation (9. 1 = .8) A(1)=E^40)> k R = R(<*).g.™ V 0 n kmAm / ^^jPim^mk' .
the observable is transformed.e.pW).4 The Liouville Equation It is natural to go one step further and to derive the equation analogous to the Liouville equation. 9. In the case of Eq.27) The measurement of an observable F in the new state with superscript " 1 " then implies according to Eq. In the case of Eq.e. (9. the observable F remaining unchanged.4 The Liouville Equation 167 pW = RpMRl (9. here this contains the dependence on a. (9.e.30) the dependence on a is contained in pW(a) = R{a)p^{a0)RJ'(a) (9.30) Ti{FRp{®I$) = Tr(F.17) (F) (1 ) = Ti{FpM) = = Tr(tiiFRpW) F' = RtFR. that the initial intensity Jo is equal to the final state intensity I\. that Io = Tr( / 0 (°))=Tr(pW) = / 1 .31).33) i. (9.28) (9. We can interpret the result either as (F) ( 1 ) = Tr(F{Rp(0)R}}) or as (F){1)=Tr({RtFR}p<0)).34) . The two cases are completely equivalent and can therefore be described as Schrodinger picture and Heisenberg picture representations. (9. on the other hand. The Liouville equation describes the motion of an equal number of systems in phase space elements of the same size. (9.q. and p(0' remains unchanged.9. The corresponding condition here is that no absorption of the wave takes place.29) where (i.e. in the state functional) in analogy to the time dependence of W(p. i. i. (9.t) in Chapter 2 in what we described as the Schrodinger picture there. the dependence on a is contained in F' = R^(a)FR{a). in other words without creation or annihilation of systems.31) (9.32) (9.
Eq. In Chapter 27 we shall see that the Poisson brackets here are actually Dirac brackets because the gauge fixing condition (9. (9. For p ( 1 ) ( a ) = p ( ° ) ( a 0 + da) follows from Eq. j£ = i[H. (9.p(°\ao)]da.p].40) Equations (9.38) H = Hl R(a)p(°\a0)R\a) (liHda)p(0)(a0)(± + iHda) pM(ao)i[H.32) in the Schrodinger picture that p^iao + da) = = = or ( w i t h p(°> —> p) (9. (2. Comparison of these equations with Eqs.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. For an infinitesimal variation of the state of polarization we have R{a) = 1 + Mda. (9.21) and (2. the matrix M is antiHermitian: Mf = .33)) we obtain correspondingly in the Heisenberg picture the equation ^ = +i[H. (9.39) and (9.46) suggests to define Poisson brackets of matrices by the correspondence {A.168 i. (9.39) If the dependence on a is not contained in p^1' but in F' (cf. = Tr(R^Rp^). i. (9.35) Thus the matrix R has to be unitary.B].B}:=i[A. We set therefore M^:=iH. .35) we have 1 = R*R = (1 + M W ) ( 1 + Mda) ~ 1 + (M* + M)da.2) has to be taken into account. (9. CHAPTER 9. In view of Eq.e.36) (9.F}.e. The Density Matrix and Polarization Phenomena Tr(p(°)) = Tr(p^) so that = Tr(Rp^R^) RlR = 1.40) can be considered as quasiequations of motion.37) (9.M .
To insure that the 169 . (10.Chapter 10 Quantum Theory: The General Formalism 10. and consider timedependent perturbation theory.1) This expectation value remains unchanged if \u) is replaced by e%a\u) with a real. Heisenberg and interaction pictures. Thus the expectation value with respect to the vector describing the state in the Hilbert space does not depend on this phase factor. 10. we establish their Heisenberg equations of motion. the Schrodinger.1 Introductory Remarks In this chapter we are concerned with the formulation of quantum mechanics in general. Thus in particular we establish the uncertainty relation for observables in general. 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).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.
such as for instance of the angular momentum operators Li. Naturally one can also write (10.3) These fundamental commutators determine the commutators of arbitrary observables A = A(p.* a conspicuous exception being for instance a spin system. B We first establish the following theorem: If A and B are observables and hence hermitian operators obeying the commutation relation [A. q).Pj] = 0.4) *Even in those cases where this is possible the operator correspondence may not be unique. and assume that A = A(p. E.Pj] = ihSlj. we assume this now.3 represent the three Cartesian coordinate directions of x. The first step in the investigation of a quantum mechanical system is to specify its dynamical variables A.170 CHAPTER 10.2.Qj] = 0. where i = 1. z and for definiteness here — which will not be maintained throughout — the "hat" denotes that the quantity is an operator. (10.2.2) (F(A)) = {u\u) <"lf. Such a correspondence with classical mechanics is not always possible. For these the following rules of canonical quantization are postulated: Postulate: The operators q~i. 10. we must have (u\u) = 1.B] = ih.y.g. (10.Pi are postulated to obey the following fundamental commutation relations (again leaving out "hats"): [Qi.q). [qi. which clearly does not hold in the case of operators. . \pi. 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. Pi • Pi. The phase space coordinates qi. Here. Quantum Theory: The General Formalism expectation value of the unit vector or identity operator is 1. classically qiPi = PiQi.1 U n c e r t a i n t y relation for observables A.Pi are called fundamental observables. however. In order to avoid multivaluedness one always starts from Cartesian coordinates in position or configuration space. y.
.B.0)1/2 = AA.2A(A) + (A)2.12) where \u) 6 "K is a ketvector representing the state of the system.5) are subject to the following inequality called uncertainty relation: AAAB In proving the relation we first set A:=A(A). we have A2 = A2 . (10. Hence (AA)2{AB)2 = (A2)(B2) = (u\A2\u)(u\B2\u). = \\A\u)\\2\\B\u)\\2 (10.13) AB = AB = (B2)1/2.8) and.9) ((A2) + (A)2 . Then [A.1) (AA)2{AB)2 = > (u\A2\u)(u\B2\u} \{u\AB\u)\2. so that (A) = (A) . Eqs.. Since A = A— (A). AB defined by the mean square deviation AC=((C2)(C)2)1/2. C = A..10.7b) = = [A.12) the Schwarz inequality (cf.(A)2 = (AA)2. (10.(A) = 0. (4. using Eq. (10.B\u) G IK as vectors) we obtain (for an application see Example 10.2 States and Observables then their uncertainties AA. Applying to the expression (10. Thus AA = AA = ( i 2 ) V 2 . Ai = = = ((A2) .2A(A)}  {A)2)1'2 (10. (3.2(A)2 .{A)2)1'2 {{A2) .5)) (for A\u).7b). (10.11) (A2) = (A2) .B] = ABBA > ^h..7a) (B) = 0.10) . (10. (10. 111 (10.(A)2)1'2 (1 °= a) ((A2 + (A)2 .6) B:=B(B). (10.9a). (A{A))(B{B)){B{B))(A(A)) (10. B}= ih.
The condition (a) is satisfied when. (10. since A and B are hermitian. R= (AB + BA).14) Hence (u\AB\u) = /^(AB where. for a complex constant c. The equaltosign applies in the case when (a) it applies in the Schwarz inequality.e. AAAB>I=h. . Zi (10.18) >R2 + I2. 4. and hence for condition (b) 0 = = tSee Sec. Then AB = \(AB + BA) + \(ABZ Z BA) (1 = 8) \{AB + BA) + \ih.172 CHAPTER 10.19) (u\AB + BA\u) = (u\AB\u) + (u\BA\u) c*(u\BB\u) + c(u\BB\u) = (c* + c)(u\B2\u). (AB + BA) This means (AB + BA) is real. Quantum Theory: The General Formalism Next we separate the product AB into its hermitian and antihermitian components^ and use Eq. We can therefore rewrite Eq.15) = = (u\AB + BA\u) = (u\(AB + BA)^\u)* (u\(AB + BA)\u)*. (10. Zi Zi (10.16) I = h. (10. and (b) when (AB + BA) = 0. (10.3. From this we obtain (uiJBu) 2 = i22 + / 2 . Zi + BA)\ + ^ih.15) as (u\AB\u) = R + iI. (10. (10. and with Eq.4). (10.17) as had to be shown.13) (AA)2{AB)2 i. A\u)=cB\u).
Ly are not "sharp". (Ly)2=0. 9 = h{u\eiikLk\u). + LjLi\u). {u\LiLj\u) = R + iQ. What does the relation imply for states u) = \lm). = 0. we obtain {ALi)2{ALi)2 > \{u\LiLj\u)\2. implying that the vector it) has to satisfy the following equation: A\u) = i($Sc)B\u) (10.e.Lj] = (LiLj + LjLi) + ih£ijkLk. = 5R2 + G 2 > 9 2 = hi2(Lk)2. i. q) = ih. (Lx±iLy) ^ 0. for which one component of L is "sharp"? Solution: Applying Eq. 9ftc = 0. i. (ALy)2{ALz)2 > ~h2(Lx)2. i.10. ALZ = 0 (these are states \lm) with Lz\lm) it follows that (Lx)2 = 0.e.3 OneDimensional Systems 173 Thus c = — c*. implying {Lx)=0.) 2 > = x. The observables like angular momentum are functions of p and q. .3 OneDimensional Systems We now consider onedimensional systems with a classical analogue as described earlier. ALy = (u\L±\u) ^ 0.20) or.e. For these we have in the onedimensional case the relation [p. Thus in this case Lx. Example 10. in view of Eq. the following uncertainty h2{Lk)2. For states \u). Separating the product into symmetric and antisymmetric parts.i relation holds: (ALi) 2 (AL.z.13) to operators Li — Li. (Ly)=0. (Lz)2^0. we have LiLj = (LiLj and ^ = y(u\LiLj so that \(u\LiLj\u)\2 Hence {ALx)2{ALy)2 > l + LjLi) + [Li. m\lm)). (A{A))\u)=i(Zc)(B(B))\u). (10. i.e. h2(Lz)2.1: Angular momentum and uncertainties Show that in the case of angular momentum operators Li. ALX 10. (10. (ALZ)2(ALX)2 > \h2 = (Ly)2.7a). such that Lz is "sharp".y.
21b) (10. Then (see below) we can derive for A the following commutator equations [q.p] = ^ihp ~ Proceeding in this way we obtain [q. (3.p}p + p[q. for q\x) = x\x) we have p\X)=\x)(th^y Now let A be an observable. We demonstrate that (a) its eigenvectors have infinite norm (in the sense of a delta function). We obtain these equations as follows. i.p2] = [q. Next we have a closer look at the operator q. q]q + # > v] = ~2ihq and \p. they do not have a common system of eigenvectors. (b) its spectrum is necessarily continuous. Eq.21a) and (10. (10.p) can be expanded as a power series in q and p.p] = ih we obtain therefore [q. Using [q.174 CHAPTER 10.22) d ih—(p2).qn] = ih^(qn).e.e. i.p)]=ih^^and \p. as remarked earlier. Q2] = [p. (10.23) = ihg(q2) (10. Quantum Theory: The General Formalism Since this says that p and q do not commute (as operators).BC] = [A. A — A(p.A(q.p)] = .i h ^ ^ . .21b) now follow with the assumption that A(q. We have (cf. and (c) the spectrum is not degenerate.pn]=ih^(pn).34d)) [A.B]C + B[A.A(q.21a) Equations (10.q).C\. This is why the operator p requires a representation in the space of eigenvectors of q. Similarly we obtain \p.
U] = ih^= aU. > We begin with (a).24) lim SK(x — x') — S(x — x').e. We have (x\x') = = where f^ J—oo dk(x\k)(k\x'} (4 = 9 ) ±.27) (10. 6K(xx') = — fK dkeik^x'^ 2TT JK sin K(X K{X — x') — x') (10. = eipa/h = Uia)'1 = U{a). an operator. (10. we obtain [q.1 how the delta function arises in the limit K —> oo.e.1 The delta function for K — oo.3 OneDimensional Systems 6K(x) 175 Fig. i. For the cases (b) and (c) we consider the operator U(a — pipa/h (10. and a a cnumber.21a) A by U.28) i.f^ ^ J—oo dke ikx —ikx' (10.26) it Here p is an observable.10. We thus see explicitly t h a t the vectors \x) have infinite norm (in the sense of delta function normalization). (10. U is unitary. . 10.25) It is shown in Fig. Replacing in Eq. 10. Since p = p\ follows t h a t U\a) so t h a t U(a)U\a) = 1.
b a cnumber).q)\i>) = (i>\U^UA(p. the spectrum is not degenerate.00). which means that the spectrum of this operator is a continuum. 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. The eigenvectors have the same norm as \x): (x\U*U\x') = (x\x') = S(xx'). 10. Evidently every eigenvalue has only one eigenvector. Quantum Theory: The General Formalism qU = Uq + aU = U(q + a). i. Eq. CHAPTER 10. (10.30) (10. In an analogous way by defining the unitary operator U(b) = eiqblh (q operator.00). WU = 1.1 T h e translation operator U(a) We saw above.29). q any other observable F(p. (10. Thus all these values belong to the spectrum of the operator q. Since this holds for any arbitrary value of a in (—00. that with U(a) = e~ipa/h = U.3. this means: With a unitary transformation (which leaves the matrix elements unaffected * and hence the physically observable quantities) one can pass to any arbitrary eigenvalue in the domain (—00.31) .29) This means: U\x) is eigenvector of q with eigenvalue x + a.q)U^U\ilj). or qU\x) = U{q + a)\x) = (x + a)U\x). cf.e. ^Observe that {iji\A{p. q) can be constructed representing a dynamical variable.e.176 i. we can see that also p possesses only a continuous spectrum (from —00 to 00). (10. or qU(a)\x') = (x' + a)U(a)\x'). we obtain qU\x) = U(q + a)\x) = {x + a)U\x).
x')5(x" = (x" + a\c*c\x' + a) x').x" . (10.33) (10.34) {x"\U\a)U{a)\x') c*(a.31) and (10. i.x") . (10.x')\ = l. the operator U(a) acts to shift the value x' by the amount a.e) ~ 5{x' .10.e.e.35) (10.32) we see that \x' + a)ocU(a)\x'). The operator is therefore called translation operator. i.x")c(a. for instance.x') = = i. (x'\p\x"). We choose the phase such that U(a)\0) = \a).36) i. (10. since q\x') — x'\x').e./ (10.x") U{x'\p\x" % pz.e. (10.x" . Comparing Eqs. (10.hm or „'i„i™" s> i ™' ~/ (x'\p\x")\ = 8'{x' x"). (10.x" . It follows that U(a)\x') = = U(a)U(x')\0) = U(a + x')\0) \x' + a).39) .37) we set (with e infinitesimal) U(e) ~ 1 Then (x'\U(e)\x") = 6(x' . i.3 OneDimensional Systems 177 Moreover.38) h 6(x' .32) Since U is unitary. In the expression of Eq.e) (x \p\x ) = . U(a)\x') = c\x' + a). (10. \c{a.5(x> . offdiagonal elelemts of the {x}representation of the momentum operator. it follows that 5{x" .e.37) With the help of the latter expression we can calculate. It also follows that (x'\U(a)\x") = (x'\x" + a) = S(x' . \x') = U(x')\0). we have q\x' + a) = (x' + a)\x' + a).a).
a conservative system. to) with respect to t we obtain ihjtU(t. isolated from any measuring apparatus which would disturb the system and hence its behaviour in the course of time) and let \tp(t)) be its state vector at a later time t > to (with no interference in between).40) 10.to) exists which is such that^ m)) = u(t.e.e. (10.° ° d(p) (ip) e"« = S'(x).42) ^Observe that 6'(x) = £ f ^ dpipe** = ± / .to) is called the time development operator. 5.to)m0)). . so that U{t. H\uE(to))=E\uE(to)).178 CHAPTER 10.43) (10.§ It follows that (x"\p\x') = d'(x' This expresses that p is hermitian.44) (10. Differentiating U(t. (10. Let \UE) be an eigenvector of the Hamilton operator H with eigenvalue E.x") = (x'\p\x")*. (io.tQ) = eiH(tt<Mh. i.4 Equations of Motion Let \tp(to)} be the state vector at time to of a completely isolated system (i. it follows that 5'(x) is an odd function of x.t0) = HU(t.4.L fdpeipx.f dpipeipx. 6'{x) = !. (5. This is in conformity with our earlier postulate in Sec.27) and (5. The operator U(t. 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)). In this case the classical Hamilton function does not depend explicitly on t.4i) We assume first of all.45) (10.e.35). i. See Eqs. since 5(x) = . 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.t0). . Quantum Theory: The General Formalism We mention in passing that.
t0) = E/(t 2 .t0) or ih—U{t. Ufa. Eq. (10.t)U(t.to) = 1.t0) [U(t + e. (10.t)l]U(t. But obviously (cf. t0) (10. Then with U(t + e.t0) = U(t + we have U(t t) = lim ^ + Mo)^(Mo) = n (1046) = U(t. (10.47) e. In the immediate considerations it is not necessary to restrict ourselves to systems with classical analogues.t) = l^eH{t).4 Equations of Motion 179 We now demonstrate that Eq.49) .t) and therefore [U(t.48): [/(Mo) = 1 . (10. (10. (10.t) = 1. = U(t0. (10.t 0 )^(*0)> U(t2.41) we obtain ^(t 2 )> = = J7(t2.e. Evidently U(t. t0) = H(t)U{t.t0)\^(t0)).44)) we have U(t.z / dt'H(t')U(t'.tiM*i.*l)^(*l)> = I7(*2.to) (differentiating we obtain immediately Eq.to) can be expressed as an integral as the solution of Eq.48) with U(to. From Eq.10. The operator H is d efined by this relation.to).t).t) = l.t0)U(t0.t0)]1 Let us set for small values of e: U(t + e. Hence U(t.45) has very general validity. i.48)).*l)^(*l.
The operator U(t. (io. the predictions of different descriptions are identical. i.52) is the operator of an infinitesimal unitary transformation (recall that U = 1 + ieF satisfies the unitarity condition UU^ = 1 provided F — F^).180 CHAPTER 10.t) = l % ^Hdt) \il>(t)) Hdt (10. (5.e. it follows that U(t + dt. The eigenvectors of these observables are also constant in time.51) Since </. the Schrodinger equation ju(t. Comparison with the equation postulated earlier.30). t)\t/>(t)) =U~ and since H is hermitian. (10. Another such description is the Heisenberg picture.t0) m0)).t0)\ii>s{to)) (10.53) .5o) ihjtm)) = Hm)). 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. 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. and these remain unchanged under unitary transformations.41). One describes as Schrodinger picture the present description of a system in which the state of the system is represented in terms of a vector \ip(t)) which changes with time t. Now let \tp(t)) be the state of the sytem at time t. Eq.e. (10. to) can therefore be interpreted as a product of a sequence of infinitesimal unitary transformations. requires the identification H(t) = H= Hamilton operator. the instant at which a measurement is performed on the system. whereas the physical quantities are represented by observables in "K which do not depend explicitly on t. Since the only measurable quantities are moduli of scalar products. In the following the subscripts S and H indicate Schrodinger and Heisenberg picture quantities.(£ + dt)) = U(t + dt. Then \Mt)) = u(t. >(*)> = i.
55) and using Eq. Here H is the Hamilton operator in the Schrodinger picture. It does not always have to be the case that As does not contain an explicit time dependence.p.57) and rf[As.With explicit time dependence. Thus. (10.48). 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.e.H]U + iW]^U. but here. However the associated observables are timedependent. As(q. we consider only cases without explicit time dependence of As.59) This equation is called Heisenberg equation of motion. t0)AsU(t. we have to take into account the contribution dAs/dt.t0)\ips(t)) = \ips(to)) = constant in time. i. t0). Differentiating Eq. = so that ihjtAH or ihjtAH = = [AH.55) Thus AH is explicitly timedependent.54) Observables transform correspondingly with a similarity transformation.i). (10. {AH.HH].58) HHAH (U*ASU)(U^HU) [AH. if ^ = 0. even if As is not timedependent.H]U = = = U]ASHU AHHH — U^HASU . For As(q. (10. for instance in cases of nonequilibrium or if part of a system disappears through absorption.e.10.4 Equations of Motion and \^H) = U\t.e.HH]. as a rule. so that a matrix element (i.p) in the Schrodinger picture dAs/dt = 0. 181 (10.56) U^[As.HH]+ihU^U. (10. a scalar product) remains unchanged: AH(t) = U\t. we obtain ih^AH at = = U^HAsU + ihU^^U at + U^AsHU (10. (10.(tf HU)(rfASU) (10. . i. in order to obtain the Heisenberg picture (description). subject to a continuous change. In the Heisenberg picture it is HH = U]HU.
In our treatment of representations. we used in essence unitary transformations of matrices. Example 10. i. since in the transition from one to the other the commutator remains unchanged. For an extensive discussion see B. We end with a word of caution. [CH. Orfanopoulos [224].61) Thus conserved quantities commute with the Hamilton operator.21a) and (10. 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): and where the expressions on the right follow from Eqs. Finally we add a comment concerning conserved quantities. An observable CH is called a conserved quantity. Solution: We have for an observable A (A) = (i>H\AH\TpH).HH}=0.21b). In particular. provided the Poisson brackets in the latter are replaced by commutators.Pi we have in the Heisenberg picture (replacing in the above AH by qi. In many cases the Schrodinger picture is more amenable for explicit calculations. We see that formally one obtains Hamilton's equations of classical mechanics. (10. In the Heisenberg picture essential properties of a system can frequently be recognized more easily in relation to their counterparts in classical mechanics — in both cases the time development of a system is given by the time dependence of the dynamical variables.13) to derive the uncertainty relation for energy and time. In the treatment of the pictures we used unitary transformations of operators (and vectors). Quantum Theory: The General Formalism Wave mechanics is now seen to be the formulation of quantum mechanics in the Schrodinger picture.2: Energy and time uncertainty relation" Use the Heisenberg equation of motion and the general relation (10. in the case of the fundamental dynamical variables qi. (10.H] = 0. This holds also in the case of the Schrodinger picture. A. We also observe that the momentum pi is conserved when \pi.e.182 CHAPTER 10. . if ffCH = 0. as in the transition from the configuration or position space representation to the momentum space representation with the help of Fourier transforms.
63) so that with Eqs.62) and dAs/dt > \WAH\i.H]) = 0. (A).7a).Such a characteristic time interval can be defined for any dynamical variable. i.67) 1 d(A)/dt is oo and AE = 0. of A\<t>) = a<£)) are then called "good quantum numbers".H}\f)\2. 2 d{A)/dt " and. H] = 0. AW = {A(A»V>. then if a measurement is made at time t'. (10. we can replace ([AH. dt in. i.)\2 ( = \{i. (10.10. and TA O C (10.65) and rA AA d{A)/dt' . this is practically the same as at time t for times t with \t — t'\ < T.68) To put it The eigenvalues a of A (i. Let \ip) = \"4>H) represent the state of the system. is displaced by A A in the interval A T = TA. (10. i. We construct the following vectors using Eq.\HAM\2) \(nA. in a different form: The eigenvalues of an observable are called "good" if [A. (10. .p) we then have 0=^M) = U[A.13) and (10. If r is the smallest such characteristic interval of time.15): (AAf(AH)2 Using Eq. (10.62) The Hamiltonian H does not depend on time.64) = 0.4 Equations of Motion Then. a position or momentum observation is independent of t. (10.e.66) This means. if we set AH = AE we obtain TAAE (10. This means that for a physical variable A(q.2 > h.e.e. 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 ) .HH]) Hence we obtain by ih(A).e. since \IPH) is timeindependent. the "centre of mass" of the statistical distribution of A. at ^±AH>h. The probability density P ( r ) =  ^ ( r )  2 = (^r>(i#> is independent of t. > (10. H\tl>) = {H(H))\f).
5 States of Finite Lifetime When E is real and P(x) is independent of t and J dxP(x) = 1. states with sharp energy (AE = 0). Recall for instance radioactive decay.69) (Observe that if ip consists of only one state with energy E in the domain of integration of E'.184 CHAPTER 10.Eo(x)eiEot/he'^2h9(t).71) From Eq. But the number of radioactive (thus excited) nuclei decreases exponentially. States with finite lifetime (At = r ^ oo) are those whose probability density falls off after a certain length of time called "lifetime".t) — ipE(x. i.70) by From Eqs. (10.72) = V'EoW 7/2h + 0i(EE )/h  oo .69): oo poo p / oo *PEo(x. the wave function ^ ( x . (10.(x)fEo(E')e«EE'Wh 2irh f dE'^E. the wave function ipEo(x.. This means. (10.70) we obtain: oo / oo TPEO (x.£ ^ + n/2)^ o ( x ) e{^1/2+i(EE )}t/h 0 o =0 . /oo (10.E1) 2irhfEo{E)4>E(x). Energy and decay length of the radiated aparticles are characteristic properties of naturally radioactive nuclei. (10. t) = ^(x) exp(—iEt/h) describes states called bound states of unlimited lifetime. have around EQ the form ipEofct) = i.69) and (10. Quantum Theory: The General Formalism 10. the particle density diminishes in the course of time (t > 0). for instance.)exp(iEt/h)). First we obtain from Eq. (10.H ) ( £ . as is observed. t)e iEt h ' dt = V>£0 (x) / JO <fte{7/2+i(i5£b)}t/ft (10. On the basis of the uncertainty relation AEAt > h these must have an uncertainty AE in their energy spectrum.70) we can determine the function fEo{E) integration with respect to t. i ) = J dE'^El^)eiEltlhfEo{E'). (x)/ E o (E')5(E .t) of such a state must be the superposition of states with different energies about EQ and a corresponding weight function / so that ^ ( x .e.tyEt/hdt = = = / J—oo dt J dEJrl>E. then fE{E') = S(EE') and integration with respect to E' yields again t()E(x. The wave function of such a state with energy EQ and lifetime r = h/'y > 0 could. This means that in the case of decaying particles.
ihU(t. to) be an approximate but unitary solution of this equation. 10. Eqs. (C/(0) unitary). Let U^°\t.e. We obtained the result (10.11).e. 185 i. This description contains effectively parts of the other two pictures and is particularly useful when Eq. (10. 7 > 0.t0). whose real part EQ specifies the energy of the state and and whose imaginary part 7/2 specifies the lifetime r = h/j. A state is discrete if its immediate neighbourhood (in energy) does not contain some other state.76) **For specific applications see e. the formula says that JEQ{E) possesses a simple pole at E = Eo — 27/2. (10.73) precisely by assuming with fEo{E)^8{EEQ) a "smearing" of states around the energy EQ (with uncertainty AE).45).e. VE(X) ^ ^ (£ > = dbfift+T/* (la73) This result is known as the BreitWigner formula. where U is the solution of Eq. i.e. which means that a nucleous with discrete energy EQ does not possesses some other admissable level close to EQ.g.74) Thus the solution of this equation is the main problem. ^(*o))> then \xp(t)) for t > t0 follows from Eq.75) (10. We see therefore that the states with lifetime r < 00 and TAE > h belong to the continuum of the spectrum. We saw previously that if the state of a system is known at time to.105) and (20.10.45) can be solved only approximately (in actual fact. £/ = £/W(Mo)^'(Mo). Considered as a function of E. W)) = U(t. (10.6 The Interaction Picture One more motion picture of quantum mechanical systems is in use. i.toM{to)). The state with lifetime r — h/'y is not a discrete state. (10. and is called "interaction picture" or "Dime picture". i. i. .e.to) = HU(t. of course. (10. that** EQ.6 The Interaction Picture Prom the last two equations we deduce for E close to ipE0(x). (14.41). They are called "resonance states". every unitary transformation of states and operators defines a possible "picture").
be defined by the relation H(°)rj(0) _ i h ^ l = 0 (exact)> n (0) = eiH(o)t/aj (1080) i. Quantum Theory: The General Formalism ih^(uWu') i.186 Then CHAPTER 10. For a "good" approximation U « [ / ' ' we have HU^ihjtU^^0. (10. (10. (10.78) at with the initial condition c/(0)t f ^ ( 0 ) _ ^ ^ 1 V \ at J U\t0. = HU<®U'.77) C/(o)t itiuWfLu^HUMu'ih^U' at or i. i.e. dt so that #(0)t ( t ) It follows that H{0\t) = dt = H<®{t). varies only slowly with t).e. In addition we have (10.t0) 0 = l.e. i.82) (10.h^U> = at ( io.e.78) (10. H(°Ht) = ih^uW. £/(°)£/(°)t = 1. [/' has only a weak ^dependence (i.83) ih^U^ ishermitian. with Eq.e.e.e. . with Hamiltonian H = H^ + H'. Now let H^(t).79) ^'~°i.81) But [/(°) is unitary. i.
80). (10.89) where ^s(i)) and As are state and observable in the Schrodinger picture.6 The Interaction Picture 187 We use now the subdivision of H into unperturbed part and interaction. 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^.87) = We define correspondingly as interaction picture state iM*)> = t/ (0). U<®\Mt)) = \Mt)).10.84) U^ where H' is also hermitian. (10.92) . lfe(*)}.78) at (10 85) (10. 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. i. (10.91) Since the Schrodinger equation is given by ihjt\^s{t)) = H\1>s{t)) = ( # ( 0 ) + H'Ms(t)). (10.85) multiplied by U') on the right of Eq.90) = U^H'\^s(t)). We now multiply this equation from the right by U' and insert the first expression on the right (of Eq.U^HU^U' H'fi'.85) = where we used Eq.86) at = U^HU^U'.88) and as interaction picture version of an observable A the quantity Ar(t) = tf(°>t AgU^.e. H = H{®{t) + H'. (10. (10. (10. (10. (10. U^H'U^U' (10.78). (10. We also set H^U^H'UW.
89)^0. i. i. cf. (10.94) _Hf)Al 1 + lhdAl+AlHf).60a) or (10. (note the difference between H' and H'j) ihjt\Mt)) = H'AMt)). (10. near zero and \tpi(t)) is almost constant in time.92) ^°)^V/(t)) + ^ ( ^ )  V ' / W > = ^ (0) ^ (0) l^(*)> + ^ . we differentiate with respect to time the operator defined by Eq. (10. the vector varies with time (different from the Heisenberg picture).55)) at + imm<Ms_u(o) at dt _umH(.94) Then we have (normally with dAs/dt dt (io=8o) — 0. comments after Eq. In order to obtain the equation of motion of an interaction picture observable Ai. i. we have i.93) Thus the vector ipi(t)) satisfies an equation like the Schrodinger equation. (10.90) and replacing the left hand side of this equation by the right hand side of Eq.e. a perturbation so that H ~ H(°)).e. Since H'j is assumed to be small (i.89). (10.95) Thus in the interaction picture the physical quantities A are represented by timedependent observables which satisfy a type of Heisenberg equation (10. H'j = UWH'UW. (10. dAs/dt=0) .o)AU(o) at +uWAsHWuW UWHWU^UWASUW + iww^uw dt +U^ASU^U^H^U^ (10.e. (with ihjtAI(t) = [AI. ^ (0) l^(*)>.H\0)}.e. so that with Eq. A^t) = U^AsUi0). Quantum Theory: The General Formalism we obtain by starting from the right hand side of Eq. (10.80) (and multiplying by t/(°)t) ih~\Mt)) = u<MH'uM\Mt)).e.60b) with iff0) instead of H.188 CHAPTER 10.e. (10. ot i.
96) in which the perturbation part H'(t) depends explicitly on time and HQ replaces H^0' in the previous discussion.98).102) We assume that before the perturbation H' is switched on at time to. (10.t0).93).97) The equation to solve is Eq.99) (1098) Instead of the original Schrodinger equation we now solve Eq.94) Hi = uW(t.e. We obtain the operator U^ from Eq.101) Iteration of this inhomogeneous integral equation yields the socalled Neumann series. (10. We have the Hamiltonian H = H0 + H'. Since . the system is in the Schrodinger eigenstate \ips ) of iJ 0 with energy Em. i. (10. (10.10.i0) = e<Ho(tto)/ft.t0)H'uW(t. ^ \<Pn){Vn\ = 1n (10.e. where according to (10.7 TimeDependent Perturbation Theory We consider timedependent perturbation theory as an application of the interaction picture. Integration with respect to t yields \Mt)) = \Mto)) ~ \ \ dt'Hi{t')\^j{t')).100) (10. H0\ipn) = En\(fn). ih±\Mt))=Hi{t)\Mt)). i. (10. We assume again that the spectrum and the eigenvectors of Ho are known.7 TimeDependent Perturbation Theory 189 10. i.80) as E/(°)(t. {ipn\<Pm) = Snm.87) and (10. \Mt)) = +' \Mto))^fdt'Hi(t')\Mto)) ft J dt' f dt"Hi{t')Hi{t")\^I{tQ)) + (10. H' = H'(t). (10.e.
103) The actual state \ipg) of the system is solution of ih\^s) = H\^s). we can write l 4 0 ) ) m = elHot/hWm) =e . n We wish to know the probability for the system to be in a state \ips ) n at time t after the switchingon of the perturbation H'. (10. 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.102) and truncate the series after the first perturbation contribution.103) for m — n and use Eq. (10.106) We insert this expression into Eq. assuming that this supplies a sufficiently good approximation: IMQ) = \fm) lz\ n dt'HW)]^). Quantum Theory: The General Formalism HQ is independent of time. i. This probability is the modulus squared of the projection of the Schrodinger state \ips) o n t o \ips )„. (10. the time dependence of \ips } can be separated as for stationary states.105) In the interaction picture the iJostate m at time t is \Mt)) = eiHot/hWP)m = \<pm).105) rt n Jto h ' Jt00 t rt  x  l h ' Jto t0 (10. (10.108) . (10.88) and obtain > n ^ V s ) = {<Pn\eiHot/h\ll>s) = {<Pn\eiHot/hUM\Mt)) = <¥>*(*)>• (10.104 ) Here we insert Eq. (10. (10.107) Jto With this we obtain from Eq. Hence we set ^> = £°«(*)l4 0 ) >n. The corresponding amplitude or appropriate matrix element is n ( 4 ° V s ) = 5> m (i) n (v4 0) V4 0) >™ = m fl "W ( 10 .^ V ™ ) .e.190 CHAPTER 10.
< KO?t .111) Different from Eq. This is the case. where the momenta of the aparticles form a continuum.10. (10. (10.8 Transitions into the Continuum In many practical applications one is interested in transitions into a continuum. for instance. if in the scattering of a particle off some target its momentum p changes to p'.110) With this we obtain from Eq. » — < (10.Em)t' (<Pn\H'(t')\pm) = Wnm(t). Thus we set H'(t) = H'0(t).109) 10. (10. (10.En (10.112) for a^0.8 Transitions into the Continuum 191 Hence the probability for the transition of the system from state m into the state n ^ m is K(t)\2 = I fdt'e^.109) the transition probability Wmn(t) = jp Jo . We assume that the perturbation is switched on at some time to. One reason why we begin with these simplifying assumptions is also to obtain reasonably simple expressions.Jr(En—Em)t _ ^ (ipn\H'\(pm) {En — E„ sm{±(En . but that otherwise the perturbation is timeindependent. or if by emission of a photon with continuously variable momentum an atom passes from one state into another.Em)t} \{tfn\H'\ipv h 2h(En .Em)t] _ sin2 at aH {±{EnEm)}H Consider the function St{a) :1 sin2 at ir a2t t 7T for a —• 0. A further example is provided by adecay.25) we here have the positive quantity (note t in the denominator) s i n 2 { ^ ( £ n . t o = 0.
In the case of transitions into the continuum we have to consider the transition probability into the interval dEn at En.Em). . though sufficiently small.192 CHAPTER 10.2. (10. we would have En — Em ^ 0. Otherwise. t—>oo Hence also (shifting t to the other side) sin2{^(ffn.Em)t} r i 7 TTT.— = o (En 2h En 2h5{En .E7m)t} = (10. Fig.2 The delta function for £ — oo. 10. Quantum Theory: The General Formalism T h e behaviour of this function or distribution is illustrated in Fig. > We see t h a t this quantity has the same behaviour (in particular for a — 0) > as the function 5K(x) we considered previously.114) It follows t h a t for large but finite times t Wmn(t) 2vr ~ — t5(En Em)\(vn\H'\tpm)\2. so t h a t Wnm would be zero. and no other state would be nearby. if the energies belonged to the discrete spectrum. Hence lim St(a) = 5(a). 10.ii5) This expression has a well defined value unequal to zero only if the energies En belong to the continuum.113) {±JEnEm)¥ or n m irtSt En — ^ 2h 1 s i n 2 { ^ ( £ n .io. We assume t h a t the matrix elements for transitions into this infinitesimal element can be taken .
pp. (10. (10.t The result makes sense as long as the uncertainty AE in the energy satisfies for finite but large times t the relation . 75. Pauli [227]. r = J E ^ W =p(£m)y(</>n#W2n (10117) The formulas (10.* In view of the usefulness of Eq. Schwabl [246]. so that initial and final states belong to the continuum. 266.10.117) provides the outgoing particle current (see Example 10.115) and (10.3: Application of test functions to formula (10.4). in the article of W.117) in many applications./ t>oc TZ J_00 dx . We evaluate the integral 0(0) = 1. for instance from a potential.W . t We choose a test function which falls off at infinity faster than any inverse power of x: <Kx) = e . p.e. (10. f See E. sin 2 (tx) ax i—( * According to F. Fermi gave this formula the name "golden rule" . _ 2irh AE > > Se. i. . Fermi [92].113) this is* hm . the transition probability into this set of states is J2Wmn(t) = fdEnP(En)Wmn(t) n •* = p(Em)2lTl{iPnlfliPm)l2t. In the case of the functional (10. Let p(En)dEn be the number of states in the interval dEn.113).8 Transitions into the Continuum 193 as equal. If we are concerned with pure scattering. Solution: For test functions <p{x) € S(R) the delta distribution is defined by the functional I &(x)4>(x)dx = <j>{0). (10. Then. Example 10.118) where 8e is the separation of states around En which in the case of the continuum goes to zero. if p(En) varies only weakly with En. F Jo . then Eq.r Z x 1 ^ ) = 4>{Q). 142.117) were derived by Pauli in 1928.116) The transition rate V is the transition probability per unit time. '"Recall also the integral J ^ ° dec sin 2 x/x2 = TT.113) Use the method of test functions to verify in the sense of distribution theory the result (10.
^ V v 4 ) = ^ T 3 e * k ° *^ko)e*° * = — v . E = h2k2/2fi..tan p 2 + a 2 . Gradshteyn and I. with v the velocity of the particles..117) for H' = V(~x) the differential cross section da _ / n V dH ~~ \2Trh) f dxV(x)e where K = k m — k n . L \3k2dkdUn It follows that T. and hence .k.( 1 / 4 ) ln(l + 4t 2 ) + t tan _ 1 (2<). ^o n . iknx . .4: Differential cross section from the "golden rule" Obtain from the "golden rule" (10. which is the number of particles scattered per unit time into the solid angle element dQ with only one particle sent into the volume L 3 (since the incoming wave is normalized to 1). . K = k 0 . 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. Quantum Theory: The General Formalism with the help of Tables of Integrals. formula 3. * ^ = .947. .( ^ V V • < ih 2 _ i k n .b2 4 p 2 + (a + b)2 2 b _j 2pa V 1 fc Urn . p = 1 we obtain I = . Then ^ . = ( — ) dk= 2ir (—  2irJ k2dkdfl. 2 . V +(ab)2 a _x 2p6 7 m —^ '. with n .e " 1 * 1 = 2 lim — . . 491. S.. § We have r Jo \p\x dx sin ax sin bx  V. E x a m p l e 10.ln(l + 4t 2 ) + — t a n _ 1 ( 2 t ) = — tan 4t7T tTV T J< (oo) = 1. 1 . Solution: We set ¥>n(x) : so that {<Pn\H'\<pm) = ^3 / d x e i K ' x V ( x ) . dEk = h2kdk//j.a 2 ' i that for a = b = t ./ cfc a n p 2 +fc 2 . 77^ + . Then p(Ek)dEk Since Ek — h2k2/2[i. M. and is obtained as ih. The ingoing particle flux is therefore v / L 3 ..„I. . 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. = 0 .<p).? 2TT / L \3k2dkdUn 1 /V(x). 2 T / ^ M. . Ryzhik. x . . . p. . Hence the result is 0(0) as expected. Jin § m = See I. L3/2 ¥>m(x) : L 3 /2 The periodic boundary conditions ipn(xi) = fn(xi ± L/2) imply kiL/2 = ±nj7r. . .194 CHAPTER 10.
122b) S e e also L.119) where H' = H'(t) is a timedependent perturbation term and Ho\ipn) = En\ipn). 193. pp. Thus we set ^(x.t)^°\E^t)dE (10.t)+ / n a{E. specifically in the case of the Coulomb potential.10.^°>(E. (21.x)^(E'. (10. (10. 148.9 General TimeDependent Perturbation The differential cross section is defined as da •• Theory 195 outgoing particle current x r2dfl ingoing particle current da / /i V (hko)/(lJL3) J dxV(x)eil where for purely elastic scattering k = ko.69).t) = ^On(t)^(x./continuum = <San(£)</4°)(x.9 General TimeDependent Method For a better understanding of the "golden rule" derived above. by which we mean without leaving the Schrodinger picture.122a) (10.^ We start from the Hamiltonian H = HQ + H'. The problem is to solve the timedependent Schrodinger equation ih^\i(>) = H\r/>). Here S stands for summation over the discrete part of the spectrum and integration over the continuum with the orthogonality conditions J and fdx.t). we present another derivation using the usual method of timedependent perturbation theory. 10. Thus we assume that the eigenfunctions (fn = (x</?n) of Ho are known.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. (10. .x) = S(EE'). We expand the wave function V( x . t) = (xl^) therefore in terms of the eigenfunctions of Ho with timedependent coefficients. (10.120) In the case of Eq. see the discussion after Eq. Schiff [243]. For a possible application of this result. 1 rfx^°)*(x)^)(x) = 8mn.121) .
t) = San(t)<pn(x)eiE^h.. (10. H'kn = J d x ^ ( x ) # V n ( x ) = (<pk\H'\<pn). (10. Next we use the orthogonality relations .t)dx.x. We multiply this equation from the left by </?£(x) and integrate over all space.129) .126) ihak{t)eiEktlh where = San{t)eiEntlhH'kw (10. .E').. Quantum Theory: The General Formalism The discrete and continuous eigenfunctions of HQ are Vi 0) (x. Then J d^*k(X)[ihSdn(t)cpn(X)eiEt/h = J d^*k{x)[San{t){En + San(t)En<pn{*)eiEntlh] (10. . / *.x.196 CHAPTER 10.„.i) = e. since f rJ>(°>(E. V>(x. into dip (10.. f 6kn in case of discrete tp's. dnptMrnW = { ^ _ En) (10. The second term on the left of this equation and the first term on the right cancel out. (10.120).„„.x). Observe that these conditions do not contradict Eqs.123) We now insert these expressions into Eq.E') = 5{E . (10.128) . ^0)(E. i.t)rJ><®(E'..127) we obtain from Eq.127) i n c a g e Qf c o n t i n u o u g £.126) + H')<pn^)eiE^h}. (10. .. (10.x.t) = eiEtlhy(E.e. = e^EE'^h5(E Thus with Eq.122a) and (10.125) where an(t) — dan(t)/dt.122b).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.. .iJ5 »*/Vn(x). (10.
(10. (10. since it is equivalent to the latter.130) H'kn by \H'kn and at the end we allow A to tend to 1. In the first place we replace in Eq. i.e. SH'kn(ty^E^l^\t) (10. We restrict ourselves here to the two lowest order equations. (10.134) .130) we develop a perturbation theory. (10. Comparing coefficients of the same powers of A on both sides. to perturbation theory in the lowest order.130) along with the parameter A in front of H'kn.9 General TimeDependent Perturbation We can rewrite the equation as Theory 197 iMk{t) = SH'knan(t)e^EkE^h. we obtain the equations: iha[°\t) iha£\t) iha[s+1\t) = = = ••• . We obtain ih{ak°\t) + \ak1\t) + \2ak2\t) = SXH'kn(t)e^. they are fixed by the initial conditions.120). i. We observe that the amplitude ak of a definite eigenfunction tpk has effectively replaced the wave function ip.^ [a^(t) E h + } + \a£\t) + A2a£>(t) + • • • ] .130) This result should be compared with the Schrodinger equation (10. 0. let's say am / 0.e.132) These equations can be integrated successively. (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. Now we set ak(t) = ak0){t) + \a£\t) + X2a[2\t) + ••• (10. In order to solve Eq. Thus first we have ak0) = const.133) The numbers ak = const.10. and we set ak°}=5km or ak0) = 8(EkEm). Here we assume (our assumption above): At time t = 0 all coefficients ak are zero except one and only one.131) and insert this into Eq. SH'kn(t)e^~E^ha^(t). determine the state of the system at time t — 0. (10.
136) we see that both "amplitudes" are Fourier transforms and look similar. (10. whereas ak is a Fourier transform in time (a distinction which is easy to remember!).137) .e. i.e. (10. ) ' X V(x')dx'. Equation (10. ifm{kl) = f J —oo (10. an expression of the following type AS?) = " 2 ^ 2 / e j ( k . In this case we obtain from Eq. (10. Later we shall discuss scattering off a Coulomb potential in terms of the socalled Born approximation of the scattering amplitude which we there write f(9.ml2sin2{^frn (En .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.Emy in agreement with Eq.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 > 0.198 CHAPTER 10. (10. What is the physical significance of Eq. The amplitude / B o r n is a Fourier transform in spatial coordinates.135) and (10. (10. We shall see that in this approximation the scattering amplitude is effectively the Fourier transform of the potential. (10.132). Quantum Theory: The General Formalism depending on whether Em belongs to the discrete part of the spectrum of Ho or to its continuum.135) We put the additional constant of integration equal to zero. H'^y^"^^/hd£.111). ipm) into the state k.k . we obtain it*™ = f J—oo SH'kn{t')e^E^'lh5mndt'. i. Integrating the second of the equations (10.e. ip) in the context of timeindependent perturbation theory.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. so that ihakl\t = oo) = 0.136) Comparing Eqs.
11. Let r : = (x. Z<i = Z) y(r) = _ M ^ .y.z) be the vector separating two particles which are otherwise characterized by indices 1 and 2 with electric charges Z\e and Z^e.m.Ti. the position coordinate and the mass of particle i with i — 1.2.2 Separation of Variables. meaning the inclusion of the creation and annihilation of field quanta). Both of these important potentials are singled out from a large number of cases by the fact that in their case the Schrodinger equation can be solved explicitly and completely in closed form. The electrostatic interaction of the particles is the Coulomb potential (later we take Z\ = 1. Both cases therefore also serve in many applications as the basic unperturbed problem of a perturbation theory.Chapter 11 The Coulomb Interaction 11. such as the electromagnetic field (the quantization in these contexts is often called "second quantization". ( 1L1 ) We let pi.i be respectively the momentum. 199 (1L2) . Angular Momentum In considering the Coulomb potential in a realistic context we have to consider three space dimensions. r=  r .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 Hamilton operator is then given by w= jt + ji_^M. The Coulomb interaction on the other hand is of specific significance for its relevance in atomic and nuclear physics.
(11. r} representation they have the differential operator forms Setting R = (X. The Coulomb Interaction We introduce centre of mass and relative coordinates R and r by setting R = m i r i + m2r2 • . Similarly d_ _dx^_d_ dx dx dx\ dx2__d_ _m^_d_ dx dx2 mi dx\ mo_9_ m 2 dx2 etc.5) The momenta P . (H4) \mi m. Y.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. y. 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 ' . (11.U7&. Solving Eqs. so that as expected we obtain the expression for p in Eq. mi + m 2 iTi'2 r2 = R ^ —r. . (11. z) one verifies that j9_ = dxi_ _d_ 8X ~~ OX dXl dx2. 2Pip2 m\m2 pip2 mi + m 2 . where m 0 = mim 2 . Z) and r = (x.200 CHAPTER 11. P = pi + p 2 . mi + m 2 va\ (11.3) for the individual position coordinates we obtain ri=R+ •^ —r. /'Pi p = m0 P2 A v.4) P = p i + p 2 ._d_ _ _d_ dX dx2 ~ dXl d dx2 nifi^i [ ' etc. p are such that as operators in the position space representation or {R. But now P2 2mo = "ip/pi 2 \mi m p2\2 m2J  = m0/pi2 2 \m\ p22 m2. so that as expected according to Eq.4). r = rir2 mi + m 2 (11.
Pj} = ih6ij.12) (11.2 Separation of Variables.y.+ ^(r) 2(mi + m 2 ) 2m 0 (H. (1114) m0 We observe that the equations of motion in R. p = W.9) ft = ftcm + ^ = centre of mass energy + energy of relative motion.15) In accordance with Eq.j = x. Canonical quantization of the classical theory implies that we pass from the Cartesian variables rj. (11. (11.13) * and = M ' 1V1 P = 0. we mean n = rx s x. r) is a canonical transformation. for which we postulate the commutator relations* [ri. (i. Pj over to operators ?i. (11.8) and so W =777 v + 7 T . The motion of the centre of mass is that of a particle of mass M = m\ + 1712 in uniform motion along a straight line.2 It follows that p2 2 P2 n2 n2 n2 (H. Hence Hamilton's equations apply and we obtain.10) we set H = Hcm + H.z). 2(mi + m<i) 2mo 2m\ 2m. p on the other are completely separated. etc.pj]=iMij.16) Although we write ri.= ^.r2) to (R. Ri. Angular so that Momentum 201 ^ ^ . .Pj).10) One can convince oneself that the transformation from (ri. pj.il) ""Op' i.pj.Ri.e.+ ^2. r = —. Pj (which in the following we write again ri. 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). ' d and 'AW OH'ATJ P =  ^ dR <9# <9r ' (ii. P on the one hand and those in r. (11. (11. with M := mi + vn^R = ^ P.11.+ P. The canonical quantization must always be performed in Cartesian coordinates.Ri.pj. [Ri.
= rcos9. as in the case of the Coulomb potential.e.20) r2MAR_ $(R) = ER*(R).202 CHAPTER 11.19) For stationary states with energy E the equation possesses a complete system of eigensolutions * which can be written tt(R. to that of the Schrodinger wave equation h2 Ar + V(r) tp(r) = Eip(r).24) .23) y — r sin 9 sin tp. (11. V(r) = V(r).21a) L*+v{r)] ip(r) = Ei/j(r) (11. 2mo (11. it is reasonable to go to spherical coordinates r. (11. where Hcm$(R) = and Hip(r) = with ^total — ER (11. (11. The Coulomb Interaction nc H = 2m. df\ 39 J + 1 d2i/j r2 sin2 9 dip2" (11. 0 < 9 < ir.21b) + E. (11.Q 2M' + V(r).17) (11. i. —n < tp < ir.r).18) In the position space representation the time independent Schrodinger equation is therefore 2M' A fl ) + h2 2mo A r + V(r) * ( R . r ) = J E t o t a i*(R. Thus the twoparticle problem is practically reduced to an effective oneparticle problem. ip: x z = r sin 9 cos ip. In these coordinates we have /\rip 1 d r2 dr \ tdil> dr + r2 sin 9 09 \ S m 1 d_ .22) If.r) = *(R)V(r). 9. 0 < r < oo.
A fc2! d ( d^ ^ = _tfU2dA r\dr and (compare with Eq. (11. (11. is the operator of the relative angular momentum. i.27) We can now write (see verification below) p V = h2Ar^ where 2. = 0.31) .33) i.j. i( i. i—>0 (1L25) (11.24)) 2 + M\ d dr2 J = tfL^(r^A\ r2 dr\ dr J d2ip (11.3. Angular Momentum 203 We can rewrite this in a different form by defining the operator * : =«. the operator —ihd/dr is not hermitian! The operator pr obviously satisfies the relation [r. 2 1 9 / 1 9 .2.12 — I2.1 it is shown that pr is hermitian provided lim n/>(r) = 0.26) However. where (r x p)j = ^eijkrjpk._0di})\ x We now demonstrate that 1.29) .i ' .pr]=ih. if i.2. that the above expression follows from l:=rxp=ift(rxV).32) (11.3. . k = 1.j. j . otherwise. with tijk — + 1 .2. = (p2 + ^ V (r + 0).2 Separation of Variables.5r = . = —1. We have I2 = (r x p) • (r x p). n ( . k an anticlockwise permutation of 1. (11.3.11.e.k a clockwise permutation of 1. (11.)In Example 11. (11. _ h2 \ .28) . (l + .
ihr • p r2p2(rp)2 < + ih(rp). (11.ihr • p ) .44) .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.ih)pr and x 2 2 9 9 2 2 Q O (11.36) It follows t h a t (note the part defined as —5) l2 = = ( r 2 p 2 .40) (11. (11. (11.35) j.k Using Eq.39) + ih).ih) — rpr(rpr With Eq.(r • p ) 2 + 3ih(r • p) .204 It follows t h a t 2_^£ijktij'k' i CHAPTER 11. Eq.k ~ rjpkrkPj) (11.ih5jk)pk (11. (11.43) r2(p2p2). This means that l2 = so that 9 9 1 (11.25)) r • p = —ihr • V = rpr + ih and hence 8 = (r • p)(r • p .15) we have fjPkTjPk = fj(rjPk and TjPkTkPj = = rjPkipjrk rjPj(rkPk + ihSjk) = TjPkPjTk + ~ ihSkk) + ihrjPj. r— = r • — or or (11.42) (11. * ' S Since r =xz we have (cf.37) . ih6jkrjpk (11. The Coulomb Interaction = &jj'&kk' — Sjk'Skj'.j.41) o = r pr. (11. (11.38) + y* + z .27) we obtain r{prr)pr = r(rpr .
reR3.2.e.e.2TT].d. S o l u t i o n : Set d£2s = sin. /'(r.Pril>) J / r= dQs rcj>(r.— (riP(r.<P) = Etl>(r.2 Separation of Variables. r := r : lim rip{r) = o i .e'iP^} .47) Example 11.11..[ 4>(r. <p)* . h f l d —(r<l>*(r.Q ' 2mor2 with the condition (11.28) and (11. r—>0 il>(r.<p)drdna " r r=0 J \ll§r~(r4'(r.! /').[ i J With partial integration with respect to r this becomes (<l>. Then {<j>. (11. 11. Angular Momentum 205 Comparison of Eqs.e. 8.<p))drdns.<p G [0.e.6d0dip. (pr4>.<p) (11.46) (11. <p)*rip(r} 0.30) I2 is to be identified with the square of the angular momentum operator. <p))r2drdns % J r or r<t>(r.e. (11.30) yields l2 = I2 = sin sine— BinO^r 09 + d^2 (11.l2} = 0. pr = . 6.(r.48) . or J .26). 0.6l.d_ r dr the operator pr is hermitian. i. ¥ p)r 2 drdn s = (p r 0. <p)r2dfls = .45) oe Thus in Eq. The Schrodinger equation of the relative motion can therefore be written V1 V 2m.e lim rip(r) = 0.1 Separation of variables We deduce from the expressions of H and l 2 that [H..<p))ril>(r. <p) d °° .1: Proof that pr is hermitian Show that if Vv lipe H = C2(R3). i. H 1 2mo pi + K) +V(r). 9. 9 € [0. 6. ip 6 X>Pr C 7i. (11.7r].6. <p)*Pri>(r.ip) = (<£. Pri/Oi V0. prip) = = I <j>{r.<p)*?(ri.
g i = x. ly] = [yPz ~ zpy.49) By cyclic permutation of x. zpx] + [zpy.e. z]px + py[z.lz (or l . We therefore first search for a complete system of eigenfunctions of l 2 . Since 1 = r x p . i. ?3 = z). y. so that [lz.53) Either of these operators is the hermitian conjugate of the other.52) This means that l . 3 . we have [h. ly\ — —tfi(lylx + Ixlyji [hi h\ = 0. they are incompatible variables.lz]=ihlx.lx] = ihly.ly) are simultaneously "sharp" determinable variables and hence have a common system of basis functions. 11. 2 2 2 = 0. H and l 2 possess a common system of basis eigenvectors. [lyA = 0.lx or l . i. (11.3 Representation of Rotation Group Our objective now is to develop a representation of the rotation group in analogy to the energy representation of the simple harmonic oscillator. /z> *xJ = l <i\}ylx + tx'j/Jj [h.l2} and correspondingly [lx. (11. and this implies in the terminology we used earlier that their determination cannot be "sharp" simultaneously.50) Thus the components of 1 do not commute pairwise. [ly.xpz] = [ypz. Example 11.qj]=iheijkqk S o l u t i o n : This can be verified in analogy to the preceding equations (i = l .q2 = V. for any two of the components there is no common complete system of eigenfunctions. 2 .ypx) = ihlz.ly]=ihlz.2: Verify that [h. The Coulomb Interaction i. [lz. zpx .e. (11. However. z we have altogether [lx.206 CHAPTER 11. We can now proceed as in the case of the harmonic oscillator and determine their eigenfunctions.l2] = 0. xpz] = y\pz.51) (11. (11.pz]x = ih(xpy . In order to determine the eigenvectors of the operators l 2 and lz one defines l± = lx±ily.e. These operators here play a role analogous to that of the quasiparticle operators .
* Thus we have (1 being a bounded operator) lz\l) = l0h\l) Then from Eq. We write or define: / .Y\l) = (lQr)h(lY\l).61) But lo is by definition the largest eigenvalue of lz.i+] = m+. (11. i + ] = [l2.+ ll.54) On the other hand we obtain from Eqs.l)hl\l). (11.lh)\l) = {l0 .i] = m. (n.r) := (Z_)rZ) so that lz\l .r) (11 = {l02)h{l)2\l) (11.h~lz(1156) (11.e.54) l 2 = \{l+l~ + 11+) + ll = l+l. (11.). With the help of Eq. since lz is hermitian.54): lzl\l) = {llz .59) = 9) h(l0 r)\lr). (11. (11. A^ in the case of the linear harmonic oscillator.54): izi+\i) = (i+iz + i+h)\i) = (lo + i)ta+\i). and with Eq. [i+M = 2%iz. y .3 Representation of Rotation Group 207 A. (11.52): 0 = [ l 2 . in particular one defines \l) € CKj as eigenvector of lz with the largest eigenvalue lo.51). (11. (11. (11. . Therefore (lo + l)h cannot be an eigenvalue of lz.57) i.55) One now defines certain ketvectors as eigenvectors of lz which span the space "K\ of the appropriate states. (11. l\l) is eigenvector of lz with eigenvalue (IQ — l)h. assuming a finite dimensional representation space.Z_] = [ l 2 . (11.11.62) 'At this stage it is not yet decided whether IQ is an integer or not! But we know that IQ is real. In the present context the operators are called shift operators for reasons which will be seen below.49) we obtain the following relations: [iz. [ig.60) Similarly for l+ with the help of Eq. Hence we must have Z+I0=0.58) {l0 : max. Similarly we have lg{l)2\l) and more generally lz(l.
+ i2z (11+ + 2izh) + 1 \ \i) = (izh + ti)\i) i0(i0 + i)h2\i). (11.65) We assumed that "Ki is finite dimensional so that the sequence of eigenvectors of lz. (11.54) (n = 57) (11.55)) we have 12Z_Z> = l.65) for r = n: l2\l n) = l2(l)n\l) = lo(lo + l)h2\l . (11.e.\ll).63) + l ) ^ 2 .\lr) terminates at some number (let us say) r = n (n a positive integer).e. (11.. [12.64) i. i. Since l 2 commutes Hence Z) is eigenvector of l 2 with eigenvalue with l_.l2\l) = l0(l0 + l)h2l\l).64) we obtain on multiplication by Z_: lVl\l) and more generally l2(f)PJ> (11.56) that l2/n) = (11 6) (l+l+l2lzh)\ln) (lll. (11. l\l . (11.From Eq.n)}h2\ln).68) .208 Thus for l 2 we obtain: V\l) ' = 11.+ 11+) + l2z\l) = i+i. (11. IQ(IQ 1 0 (11.60) and (11.n) (1 = 60) {l)n+1\l) = 0.. Eq. (11.n).h)\ln) {(lQnf(lQ. l\l) is also eigenvector of l 2 with eigenvalue Zo^o+l)^ 2 .y\i) l)h2\lr).e..56) CHAPTER 11.66) It then follows from the second of relations (11. The Coulomb Interaction (l+l. i.. i.60) = Vli\l) = l0(lQ + l)h2li\l) i0(io + l0(l0 + i)h2(i. \l).Z_] = 0.e. (cf.67) i (11 60) = But according to Eqs.
. \l — n) into one other.70) they satisfy the eigenvalue equations (with m = l0Jo l.n) 2 . lz) are then: (l.m).l)h.. p. Messiah [195].m) = mh\l. M. Brink and G. we skip this here.  U o .11. i. The basis vectors \l).. —loh. The normalization factor in Eq.l)h \/(lo + m + l)(l0 . (11.. *D.m .Z 0 )... lo = l0n. According to Eqs.71) can be established in analogy to the method used in the case of the harmonic oscillator. 24.m / + l)h . Vol.m) — mh\l.{lo .l_. The dimension of the representation space is therefore n + 1 = 21Q + 1. (11. We write these* for integral values of IQ: \l.67) we obtain (observe IQ(IQ 209 + 1) = —lo(—lo — 1)): *o(*o + 1) = Wo . Satchler [38].m±l\l±\l.n)} = (l0 .. lz\l.m) : Mo).m).63) and (11.lz transform the vectors \l). lQ = .. (11. m + l)h.3 Representation of Rotation Group so that with Eq. We therefore write the 2/o + 1 orthonormal vectors: \l..m) (l. where lz\l.m) = IQ(IO + l)h2\l. In the following we consider primarily the case of integral values of IQ. y (lo .n){l0 .n .69) Prom this it follows that IQ must be either half integral or integral.1).m).65) these are eigenvectors of l with eigenvalues Zo(^o + l)h2.m) = = mh. As in the case of the harmonic oscillator we introduce normalized basis vectors..m)\l.e..m) = = = = y/lo(lo + 1) — m(m +l)\l. § See e.l0): l2Z. (11.. . \l — n) are eigenvectors of lz with eigenvalues loh.g..m + 1)(Z0 + m)\l. A.m\lz\l. y/(l0±m + l)(l0^m)h. II. The phases are chosen such that§ l+\l. The nonvanishing matrix elements of 1 = (l+. p.. R.1). (IQ — l)h. (H71) These expressions should be compared with the corresponding ones in the case of the harmonic oscillator.. The operators l+.70) According to Eq.2. 17.m y/lo(lo + 1) — (m — l)m\l.. (11.l.m) l\l.m) with — lo<m<lo2 (11..
ip) = e(9)$((p). (11.e. In these coordinates we obtain: h l± so that h2 = xpv . We now perform the analogous procedure for the angular momentum or. $ oc eim*.<p). 9.1. with i2Y£{e. l_\l. i. .75) the variables contained in the expression (11. . . and m = —IQ.<p) = i0{io + lzY£(0.74) can be separated: 1 d sin 9 36 2 1 sm6— 89 j — sin 0(9) = h(l0 + 1)0(9).210 Then.yp.<p) = i)n2Y^{e^). d2 sin2 9 dp2 _' (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. where IQ = 0. for which the eigenfunctions \ip) are given by l2 l0(l0 + i)h2\ip).. ip. (11.2.72) In view of the spherical symmetry it is reasonable to use spherical polar coordinates r..77) $((p) = m2<b(y). ih d_ dip' lx ± ily = ±he^ d I ^ ± zcotfl^ . —lo + 1 .. We .76) d2 (11. .4 Angular Momentum:Angular Representation In the case of the harmonic oscillator we arrived at the position or configuration space dependent wave functions by considering the position space representation of states. for the simple "rotor".l0)=0. CHAPTER 11. 11..With the ansatz Y(9. /o. The Coulomb Interaction Z+Mo> = 0.74) We choose the eigenfunctions of l 2 and lz as basis vectors of the ("irreducible") representation.73) 1 d d sm0 06 \Sm9dd) + (11. as this is sometimes described. as remarked earlier. (11. as above.
<p) = (iy 47r(Z + m)! In particular for m = 0: 1/2 PI11 (cos 6)eimi?. (11.11.(5cos 03cos0). (11. m > 0. (11.76) and (11. and that their phases are such that Yt0(0.n>). We mention in passing that in spherical coordinates d lz = xpy .^ = *(n . / 3 / 5 J — casO.\ = W . (11.i».=o m=l sin 6 The connection of these hyperspherical functions with the associated Legendre functions P™. These functions are defined in such a way that they are normalized to unity on the unit sphere.r o ) ! YT{e. y2° = \ (3cos 2 01) 2 V V 4vr V 16?r 3 W—.77) and o these are determinable by the requirement of uniqueness of the original wave function and its square integrability.78) and the completeness relation oo I ' YrVMYTVrf) = °^~' ± J2 Y7n*(f}.(p) are called spherical harmonics.ypx = tfrK(11. orthonormalized system of eigenfunctions.V>)Yir'(8. This is the case if one demands that l 2 and lz possess a common complete.t»'\Y7n((l'. so that im(ip±2ir) The functions Ylm(8. Uniqueness implies immediately that m has to be integral. is given by (2Z + l ) ( Z .* ? ^ .4 Angular Representation of Angular Momentum 211 observe that Z and m define the eigenvalues of Eqs.0) is real and positive.82) .<p) Jo = 6mm'5w.80) y? Furthermore Y°l = V^fcosO).79) . The functions Y™ satisfy the orthonormality condition dfilimT' = / Jo < W sm8d9Yr(9. (11.81) 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.
Setting y = (lx2)m>2Pr(x) m .88) we obtain for every value of x ^2 ai(K + 0( K + * .e. The associated Legendre functions or spherical functions PJn(x) satisfy the differential equation (1 . (11. (11.2. oo (11.Y^O»[(K + i)(/e + i ~ 1) + 2(re + i) .2xy' + 1(1 + \)y = 0.l)xK+i~2 i .87) we obtain the equation (1 . i.m = 0.1. i.89) Inserting this expansion into Eq.d. i=0 (re lowest power > 0).m(m + l ) ] i f = 0.e. 22 (11.h. If we did not know yet that I = IQ = an integer > 0.2(m + l ) x i f + [1(1 + 1) .e.2xy' + 1(1 + 1) 1x The number m must be an integer already for reasons of uniqueness of the wave function. in the replacement ip —> if + 2K.x2)y" .x2)Pf . exp(imip). we would now have to search for those values of I for which the solutions are square integrable.g.x2)y" . is also in use.p.. of Ytm..212 CHAPTER 11. f.83) Their connection with the Legendre functions Pi (x) is given by the relation pr{x) with the differential equation =£^ Pi{x) (iL84) (1 .x2)P[' .90) . i.86) y = 0. Instead of the values I = 0.e.2xP[ + 1(1 + l)Pi = 0. (1 . the designation s. For the solution we use the ansatz of a power series. It is instructive to pursue the appropriate arguments..85) (11.88) y = ^aiXi+K.. i. (11. The Coulomb Interaction The number I is referred to as the "orbital quantum number" and m as "magnetic quantum numbed. (11.1(1 + l)]xK+i = 0.. (11.. respectively. Consider the equation for Pi(x).
95) Vl = rxiThe hermiticity condition for pr then requires that (cf. l(l + l)h2 _2mo + 2m. when the series terminates after a finite number of terms. (11. We have therefore (multiplying Eq. first line of Eq. 93. p. these are of little interest to us here.26)) yi(r = 0) = 0. For the solution in accordance with (11.46)) p.E Xl(r) =0 (11. Eq.1(1 + 1)]. M.2 + 2mo dr 2mor V{T)E Viir) = 0 (11. (11. . Murphy [190]. l 2 and lz are the solutions of the Schrodinger equation of the form (11. Comparing the coefficients of xK+3 we obtain aj+2(K + j + 2){K + j + 1) = OJ[(K + J)(K + j + 1) . (11. they have to be polynomials (recall the orthogonalization procedure of E.1. .91) from which we deduce that K = 0. Margenau and G. but also those for a different series for a. We see that this happens precisely when I is a positive or negative integer or zero. 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. Eq.29)) 2 fe2l 92 Pr = & r 7Tororz We set (11.Qr2 with (cf. For \x\ < 1 we can obtain from this the coefficients of a convergent series.11.5 The Radial Schrodinger Equation for Hydrogenlike Atoms For i = 0.1 we obtain by comparing coefficients QOK{K 213 — 1) = 0.78) to be square integrable.96) + V(r) .94) ^For further details see H. ai(re + !)« = 0. (11.*" 11. (11.92) This is a recurrence relation.93) where Xl(r) is the solution of the radial equation (cf. > 1. Schmidt).94) from the left by r) h2 d2 h2 2 + H! + \)IL. (11. However.
L i + + . He + .214 with CHAPTER 11.1)). We assume that V(r) has at r = 0 at most a singularity of the form of 1/r. Equating the coefficient of 1/r2 to zero. 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. The equation also has an irregular solution. We now consider the hydrogen atom with V(r) — —e2/r (i. and others.96) can be shown to possess a regular solution there which behaves like rl+1{\ + 0(r)) in approaching r = 0. In these cases the behaviour of yi near the origin is dominated by the centrifugal term and thus Eq. (11. The Coulomb Interaction (11.98) 2mo e2 _ 1(1 + 1) yi = o e+ 2 y'l + h r r2 2m0E e = (11.100) with (11.101) 2 r ^ We set  6 = ( _ e ) 1 / a = ( = (11. Jo Jo We thus have a problem similar to that of a particle of mass m in the domain (0. This applies to the Coulomb potential but also to Coulomblike potentials like screened Coulomb potentials or the Yukawa potential. and substitute this into Eq.e. Z\ = Z2 = 1 in Eq. since it satisfies neither the condition yi(0) = 0.e. (11. which behaves like 1/r1 in approaching r = 0. nor the condition of normalizability for l ^ 0. s = 1 + 1. an analogous consideration applies to hydrogenlike atoms like. Obviously the irregular solution has to be rejected. In order to verify this behaviour we set <VflV>r> yi(r) = rs(l + air + a2r2 \ ).99) e ip(r) = Eip(r). for instance. In the first place we investigate the behaviour of yi in the neighbourhood of the origin.102) .96). i. r m0 = meMp me + Mv (11.1)+1(1+ 1) = 0. (11.1. oo) of a onedimensional space. we obtain the socalled "indicial equation!'' s(s .97) / r2dr\Xi(r)\2 = / \yi(r)\2dr.
We set therefore Vi X J.e. (11. (11. . *The factor 47reo appears in SI units. behaves there like xl+1) extended to an exponentially decreasing branch at infinity. We set x — We set as Bohr radius* a = and 1 v = rea rriQe ft V (47Fe 2KT. We can see from Eq.a * = 0.11. Magnus and F.+if>x/2 vi(x).103) °y moe2 ( = 0. (11.106) 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. p. (11. see e. Oberhettinger [181].x ) $ ' .105) we obtain the equation for V[(x).100) as^ d2 dx2 1(1 + 1) x v___\ 4 2/1 = 0. for E < 0 they > decrease exponentially.107) equation. W.6 11.105) t h a t for x —• oo there is a solution behaving like exp(—x/2).6 The Discrete Spectrum of the Coulomb Potential 215 For E > 0 the solutions oscillate in the region r — oo. i. T This equation is known as Whittaker's equation.108) This equation has the form of a confluent x$" + (6 .e. (11. (11. Inserting the ansatz into Eq. It remains to determine the function vi(x) for a complete determination of the solution.105) 11.g. hypergeometric (11. (11.88.e. i.104) W i t h these substitutions we can rewrite Eq. d2 x—j 1 + (2l + 2x) ax d j dx (/ + 1 .529 x 1 0 " 1 0 meters) 1 V/2 e2f hc\ m0c^1/2 2E 2moE J (11.6.v) vi(x) = 0.1 Discrete Spectrum of the Coulomb Potential T h e eigenvalues Here we are interested in the solution which is regular at the origin (i.
called dupZication formula. more precisely. 02zl / i\ (3) r(2z) = — ^ r ( z ) r ( z + . (2) r ( z ) r ( l .109) In our case we can write the series The expression* r(x + l) = x\ is the gamma function. However. (11. The Coulomb Interaction The regular solution of this equation is the confluent hypergeometric series (also called Kummer series) 1 .z) = ^f^. .z(zl)\ = z\. (11.e. The function T(z) is defined by the integral /co T(z) := / eHzldt. and for both cases. (4) r ( i ) = V5F.111) Important properties of the function T(z) are: (1) zF(z) = T(z + l). 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. ^ ( 2 ^ ) ! = 2 2z z!(z . (11. i.112) The gamma function is a nonvanishing analytic.(0_i)!(6 + n l)!n!' (11. particularly pure mathematicians. + ' ' ~ 6 1 ! + 6 ( 6 + l ) 2! + Z. it is convenient to use only the factorial notation. with many of these around in some calculation.106). the factorial for an arbitrary argument.J. and hence the *Some writers. 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. meromorphic function.! ) ! = imlFI' c a l led inversion or reflection formula.216 CHAPTER 11.1/2)!. (z)!(z . (5)r(n + l ) = n ! . since the exponential function (generated by the infinite series) would destroy the asymptotic behaviour of the wave function which we separated off with Eq. are very strict in reserving the factorial exclusively for positive integers and the gamma function for all other cases.
V))T(v . which is thus seen to be a consequence of demanding the square integrability of the wave function. every n.114c) (Recall that for an arithmetic series a + (a + d) + • • • + (a + (n — l)d) = na + n{n — l)<i/2.104)) 1 n =v = K.) 'Observe that with property (2) of the gamma function we can reexpress the ratio of gamma functions in Eq. if the series terminates for certain values of u.1) + n = n(n . • • • • <»•»*> It should be noted that the magnetic quantum number m does not appear in this expression. for z.1) + n = n2.e. . in our case: a = 0. 2 .v) T(l + lv) _ ~ K sin{7rQ/ .I) r(vl p)sm{n(i/ I .. 1=0 (11.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. which are also described as its nodes. One defines as the principal quantum number the number n. i. = n = Z + l + n'. the orbital or azimuthal quantum number I can assume the values 0.11. i. d = 1.101) to (11. This is achieved. Thus the infinite series must break off somewhere. 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 . . . With (cf. The quantum number n' is called radial quantum number. This number is equal to the number of finite zeros of the radial part of the wave function (excepting the origin). and so to the number of different values of m.. i. . n' = 0 .110) as follows: T(l + 1 + p .e. 1 ..6 The Discrete Spectrum of the Coulomb Potential 217 latter would not be square integrable.113) This is a quantization condition.2. (11. Eqs.1.. (11.a " \ 2moE we obtain the energy spectrum of the discrete states with angular momentum I ie t m0e2 / 1 N1/2 * =  ( s ) ' ^ . . For every value En.e.» • » • » . (11. must be a polynomial.
The following spectroscopic description of states is customary: nl —> ns. and from n. where the superscript 21 + 1 at the top left of I gives the degree of degeneracy. hence p). .d.p. The Coulomb Interaction E=0 degeneracy lEl 3 31s 21s 33p 23p 3 5 d 9fold IE I 4fold IE I 1 1 1s none Fig. from an initial state n > 2. (c) The nucleus was treated as pointlike and not as something with structure. 11.l = 2 to n = 2.1. 0.1 = 0 (called sharp series. hence d).... (b) The spin of the electron was not taken into account.'. I = 0. np.. nd.1 = 1 to n = 2. The spectrum of the hydrogen atom thus has the form shown schematically in Fig.§ One also writes n2l+1l. . stand for I = where n denotes the principal quantum number and s. One describes as Balmer series the final state n = 2.4. . The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically..1 = 1 (called diffuse series. 2 . . 11.\ I = 0 or 1. / = 1. .218 CHAPTER 11. I = 0 to n = 2..1 = 0 (called principal series..1. from an initial state n > 2. nf .. ' O n e describes as Lyman series the final state with n = 1.'.1 Hydrogen atom energy levels. hence s). respectively. and hence the fine structure of the energy levels is excluded. ne. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2) as we illustrate in Example 11. from n.
We distinguish between different definitions in use. (11. S.6 The Discrete Spectrum of the Coulomb Potential The associated radial wave function is yi(x) 219 = xl+1ex/2<f>{n'.2l { + 2.115) 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. Messiah [195]. Gradshteyn and I. Unfortunately the definition mostly used in mathematical literature differs. p. Each can be recognized by reference to the generating function.117) The function *Lk(z) is a polynomial of degree r. 84. Magnus and F. (u. W.2 (11. formula 7.6. it is given by^ *Lkr(z) = ^ ^ f 1 *ir. Vol. . For clarity and later use we denote the function used there by its normal form L^. 2 . (11.114b) yields the binding energy of the states. 3 . • fc. see. in fact. Unfortunately. . and I. • (11.= t . for instance. 11.11. .z) = (r + k)\ Lkr(z). several different definitions are in use. (11. however.115) ' =0 (ri . 844.118a) where Lk. Ryzhik [122].1.115) is effectively the associated Laguerre polynomial normally written Lk(z). Oberhettinger [181]. M. The mass of the hydrogen atom is given by massif = Mp + m e + E < Mp + me. l . I. Appendix B. With the help of Eq. We mention here three different definitions used in the literature.414(3). .{z) is the associated Laguerre polynomial as normally defined and mostly written Lr (z). • ( .r = 0 .2. "1'f„+11" P] „. . so that one always has to check the definition.110) is usually reexpressed in terms of orthogonal polynomials known as associated Laguerre polynomials. Apart from factors. the polynomial contained in y\ of Eq.x).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. k + l. a + 2:*) = t v . Thus Messiah [195] defines the original or stem Laguerre polynomial as : >Lr(z) = *L°r(z) = e'^(e'zr).r f . p.p)\(2l + 1 + p)\ The spectrum (11.
p.3): e2t/(lt) (1 . which we write ^Lk.d z—^z + {k + l~z)—+r Lkr{z) = 0. I. 163. Sneddon [255]. Messiah [195]. is defined by^ dk (—'\\k(r^\2 (11. .2. Example 11.„\*Kk. 162 .119) dz dz In particular one has (we cite these for later comparison) LQ(Z) = 1.** The associated Laguerre polynomial here. and Ll(z) = l. r r Z r=0 v tr k k (11.3 we show in a typical calculation how this normalization condition can be obtained with the help of the generating function. 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. L\(z) = 2z.121) In Example 11. There at some points n + 1 is misprinted n + 1. .120) ' r=0 It will be seen that the orthonormality condition is" z ezJc*Tk. It is said that this is the definition preferred in applied mathematics. The reader is there warned that '•'•care must be taken in reading the literature to ensure that the particular convention being followed is understood'.. .(z)*L TJz)dz 3 = KP + kV}Opqit Jo (11. N.z .. L1(z) = l . ^ F o r this definition and the following equations we refer to I. N. Appendix B.220 CHAPTER 11. Sneddon [255]. The Coulomb Interaction The differential equation of the associated Laguerre polynomials is (same for *Lk{z)) d2 .164. For our purposes here it is convenient to refer to yet another definition of the associated Laguerre polynomials. Ll(z) = 33z + ±z2. **I.1. pp.122) "See A. Vol. (11.„M„ z"*L. L2(z) = l2z + z2.
8): (11.1 .119): ty"(t) + (a + 1 . (11. (11.119) derive the generating function of associated Laguerre polynomials (11. (11.124a) (i*) pi The orthonormality condition is (cf.. with y'{i) = — / ps(p)e~ptdp ty'(t) = t f ps(p)eptdpp2s(p)ept s(p)e Pt J ^(ps(p))eptdp.121). and partial integration.122) we see that tLkr+k(z) = (l)k*Lkr(z) = (l)k(k + r)\Lkr(z). (11.z . and obtain with this the normalization condition (11. *L\(z) = . ^L2(z) = 2 . (11.4 + 2z.11.120)) it)k& zt/{lt) ~*P p=k fc (11. The generating function of the associated Laguerre polynomials so defined is (observe (—t)k as compared with (11.120).124c) Example 11.. Then. (11.124b) Comparing Eqs.107) implies k = 11 + 1 and r = n + I. N d rk 0.. i L1(z) = l . ^L\(z) = .. Solution: We consider the following equation of the form of Eq. '"''Comparison with Eq.% ' ( * ) + by{t) = 0 with ansatz y{t) = I s(p)e~ptdp. and *L\(z) = .4z + .123) For the purpose of clarity and comparison with the polynomials of the other definition it may help to see that here in particular % ( « ) = !. . Example 11. Jfpd>2s{p))eptdp.1 8 + 18z 3z2.3: Generating function of Laguerre polynomials With a Laplace transform ansatz for the solution of a second order differential equation like Eq.118a) and (11.6 The Discrete Spectrum of the Coulomb Potential The differential equation of these associated Laguerre polynomials is** z 221 ~r^z + (k + lz)— + dz dz d2 n . The limits will be determined later.
It follows that 1 d i I _i. Case (2) implies the exponentially increasing solution ~ e + t . We demand (to be considered later for integration contours) the following condition..p ) o + i 7 t = 0 .q) and P~ t ( . Case (1) with p —• oo implies y(t) ~ / dppa~1e~pt ~ t~a. agreeing with Eq. s bK> (lq)°+i (11. Then the preceding two equations (inserted into the original differential equation) leave us with I eptdp d {p(p + l)s(p)} ." b + 1 e~ptdp = coefficient of pb in g(p) := (p + \)a+be~pt.p + 1 = 1/(1 . Hence one defines now as associated Laguerre polynomial L£ and as the generating function g(p) of these (Cauchy's residue theorem requires the coefficient of 1/p): L b(t) •= — <£ 27T. with requirement of a finite solution. n! (6 — n)!(a + n)! An alternative integral representation is obtained by setting p = q/(l — q) or q = p/(p + 1). and (4) around p = 0 if b > 0 and integral.^ . since for n > b : (b — n)\ —• oo. j p = (P 0 "V?. t > 0. .124d) In order to obtain the orthogonality and normalization of the associated Laguerre polynomials.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 = 0.222 CHAPTER 11.] from the above equation: C(p) := p(p + l)s(p)ept = 0.119) with a and b as there.. this is the rejected irregular solution. (11.( a + l ) p s ( p ) + 6s(p) dp 0. p and one has (as polynomial of degree 6. we insert this contour integral representation of the polynomial in the following integral and obtain __?(_) = P da <* .p ) « + i q c + i ( i . The Coulomb Interaction since d(p2s)/dp = ps + pd(ps)/dp. 2TT_ Jq=0qb+l(lq)a+l 1 .118b)) Ll{t) = = coefficient of pb in ^ (p + l ) a + t > n=o £ ( ~Pf" "" coefficient of pb~n in (p + ! ) « + ' ^ = /g <*>" (a + 6)! n! —(. Thus in the case of Eq. p° Possible paths satisfying this condition are: (1) If b < 0 from 0 to oo. which removes [. Then dp = dq/(lq)2.a = 21 + 2. the only possibility is the last. case (3) similarly. 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./ ( l — <?) —— = G(t). Hence the expression in square brackets vanishes. (11. (3) around p = — 1 if a + b is a negative integer. (2) from —1 to oo.) 2 J J p 6 + i ( l .
3 T h e eigenfunctions According to the previous discussion we obtain n2 orthogonal eigenfunctions in association with the eigenvalue En. f a p)(i 1) r + 1 l (1P9) J The coefficient here required is obtained from the expansion as . 11.125b) 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.11.c (al)\ d(col)! c(c + a)\ c!a! with use of the inversion formula (11.(x).111).—: <f h_nJ_[ dp = (a + c)\ T^ if c = o. The wave function of the state (n. and 0 if cj=b.6. Clearly only the second form is sensible.112). .6 The Discrete Spectrum of the Coulomb Potential 223 We evaluate the integral on the right with the help of the integral representation of the gamma function. I. where Fnl(x) = yi(x) (11. and obtain r taet[P/(iP)+q/(ig)+i]dt Jt=0 With this the integral I becomes = . (11. Then Nni is a dimensionless normalization constant and (note that we use the associated Laguerre polynomials as in the book of Sneddon [255]) Fnl(x) = xlex^LZ. Jo (11. is introduced for dimensional reasons.125a) and the factor a~ 3 ' 2 with a = Bohr radius. Eq.?). m) is 3 2 &nlm = a~ / NnlFnl (—^Y^O. so that I = (c + a)! 1 f . (11.127) . /•OO / Lg(t)£?(t)taefctt: (a + c ) ! .—i.
3(2Z + 1) + 2} 2 {2(n + 0 . (11.! ] + (21 + l ) 2 .3fc + 2). This behaviour is indicated in Fig. 2 (11..129) n2a' Analogously we have roo j = p2 I r n \ 2J ° J0 'V^yr nl _ na J^° xF^(x)x2dx _ na f^SJn+K 3 ) \na) na {6(n + Q2 .6(n + Z)[(2Z + ! ) . (11.1 + 1) (11. and for the explicit derivation of the second case below Example 11.v (6n2  6nk + A:2 + 6n . = 0) we have ) r /u =.(2Z + 1) + 1} = ^[Zn21(1 + 1)]. The Coulomb Interaction One finds in particular (see also Example 11. from which we can deduce information on the behaviour of the electron: rF"\nrg)r2dr ^ _ 2 £> n iF&ztfdx *%(*)*** 1 Jo°° Fix ()r2dr 2 ^i + /„. For the ground state (n = 1.131) .(l) n+l.7.128) These expressions permit us to obtain the mean or expectation values (1/r) and (r).2Z . a (r)is = ~a. 11.8):* OD  <n!>3 (nk)\ (nfc) (n\)3 i£ n (3) = _ .224 CHAPTER 11.n+l na l2 ° ^ 2 na n+L+lV) (2n + 2/ .2.130) We see that the mean value of r is the larger the larger n is. + . thus on the average the electron is the farther away from the proton or nucleus the larger n is.
In the Kepler problem of a particle of mass mo in Classical Mechanics with the Newton .11. or (r ) ~ n a 2 2 ^ [ 3 n 2 .( n . with n +1 replaced by n and 21 + 1 by k. 173/174. Sneddon [255]. In this case (r) = = For (r 2 ) one obtains < r 2 ) = n 2 ( n + ^ ) ( n + l)a 2 .l ) n ] =  [ 2 n 2 + n] an[ n + (11. i.(r) = x W 2 n + 1 =.j . This corresponds to E ~ — e2/2n2a like classically with circular radius n 2 o. 11.134) *These are taken from I. Problem 5. = l a nl n N'nll rFnl. / ! J — .2 Hydrogen atom wave functions.^ We obtain therefore for the quadratic deviation (r) .. In the case in which I assumes its maximal value n — 1. pp.133) for large r.6 The Discrete Spectrum of the Coulomb Potential r. S.132) (11. v ^ V2n + 1 (11.J n 1s 2s 3s *r Fig.e. \f{r ) ~ n a.2 I 3/2 Kl r 225 \2 1 R„. we have and tp becomes particularly simple.20.2 1 A r = V (^2) .
Thus we have to evaluate the following gravitational potential written V(r) = —mofi/r.4 the relativistic equation in cylindrical coordinates.— > where ampler is the length of the semimajor axis of the elliptic orbit. i = 0. we have pg = const.r. p2 = p2 + — pg • eV. but rather like classically in a plane (i. i. In order to obtain an impression of relativistic corrections in a simple context. in which periods are obtained from the BohrSommerfeldWilson quantization condition. Example 11. 6. pJ = —— and j> Pidqi = nth. for these values of I it is not distributed with uniform probability over the spherical surface. Solution: We have H = yjm%c4 + pIn cylindrical coordinates. . the states Ytl are predominantly concentrated in the xyplane).3.135) This expression becomes very small for large values of n.226 CHAPTER 11. we consider in Example 11. Since 6 is a cyclic coordinate.e.e. Thus. identifying yj (r2) with aKeplen w e see that T K ep l e r = ^(r2)^ = ^(n2af/2 = *^a3'*n*. This means. relativistic corrections are of the order of v2/c2 ~ 0(En/moc2). r. However.6. determine the energy E by evaluating the BohrSommerfeldWilson integral of "old quantum theory". „ _ 'Kepler — the Kepler period is given by o Z7ra 3/2 Kepler . The above treatment of the hydrogen atom has obvious shortcomings: (a) The hydrogen atom was treated nonrelativistically. \/v vT 7 v^ See also Example 14. = ngh. (c) The nucleus was treated as pointlike and not as something with structure.4: The relativistic eigenenergy using cylindrical coordinates Using the formula for the relativistic total energy E of a particle of mass mo and charge e moving in the Coulomb potential V = ~Ze/r (with E numerically equal to the corresponding Hamiltonian H for a conservative system). the electron remains practically localized within a spherical surface of radius (r). and separating this in cylindrical coordinates r. (b) The spin of the electron was not taken into account and hence the spin fine structure of the energy levels is excluded. The Coulomb Interaction or TT = ^ = (11.
r\ = r ( l — cos#) = r — z. but depends very slightly on ng alone which gives rise to the "fine structure" of hydrogen lines. . ( v c + i ) .136a) (11.6 The Discrete Spectrum of the Coulomb Potential 227 integral which we achieve with the "method of poles at infinity" explained in Example 16. ip are the spherical coordinates. It follows that (with h = 2nK) Z2e c h?ni 2 2ixnrfr = — 27r nehtjl  Ze2E >^m2c2E2/c constant'. <p = <p. Whenever there is degeneracy. 2 2 Therefore the energy is no longer a function of n = nr + ng. Thus E2 2EeV(r) e2V2(r) nrh = a> prdqr = ilc2 .d r Kmlc2 + T U^\±AnlK2^Ur 2 2 \ c / r * + * ! + $*—».2.3. the "fine structure obtain the ordinary nonrelativistic energy) Z2ei c2h2[nr + ne^jl 1/2 E — mac + 1 Z2ei/c2h2n2] 1/2 VTIQC 1+ •+ ng^la Z /r. which are related to Cartesian coordinates in the following way. which correspond to a new choice of polar axis. 11.* See L. (p). it is possible to separate the Schrodinger equation in other coordinates.6. where the constants are identified by comparison. (11. p. y — 0: V ^ c o s tp.136b) £ = r ( l + cos#) = r + z. Schiff [243].(p . The deeper reason for this are symmetry properties of the problem. 11.. where r = ^/x1 + y2 + z2. (and subtract moc2 to We then obtain with a = e2/hc ~ 1/137.T). z=^rj).11.4 Hydrogenlike a t o m s in parabolic c o o r d i n a t e s The Schrodinger equation of a hydrogenlike atom can also be separated in parabolic coordinates £. as also indicated in Fig. where ip = 0.9. In the case of degeneracy with respect to the magnetic quantum number m (as in the present case) one can find linear combinations of hyperspherical functions Yj^(9.V . y=y/%qsanp. 86.
. .22) becomes — as we show in Example 11. We rewrite the first of these equations with the help of the relation M (w ^  m + e) .3 — h2 ( 4 r d (rd^\ d_f df\~ + 1 d2xl>\ £?7 dip2 2Z1Z2e< Z+V •ip = Eip. so t h a t the variables can be separated. The Coulomb Interaction Fig.3 Parabolas £ = const.137) One now sets the total wave function ip = £(£)!N"(r7)$(</?) and divides the equation by ip.228 CHAPTER 11.136b) the Schrodinger equation of relative motion (11. for uniqueness of the wave function. with zaxis as axis of rotation. there is really only one basic equation to solve here. . Multiplying the remaining equation by (£ + r /)/4 and setting the £part equal to —A. 2 . one obtains the following two equations: £ ) + TJIQE A m2n  « ' m = o. = 0. Then 1 d2$ izimip $ oc e $cV = —m where the separation constant m2 must be such t h a t m = 0 . 1 . (11. and r\ = const. In terms of the coordinates (11. 11. d_ dirt) + TUQE fm0Z±Z2e2 2h2 V ^2 A V m (ri'N) Arj1 2n Since these equations are alike. a separation constant.
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 / . . 2 .114b). Eq. .o. In the spherical polar case we had (cf. m0Z1Z2e2 ^ 2 = n (say). ds2 = dx2 + dy2 + dz2. . (11. p = /?£ and cr = /^r?.e. . E x a m p l e 11.' / 2 p m / 2 L 2 Z + i = m ( / 9 ) ) Analogously we have N oc eCT/2<7m/2L™ (a).105)) Hence we write I = (m — l ) / 2 .11. T h e n the first of the two equations becomes 1 £) n' m2 — 1 4 p + 4p" ( P 1 / 2 £ ) . To obtain a closer analogy with the separation in the spherical polar case. as well as the Laplacian A = V and the Schrodinger equation.6. . " . n 2 = n" .125b) (dividing by p 1 / 2 ) £ oc e . We leave the calculation of the degree of degeneracy to Example 11. from Cartesian to parabolic coordinates. ph h2p2 _ h2 m2Z2Z2eA hn 4 2 _ ~~ m0Z2Z2e4 2ft2 n2 ~ 2m 0 ~ 2m 0 This agrees with our earlier formulas like (11.5: The Schrodinger equation in parabolic coordinates Perform the transformation of the metric. . . . 2 . . i.106) and (11. T h e n by comparison with the spherical polar separation the answer is as in Eqs. .6 The Discrete Spectrum Then of the Coulomb Potential 229 A 22 (£ 1 / 2 £) + moE 2h2 A m2 — 1 +£  4£ 2 (cT1/2£) = 0. we set (since E is negative for the discrete spectrum we consider) ITIQE =*"• H J?P ^( <9p: 1 / 2 i m0ZiZ2e2 A = n". . 1 .± ( m + 1) = 0 . (11. 1 .
e. Beware! For m = 0 there is only one mstate. = e±im^e("+^'2{par'2L^{p)L^ (a). z = (£ — rf)/2.e.2)d/2. The Coulomb Interaction Solution: We begin with z of Eq. Observing that the highest possible value of m is n — 1 (i. The general formula for the Laplacian is§ ^ follows that 1 [9£\ A = ' 9 \ gr.2) and summing over m. " . are elements of length. We use the method of selector variables. .e. 0 there are two. and attach a factor to to each term introduced in the wave function.[df + dr]2 . 2 2V? V^ One now performs the square of this and obtains similarly dy2. 2 _ = (9id02 + (gvdvy + /„ .2d? dr)}. Similarly we proceed with dx and dy.g<p 9 \ .?e 9£ J 4 t d \ g^gv 9 \ dri \ gv dr] J t d \ g^gr.114c). + l 4 {^^)drl f i + v \ . \— < / — cos ipdr] — \/£r] sin ipdtp.2 = n — m — 1 and therefore the number of the coefficient of u}n'm1 . {g^Y. 9 \ 9<p\ gv 9f\ _ 1 a2 a / a\ a / ay £77 dip2 E x a m p l e 11. We consider first the case m = 0. as t h e two signs of t h e exponential indicate. </. Here we have (see preceding text) 711+77. see Eq.tx2 .2 .. . and so dz = (d£ — drj)/2.2 = n — 1.l)a + (n . W.1) ~ . and one obtains 2 = _ 1 4{ £ ) * (t + 7 )\ j.2)d} = (n. i. 'Recall that the sum of an arithmetic progression is given by a + (a + d) + (a + 2d) + • • • + {a + (n . for m / . etc.136a).l ) ( n . 712 and m which give n. Solution: For the wave functions ip = £(£)J\f(r])&(ip). Eq. Magnus and F. last chapter. / — cos ipdl. 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 = 1.6: 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.* 71 — 1 n—1 n— 1 n n+ 2 E ( n ~ m ) = n +2 E n _ 2 E m= n + 2n(n . 2 + ^dV . we have to find the number of values of the quantum numbers n\. dx = d{ y £77 cos ip) = . (11.E E n l „ .. when m = 0 = 77.230 CHAPTER 11. (11. n 2 (1UJ)2 The coefficient of a . i. Oberhettinger [181]. In the case m ^ O w e have n\ + 77. n the double sum ]T]2 is TI — m. all cross terms cancel in the sum of 7 the squares. Since the lines of constant £ are orthogonal to lines of constant 7 (they cut perpendicularly). . (n _ !) n = n2 §See e.g. Thus dz2 = .1 in this expansion is seen to be n.136a). (11. / J„^2 . Thus. Thus we want to obtain the coefficient of wn~1 in £2 . > .e. for instance. We then sum all possibilities and select the coefficient of ojn~1 in the result. i. N2 where g^dS.
7: Stark effect in hydrogenlike atoms 231 Consider a perturbation v = eFz. (11. Ikr+k. (11. where F is the electric field. Sec.e. with RayleighSchrodinger perturbation theory).. with Eq.•* n (l) = / Jo It follows that .124c).e. which will then give the change in energy of that state to the first order (i.6 The Discrete Spectrum of the Coulomb Potential Example 11. With the help of the wave functions we derived above. with z = r cos 9.™(p)] 2 By contour integration (see Examples 11. . Z2 —> e.—r~n = 3(mn 2 ). Thus e. we can relate the expressions Iq to those of Eq. with the connection » (11. with Z\ —> Ze. s 0/ . L ni\ ny.{ n i {2ni + TO + 1} + {ni j=± n 2 } 6(nin2)(ni+n2+TO+1) ^±n2} ^7—.g. I.1 (cf. we would have to use degenerate perturbation theory and obtain the same result. if we expressed the perturbation in spherical polar coordinates. However. these expressions agree with those of Eq. { 6 ni2 + 6 n i ( m + l) + (m + l ) ( m + 2 ) } . Apart from the power of the leading factorial (which results from the definition of the associated Laguerre polynomial in the generating function). . and determine the average value of v for a given state.a). .6).11. . We obtain the volume element dV in parabolic coordinates from the above results as dV = g£gvgvd£dr)dip = — (£ + rj)d£. i.e.) = ^eF(p .8 and 11. fc — m.128) if one makes there the substitutions n —> n\ (or 712) + 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. ( ! ^ ± ^ ) l { 6 n 2 + 6 n i ( m + 1 ) + ( m + 1 ) ( ? n + 2)} ny. = j (2ni+m+l). eF{z). 8.(p)L™ (<7)]2(p + a)dpda rev where JJ = / Jo dpe~ "p"[I.drid<fi oc (p + o)dpda.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. Solution: Using formulas and results of above we have to determine (v) = eF{z) = leF(e. . In fact. i. (11. 2(ni +712 + TO + 1) Therefore the change in energy of the state to first order is given by.eF / nh2 \ {v)=3(n1n2)^—lr2j Thus in this case nondegenerate perturbation theory could be used as discussed in Example 8.127) and setting n = r + k. {p — a) — = dxe~*xkvLkn (z)] 2 .r.128).
In Example 11.p ) r + 2 [(lp)(lg)]"l+1(lpg).1 ( n + m)!"[ _ (n + m + 1)! (niy J n! (ra + m)! _ (n + m)! (n .8: Laguerre linear expectation value integral Evaluate the following integral (which is different from the normalization integral of Example 11.x ( n + m)! ( n . It follows therefore that the coefficient of (qn — qn~1) in (1 — pq)~~(m~'~2' is p n ( n + m + l)! n!(m + l ) ! Finally / is the coefficient of p n p n . .l ) ! ( m + l)! in the expansion of p " ( n + m + 1)! p " " 1 ^ + m)! 1 _t .b.9 we consider integrals with an arbitrary power of t.3 by one power of t. The Coulomb Interaction E x a m p l e 11. L™(t) = coefficient of qn in G(t) e tq/(lq) (1 .a.7 and (remembering the definition of the associated Laguerre polynomila there!) the second of relations (11. (11. _ „ ..9: Laguerre expectation value integrals Evaluate the following integral with the help of the generating function of Laguerre polynomials: /•oo rSc(i):= Jo dte[Ll{t)Lac{t)tae\ i.p ) ( l . the integral / is the coefficient of pnqn in the expression _ / .124d) of the alternative integral representation obtained in Example 11.3: /"OO I := Jo et[L^{t)]2tm+1dt. (n + r .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 .9 )« Solution: Using the generating function for both associated Laguerre polynomials.l)!r! X We now evaluate the integral / as the coefficient of pnqn _ ( m + l)! [ ( l.9 ) ] m +h 1 yJo 0 ' 1 —pP 1 q 119 lpq _ (l9)(lp) = a With (see Eq.pK — — —L .0 ° e *[l+P/(lP)+9/(l<j)] t m+l di) t  P  9 l [ ( l . dt = ds/a. s = at.1 ) ! (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 .1)! n\ This verifies the expression I™t in Example 11.128). ) (l_pg)(m+2) Now the coefficient of qn in (1 — g)(l — p<j) . tt„:_ (1 .g )(l.232 CHAPTER 11. the integral arising here becomes with Eq. (11. and is a special case of Example 11. c positive integers.9) with the help of the generating function G(t) of associated Laguerre polynomials obtained in Example 11.p n _ 1 ) in (n — 1)! J p"(n + m + l)! n! p n . E x a m p l e 11.^~T^——^ ) = coefficient of (pn .( m + 2 ) is equal to the coefficient of (qn — qn_1) in (1 — pq)~(m+2\ The coefficient of qn in the expansion of (1 — pg)~ M is pn(n + fi — l)!/rt!(/i — 1)!.3).p ) ( l .
(*) from the coefficient of a1: />00 233 1(a) = / Jo dte.e.l ) * x coefficient of a* in the expansion of 7 ( a ) . i.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 «.p) .q)(l .a t [Lg(i). we have to have y + (c + b — 2r) = 1.™( r > (1 + a)P r £^0y^w Jc+6_2r = .11. so that = (q+fc±£) .e. c + b — 2r = l (i. In case (a) we have u \ 7 V^ ir fb + c\ 7(a) = Jo / Tli u.6 The Discrete Spectrum of the Coulomb Potential Solution: One considers the following integral and obtains /£.^ ( l . — (cr)!r!(6r)! Kv.124d) of an associated Laguerre polynomial. where (a+**£)! c j n Jo = .e. Using the contour integral representation (11.L. 7£. ° ^ + a ( l . i(a) = / Jo dttaet{p/(lp)+q/(lq) + l+a} = Q . = c (q + fe)! W ° (^)!(^)!(^)! ~" (2) In the case of one power of a.w(r) = and / a ) = r\(P ~ r . 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 odd). .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 .111) of the gamma or factorial function.f ].p) _lpq It follows that (using Cauchy's residue theorem in integrating around q = 0) 1{a) = (2^ / / ^ ^ [lpq c [ 1 .Pq + Q(1 " 9)(1 " P)1_a_1 = ti f p^ X COeffident f ap)]"01.p ) ] . we have 2r = 6 + c = even.a _ 1 = [1 + a ( l .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. . (lp)(lq) + a(l q)(l .?(t)t a e. (a) j / = 1. i. c + 6 — 2r = 0 (i.r ) ! a ' .e.(i) = i ! ( .1 + y w)\(a)y (r — w)\w\(j3 — r — l)!(j/ — w)\ Special cases: (1) In the case a = 0.
b\ V cited in Example 11. the standard principles of nonrelativistic scattering theory. . The Coulomb Interaction b + c\ 2 J (b¥)\(Pb¥w)\{a) ($±£ w)\w\(J3 ^±£1)1(111. ^A very readable source to consult. R.^)(. 2mo Here we set (this defines the velocity VQ) j? E ^ 1 (11. cannot really — and this means in a rigorous sense — be applied in this case.)!' f^)(. t We have the Schrodinger equation of relative motion of particles with charges Z\e. + (!±<)«. Taylor [267]. . however parallel to the Coulomb case.234 and (only w = 0 and 1 arise in the sum) CHAPTER 11. Zie given by h2 _ A _ ZiZ 2 e 2 V>(r) = Ei/>(r).7.~* Hence (recall that 7£ c (i) = i!(—l) l x coefficient of a% in / ( a ) ) (a+^)!(/3a) r r in (£f^)!(^)!(^±£)! t J = ( a + b)\((3a) > 6! _ (a + ft)! = [a + lb + l ) a . Since such a rigorous treatment is beyond the scope of this text. like the asymptotic condition and definitions of the scattering amplitude and the phase shift. These results are seen to agree with the expressions 1^. is the work of J. One should actually consider a screened Coulomb potential* and consider quantities like the partial wave expansion (to be defined in a later section) as distributions.138) " 2 ^ =2 ° ' 2 m Uo 7 = ZtZ2e2 ^T (1L139) * Screened Coulomb potentials are considered in Chapter 16. In view of the slow falloff of the potential at infinity.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.1%^ 11. also with regard to earlier literature.
i.b. . (11.M) = l + .. 86 .141) Inserting this expression into Eq. a(a +1) z2 ] (11.136b).^ W ) = 0.146) See.z). writing Vv = Aeik^v)/2f(ri) we obtain the equation d2 x d f(v) = 0. (11. for r — oo) • ipr = Aeikzf(r . for instance. Our question is now: How does ipr behave for r — oo. that we encountered previously.142) This equation is of the form of a hypergeometric equation. (11.7 The Continuous Spectrum of the Coulomb Potential Then Eq.109) we can therefore write the solution ipr = Aeikz$(i>y.z)).11. with the asymptotic expansions (11. . „ az 1.z).z) = Wi(a.z) + W2(a. (11.87 under Kummer functions.+ ^  + . (11. According to Eqs.145) U ' ' j " T(6a)1 J ^ 71=0 r(a)r(a6+l) n! ' (7T < axg(z) < vr).138) becomes 235 ^A + k2 . (11.108) and (11.. b. (11. Oberhettinger [181].b.140) and using (11.140) The simple Coulomb potential permits particularly simple solutions.143) and v = ikrj.. (11. Magnus and F. what is the > behaviour of 3>(a. where ^..e. . pp. A = const. i. z) for \z\ —> oo? We can find the appropriate formula in books on Special Functions:* $(a. W. In particular the equation possesses a regular solution of the form (also ip ~ exp(ikr). v = ik(r .e.144) *(a.b. z = rcos<9.
kz+l^ktrz)) C .l.148) .ik(rz)) where ^ ~ Aeikz——^—f't^ e ife(rz) + W2(n. > A = Aeikz[W1(i1.153) The asymptotic solution represents t h e stationary scattering state (E = const.7 ln*(rz)) k(r . if tbr were „ikr ibr ex el"z + f{6) r (11.147) (—7r < arg(z) < 7r) It follows t h a t for r — oo.a) (z) na)"~ r(la) T(ba) n\ ' (11.152) Since z = r c o s # . = r cos 0.152) 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) .151) exp[—ryln fc(r — z)] — exp[—ijIn = kr(I — cos 6)) exp[—ryln 2kr — ry In sin 2 (0/2)].z)T(l . (11.l.149) (11.e.154) . Without t h e logarithmic phase factors. (11.i 7 ) (11.236 n\Zzab CHAPTER ^T(n '£ fo 11.150) ^(rz)]1"^ 1 + 0 T(«7) i.e. We now define a quantity f(9) by t h e asymptotic relation lb O < ^(. J\") ci(kr~/\n2kr) (11.151) with Eq. a comparison of Eq.^l*7 l + O (11. The Coulomb Interaction W2(a. i. with 2.b. A(j)11 A where \ " r(i + \i) 1 j(kz+"flnk('rz)) 7 f ( l + Z7) _ e i(fcr.z) + la)T(n + b.(11. (11.ik(rz))} =^ + ^d.) of a particle with incident momentum hk in t h e direction of z for large distances away from t h e scattering centre.
2im. (11. This cross section is defined by the ratio of the outgoing particle current (in direction 9) to the current of the ingoing particles.156) . z — — oo. The quantity /(#) 2 would then be a measure for the scattering in the direction 9.11.n . (11. For the incoming wave ipi = exp(ikz). As a consequence of the masslessness of the photon. do dQ K 12 K Fig.. the effective range of the Coulomb potential is infinite. The problem of the logarithmic phase.155) Since the gradient is proportional to the configuration space representation of the momentum. the Coulomb potential has an influence on the incoming wave even as far as the asymptotic domain. also called phase anomaly.7 The Continuous Spectrum of the Coulomb Potential 237 the wave could be looked at as the sum of an incoming plane wave exp(ikz) and an outgoing scattered wave.1 The Rutherford formula The above considerations permit us to calculate the differential cross section which is a measurable quantity.4 Variation of the observed differential cross section with 0.7. In other words. We define as current density the quantity h J := 2imo [^(v^)v(v^n. 11. 11. the current density is * h . that we encounter here is a characteristic of the Coulomb potential. we see that this expression is the exact analogue of a current density in classical considerations.Q hk mo v0.
In Eq.152) yields as density of the outgoing current Jr = ^\f(0)\2vo(11.157) The differential scattering cross section da into the solid angle element dVL is defined by the ratio da ir which in the present case is with \f{6)\ obtained from Eq.e. This implies effectively that the Coulomb potential becomes (virtually) a screened potential or Yukawatype potential of the form V .238 CHAPTER 11. The Coulomb Interaction In the case of the Coulomb potential we now ignore the logarithmic part of the phase (see discussion later).159) as the Rutherford formula which can also be derived purely classically. since in actual fact the Coulomb potential occurs only in a screened form. for 0 —> 0. i. thereby screening itself off from the surroundings. (11. the square 7 2 ).154) we chose the prefactor of the incoming wave in (11. is called scattering amplitude. That this formula retains its validity here in quantum mechanics is something of a coincidence. In the context of quantum electrodynamics one refers to "vacuum polarization" and relates this to the idea that a charged particle like the electron polarizes the otherwise neutral vacuum by attracting virtual charges of opposite polarity and repelling virtual charges of the same polarity. We see that a depends only on the modulus of the potential (i. 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. which yields the scattering cross section.153): 4fc2 s i r  The expression /(#).152) as 1. Correspondingly the other part of (11. We recognize the result (11. We make a few more important observations. in the current this would in any case contribute only something of order 1/r. This implies that scattering takes place for the attractive as well as the repulsive potential. 'screened ZlZ<2e2 cr/r0 ° . In reality this divergence does not occur. (11.e.
(11.4. The wave packet is a superposition of waves ip(x. / dk 7^340(k)4(xxo.166) Let the incoming free wave packet at time t > t$.165) ^ ( x ) = £ fe Vk(x). (11163) (11. A wave packet with momentum maximum at k = ko (with momentum p = hk) is therefore given by the following expression in which xo can be considered to be an impact parameter. m0 m0 and V.2 fr2 . t 0 ) .t).*'^ e1 piEk{tto)/h (x x i)= (2^3^ ( k ) roT ^ k o l * . 11. t 0 )^k(x. i. (k). t0) = e. which in the case of free motion (zero potential) are simply plane waves. 11. _ x o _ v o ( t _ toh io)_ em0^VoKtt0)^o)ix (1L167) 0 l/> () ) fxXn t)~ gik(xxo) .11.k(x) is a solution of the continuum with energy h2k2 E = Ek = — >0. 2mo V + F(x) k (11. be given by ^ f dk AL. the wave packet in the absence of a scattering potential.8 Scattering of a Wave Packet We saw earlier that in quantum mechanics a particle is described by a wave packet. of continuum wave functions.e. t) = U(t.161) where Ak0(k) ~ A(k — ko) and ^ k (x. whose maximum moves with the particle velocity.yJ ^ j A k o W 727)3/2 e 0.t) = e .i i f (''°>/Vk(x. The particle velocity v is v = — . and if H is timeindependent one has the stationary wave function ^(x. and which provides the uncertainties arising in quantum mechanics.e. i.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.x° ' r j . 2 (11162) where Vk(x) = ^ ( x .t). (11.e. t 0 ). of the equation r fi. v0 = .^ ^ % ( x ) .164) Ek~Eko + {kk0)v0h. i.
167) we have ^k0(xx0. The Coulomb Interaction In Example 11. this solution > has the following asymptotic behaviour in which f(6.172) h h k0 {rv0(tt0)} (11.ko) k0 (11. (11.x + _ / k ( e j ¥ ) ) ikxo (2TT)3/2 zkxo = <>(* r y (2vr)3(27r)3/2e M ^ ) e *o)/^] e ikx 0 (11. i.ip) is the scattering amplitude with respect to the incoming plane wave. can be written A (X) = ( 2 ^ + W J ^ ^ ^ ( ^ ( x ' )  eikx 2mn f P»fexx' (11168) We saw earlier (cf.xo. From Eqs.t o ) / V k ( x . (11. i. k = kez. ip) could be carried along but we ignore it.171) The phase of fk(0.154)) that for x — oo.173) . (11.k0) ~ k0 + and the expansion of E^ in Eq.<p) + Ol Akr l (11.240 CHAPTER 11.166). With Eqs.165) ko • (k . (11.k 0 ) 2 ~ Jk20 + 2k0 • (k .10 we show that the solution V'k(x) of Eq.i) dk (2TT (11.169) The wave packet in the presence of the potential is obtained from the free form (11. (2TT)3/2 eik* + —fv(P.t) = J ^ A k o ( k ) e . Eq.170) A ko (k) e .x 0 ).x °Vk(x).163) we obtain </>ko(x .161) and (11.167) by replacing the plane wave exp(zkx) by this scattered wave.^ ( * .169) and (11. (11. the stationary scattering wave.iB *(**°)/ R °'tj + 1 pikr e * . ^k(xxo) ^k(x) ~ = e.e.ik . Then with k = ^(ko + k .e.
r ' ) = 47r<5(r .10: Schrodinger equation as integral equation Demonstrate the conversion of the Schrodinger equation into an integral equation. .r')U(r')ip{r')dr' /4T7 (as we saw earlier) . r 241 (0 LO) f ^ Afc o(k)rikxnri[fcor£fcoato)/ftl i(kko)x' = /k0^.(r) 2rrao V(r)V(r) 7/>(r) (2TT)3/2 ^dr'G(r.154). Here ^ (x' — xo. U(T) = ^ V ( r ) . Solution: We define the Green's function G ( r .(«)/ ipko(x .y)eiko'Xoe'[fcor£.^e (2^3(27r)3/2e y (27r)3(27r)3/2e (11.5.11.e. E x a m p l e 11.r')l/(r>fc(r').r').r ' ) . '.5 Scattering of a wave packet with scattering centre O. 11.171). and determine the Green's function G(r. t) .z Fig. r ' ) by the equation (A + fc2)G(r. 11. i. T h e scattering amplitude is t h e same as t h a t given by Eq.x 0 .174) r /k0(^. Adding a suitably normalized solution of the free equation.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.167).to). *.r')V(rW). however. (11.*0ko (x . the solution can be written iMr) : (2TT)3/2 ^dr'G(r.8 Scattering of a Wave Packet so that with Eqs. (A + fc2)t/. A particular solution of the Schrodinger equation is — J G(r.^y /k0^.x 0 . t) f. (11. (11. with the energy of the maximum momentum of t h e wave packet.
The Coulomb Interaction We find a set of Green's functions G(r.6. 11. Thus we consider the contour integral taken around the contour C+. .fc2 ' 2 ^ i fc72^"fc2 Integrating out the angles we obtain 1 G roo />7r 2n fc' rZn j. displaced slightly away from the real axis as indicated in Fig. one obtains another Green's function given by G_(r)= r eikr. These are the socalled outgoing and ingoing Green's functions. 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. Imk .e \ r dr\ k ikr 1 ikr 0 With the choice of an analogous contour C— in the lower half plane. we relate it to a properly defined contour integral with the two simple poles at fc' = ±Vfc 2 + ie = ±(fc + ie/2k). e > 0 small. 11.=k+ie/2k 1 d f. Hence we write G(r)G+(r).fc ) oo dfc' 1 d nik r 7rr dr / . 9{k') = 1 1 S(T) = j G(r) 1 ^ A j dk'e i k ' r . r ' ) = G ( r — r ' ) with the Fourier transforms G(r) = J dk' 3 (k')e ik ' r .Rek' kfc/2K Fig.O O ™ fc2 (e*fc'r _ e ifc / r 1 fc ) =  / fc'dfc' i(fc' 2 fc 2 ) Since this integral does not exist. which is such that the contribution along the infinite semicircle vanishes. « = 52 / J _ Z100 7rr 7o / A fc'dfc' 2 / 2 • fc2 (fc' *(fc' .242 CHAPTER 11./2 dk'd(—cos 6)dip iki.6 The contour C+ in the upper half of the k'p\ane. dk' ik'*r 2TT2 fc'2 .
2l + 2. i.140) (previously we solved this only for the bound state problem) § . whose continuation to infinity is again given by (cf.176) " 0—A ~ (I + 1 + iy)<t>i = 0. .9 Scattering Phase and Partial Waves 243 11. 21 + 2. 2ikr)] [Wi{l + 1+11.178) (11.e. z) + W2(a.g.z2 27fc 1(1 + 1)' yi = 0. b.21 + 2. Before we introduce the general form of the socalled partial wave expansion.179) = = elkr{kr)l+1F(l e (kr) ikr l+1 + l + ij.0 + W2(l + 1 + iy.145)) F(a.e. (11. If we proceed from the radial Schrodinger equation. (11. we define first the scattering phase for the case of the Coulomb problem proceeding from our previous considerations. b. b.£).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.100) — and we saw that in the particular case of the Coulomb potential this is not essential — we have to solve the following equation for the scattering case obtained from Eq. (11. Then yW (11. Then £^<fc + (2l + 2 £ = ~2ikr.175) yl' + k Here we set (as in the case of the hydrogen atom) yi = eikr{kr)l+lUi). Eq. Thus we proceed as above. (11. i. z).11. (11. the logarithmic phase as a consequence of the slow decrease at infinity and a special symmetry which permits separation in other coordinate systems. Eq. 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. (r) (11177) Similar to above we obtain a regular solution y) ' with 0j = F(Z + l + i7. z) = W^a. e.2Z + 2. where Mo is a parameter in Chapter 16. under the stated conditions 2/07 = Z\Zie2 = —Mo.
183) r(Z + 1 + 17) := Reie.147) this becomes asymptotically (r) Vi lkr eikr(u„\l+l (kr T(2Z + 2) '—(2ikr) \T{1 + 1 .181) so that eldlT{l + 1 . T(2l + 2) (2fcr)*T« (Alliy ikr .180) ^ ' (11.«7) „ . (11.244 CHAPTER 11. . Then (r) r—»oo Vi 7T7/2 T(2Z + 2)e*7'*e i8i 2'+1I\Z + l + i 7 ) i5.l e*"" + T(l + 1 + H ) T(l + 1 .i 7 ) I T 2l l—X—i^j ( + 2) 2ikr( ?.' .j syyll iy il( _ . +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.n) • e „—ifer+i7ln2fcr _eMr(i7Jl)/2 r(z + I +17) T(2Z + 2) /2 • ikriy In 2fcri7r(J+l)/2 r(z + 1 . Then e2i5< = Reie Re~ J2i9 9 = Si. The Coulomb Interaction Using Eqs.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 .ii)' (11.i 7 ) „ikr—ij\n2kr T(2l + 2) e e* jTv(lliy)/2 2l+i —ikr r(z +1 .ry) = e"^r(Z + 1 + ry). (2kryr{i" V 7 .146) and (11.
The expression e 2i5j _ 2i5i—iirl is called Smatrix element or scattering matrix element. indicates that — looked at analytically — the scattering amplitude f(6) possesses at E — 0 a branch point of the square root type.e.e.n ' .e.139) n' = 0. One defines as regular Coulomb wave Fi(j._W2)_ ( 1 L l g 6 ) The factor e2lSl between the Jost solutions (in the regular solution continued to infinity).11. Squaring this expression we obtain the wellknown formula for the binding energy of bound states (as a comparison with Eq. i.e. Analogously one defines as nonregular spherical Coulomb function or Jost solution the outgoing or incoming wave u\ . as we saw. In the present quantum mechanical considerations I is. considering I as a function of E extended into the complex plane.9 Scattering Phase and Partial Waves Hence 5j = argr(Z + l + i7). a positive integer.2.. (11. g = e«* = rSi + 1+ *V*ri (11. (11.185) J F} ^° sm(AT7ln2£T + ^ ) (also called regular spherical Coulomb function).1.187) contains almost the entire physical information with regard to the Coulomb potential. 245 (11. However. i. For instance the poles of S are given by Z + l + z7 = .188) in the form y/E.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. That the energy E appears in Eq. i... where the continuum of the spectrum starts.184) The phase 6i = 8i — lir/2 is called Coulomb scattering phase. (11.114a) reveals). with Eq. . considering I —• an(E). i. .
(11. The large r asymptotic behaviour of the solution which is regular at the origin in the case of the scattering problem can be written ^ r e g = V'in + V'scatt. 11. lman(E) jump from E<0 toE>0 o A LU infinity  E<0 Rea n (E) Fig.246 CHAPTER 11. and thus in general. The Coulomb Interaction and plotting these for different values of n.7.7 A Regge trajectory of the Coulomb potential. (11." These Regge trajectories can be shown to determine the asymptotic high energy behaviour of scattering amplitudes and hence cross sections. Singh [252]. For the Coulomb potential a typical such trajectory is indicated in Fig. we obtain the partial wave expansion of the scattering amplitude as follows. we obtain trajectories which are known as Regge trajectories. 11. We mentioned above that the logarithmic phase does not arise in the case of shortrange potentials. In these cases.191) .190) and the scattering solution „ikr Vwt = / ( 0 ) — + 0 V.189) where as before the incoming wave in the direction of z is ikz ipin = e (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.
r e. (11. 21 . 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 ' .r ^ V ^ Wostf).g. When we solve the Schrodinger equation and calculate the asymptotic behaviour of the regular solution we obtain correspondingly V'reg .k)=me2iS^k\ in which r\i^\ indicates absorption. .196) "See e.eikr\P^cos9). The mathematical expression can be found in Tables of Special Functions (there cf.195) with the scattering matrix element S(l.i f c r . The potential disturbs this phase relationship (by delay or absorption or redirection). (11.22.9 Scattering Phase and Partial Waves 247 For a spherically symmetric potential the amplitude / depends only on 9. 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„ ' .^(cosfl) (1L194) This therefore yields the following partial wave expansion of the scattering amplitude = ^lY. Magnus and F. The incoming plane wave can be reexpressed as a superposition of incoming and outgoing spherical waves with appropriate components of / and with a definite relative phase. pp.192) Thus the incoming and outgoing spherical waves have a definite relative phase.i * r 1 v^ f ne2ibl — 1 \ = k^2l W +1 \^r^)Pl{cosd)—. W. 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.11.^2l + l^S^k) lik (11. (11.193) where r\x e2l5t determines the change in amplitude and phase of the outgoing spherical wave. Oberhettinger [181].~ J > Z + l)Lil*e~ikr .
This page is intentionally left blank .
in general. which in one form or another. was developed only in the last two to three decades. as — for instance — for trigonometric or double well potentials.* *See e. In particular we consider cases which illustrate the occurrence of tunneling and that of resonances.e. 249 . Razavy [236]. which a particle in classical mechanics would never be able to intrude. therefore. M. that the explicit derivation of such quantities in nontrivial contexts.e. states of a finite lifetime. i. i. For various general aspects and applications of tunneling we refer to other literature. in texts on quantum mechanics is rare. These newer methods will be presented in detail in later chapters. that quantum mechanically a particle may have a small probability of being in a spatial domain. It is not surprising. In fact. to explore first much simpler models in order to acquire an impression of the type of results to be expected. can be found in any traditional text on quantum mechanics. Thus in this chapter we consider traditional "square well" potentials in onedimensional contexts. namely the instanton (or more generally pseudoparticle) method. Similarly difficult is. the computation of finite lifetimes of a quantum mechanical state.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. one of the main and now wellknown methods to perform such calculations. It is very instructive.g.Chapter 12 Q u a n t u m Mechanical Tunneling 12. Such a vital quantum mechanical phenomenon is socalled tunneling. however.
2 Continuity Equation and Conditions The statistical interpretation of the Schrodinger wave function ^(x. so that p = T/>2 is the probability density for a spatial measurement at time t. M^ Then dt + Vj 1=0. (12.t) = HiP(x. Correspondingly one interprets as probability current density the expression j x ( < ' >=iH v ^7 v <4 = [W{x.1) as the probability described by ijj(x.2) we have = o + ~v[(vv>*)V'V>*(W)] = vj. i.4) (12. Quantum Mechanical Tunneling 12.e.t)\* (i2 2)  With the help of the Schrodinger equation ihiP(x.250 CHAPTER 12.. ihr(x.5) L j • dF IFo is the probability that per unit time a particle passes through the surface FQ.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. According to Eq. t) was developed by Born in 1926.t) (12. .t) that the one particle concerned is somewhere in space at time t. We assumed this already in the preceding since we interpreted the normalization integral / Jail space dx\ip(x.t).t)\2 =1 (12. r (12.
However. Such a region is for instance the domain where V is larger than the total energy E. Delta potentials in more than one dimension do not permit bound states and scattering.2 . Gosdzinsky and R. as we mentioned above. ~hr> k = n^ (12 7)  . where „ mpVo V(x) =+V0S(x). which is the case we consider here. If ^>(x) and V^>(x) are known at every point x. even for discontinuous potentials. If there is no absorption.33. Nonetheless. From this follows that both p and j have to be finite and continuous everywhere. Thus it is possible that the probability for the particle to be in such a domain is nonzero. we can do this also here. The 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. This case plays a special role. pp. t For the singular delta function potential a modification of these demands is necessary. 2m E 2 V0 > 0. higher dimensional delta potentials do not have properties like that in one dimension. (12.12.3 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. 1 0 ~x o a= f See L. as shown by P. It is therefore quantum mechanically possible for a "particle" to pass through a classically forbidden region. 29 . It is therefore sensible to demand that V( x ) a r i d V'0(x) be continuous. 'Maybe the reader dislikes being confronted again with the onedimensional case. 12. This effect is known as "tunnelincp. Tarrach [119].6) . The Schrodinger equation for this case is* ip" + [A. although in general small. 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.3 The ShortRange Delta Potential 251 The timeindependent Schrodinger equation is a second order differential equation in x. with regularization their study is very instructive for illustrating basic concepts of modern quantum field theory. In much the same way that we define there reflection and transmission coefficients. since this potential implies a discontinuity of the derivative at its singularity (as will be seen below). 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. Schiff [243]. the equation can be integrated at every point. the problem is analogous to that of the scattering process from a potential in a onedimensional case.2a6(x)]il> = 0.
Quantum Mechanical Tunneling Integrating the equation over a small interval of x around 0 we see. a — ik £Qr < { eikx a eikx •t" .e.(ik .11b) 2aC. that the derivative of i/> is discontinuous there: W\U + k2 J i>dx . i.12a) (12. (12.' : It follows that ik .14) One defines respectively as reflection and transmission coefficients R and T the squares of the respective amplitudes.8) (12.2 a J 5{x)4>(x)dx = 0. (12.ikC = i. (12. a 2 — a 2 a — ik a2 + k2 ' .12b) a=—^— C = a + ik and hence B = C1 Thus for i/> we obtain = ?—.ikB) = 2aC or 2a(l + B). .252 CHAPTER 12. 2 (12.13) ~A R= for x > °(12.e.= 2a^(0).[^']o.ik(C . The continuity conditions at x = 0 are therefore for V : and according to Eq.9) for ij. Consider the case of a purely outgoing wave in the region x > 0 (no reflection back from infinity). except in domains of the potential V. In the region x < 0 we then have both types of waves: the incoming or incident wave to the right as well as the wave reflected from the potential. _ fQr x < Q^ eikx _ geikx ^=\ ~ Blkx *x . [<//]0+ .e. (12. i. e Q (12.11a) ikC .1) . i.: (i2iQ) Ce for x > 0.e 1 + B = C.9) States with k > 0 correspond to those of free particles that move with constant velocity v = hk/mo.
2 o ^ ( 0 ) . the quantity T.19) Thus the delta function potential supports exactly one bound state. R. Newton [219]. K = a.i 8 ) .e «x = J \ CeKX for for x > 0.) and V>'(0+) .K C . We write the solution ^ so that ^(0+) = ^ ( 0 . . = .e.16) The corresponding wave function proportional to etkx for x < 0 associated with the pole for k = — ia is^ rl){x) = V^e~aW.(K)C = . ( 1 2 . 'Note that the physical pole has negative imaginary part. at k2 = —a2. possesses simple poles at a ± ik = 0. Thus ^ = Ce. Proissart [223] or other books on scattering theory like R. See e.a l s l. or h2 E = —a2. 2m 0 (12. when normalized to 1. f Ce~KX c. i. z<0.3 The ShortRange Delta Potential and T= ^rr a — ik 2 253 =1R.g.2 a C .K 2 < 0.ip'(0) = . . as coefficient of the outgoing wave. / oo In the bound state problem the Schrodinger equation is ip" + [k2 + 2a6(x)]ip = 0. and C follows from the normalization: OO A.15) Furthermore we see that similar to the Smatrix which we encountered in the case of the Coulomb potential. Omnes and M. oo (12.2 = . i.17) da#(s) 2 = 1.12.e. poo nc s~t2 / oo dxe dxC2e~2aW = 2a\x\ _ u 2C2 / JO a Hence C = y/a and therefore ^ = VaeQK (12. (12.
x G [a. Quantum Mechanical Tunneling 12.254 CHAPTER 12. \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. V(x) V=E a v=y Fig. 55. Schwabl [246]. p. We want to consider the case of stationary states for E > 0. p. 58. The latter contains also specific applications in nuclear physics. (12.20) for for \x\ > a. .1 The square well potential. (12. The Schrodinger equation is iP"(x) + 2 ^[EV(x)}iP(x)=0.1. x€[a.a]. 12. or in the lectures of E. Fermi [92]. oo]. Similar treatments of square well potentials or barriers can be found in most books on quantum mechanics.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. 12.22) + De Feikx + Ge~ikx *We follow here largely F. —a]. We set 2m0(E + VQ) h2 so that ip" +fc2V>= 0 ip" + K il) = 0 2 (12.
fc Af(a) = .f )e" t\ K\0—tKa—ika . (12.26) K^gifoa—i/ca fc. ) A l ( pika p—ika pina K ina kC p—ina \ _p—ika J \ Kp—ina J ( c\ U/ (12.iKa ? ikAe~ika and . (12. M(a) . (12. .ikGe~ika. (12.12.ikBeika = inCeiKa + DiKe~iKa. ( V C n vD ).e.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.. I (12. Fikeika .24) yields F ^ G. ).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 \ . 2 l ^i i K\ iKaika O \ (1 + %)e ina—ika (1 .25) ^\=M{a)lCD where 1 / f l — « \ irea+ifca /i . (12.27) v ' Similarly Eq.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.23) Ce~iKa + DeiKa iKCe~iKa + DiKeiKa = = Fe ifca + Ge"*fca.
35) with with with x G [oo.e sin 2na IJlika e B = r)F sm2na. \R\2 = T(E)^sm2Ka (12.oo].4T(£)e ifci.256 CHAPTER 12.a]. the quantity T{E) := £ = A lika cos 2/ta — je sin 2/«x (12. we obtain A = F( cos 2na — .e.(1 + e 2 ) sin2 2na 1 . . (12.e.i^ 2 sin2 2KO 1 + ry?7* sin2 2/ta Inserting Eq. x € [a. i. (12. 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. (12.33) One defines as transmission amplitude the ratio of outgoing to incoming amplitudes.—2ika [cos 2/ta + \e sin 2«a]e~ (12.37) .a . xe[a.32) [cos2«.a ] .31) we have A AT(a) = )MFo B ^r)sin2na .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) (12.36) One defines as reflection coefficient \R\2 the modulus squared of the reflected amplitude divided by the amplitude of the incoming wave. (12. = < Ce~lKX + DeiKX . Quantum Mechanical Tunneling Hence with the complex quantities IK (12.34) into Eq.
4 Scattering from a Potential Well On the other hand according to Eq.e.12.39) 2 i. an equation of the type f~1f = g19. . (12. Dwight [81].e. (12. (12.38) as expected.e. 83. 2\k K.e. in the domain of A imaginary and K real. for VQ < E < 0. i.42) These equations have no solution for K.31): 1\T(E)\2 257 1 . However. cot(fca) = i— or tan(fta) = K K . (12.44) .40) we obtain cot(na) — tan(/«j) — i K IK . which is satisfied if either / = g or / = — g'1.43) Thus we set k — —i ' H . (12. formula 406. and k real. Li ( — H— I sin 2na = 0.12. i. 2m0\E\ h? ' (12. p. nor when both parameters are purely imaginary. B. (12.35) and using Eq. (12.34) that the amplitude F = AT{E) of the outgoing wave has a simple pole where cos 2KCL e sin 2KCL = 0.41) i. they have solutions. (12. where cos 2/ca With the relation§ cot(2«a) = [cot(«a) — tan(/ta)]. We see in particular from Eq.\ sin2 2KQ •n 4 sin2KaT(E) — \K\ >  IRI2 (12.
(12.34). "Note that dtan(2K<z)/'d(2na) evaluated at ER is 1.42) which then leads to bound states with associated even and odd wave functions.fl = ( . (12.l .g. t2 2 (12. .46). Schwabl [246]. . ± l .20).e. n = 0 . 64 . Quantum Mechanical Tunneling . (12. t&n(2Ka)ER = 0. (12. ± 2 .45) These are the equations that one obtains from the connection relations for the case of Eq. At their positions cos(2/«z)£.46) This condition implies the eigenvalues E^ER = n2^^ V0>0. around the values given by Eq. (12.e. Also.e.35)) for E > 0 sin2Ka = 0.49) Taylor expansion of the denominator in powers of (E — ER) yields .65.50) See e. (12.c o . i. This can be understood. (12. = 2/rn (EER). Then according to Eq.258 and obtain the equations CHAPTER 12. (12. „ . from Eq. T(E)e2ika = — r . i. pp. The remaining wave part becomes exponentially > decreasing. The vanishing of A implies that because k is imaginary the factor exp(ikx) does not diverge exponentially for x — . 2m0\E\ and Kcot(Ka) = y—^—L.l ) n .48) We expand T{E) around these energy values. . Eq. The amplitude of the transmitted wave assumes its maximum value where (cf. We now return to the scattering states. i. The infinity of the amplitude of the outgoing wave corresponds to a zero of the amplitude of the incoming wave. A = 0. k\ d(2Ka) K) dE \ E R 1v/2^a 2 h 2ER + V0 ^ER~(ER + V0) r 1 (12. 2Ka = nn. (12. F.+ \tan(2Ka) where T is given by" 2 1 (K 2\k 1 (K k\ /n . .' We observe that T(E) has poles at those values of E which supply binding energies.47) These states in the continuum are called resonances. and is itself infinite there. 2m0\E\ Ktan(«a) = — y——^— .
The Schrodinger equation to be solved is ^"{x) + ^(EV(x))i.5 Degenerate Potentials and Tunneling We obtain therefore T(E)e2ika = = (^ 259 iT/2 EER + iT/2' (12. We restrict ourselves to the period — la < x < 2a.' ''' 5a 3a \ a ~~~.^ l / . a 3 a. 12. . A potential of this type is known as a PenneyKronig potential.12. This result is physically very plausible.e.52) Here ER is the energy of the resonance and 2/F the lifetime of the state. T(E) has poles at E= ERi: (12. 12. 2a<x<2a. s \ V0 /• ~\ 'X . 12. V(x) x=2a x= =2a .' ' 5a \V / Fig.5 Degenerate Potentials and Tunneling In the following we consider a potential which is very similar to a periodic potential and can serve to illustrate a number of aspects of the process of tunneling in quantum mechanics. = 0.2 The PenneyKronig potential. We consider a potential consisting of a chain of rectangular barriers as depicted in Fig. V(x) = 9(x + a)VQ6(ax). These are precisely the resonance states. .2. Since we have reflection as well as transmission there are states with repeated reflection and finally transmission.51) i.
2..53) *w«w^( « j)'*^{i i ^ . and hence for = Eip(x).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. . i. We set ..55) The Hamilton operator of the problem.Vo) n.e There is the additional boundary condition (for Vo = oo) ^ ( .260 CHAPTER 12. In view of the periodicity we have ^(x) = ^(x + 4a). HV^(x) = H.2 < a . (12. i. .e. with V as parity operator... = EVip(x). (12. Quantum Mechanical Tunneling (a) We consider first the case of VQ = oo.a ) = 0 = ^(a). E^En = ^ ^ t n = 0. and we choose (12.. Hip(x) so that with Hi/)(x) — Eip(x): H[^{x) ± t/>(x)] = E[^{x) ± xl>(x)\.1. VHV1 Hip(x) = Eil){x): VHV~lV^){x) = EVtl>{x). This boundary condition implies the quantization of the energy with ^ ( 2 a ) = n7r. x + 3a <> ™ since the particle cannot penetrate into the infinitely high wall. the differential operator is even in x.e. 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.2 = W h 2 _„2 = 2mo(E .
^kAcos{k2a). We expect. — a]. (12.a]. are ^_(0) = 0. (12. This wave function is no longer restricted to the interior of a box. We can write the solution as consisting of the following pieces: { Asink(x + 3a) BeKX + Ce~KX with with i £ [ .2). we shall not exploit here. (12. however.60) The boundary conditions that ip± have to satisfy. xe[a. At x ~ — a the connecting relations are Asin(k2a) kAcos{k2a) and at x = a: BeKa±Be~Ka K. 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. We therefore have different boundary conditions.2a..59) Asink(x + 3a) with x & [~2a.Be Ka (12.58) D sin k(3a — x) with xE[a. V+(0) = const.5 Degenerate Potentials and Tunneling 261 We restrict ourselves again to the period — 2a < x < 2a. of course. We > obtain even and odd functions ip± by setting ^ = ±1. This means that i>±(x) = <  = ±1. The central region of the solution is the classically forbidden domain.61) = Be~Ka ± BeKa = ±[BeKa ± Be'Ka].. V'(O) = const. (12.12. — a].2a].63) = KBe~ F = . BeKX±Be~KX with xe[a.62) = nBe~Ka T nBeKa = ^n[BeKa q= B e " M ] . so that (W meaning Wronskian) W[tl>+.2a]. which. = Ka ±Asin(k2a). ±Asink(3a — x) with x€[a.il>]?0.a]. that these are such that for VQ — oo they approach those of the previous case. (12. 12. VV(°) = °.
For large values of Vb.2 e .nir)2}} = 1 ± 2e~2Ka. Quantum Mechanical Tunneling Fig. we set tanhx = l . say.even.262 CHAPTER 12. and is — as we see — less than that belonging to the lower sign case. for VQ — oo. i.2 a . However.+ • • • 1! cos^(n7r) 2ak . » we expect the eigenvalues to approach asymptotically those of the case Vo = oo. (12.e.mr)2}. the limiting value for K = oo: tan(2a/c) = = tan(n7r) \ (2ak — nir) 1 — ^—.nir + 0[{2ak . say A. i.Thus we observe a splitting of the asymptotically degenerate eigenvalues: kesympt. (12. (with reinsertion of nir/2a for k) 2ak ~ nir .67) The value of k belonging to the upper sign.nir) + OU2ak .e. In general one now rewrites the equations in matrix form.e. where Akoce'2™.3 Level degeneracy removed by tunneling. in the present case it is easier to continue with the above equations. is the one belonging to the even solution. It is plausible therefore to use appropriate expansions.66) Inserting these expansions into Eq. i.e. (12. k = fc0dd.64) we obtain ~[(2ak k . the odd solution with. i. F ^odd (12. Taking the ratio of the equations in both cases. 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.65) and we expand tan(2afc) about nir. 12. + 0 ( e . (12.— (1 ± 2e~2A + O (—)..4 x ) and cothx = 1 + 2e~2x + 0 ( e " 4 x ) .68) . = AA.
2mn. . exp r rb rb / •J a 2mo (V(x) . Compute this probability. 12..V{x)\i. .70) xo = (a + b) 2' . exp . a sinusoidal bump in the road with a height of one meter at its peak as indicated in Fig. and is a reformulated version of a problem set in lectures of Fermi. Solution: The wave functions i/>o>Vv in regions V = 0. 1pV 2ro 0 4>o '.4 A car meeting a bump in the road.69) where the potential V(x) is (with g the acceleration due to gravity) V(x) y'(xo) Setting w(x) = 7r**E.5 Degenerate Potentials and Tunneling 263 Thus the effect of tunneling — here an exponentially small contribution — removes the degeneracy of the asymptotically degenerate states.3.v = 0.. w(x0) •• = = mogy(x). as indicated in Fig. .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. 12. 12. Example 12. b > a. i. The following example is an attempt to transcribe the quantum mechanical effect into a macroscopic situation. '. p. (12.e.2 dxy/V(x) (12.** " W ~^7~ Fig. Fermi [92]. .^ 0 = °' Hence in dominant order •00 — exp .12. V ^ 0 are given by the equations ^o + " J .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. y(x) = sin and y(xo) = 1. 0. Classically the car is unable to overcome the bump. but quantum mechanically there is a finite probability for this to succeed. ^v + ^[E. 2m0E „ 2mnE . 57.
264 we have ro / dxyfV(x) Ja CHAPTER 12.1 / 4 ) ! = (4/3)(3/4)! = (4/3) x 0.2 [fr/2 ~ I)'] (Ml)! 2 ' where B(x. h = 1. and assuming a length of b — a = 100 m of the base of the bump. We have 8m 0 : exp ^(ba)/4 x 2_1 ^^W(s x °' 9191 (12.1.055 X 1 0 _ 3 4 J s . (12.384.71) The integral can be looked up in Tables of Integrals.**.2. Gradshteyn and I. p.621. 14.9191 ~ 1. p. Jahnke and F. 950.l . p. (15.807 m s . ^ ) = 2 ^ . we obtain ( . 1000 kg) and using the following values of the natural constants: g = 9. Ryzhik [122].17)). formulas 3.4 x 10 3 9 ].1.72) Assuming a mass of 1 ton of the car (i. 369 and 8. . Quantum Mechanical Tunneling = y/mog 2(1) — a) /""'' / dw sin1/2 n Jo w. **For instance E. dx sin ' x •• Looking up the value of the factorial in Tables.2 . which give /•TT/2 dx sin'M . Emde [143].1 / 2 ) ! = v^r) TT/2 /. We can now evaluate the probability.e. : 2 ^ B ( ^ . one evaluates an approximate probability of exp[6. n I . M. S. We find™ (with ( .y) is the beta function (see Eq.
144. Siissmann [266].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.Chapter 13 Linear Potentials 13. we consider the linear potential in three space dimensions that forbids the particle to escape. 265 . 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. p. for instance.* In Chapter 14. An additional reason to consider the linear potential at this stage is its appearance in the following chapter in connection with the matching of WKB solutions across a turning point. Example 14. This is the problem of Galilei in quantum mechanics. This is the quantum mechanical version of the problem of Galilei in one space dimension. in Chapter 15. We shall see that in this particular case the wave function can be written as a superposition of de Broglie waves. in the book of G. each of which propagates with the classical momentum and energy of a Galileian particle.3. Then considering the probability distribution determined by the squared modulus of the wave function. Thus. Finally. 13. whereas the first case permits only a continuous spectrum.1 Introductory Remarks We distinguish here between three different types of linear potentials. we consider the corresponding case of a particle above the flat surface of the Earth. 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. * A highly abridged treatment of this problem is given. the last case allows only a discrete spectrum.
3) In general the momentum representation is of little importance. The equation of motion follows for instance from the derivative of the constant energy. x > 0.t). (13. i. V(x) = —mogx. We see " that the case g < 0 can be related to the reflection x — —x. Linear Potentials we shall see that at large times t its behaviour is determined entirely by the parameters describing the classical motion of the Galileian particle.t) = mogx + V0.t) = m0g with potential V(x.t).e. and yields rriQX = mog. We consider here the case of Eq. (13.1 S u p e r p o s i t i o n of de Broglie waves We consider a particle of mass mo falling freely in the timeindependent homogeneous field of the gravitational force F(x. In the present case of quantum mechanics we have h2 d2 2mo dx2 m0gx ip(x.t) = .= l V 27T Joo dxelkxip(x. (13.1) where g is the acceleration due to gravity and we can put Vo = 0. from d_ m x2 0 dx + V(x) 0. (13.t).1).t) = c\x\ + V0. c> 0. Recall the classical treatment of the freely falling particle.4b) . A different > "linear potential' is the socalled "confinement potentiaF V(x. Therefore we use the Fourier transforms il){x. The present case is an exception and a good example of its applicability.2. g > 0.2) which is the case with a discrete spectrum. (13. 13.t) = .4a) f°° ${k.266 CHAPTER 13.= l V 27T J_oo 1 dkeikxijj{k. (13.
t) the expression (13. d \ ~f1 dkr[ k + m0gt ^r't (13.6). Equation (13.3) yields us now by inserting for ip(x. The vanishing of the boundary contributions is to be understood in the sense of distribution theory (i.3) as a partial derivative d/dk: d lKx dkeAkx{imQg)—4){k.8) With this result and the substitution (13. (13. We rewrite this equation in the following form th (13.t) dk nikx k=oo '2K (im0g)il)(k.6) and wish to solve it. (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.e.6) (and to avoid confusion we do not introduce new functions and variables): k > k + mogt (13.e. (13.t). multiplication by a test function and subsequent integration).7) applied to every quantity in Eq.9) .t).2 The Freely Falling Particle: Quantization 267 and consider the following partial integration in order to reexpress the last term in Eq.d m +imo9 .t)im0g^k. • fd ~(' m0gt %h .< (13. (13.7) In the wave function on the left hand side of Eq.t) h2k2 d ~J(k.6) the latter equation becomes d ~( m0gt ^(^M<=+=?. (13.t) dketkxiP(k. we obtain the differential operator on the left of Eq.5) m+tm9Wcr{ktt)=2^^ktt)t (13.6) the argument k is then replaced by this.13.4a) d ~ f I dk ikx dk 2vr (im0g) iP(k. i. To this end we consider the following substitution in Eq.t) m0gx /2TT m0gxip(x. Applying the total derivative d/dt to this function.
the following result h 2moi tp[k\ ^—. We have thus obtained the solution of the transformed equation. To obtain the solution of the original equation (13.11) *[k?f. (13. On the other hand the second operator expression implies dtj>o imog^—ip(k.e.i) . Our next step is to reexpress the solution (13.1 we verify t h a t this solution indeed solves Eq.gh [ d t ' j f c + ^ ( t ' .t).t) ok lm og^— exp dk •ghf dt'fc+^(t't)W(fc.t ) = ipQ(k)exp Jo fdt' k+ m0gt' H (13. E x a m p l e 13.6) to the solution (13.6).1: Verification of momentum representation solution Verify by direct differentiation that the solution (13.11) in a neater form.11). We define the wave number kt as kt.10) k by m0gt back : k —> k h The solution of Eq.6).= k rmt. t ) .t) = i>0[k 1— ) exp 2m0i J0 k+ ^(t't) (13.11) solves the partial differential equation (13. Linear Potentials This is now an ordinary differential equation of the first order and can immediately be integrated to give an exponential. (13. (13.12) . (13.6) is then $(k. (13. We observe t h a t t h e solution is actually a function of kt — k — mogt/h.10) Note the amplitude function ipo(k) in front. we have to reverse the substitutions and replace in (13. Thus we have first (the second term arising from the upper integration limit.t ) ' U ( f c .6).6). the third term from the integrand) > / mog \ o exp dk h J h dt y ' 2mo k2i>(k.268 CHAPTER 13. i. Thus addition of both indeed leaves the expression on the right hand side of Eq.t In Example 13. Solution: We apply the two operator expressions on the left of Eq.
+^f (13. . Inserting ip(k. we can write the integral / = l h l h (hr + 7 7 (h 3m0g 3 m0g \ 1 H 3m0g With the definition « * ) • ! ( * = * .o o 2 —  exp iktx h mZgH' .13.19a) .17) = J_/ C X ) h2 [It(k) .13) rn0_t+2 + mlg2t3 .= 1 / f°° dkelktx^{kut) dkipol k V —t 1 exp iktx K ) (13. 27T ^ . (13.. we have ip(x.16) Next we evaluate the Fourier transform of ip(k.2 The Freely Falling Particle: Quantization 269 Then the integral in the exponential of the solution (13.t) defined by Eq.— [It(k) .14) (13. .I0(k)].4a). m0g (13.18) h =k mogt p h S' h . (13.2„2+2 mfaH 3^3+3 . we obtain rp(x.15) we obtain / = . _ k t  h2 3 Inserting here the explicit expression for fct./„(*)] 2m^ig Inserting the explicit expression for the argument of the exponential.t) = .gfci + 2 g2m0 2/i 2 ' (13.o ut = 2mo —kf 2mo A.11).t) = 1 / dkipo ( k mogt 2 h ( 1m\ig \ With . m0g ~7T' k Z k _ r ^ h t Y + i n Smog (13.} fA ' _ _tkt 2 k+ t) 2 k .t) into this integral and recalling the property (13.11) can be written j'A ^.
the relation (13.ut> = Jo t2 g2m0 i 3 k t .gk— + 2h 3 2?7lo 2 2 h mQgk ^ _ mjfo2 2 i + ft2 fcr 2mf)g 03mg £3 . (13.20) expresses the wave function tp(x.t)\2 for t — oo (distant future or past). The fcdependent terms in the argument of the exponential are T:=ikX^{^fk2+r^k 2rriQig [ n nr (13.The associated kinetic energy of the particle is mv+ = huJf. To this end we convert Eq.o o Jo This result can be interpreted as follows. These wave vectors vary in accordance with Eq. Such a decomposition is possible only for homogeneous force fields.2. fm0gt\ \ 2h ) (13.t) as a decomposition into de Broglie waves.18) into a • different form in order to be able to perform the integration with respect to k. 13.20) (13.270 and CHAPTER 13.t) as 1 I00 ip(x.21) In other words. (13.19b) h3 3 we can rewrite ip(x. 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.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 .2 Probability distribution at large times Next we inquire about the behaviour of the probability density \ip(x. Linear Potentials f dt'.t) =y= dkip0(kt)) exp i< fctx V27T J .v = kh/mo.19a) exactly as for a falling particle. (13. With the help of the de Broglie relation this wave vector corresponds to the particle momentum Pt = P — rnogt or to its velocity vt = v — gt.23) . 7 u>t'dt' (13.
TOO#A ft and.t) = 1 1 oo 271 27T /.26) with the formula 1/2 (13. fi > 0. (13.t) 1 exp r . We evaluate the Fresnel integral in Eq.e.29) .28) (for a verification replace a by a + ifi.QX 2 ^ 0 7 (m.t) exp i\ mogxt mox2 — h 2h 2ht mo<?2£3\] r (vn§x 24ft mogt\ (13. may be put in front of the integral.26) d£e ie £ kt^ m0' 0.2 The Freely Falling Particle: Quantization The integral (13.QX ht V ht 2ft For i — oo the definition of £ defines a distance xc\ given by > k i.25) The amplitude ipoi^t) then becomes for i — oo: > 4>o(h) \t\—»oo 4>o[ k \ ./oo dkip0(kt) exp l x ixI exp oo . integrate and then take the limit fi —> 0). 2h h 24ift ' mogt^ r m. /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. Then ip(x. ftf fti m0gt\ 2h J' " J' m0x ht mo5* 2ft + 2mo ~ftT f. (13.13./m 0 a. (13. being now independent of £.27) xd = vt Thus xc\ is the distance travelled by the particle as determined in classical mechanics.18) thus becomes ip(x. It follows that ip(x. rnogt m0g2t3 tX^r~. m0xcl ht m0gt 2h gt O v: °° 2hti (13.24) We now set fit / 2m 0 V" 0 V m0a.
Linear Potentials Replacing here x by the large time value x c i. (13. 13.dA 4>o(h (13.e. . 1/2  ±^(2mlg/h^x^ / 2i (13. kt and ut 13. xc\). These are incoming and outgoing waves. V(x)=m 0 gx V=E Fig.t) = eiEtlhct>{x) (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. For stationary states we set xl){x.t) ~ W —exp i V irlt m ktxd J cot.3) the timeindependent Schrodinger equation i^ + ^ [ E + mogx]<l>(x)=0.31) and obtain from Eq.30) We see therefore: The large time asymptotic behaviour of the wave function is determined entirely by the classical values of x (i.33) oo for E <C \mogx\.272 CHAPTER 13.32) For x — oo the function cf){x) behaves like > 4>{x) ex exp exp dx { 2mo (E + mQgx) \ H2 .1 A turning point at x = —XQ. For x exponentially increasing and decreasing solutions are possible. (13.1) has only a continuous spectrum.
t) = — = / V27T J . cf.e. so that ih dtp{k. The point — XQ.t) = dt (13.34b) It therefore follows from Eq.1.t).t). at which the classical particle bounces back. H2k3 4>{k) — cexp Ek (13. E)dk. 1 f01 dketlcx^{k. probability. t).6) that also for 4>{k.36) m0g \ 6m0 with c = const.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. we .0 0 dkelkxE^(k. i. However quantum mechanically this is possible with a small.t) — (p(k) > E +^og~Mk) = 1 (h2k2 ^m. i.34a) Then d &k 1 f°° ih—il>(x.e. 13.e.4a) and (13. Akx (13. its momentum p would be imaginary there.e.e. (13.13. and we can establish for the resulting continuum solutions the orthonormality and completeness relations. i. Fig.35) It follows that /W tm0g \ 2m0 i. i. We determine the constant from the condition that the continuum solutions are to be orthonormalized to a delta function. exponentially decreasing. i.e.4b). 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. t) = Eil)[x. Thus the spectrum does not contain states which are normalizable over the entire domain x e [—00. In view of the simple form of the potential in the present case the equation can be integrated in the momentum or k representation. at which E — V = 0. is reflected.00]. but also d 00 dt Eijj(k. which is not possible for real particles. The Fourier transform is given by the relations (13. and hence is continuous and extends from —00 to +00. J \ „ (13.
h? h2 . / dE(j)E(x)(t>E(x') = ^ s .. apart from a constant phase which we choose to be zero. Linear Potentials OO dx<ffE(x)<i>E>(x) = oo 8(EEf).M *K Joo Setting dt cos ( ts n Jo st).37) (eikxdk.L dxeikx(f>E(x).43) y=2^x+2^E.£') = <*(#£')• 1 27rm0g' = /o .a : ' ) .41) J—oo Next we insert (13. (13.e.36) (with c replaced by (13. the relations oo /*oo dE4>E(k)4>E(k') = 6(kk'). / <>~ikx VZ7T J (13. i.38). so t h a t oo / <fiE(x) J dke ikx s/2nmog exp m0g \ 6m0 Ai(s). (1340) i. </>E(x) = .274 have / Using d(x) = — one verifies t h a t with CHAPTER 13.39) we obtain dkcc*exp m0g It follows t h a t cc = &(# .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 ..39) dk4>*E{k)4>E. V 2lT J Mk) = .e.38) we have for <j)E{k): OO /"OO / oo (13. (13.36) into (13. 13. (13. In a similar way we can verify the validity of the completeness relation.(k) = / J ~oo dx<t>E{x)<\>EI(x) = S(E  E') Inserting (13.40)) into (13.L [ dkeikxMk).
^x + ^ E (13.45) One can now derive normalization and completeness integrals for the Airy functions.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.. the function 4>E{X) i and setting s y = — = 1 .46) This follows since with Eq. whereas the other is of exponential type.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. so t h a t the integral becomes 4>E{X) = 2vr ^/mQg / n oo dtexp oo h? * 2m£g y \ 7 5 * " 3 M' Choosing /2mfo\1/3 A= ^ .J * .^ . / s M . 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. (13. For instance dxAi(y / • + x)Ai*(x + z) = 5(y — z). . and we shall see t h a t one branch has periodic behaviour. For a further solution Bi of Airy functions we refer in Chapter 14 to Tables. (13.13.44) is then expressed in terms of the Airy function: .
e.\zz.50) Im z domain (b) / ' ^ ^ ~C domain (a) 0 ^ ^" ~ ^ \ x Zs=i^S Fig. . i.48) We have $(z) = sre1 1 A S„3i6 r e sr cos 6 . We want the integral to exist and so desire that lim e**^ = 0.51a) and since with sin 30 < 0 also sin(30 + 2ir) < 0: TT < 36 + 27r < 0.f dzJ*W 2TT JC with z = reie.4 The Saddle Point or Stationary Phase Method The integrand of Eq.e.51b) . (13. Thus we must have sin30<O.r 3 sin 30 ). 13. or n <6 <TT o for r > oo. (13.47) Consider the following integral along some as yet unspecified contour C in the plane of complex z: / := i . (13. (13. Linear Potentials 13.49) (13.TT < 30 < 0.43) contains the phase (with x — z): > *(*) := sz ..276 CHAPTER 13.TT < 6 < 0 for r > oo.. .2 The contour C for s < 0 with ends in the angular domains (a) and (b). i.r 3 cos 30 J + i ( sr sin 9 . o (13. (13.
Then 6S = ±7r/2.58) . oo] as indicated in Fig.55) into (13. 13.57) $"(z s ) = 2% s. we choose zs = iy/^s. Then the ends of the contour lie in the domains specified by Eqs. so that exp(±2i9s) — 1 and — s — rl. i. we obtain da 2KJC (*)V4 c 1 e **(Zs) exp i{$>{Zs)+l(zZsf<$>"(Zs) + (13.4 The Method of Stationary Phase 277 The integral (13.48).52) = exp(±i7r) Let s be real and s < 0.e.54) i ¥~s)3/2' (13.51a) and (13. (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.56) 2n(s)^Jc But I daefr"?*"^. —i ${zs) = szs . (13.55) s < 0.13. (13. (13. It is reasonable to choose the contour in such a way that only a short piece of it contributes substantially to the integral. i. (13. where the derivative of the phase function vanishes. at which the phase becomes stationary.2.S )V2 (13.51b).45) exists for the contour C. and the main contribution to the integral comes from the a region around the saddle point. Hence _ 2 / s)3/2 /•oo 2TT( s)!/4 doe J—oo CT2 i*3 3(. Then Ze TeC : V^se±i7r/2 = ±i\fzrs.e. provided its ends at infinity satisfy these conditions.54) where a is the new variable with —oo < a < oo. Inserting (13.z3s and V^s + . Such a piece can be found in the neighbourhood of the socalled saddle point zs.53) Since we shall have the contour C in the lower half plane. 0 = &(z8) = s~zt s = za „2„2i6>.
45) 4>E(X) ~ Ai(z).2/32u = 0 (13.2 we show that for s — oo: > Ai(s) 1 . Gradshteyn and I. V 2h2 (13.s ) V 4 exp \(s) 3/2 (13.278 CHAPTER 13.42) into an equation with a more appealing form. . We observe that the Airy function has a periodic behaviour in one direction but an exponential behaviour in the opposite direction. (13.44) and (13. With the help of the substitutions (13. 2 s 3/2 .491.64) tSee I. (13.56) decreases exponentially around a — 0 for real values of a. M. (13._ 1 . Sec. S.59) In Example 13.. > so that Ai(s) 2 v ^ ( . This is an important aspect in the WKB method to be considered in Chapter 14. where — with e = (2/h2mog2)1'3 — we have /2^m3g3y/3 x\ = e[E + mogx}. but would increase exponentially for imaginary values of a — thus describing a surface around that point very analogous to that of a saddle. V^s / 4 V3 (13.63) The Airy function can be reexpressed in terms of cylinder or Bessel functions Zv(z). (13. as one can see from the following equation^ ri' + Ffz •2^2.61) 2m0E = f 2 \1/3r 2 2 2 fi h \h mog2 Equation (13. Linear Potentials where the contribution oc er3 is obtained from &"'(zs) whose evaluation we do not reproduce here. 8. For (—s) — oo the integral converges towards T/TC.32) with the solution (13.62) This is the Airy differential equation with one solution 4>{z) = Ai(z). Ryzhik [122]. According to Eqs. cos .60) The reason for describing the point zs as a saddle point is that the integral in Eq.32) therefore becomes dz2 + zd) = 0. (13.44) we can convert Eq.
* " ( z ± ) = =F2Vi.^ e ^ / V2 4 [°° Jo dae'2^.59) and (13. Thus there are two symmetrically placed real saddle points on the real axis.z ± ) 2 * " ( z ± ) ._ .13.60) we have thus obtained (with the help of the saddle point method) the asymptotic expansions of the Airy function Ai(z) which play an important role in Chapter 14. cos  . 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 .51b) are satisfied.. 27T Ic Jc 3 This time s is real and positive.43)) Ai(s) = — [ dzeiit(z\ *(z): 1 Ai(s).4 The Method of Stationary Phase with solution u = z1/2Zi/2/j(7A 279 (1365) With expansions (13. we obtain f° Joo dp=—[a+ i\a\]da. Then. with *"(z) = 2z.s 3 / 2  . (13.51a) and (13. . changing to the variable p. the conditions (13. Hence the saddle points are given by 0 = * ' ( z s ) = s — z2. It follows that for s —> oo: Ai(s) ~ . Choosing the contour C to lie along the real axis. Example 13. Eq. kl dpe*' = ~e^'4 V2 f° Joo daeW2 and f°° dpe'^ JO = .2: Periodic asymptotic behaviour of the Airy function Obtain with the saddle point method the asymptotic larges behaviour of the Airy function Solution: We have (cf.
/ " ( z o ) = —n/z2 = —1/n. .280 CHAPTER 13.*. and hence integrating as described above through the saddle point indicated in Fig.111). i. Linear Potentials E x a m p l e 13.Be 2 Fig. ~ ef<z ex 3f(*)dz / x2f"(z0)/2 dx = i 2TT P/(*o) c The gamma function Tin + 1) or factorial function n! is defined as in Eq. Solution: Briefly. ^/2KJW1.e. the method of steepest descent evaluates the integral J^dz by expanding f(z) about its extremum at ZQ with z — ZQ = ix. since f'(zo) = 0 and (z — ZQ)2 = —x2. (11. 13. Thus here / ( z ) = — z + n l n z . z o = n . one obtains n! ~ where we used f^° dxexp(—w2x2/2) = V2^nn+1/2en. Then. Jo Jo \ .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. 13.dz = idx (observe that this implies an integration parallel to the axis of imaginary z through the saddle point). n!= / e~zzndz= / e~z+nlnzdz.3.3 The saddle point at z = n + ix.
* The method had. *R. where *G. N.1 Introductory Remarks One of the most successful methods of solving the Schrodinger equation is the WKB method.§ 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. ^ The central problem of the method is the topic of "connection formulas". Dingle [70].317.* Descriptions of the method can be found in most books on quantum mechanics and in some monographs. for misunderstandings pp. the linkage of solutions above a turning point to those below. Jeffreys [144]. however. The WKB result is therefore frequently described as the semiclassical approximation. ' T h e most prominent earlier expounder is H. 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. Brillouin [37]. 292 . Kramers [153] and L. i. For the history see pp.e.g. B. and the role played by h in our considerations here. named after its first promulgators in quantum mechanics: Wentzel. It goes without saying that such comparisons are very instructive.Chapter 14 Classical Limit and W K B Method 14. § See e. of misunderstandings and other aspects may be found in the monograph of Dingle. A critical overview of the history of the method. been developed previously by others in different contexts. A.e. 316 . 281 .295. O. The central issue of WKB solutions is their continuation in the sense of matched asymptotic expansions across a turning point (i. H. Froman [99]. Kramers and Brillouin. Wentzel [282]. will there be that of a coupling parameter. Froman and P.
2 Classical Limit and Hydrodynamics Analogy In this section we consider the limit h —• 0.7. It is then possible to derive in a general form the relation known as BohrSommerfeldWilson quantization condition.282 CHAPTER 14.2) dip(r = I VA{v) + ^ ( r ) V 5 J eiS^lh.* We illustrate the use of this condition by application to several examples. It is for these states that one obtains the timeindependent Schrodinger equation A^(r) + ^ [ E . Classical Limit and WKB Method E — V changes sign).t) = eiEt/h7P(r) and \V(r. For stationary states ^/.g. We start from the Schrodinger equation ih <9* =HV at with H=£— p2 2mo + V(T).V(r)]^(r) = 0. the function S being called phase function.3) . the energy has a definite value E with V(r. In this equation we set. so that _ Aip AA+^V• h AA+CVAn (AVS) + IVS h VS + AAS) iS(r)/h (14. the derivation in Example 18. (14. We shall see that in this limit the > Schrodinger equation describes a steady stream of noninteracting Newtonian particles. • {VA + %AVS [ h B iS(r)/ft +^{vSVA+^A(VS)2 *See e. 14.1) (14.t)\2 = \iP(r)\2. A(r) and S(r) being real functions of r. for which the observation of position and momentum is independent of the time at which the observation takes place. V>(r) = A(r)eiSWn.
11) We define as probability current density the vector quantity J:=K where V* = ijf* V* rnio i_ „ *.t)}2.5) 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.* VAVS+AAS + V(r)A.13) . These equations are equivalent to the Schrodinger equation.7) h2 AA 2mo A 1 (VS)2 2mo = 0 (14. (14. (14. V(r.+ VAVS + AAS = 0.10) m 0 ^ .Q A 2i 8A 1 (14.6) 2mo dt now with A also a function of t.4) (14.t)eiS{r't)/h.1) and separating real and imaginary parts. (14. We set in this equation. h2 ih— = H$ A + V(r) * . (14.10) with a continuity equation by defining as probability density p:=*** = [A(r. (14.t) = and obtain dA i_dS_A + dt h~dt 2 h r A(VS)2 AA h2 2mo ih A(r. h h (14.V(Y) + and 2VA • VS + AAS = 0. h h Taking real and imaginary parts of these equations we obtain the equations: + dt and 2rriQ 2m. we obtain the following two equations: E .12) VA A (14.14. (14.2 Classical Limit and Analogy with Hydrodynamics 283 Inserting this into Eq.
10).284 CHAPTER 14. TTIQ (14. (14. (14. (14.14) A2VS. the implications of this equation are also those of the equation of continuity. we have mo—(A2) + V(A2)VS + A2AS = 0. Classical Limit and WKB Method This means # * — V * = — AVA + — and so J = — A2VS. ot 2 Multiplying this equation by 2 and recalling that the function A was introduced as being real. Next we investigate the significance of Eq.n. + h2 AA V = — ^ .01.10) we obtain = — [V(A2) • VS + A2AS}.16) BA 1 moA—.15) and (14. (14.J = 0.21) . Since J = — A2VS = PVS.17) With Eqs. ^ + V • J = 0. mo <™> (14. ie. Since we obtained this equation from Eq.. (14.9) as 9S 1 mo^ 2 + T. m0 But now (14. %(1"1' I<2) and V J = — V • (A2VS) mo Prom Eq.16) this becomes m 0 1 rr + m o V .19) m0 m0 it is suggestive to define the following vector quantities ± = at v := I p = JVS. (14.18) ot ot This is the equation of continuity that we encountered already earlier. (14.9) for the phase function S.20) With this we can rewrite Eq. (14.+ AVA • VS + A2AS = 0. . (14.
22) (14. This result is very general. V • J = 0. Such series are typical for expansions in the neighbourhood of an essential singularity. i. However. d — = (14. An expansion in ascending powers of h therefore starts with contributions of the order of 1/h2. with the help of Eq. . and hence''' ^(m 0 v) + m 0 (v. that the quantum theory implies an essential singularity at H = 0 (i.V ) v + V 7 = 0. In the first • place we require the ansatz (14. or that of equilibrium in electrodynamics) and dS/dt = —E (from ^ oc exp(—iEt/K)). they are asymptotic expansions. We deduced this equation solely from the phase S in the limit h — 0. 'Recall the formula V ( u • v) = (v • V ) u + (u • V ) v + v X rotu + u X rotv.20).e. One should note that in no way do we simply obtain the classical equation of motion from the Schrodinger equation in the limit h2 — 0.14.e. The particles have no interaction with each other.2 Classical Limit and Analogy with In the limit h2 — 0 this becomes > Hydrodynamics 285 ^ + W 2 + F = 0. This expression with h in the denominator shows. where. (14.2) for the probability amplitude. > The motion of the particle or rather its probability is altogether given by the above equation of continuity.25) dt We recognize this equation as the classical equation of motion of the particle of mass rriQ. (14.24) d 1 dr d •— = d Lv _ . This is the analogy of the Schrodinger equation for h2 — 0 with hydrodynamics. a singularity different from that of a pole). so that V ( v • v) = 2(v • V ) v + 2v x rotv.1/h. which for very small values of h yield the dominant contributions. we obtain ^ V 5 +  v ( v v ) + V V = 0. r o t v = ( l / m o ) r o t g r a d S = 0. v dt dt so that the equation can bedt written dr dt ~(m0v) + W = 0. This means.23) (14. The equation of continuity then describes the stationary flow of a fluid. the wave function \I/ describes effectively a fluid of particles of mass mo. We 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. > In the case of stationary states we have dp/dt = 0 (which is analogous to the condition of stationary currents. dt 2 Constructing the gradient of this equation.
(14.286 CHAPTER 14. similar to our earlier procedure with A and S real functions. Dingle [70]. As remarked at the beginning. 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.5)) and 2 ^ f dx ax + All=0. There are also other more specific descriptions. Kramers and Brillouin. such as "LiouvilleGreen method' and "phase integral method'. Sec. is an asymptotic expansion. B.4) and (14. p. .2.§ y = A(x)eiS^'h where W(x) = S(x) + . h2 l 2m0 {EVy corresponds to an expansion in decreasing powers of (E — V).26) We observe already here that an expansion in rising powers of k. We consider the onedimensional timeindependent Schrodinger equation y" + ^. 318.3.27) (14.[EV(x)}y = 0.1 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. 6. z (14. i.30) ax *For details see R. or h2/2mo(E — V).28) The calculations we performed at the beginning imply the following equations (compare with Eqs. Classical Limit and WKB Method 14. Messiah [195]. We set.3 14. (14.e. §We follow here partially A. (14.$ The basic idea of the WKB method is to use the series expansion in ascending powers of h and then to neglect contributions of higher order. The expansion in powers of h2. i = eiW(x^h.In A(x). the method is widely familar under the abbreviations of the names of Wentzel.
32) s" = sz + h2s'( + •••.33) 3 (S")2 .4 (S>)2 1 S'" 2 S> (14.A = const. (14.14.V) = h2 In the WKB method one now puts S := S' = S0(x) + h2S1(x) + (h2)2S2(x) S'0 + h2S[ + ••• . (14.2rn0(E V) = h2 lS'A 2S'0 (14. r//\ 4V^ 2 Up to and including contributions of the order of h2 Eq.2mQ(E . (14.28). Integrating this second equation we obtain ll[dX so that = \l^dX' A = c(S')" 1 / 2 . Eq.35) .29) becomes (note that this is still exact!) (S')2 . .e. Li This expression can be substituted into Eq. Since the expression (14. 3 / 5 .\ In S'.32) is therefore (S'0)2 + 2H2S'QS[ . 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.34) Equating coefficients of the same power of h we obtain (S'0)2 = 2m0{E . (14.10)).31) \a.3 The WKB Method 287 Equation (14.V) and (14.35) can be integrated.36) Equation (14. + (14.30) corresponds to the equation of continuity (i. (14.
Classical Limit and WKB Method has the form of a momentum.41) We now ask: What are the conditions of validity of Eqs. (14.± ln(S'0 + h S[) + In c (s'0y/i and hence exp :5 0 + 0(h2).37) For Eq.39) where a and /3 are constants.19). we expect that in any case So » ft2Si. E > V(x).E)' Then y(x) = V^(a 7 exp dx + 6 exp T(x) (14. .40) rx dx v "7 W)\.39) and (14. Dingle [70]. (14.31): y = exp = ~ ~iW(x) h exp L nH™ exp ^S .and So = ±h / — — A J k(x (14. B. Chapter XIII.41)? We note first that the expansion in powers of h2 does not converge.I In S' + In c h 2 2 2 exp ^{S0 + h Sx) . it is reasonable to set (cf. For V(x) > E (this is the classically forbidden domain) we set l(x) = ^==^====.288 CHAPTER 14.33) to make sense as an asymptotic series.27) we obtain now with Eq. (14.38) h y/2mo(EV)' (14.e. and instead is asymptotic which we shall not establish in detail here. The solution is seen to be periodic provided k(x) is real. y/2mo(V(x) . p = h/X) k{x) :Then S'n +. i. Eq.' For the expansion (14. (1. (14. y = c'{k)1/2 exp or y(x) = a \ A ( z ) cos I / dx W) dx + 13). (14. 1 S e e R.
14.36).36)) 2s'0s[ = [(s&r 1/2 ra) 1/2 With S£ = ±ft/A. E > V(x). (14. we obtain 1/2 1/2 2S[ and hence ±2hS[ = [kL/2]"kL'2 = lk~1/2X = Al/2 [lAl/2A//_lA3/2(A/)2lAl/2 so that It follows that (14. Setting R := A' = \moKV'{x)\ \2m0{EV{x)\W .V{x)) ^ This condition is satisfied provided E or E — V{x) is sufficiently large or (14.1 / 2 ]" [§0SS). Eq. Since k(xy 1/52\S'0 (14.43) The condition So 3> ^2<Si therefore implies that /f>^i/^)k = H/y/2m0(E is sufficiently small.45a) .3 The WKB Method We have 289 S0 = ±h and Si follows from (14.3/2 s&T + 3 {So Il« we have (cf.42) [(^).
290 CHAPTER 14.a)V'{a). In the neighbourhood of E = V(x). V(x)) (14.V(x) = V(a) . 14. Classical Limit and WKB Method we can rewrite the inequality (14. *2 A(z) = V / 2m 0 (£ . i.e.44) as Idxsj2mQ(EV{x))l 1 + ^R2j » ^Rh.3. (14. around the classical turning point. In a similar way we have for E < V(x): l'(x) <C 1.48) The (dominant) WKB approximations are exact solutions of this equation.47) Differentiating this twice — we omit the details — we obtain the equation y" + l A2 (A 1 / 2 )" *i/2 y = o.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.46) The desired equation has the solutions y(x) = aXll2 exp ±i dx / (14.45b) so that the condition So » h2S\ is also satisfied by demanding that i ? C l . We can derive the equation whose exact solutions are the WKB approximations of the Schrodinger equation. h = 1) .48) behave in the neighbourhood of such a point called "turning point'. at which E — V{x) changes sign? We note first that in this case near x = a: E . and hence (with UIQ = 1/2. In terms of X the original Schrodinger equation is y" + = y = 0. (14. the WKB approximation is in general invalid (see below). How do the solutions of Eq.V{x) ~ Or . (14.
since this requires different approximations there.1.14.1.3 after stating them first here. in the domain E > V the WKB solutions are oscillatory. (14. The transition is provided by matching relations.1 there are WKB solutions on either side of the turning point and some distance away from it. 14.48). and on the other side. 14. We assume we have a turning point as indicated in Fig. We define solutions ?/i and ?/2 with branches to the left and to the right as follows (for . It is therefore necessary to find other solutions in the neighbourhood of x = a and then to match these to the WKB solutions in adjoining domains as indicated in Fig. i. V=E Fig. On one side. 14. in the domain E < V the WKB solutions are exponentially increasing or decreasing. (14. which connect one domain with the other. We shall not derive these conditions here in detail but rather make them plausible in Sec. 14. In the immediate neighbourhood of such a point the Schrodinger equation can therefore not be replaced by Eq.48) possesses a singularity of the form of l/(x — a)2 at every turning point x — a.1 The regions around the turning point at x = a. As indicated in Fig.3. 14.3 The WKB Method with the derivative relations 1 1 a)5/4' 4(x 1 5 16 (x.e.a ) 9 / 4 ' 291 (v^r 1 (x — a>2 These relations imply that Eq.
50) x S> a. The first W K B matching condition in the direction indicated by the arrow (and for the potential decreasing from left to right. a interchanged. one could simply replace everywhere /•a / pa • • • b y / Jx J x .49) x 3> a Vxsin and 2 exp V2 VACOS (L X dx A ra dx X for for i « a . (14.1) is then: VI exp {~[T) Vxcos fx dx TV Ja ^ _ 4 (14. This means the asymptotic part suffices only in t h e exponentially decreasing case. in order to make the formulas independent of whether the potential is rising or falling. (14. and E ^ V for a ^ x the same formulas apply with the same direction of the arrow. as in Fig. Thus.52) For a potential rising from left to right. has the same asymptotic behaviour in the domain i « a a s Ay\. however with the limits of integration x. P u t differently: We could add a lot of contributions to Ay± without affecting the asymptotic form of Ayi. (i x dx 7T ~~k~ 4 / One should note t h a t the solution ^4?/i + By2 for A ^= 0.51) 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. Note also the extra factor 2 in yi.H— J =^ v/exp + m (14.292 CHAPTER 14.1) is given by the following relation with the + sign in the argument of the exponential on the right and no factor of 2: VACOS I / —. 14. 14. Classical Limit and WKB Method definiteness note the + signs): Vtexp yi ~ S + Jx I X for for i « a .
a). y" + Toy = o.3 The WKB Method 293 In that case Eq. is singular at a turning point x = a.2 The overlap regions around the turning point at x = o.54) .46).Airy overlap* Fig. One should remember that the matching relations above are those for the dominant terms in the WKB expansion. as we saw.e.3 Linear approximation and matching How do we arrive at the matching relations given above? We observed above that Eq. Therefore we consider now this original equation in the domain around x = a. 14. 1 ~ (x . 2 1 X (14.3. the matching relations have to be altered accordingly.48). If higher order contributions are to be taken into account. Thus around that point we can not use the WKB solutions and hence have to find others of the original equation (14. (14. whose exact solutions are the dominant WKB approximations. (14.53) Since at x ~ a. (14. the equation V approximately linear around turning point V=E Airy solution WKB . the equation becomes approximately y" = (xa)xiy.51) — always with the decreasing exponential on the left — is valid in both directions. 14. i.14.
(13. In particular we encountered a solution written Ai.1 / 4 by \^.5960. i I 772 \3 + > for ia>0.59). we have r x * f x ' /2(x a)i/2d*=^(x . (14.57b) To the same degree of approximation as the equation y" = (x — a)xiy.a)X\/3.4.56) We have discussed some aspects of solutions of this equation in Chapter 13.57a) COS 2^2/3 _ SZ 1 1 ?(z) /2 3 es^ Bi(z)~< 1 1 for x — a C 0. With the substitution z = (x . A second solution is written Bi. R. (14. formulas 10. or to the literature:H 3/2 2^{zy/* Ai(z) ~ < ~~FL~T77 y/TTZ1/* 1 1 s e K*) n for ~. . > for a. p.a)3/2 ^ /2 =^ 3/2 o Q. J (X (i4 58)  Replacing in Eqs. — a ^> 0. Classical Limit and WKB Method where xi is a proportionality factor which we choose to be constant. (14. 448. Abramowitz and I.60). (14. In the following we require the asymptotic behaviour of the solutions Ai(—z).57b) UV2 3 11 by J dx . Cf. B. This is given by the following expressions. as > may be seen by referring back to Eqs. Dingle [70]. p. (13. A.294 CHAPTER 14. Stegun [1]. or M.55) the equation becomes the Airy differential equation d2y dz2 zy.Bi(—z) for \z\ — oo.^ 4 I 7T x — a <C 0. (14.291.57a) and (14.
1 to the quartic oscillator. 14. In this sense we have now verified these relations. In the direction to the left as indicated in Fig. e. by Ai(z).(zf2 by J we obtain the matching relationsX (14. 14..2.g.2. In the neighbourhood of a turning point the Schrodinger equation becomes an Airy equation (in the leading approximation).l. (14. 14. We summarize what we have achieved. we enter the domain of validity of a WKB solution. F T' dx (Z)1^ by ^ harmonic oscillator V=x 2 +x 4 V=E linear approximation Fig. As an illustration of the use of the WKB solutions we apply these in Example 14. are only valid in a small interval around the turning point at x = a as indicated in Fig. As we approach a limit of this interval in going away from the turning point. These solutions. and hence must be proportional (in view of the uniqueness of the solution there).14. Example 14. n = 0.1: 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 / — = .3 The anharmonic potential well.1 the Airy solution becomes proportional to the exponential WKB solution and to the right to the trigonometric WKB solution.59a) .3 The WKB Method and 295 •. At and around the turning point the solution of the Schrodinger equation is therefore given by Airy functions. In this way different asymptotic branches of one and the same solution are matched across a turning point. however. The latter domain is a small region where the solutions from either direction overlap.52).l v/2m0(EV(x))dx 7r = (2n + l).51) and (14.
as in the case of the harmonic oscillator itself. /n = s u ^ TJcos^ya TJ+cos^a T J .y o dx TT T provided 6 _ 4 /. Classical Limit and WKB Method where x = a. 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 ( . According to Eq.59b) Solution: We consider an anharmonic potential well as depicted in Fig. (14. . n = 0 . . .3.61) Evidently these functions have to continue themselves into each other (since the wave function has to be unique).l. 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. Consequently only the lowest eigenvalues would be reasonably well approximated by those of the harmonic oscillator. 14. (14. regions 1 and 3 in Fig. . 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. j/i.296 CHAPTER 14. 14. 1 .b). The potential now has to be inserted into this condition and the discrete eigenvalues En of the problem are obtained. Then around these points the potential would be well approximated by a linear potential (as we saw above). 2 n = 0. b are the two turning points. . In these domains the dominant WKB approximations are 1 /j/i(x) = cVlexp / fa dx\ I —/ — \ r for i « a.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 .e. In the neighbourhood of the minimum at x = 0 we could approximate the potential by the harmonic oscillator. yz{x) = c'vlexp ( [x dx\ I — / — ) for x 3> b. (14. i.3.2.^ . the continuations of these functions into the central region 2 are: ya(x) = c v A c o s I / J. I " ' (K rx dx iz\ _ ( (b dx_ 7T _ fx dx _ • f / fb dx\ f* dx\ ( fb dx\ .44) this implies large values of n.51). whereas the approximation of the eigenvalues by comparison with the harmonic oscillator in the domain of the minimum. (14. for which the WKB approximation surprisingly yields already the exact energy eigenvalue as one can verify by evaluating in its case the BohrSommerfeldWilson rule. ^ A = (2n + l ) . (14. The corresponding eigenfunction would be approximations of the proper eigenfunctions around the origin. The condition for this is ya{x)=yb{x) for xe(a. Here we are interested in the discrete states (the only ones here). Hence we search for solutions which are exponentially decreasing in the far regions.60) Using the matching relation (14. extends only over a length of very few wavelengths and is therefore valid for n small.(x) = c ' v Acos I / — I for a < x < b. 2 .. Sometimes this distinction becomes imprecise.. We argued earlier that the WKB method is suitable in the case of large values of E. Let E = V(x) at x = a and x = b.
2 . 2. (c) V Fig.64) (14. n = 0.1. 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. . Consider the Schrodinger equation in the abbreviated form d2iP(q) dq2 f(q)t/>{q) = 0.14. and knows of no published form or script. with n — 0. . ri2 I Jqi :pdq n + 1 — . equivalently.x number of turning points h.63a) or. pdq = n + 1 — . 14. . (14. . Wilson by supplementing classical mechanics by Planck's discretization. Bohr.x number of turning points TTH. The corrected form* is.e. (14. . This relation is sometimes wrong. Sommerfeld and W. Dingle.1. In this socalled "old quantum theortf (i. B. A.4 BohrSommerfeldWilson Condition 297 14.4 Bohr—Sommerfeld—Wilson Quantization A quantization condition which is very useful in practice and in a wide spectrum of applications is — and remains in spite of its old fashioned reputation — the quantization condition established by N.4 Three cases with different pairs of zeros.63b) *The author learned this in lectures (1956) of R.
i.e. / • n an integer > 0.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. In quantum mechanics we have dq2 + Ki/j(q) = 0. as illustrated in Fig.4(b). i. We have the case of trigonometric solutions. if /(g) is negative.67) Case (b): Next we consider. Case (a): The case of two trigonometrical zeros. 0(g) ex . i. together with q — — oo.4(a). Since sine and cosine have zeros spaced at intervals of TT. 14. The WKB Method One pair of solutions is (as we know from the earlier sections of this chapter) tpT(q) oc 7T7Iexp 1 sin (_J)l/4 Cos 1 = JdqVJiq) F if f(l) is positive.e.e.66) J<1\ Thus in this case the socalled old quantum theory is correct. There are three cases. where n —. (14. Before we continue we recall from Eq. Jqi ^ fpdq J =2 pdq = (n + l)h. (14. the case of tp(q) — 0 at q = gi. 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. In this > case the wave function for g > go is . —/ = K J h It follows that / pdq = (n + l)Trh= (n + l)h. (14. this means dqyf—f = (n + l)7r. this is a trigonometrical zero. i. where /(g) is positive. where /(g) is negative.298 CHAPTER 14.65) We use the knowledge of these solutions to find the eigenvalues of the equation.51) the linkage between the trigonometrical and the exponential solutions across the position go in space at which /(g) = 0. this is an exponential zero.
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 . (14.3.67): • the wave function for q > qo must be proportional t o (Z)1/4 sin [ ^W)dq+>x Jqa .4 BohrSommerfeldWilson This will vanish at q = q\ if rii / V^f{q)dq+j^ Jan It follows t h a t i 4 Condition 299 = {n+l)TT. E x a m p l e 14.63b). Hence we obtain the condition I'o Vf(l)dq 90 (n + 1) IT. As illustrations we consider Examples 14. the sines must be proportional. Case (c): Finally we consider the case of i/j(q) = 0 at q = ± o o exponentially as illustrated in Fig.4(c).70) We can therefore summarize the results in the form of Eq. or Pdq (n + l ) " 2 h.14.68) (14. For the wave function to be exponentially decreasing to the left of qo and using Eq.2 and 14. ^~f(q)dq Jqo LJqo + 7T J'% On fq / i 1 V~f(<i)dqjK (n + l)7r. Solution: As in the case of the quartic potential.63b) to obtain the eigenenergies of the quantized harmonic oscillator. i. Therefore their arguments differ only by (n + l)ir. this is the case of two turning points.77T (where we reversed the order of integration to obtain a positive integral and multiplied through by minus 1). Since both cases refer to the same region qo < q < q'0. and • • for the wave function to be exponentially decreasing to the right of q'0.e.69) pdq = (n + l ) " 4 h. (14.e. (14. Hence we have pdq= i n +  .63a) or (14. ri\ / J do y/f(q)dq (n + 1) 1" (14. 14.2: The harmonic oscillator Use the relation (14. Thus in this case the socalled old quantum theory is wrong. i.
63 x 10~ 3 4 J s . 14.300 With potential V(q) = 2ix2mov2q2. g = 9. as illustrated in Fig. ^Thus a trigonometrical zero is the condition of absolutely no penetrability beyond it.71) For m 0 = l x 1 0 _ 3 k g .5.28 x 1 0 . so that E = hu{n + 1/2). whereas a turning point does.2 . 14.V(q))dq The result is E/u. Solution: This is the case of one trigonometrical zero of the wave function ip at q = 0 and an exponential zero at q —> oo. one obtains E ~ 2.3: 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 . in principle. Thus there is one turning point at E = mogq.2 3 J x (n + 3 / 4 ) 2 / 3 . . and h = 6.5 A particle above a flat Earth. 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. 4 / x CHAPTER 14. :: V=m0gq Fig. Example 14.807 m s . permit this although with rapidly diminishing probability. where g is the acceleration due to gravity. Calculate its quantized energy.
2.( 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. Chapter 16) one obtains the quantization relation (for Mo = const.5 Further Examples 301 14. In the case of screened Coulomb potentials (cf. d dE V)={n+)n.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 .) M0 N = l + n+1 : + • 2/B Hence dE K 2 d dE[ M0 2VB.14.4: W K B level splitting formula Derive from the WKB solutions of the Schrodinger equation with symmetric double well potential the WKB level splitting formula A f B £ = . (14.= — <*— = h(n+\. (14. 1 . 9 .75) which verifies the formula immediately. E x a m p l e 14.AE(q0 2 = 2n + 1) = — exp 7r dzy/E zn + V(z)/h where q=zo are the left and right barrier turning points.73) Jx! rp2a dx I^<E 2 v) = ^ n — = .l.f V ft h JXl h 2TT (14.72) \\2dxp=Qdx^{Eh Hence.5 Further Examples E x a m p l e 14. 2TT W 2K J p dE V where p = ^/2mo{E Solution: We have — V). and u> is the oscillator frequency in either well. h M0 _h 1^3/a ~ 4 0 / 2 J V \ 3 _ 2N3 \M0) ~ ~Ml ' .5: Period of oscillation between two turning points Use the BohrSommerfeldWilson quantization rule to obtain for the period T of oscillation of a particle of mass mo between two turning points the relation T 1 m0 f dx d ( — = . ra 2/ = 0. with p = modx/dt. Solution: The proof is contained in Example 18.
and (b) evaluate the WKB exponential exp[—2/b arr j er ] and the WKB prefactor 2/ w e u. Example 14. Eq.6 The cubic potential. q = y+ a .e. (11.77) Determine (a) the turning points.78) (a) We determine first the three turning points at qi = <?i.a . where .6.<?2i<?3 for E > 0 given by E — V = 0. (14. which determine the imaginary part of the energy. Fig.~ \ (14. (14.79) .y2)(y . where 1 2 • 1 2 y = Q. 14. 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 .76) in agreement with the result obtained from the classical Kepler period for the Coulomb potential (cf.6: WKB method applied to the cubic potential Consider the cubic potential V(a)=1b\\a*~q). The q term in this equation can be removed by transforming the equation to a cubic in y.302 and T CHAPTER 14. i. 14.133)).y3) = 0. The WKB Method ^ 2 JV 3 2TT ~ ~Mjf (14. Solution: The potential has a finite minimum at q = 0 and a finite maximum at q = 2a2/3 as indicated in Fig. It follows that the energy between the minimum and this maximum lies in the range 0 < E < £ m a x = ^aSb2.
(l_fc' 2 a + fe'4)l/4. (14.82) In o r d e r t o d e t e r m i n e t h e r e l a t i v e p o s i t i o n s of t h e r o o t s qi.g i < 1. 78: sin 120° = v ^ A c o s 120° = .2 c o s 0 ] . W e o b t a i n ^ fbarrier : = 6(92 ~ 9 3 ) 2 ( 9 2 . M i l n e .85) T h e angle 9 = 60° c o r r e s p o n d s t o E = 0. V cose' + sin 8 (14.gi = 1.{ V 3 e } > 0.5 Further Examples 303 T h e r o o t s of t h i s e q u a t i o n a r e k n o w n * a n d for 4 P 3 > 27Q2 a r e r e a l a n d e x p r e s s e d in t e r m s of a n a n g l e 6 given b y (observe t h a t w i t h t h e choice of t h e m i n u s sign.04.08.9 3 a n d t h e elliptic m o d u l u s k a n d t h e p a r a m e t e r g a r e given b y k2 k'z = 92 — 93 _ 92 . T h e r e f o r e E = 0 c o r r e s p o n d s t o fc2 = 1. D .120°)] = . D w i g h t [81]. H e r e cos(<9 ± 120°) = . 6 = 0 i m p l i e s E m a x of E q . p .1 / 2 . B y r d a n d M . 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. 14. 37. 92. V3 + t a n 0 2 .78)) c o s 30 •• 4P3 ( ' \ aeb2} . 92 = a2[l .2 c o s ( 0 .§ ( c o s 0 ± \ / 3 s i n 6 0 a n d § sin 6 sin# (14.< 1.a 2 [ l . g 2 ~y{3}>0.= s i n t 3 v 3 (14.c o s ( 0 + 12O°) + c o s ( 0 . § T o h e l p we cite H. (14.ui=K(k). ui=sn 1 \l— — = l. 0<6O°. 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 . F . 80 a n d 361.6. q3 ~  . B .e. (14.T h o m s o n [196].120°)] = a 20? 2a2 3 2a 2 • cos 0  91 .c o s O + cos(6» . 93.93). ' P .\tan0^ v/3.2 c o s ( 0 + 1 2 0 ° ) ] .1 2 0 ° ) w i t h 0 < 6 < 60° for 0 < E < £ .g i ) ( g 2 .83) T h e i n t e g r a l m a y n o w b e e v a l u a t e d w i t h t h e h e l p of T a b l e s of I n t e g r a l s . f o r m u l a s 236. we h a v e qx = .cos 0 + cos(6> + 120°)] = a2 92 . q3 = a2[\ . F r i e d m a n [40]. h = 1) ^barrier := / d( }J^2~(E~v) = ^~= dq^/(q . p . p . 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 .q)(q .93 2a2 [ . . 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.80) I n s e r t i n g t h e r o o t s yi. _ \ / 3 — tan(9 2 V92 . 91 < 93 < 9 < 92(14. we set 0 = 60° — e. (b) W e see t h a t gi a n d 93 m e r g e t o 0 w i t h e — 0. gi ~ y {V3e} < 0.2/3 i n t o q = y + a 2 / 3 . (14.84) 92 .79) a n d t h e n s e t t i n g m o = 1. p . 212. M .81) m a 91 . tan60°=%/3.' : ± 6 0 ° w h e n E = 0. a n d t h e a n g l e 0 = 0° t o E = £ m a x . e > 0.14. i.9 l ) f f / du s n 2 u en2 d n 2 u .3/2.9 3 — = = 2 " 3 " [ .0^6O°.9 2 [ .k 2 2tan# \/3 + tan! .
we obtain in the limit of E = 0: j d ( ^ ( E „ y ) = 2/barrier = ^ ! .92) ''P.2)*w+2(fe4+k'2)E{k)] v ' G(k) For one complete round from one turning point back to it. (14. E(u = 0) = 0. The WKB Method elliptic integral of the first kind.~ y ^ ) a< [ COS 5 + . V3 cose + ^ = V3 2 1 + 72 v/l + 3 7 4 C (cose) 2 = 1 + 3 7 4 = 4 d . V?2 . 72.2(fc2 . G(fc) =86aS (ITS£ li[fe. 2 2 . V 93 .90) hi JlnynEv) V(qqi)(q2 Vz b Jq q) I" bJqi where P fe = 9391 9291 = . .fc'2 + fe'4)"1/4. Thus Iwen=lgF(ip. as in (14.32)). g= 2 . Next we evaluate the required integral across the well from 91 to 53 at energy E. sinV = q3 qi ~ = 1.?)(?3 .) .qz) (14. with K(k = 1) = oo. Friedman [40].84) we obtain Carrier = 2a 2 sin6> 2 b( .91 . Eq. the integrand of the integral across the well is effectively the inverse of this momentum. 1+72 = and J o s 0 + ^ = (1 .87) Inserting G(fc) together with the expressions for the prefactors into (14. 0 (14. . Whereas the integrand of the above barrier integral is effectively a momentum. D.85). E(u) the incomplete elliptic integral of the second kind. Thus Jwell : = [13 1 f13 rf 1 / dq J^^EV) 1 y/2 1 V 2 f[13 = — —. 91 < q < q3 < 92 bJc. p.k). (24.86) 15 tan.91) a2[cos0^] y/3tanfl ? L .S . A:') = # ( * ' ) • (14. (14.sm t %/3 £\. Byrd and M. E(k = 1) = 1) (•u\ = K{k) i Q G{k) = / Jo Next we set s dusn2ucn2dn2u = j[k'2(k2 15fc4 . E(u = K(k)) = E(k).89) This expression will again be obtained later with the help of configurations called "bounces" (cf.304 CHAPTER 14./ dq~ ^2 b JQl 1 \/(qqi){q2q)(q. We can evaluate this integral again with the use of Tables of Integrals. formula 233.^ V3 / V V3 y a / c o s e J.^ + ^ ) (l + fc'2)2 .00.2)K(k) + 2(fc4 + k'2)E(k)) k ^i — . (14. F.91 F(TT/2.= —^ = k a2[cos0+^] v ^ + tanfl V=".
85) (14.! With these expansions one obtains ^barrier = 15 ahb 2 + 8A .3.97) 4 32 4 32 Here we insert the corresponding expansions of the complete elliptic integrals'^ (note the argument of K and E is fc): K(k) E(k) fc'2 = = < $ ) fc'2 1+ f l n\k'l f H + 1 4 5 3fc' 1T Hfc^J ~2_ 1^ \k'J 12 (14. With q = 1/z the integral (14. .114.** However. S. we obtain > . Thus one obtains from (14. we obtain 15 .88) in rising powers of fc' . formulas 8.93) where w is the harmonic oscillator frequency in the well..2. we cannot expect the ratio exp(—2Jt>arrier)/2J'well oc w exp(—27barrier) evaluated at k = 1. 2a 5 b K(k) + 2 E(k). we obtain E a6b2k'4(l + k'2). 905. 32 (14. so that even in the case of the ground state the zero point energy will contribute to the prefactor.94) C °S3 2 (lfc2 + fe4)3/2  l+ g f c ( l + fc) + Comparing this equation with Eq.80).2 IE 4 fc' ln 8 (" «V iU^fc' 4 . The same expression for E is obtained by applying the "method of poles at infinity" to the BohrSommerfeldWilson condition (14.4 2n + 1 fc' = 4 — ^ — .3 and 8. M.89) can be expanded to exhibit a simple pole at z = 0 allowing evaluation with Cauchy's residue theorem: 2 dz 2Ez2 a2b2 %2dz Ez2 2nib\ o?E o?b2 2KE ab tt I.l.99) **For the potential approximated around the origin as V ~ a2b2q2/2 the eigenvalues are E = fat)(n + 1/2) with it) = ab. 906. and hence at E = 0.113. Ryzhik [122]. (14.96) Expanding the coefficients of the elliptic integrals in Eq. 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. (14. pp. (14. (14. a°b n = 0. To obtain the latter we have to use the quantum mechanical expression approximated by E = hw(n + 1/2). Gradshteyn and I.95) Comparing this with the harmonic oscillator approximation E — En = ab{n + i ) . in fact through logarithmic contributions contained in the argument of the exponential for k ^ 1. to represent a physical decay rate.5 Further Examples Hence (using K(0) = rr/2) 305 ab ab (14.14. . (14.63b).
101) (14.. this has not yet been done.178)) call the Furry factor set equal to one. 8 Bfc .96) for the imaginary part of the energy as in the Breit.Wigner formula (10.11)) dq 2TT y/2§?(Ev) It would be interesting to derive the same quantity with the perturbation method and to compare the results. (18. It follows that (for one complete orbit back to the original turning point) . V (14.306 CHAPTER 14..2 7 b a r r i e r ] Z Z7T where (cf. so that we obtain 4 . 7 ~ —.J + 0(VoJb).e x p [ .> tf = = 2/wfiii = ±tv8a°b——P.73).j . . (14.— (14..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. (20. 15 a . _ exp[2/barrier] >.ba / 25a5b\n+2 ±l_l eT5 a f >. . /25a5b\™+5 e exp[2/barrier] ^ ( r ) Correspondingly we obtain for the full period 2/ w e ii: 2/well = ? K ( f c ' ) ( l .7 . oa ba Thus with the W K B method we obtain exp[2/barrier] — barnerj = 8 s.e. We observe that with \j2je as 1.( n + l ) e _ JL a * 6 and we obtain for the ground state (ro = 0) .„.100) 'barrier = ~0°b In [ ^ — j . +3 the result becomes exP[2Jbarrier] = ± i _&g_ (25a5br+J e . (0) i hw En = E\> . 2n + 1 — / 26a5b \ . n ~ V27T . also Eq. the result agrees with the ground state path integral result (24.102) . i.103) 27 w e l l 27rVn+y ' fn With Stirling's formula in the form of what we later (with Eq. _ T T • (14. i./^~^l^ab ^aub . The WKB Method For k' —• 0 the first and the third terms dominate.e..fc'2 + fc'4)1/4 * ± t ^ .
g. Quigg and J. Lane and T. along with inclusion of angular momentum and spin effects and their interactions. This would mean that the force binding the quarks together would not decrease with increasing separation as in the case of e. i. K. the Coulomb potential. L. This potential became widely popular in the spectroscopy of elementary particles.M. [232]. [233] and H. at least indirect. Rosner [230].Chapter 15 Power Potentials 15.1 Introductory Remarks The particular linear potential we considered in Chapter 13. T. 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. after the realization that quarks as their constituents might not exist as free particles and.* 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. Gottfried. Kinoshita. L. [231]. Eichten. describing free fall under gravity. C. The study of this potential. that it may not even be possible to extract individual quarks experimentally with any finite amount of energy. B. K. indications from the nonabelian generalization of quantized Maxwell theory. that this is indeed the case.e. the classification of the various states of nucleons and mesons and other particles. In the present chapter we consider briefly the threedimensional potential V(r) oc r = r. D. led to a spectrum which contains only scattering states. which permits only bound states. C. in fact. Quigg and J. 307 . Thacker. Yan [82]. See particularly E. Rosner [269]. There are numerous. now known as quantum chromodynamics.
2) tt(r) = Ylm{9. and u'(0) > [u(r)/r]0. We leave consideration of the logarithmic potential to some remarks and Example 15. we return to the Schrodinger equation in three dimensions. and obtain 2/x 1(1 + l)h2 u"{r) + EV(r) u{r) = 0 2 2/ir (15. A > 0.£ ] * ( r ) = 0. For a central potential V(r) we write (15. Power Potentials general power potentials.1.r acting between two particles is constant and directed towards the origin (of the relative coordinate). and the normalization [ dr\y(r)\2 = 1. ~ (I + l)rl (15. (15. </>)R(r).u ~ r i + 1 (/ + l)R(r). we consider immediately the more general power potential V(r) = \r\ 'We have u(r) = rR(r). In order to have a clear starting point. This force maintains this value irrespective of how far apart the particles are separated.m2. R(r) = ^ . 15. m\m<i M= : . (15.308 CHAPTER 15.6) = (I + l)u(r)/r = R ~ rl.V 7 = const. i. where \x is the reduced mass of the two particles of masses mi. h2 2n V 2 * ( r ) + [V(r) . (15. v > 1.5) u'(0) u[r) r=0 = R(0). f dr{u(r)}'2 1.1) i.2 The force The Power Potential F = .e. and so u'(r) .4) Instead of considering the linear potential.e.3) with the boundary conditions''' u(0) = 0. mi + m2 and r is the relative coordinate.4>)ru{r) = Ylm(9. These particles therefore cannot be separated by any finite amount of energy.
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.8a) ^ JO = (2Ar + l ) £ .2 The Power Potential 309 In Chapter 14 we obtained the BohrSommerfeldWilson quantization condition (14. We also assume that V'(x) > 0.2. or f l H " ^ ) ! .9) into the threedimensional case in the following form: / Jo C dry/2n(E . . (15. (15. Onedimensional case: We consider first briefly a particle of mass TUQ in a symmetric onedimensional potential with the symmetry V(x) = V(x) without loss of generality.7) k(x) = h/y/2m0(EV(x)). odd integer ^=(nll)vh. .63b) for the onedimensional Schrodinger equation. d5. (14. (15..63b): 2 and V(0) = 0 (15...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!). is to be interpreted as the principal quantum number in three dimensions. N = 0. We now want to find a quantization condition for the threedimensional problem.2.10) Here rc is the turning point and n — 1. so that with Eq.15.4) in the threedimensional case.V(r)) = (2nl) +1 N. This means that we transcribe Eq.8a) we can write the quantization condition (14. We therefore proceed as follows."MA. 3 . (15. which has been reduced to an effective onedimensional problem in the polar coordinate r. Then k(x) = k(x). x > 0. and the turning points are at a — — b := — xc. In order to to obtain the .1.8b) This condition implies that we have only one pair of turning points. (15. (15. .63a).
c (nj\irh.6) we have to evaluate the > integral / : = [ ° dr^2n{E\r»)= Here the turning point rc is given by E .13) we obtain _ X_ „ _ so that A£_ (15. (1516) (15. _ —vr" *dr. dr = .y) T(x)T(y) _ T(x + y) ("!)/" /•! / d"(1"")/. .'(1*)1/2= J1F\lt)y1dt.17) The integral on the right is the integral representation of the beta function B(x.310 CHAPTER 15. (15. (15. E A fv.11) rc = (15.y) defined by (xl)\(yl)\ y ^ (x + y . 1W or Xv \E I It follows that (I) / ^""^(l*)1/2.12) and (15. Power Potentials eigenvalues E —• En for the radial potential (15.15) I ' = ^ 1 (( 1) B{x.\rvc = 0 . Now. (15. s.12) i/z Jo We set dr 1 l A v ~ E \ (Et\l/v so that (15.13) dt .14) Prom Eqs.
55). for instance simply graphically as indicated in Fig. . $T(z) = {z.1. and n — — ]7r/i. 15.2 The Power Potential 311 so t h a t by comparison y = 3/2 and x = 1 + (1 — u)/i/.* In this limit the eigenvalues become:^ E„ (nh^ni) Lr(f)r(i)^7^ rv"ir •n2 2/i 2^2 (15.19) or En = _ 1)^(3 + 1)^/^2^+2) r()ir(i)^7^ • v 0 1 Fig. t h a t in the limit v —• oo the potential approaches the shape of an infinitely high square well.18) \v \E or 3 E2 r(§)r(i)V2M (n ' (15. (12. 15. ^ '"Compare with Eq.15. One can convince oneself now.1 Approach to square well with v —> oo. T(z + 1) = z\.1)!.20) These are precisely the eigenvalues which one obtains from the vanishing of the (periodic) eigenfunction at the wall of the square well (of course n — 1/4 has t o b e replaced by n since the W K B approximation is only accidentally correct for small values of n ) . The case v = 4 is that of the pure anharmonic oscillator also discussed by M. i r ( i ) = r ( l + i ) = ( i ) ! . Weinstein [281]. (15.
since these are the ones we are interested in (cf.2. 15.25) "Comparison with Eq.1 + \)hx> = {An • l)£fiw. the case of the linear potential. But for our present purposes this applies only in as far as the solutions (j)(x) which are exponentially decreasing at infinity vanish at x = 0.41) for the . (13.//L This result agrees with that for the onedimensional harmonic oscillator r2 in the form" cf>"(r) + ^(E\r2)cf>(r) =0 (15. the harmonic oscillator.g — —2. The potential V(x) = X\x\ (15. (6.oj2/2.312 CHAPTER 15. we obtain En = (n . > and with Eq. For the following we set in Eq. we obtain E„.oj = ^/2X/JX. x > 0. Power Potentials In the case of v = 2. eigenvalues (JV + h)tkj — (2n . (6.23) The eigenvalues for the linear potential can also be obtained from the zeros of the Airy function.> r ( 2 ) A V 2 in \)*\W \it^T& (15. so that the appropriate Schrodinger equation becomes d2(j) + (Edx2 x)<t>(x) = 0 .mo = \. discussion at the beginning of this section).2) implies A s fj. Physically it does not make sense for the probability amplitude to have a discontinuity there. = (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.24) is discontinuous at x = 0 (this means the derivative there jumps from positive to negative).21) (4nl). 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). In the onedimensional consideration these are precisely the odd wave functions as illustrated by an example in Fig. In the case v = 1.61) h2 = I. (15.
\(s?» (15. W i t h z :— E — x this is d24>{z) + z<f>{z) = 0.27) i. Eq.2 Behaviour of an odd wave function at the origin. and for s —> oo the trigonometric behaviour of Eq.56) and (14.e. we must have as a result of the boundary condition <p(x) = 0: 4>(x)\x=0 oc Ai(E . (15. i.x)\x=0 = 0. Ai(s) 1 2v^(s)V4 exp x —> +oo. this wave function has the required exponentially decreasing behaviour at infinity. i. s = E. dz2 According to Eqs. (13. 15. i.e.x).60) applies.2 The Power Potential 313 V(x) V(x)=x turning points exponential fallofl Fig.59) applies.57a) one solution of this equation is the Airy function <l>(z) on Ai(z) = Ai(E .26) For z = s := E — x —> —oo.29) . (13.15. i. (15. E > V or s > 0.e. (14. In the domain —a < x < + a .e.28) In particular at x — 0. Ai(s) "v^ G ' cos s 3/2 _ t (15.e.
29) implies that cos (  ^ . [231].e.82878 1 2 3 4 5 6 7 8 9 10 2. so that n — 1/4 ~ n. Quantum Mechanics with Applications to Quarkonium.51716 6.32025 4.02137 10.33811 4.78445 7.03914 11. n = l. Rosner. i..30) ! ^ / 2 .28) is really only valid for s large and s — oo.7 T Tl * 2 V 4 Actually this expression is only valid for E and hence n large. of course. (15.2.1 demonstrates the quality of the WKB approximation by comparison with the exact values. 201. (15.93528 12.93602 12.94249 9. Co.04017 11.00767 11. The expression (15.^ = !(2nl).08795 5. Table 15.94413 9..314 CHAPTER 15. (15. with permission from Elsevier.. L. to obtain the zeros of the Airy function numerically. but for large > values of E Eq.e. Quigg and J. the eigenvalues En are determined by the zeros of the Airy function.3. *This Table is reprinted from C.78671 7. (2/3)E3/2 or 3 / 1 2/3 E — En — . . Table 2.00852 11.= ) = «. i.82814 i. = im — (7r/4). p.1: SWave Energy Eigenvalues for V(r) — r (with h = 2/x = 1) from f 3 7 r ^n 2 ( i>j2/3 .02265 10.En from Ai(£„) = 0 2..e. It is interesting to note that in the dominant approximation both methods agree.08181 5. It is possible.52056 6. Power Potentials Table 15. copyright of NorthHolland Publ.31) (an odd function has an odd number of zeros!).
In order to be comparable with experimental data these investigations. the charmanticharm meson ^ . We shall encounter WKB normalization constants at numerous points in later chapters. e. Weisskopf [239]. (14.e. See also J.3 The ThreeDimensional Wave Function 315 15. ee. P. can be shown to be proportional to the modulussquared of the particle wave function at the origin. dr[u{r)f " f** «" (i £>i)4 Here we replace the oscillatory part cos (.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. had to be supplemented by inclusion of relativistic corrections as well as other contributions arising for instance from spin and angular momentum interactions. van Royen and V. We indicate briefly here — without entering into further details — how these quantities can be calculated.e. Jackson [140]. The resulting masses agree in most cases very well with those extracted from experimental observations. which are inversely proportional to the lifetimes.g. have only a finite lifetime and decay into other particles such as electronpositron pairs. *Note that this makes N a W K B normalization constant. e. In general these quarkantiquark pairs. .e. i.** r ( # ^ e e ) o c *(0) 2 .• •) by a mean value. in Chapters 24 and 26. i. of course. We saw previously (see for instance Eq.15. F. cos2()^— / 1 fr° 2 > w .3 The ThreeDimensional Wave Function The linear potential has in particular been used in the investigation of the spectrum of heavy quarkantiquark pairs.34a) **The most frequently quoted reference for this result is R.g.* We can determine the constant N in the leading approximation by demanding that = 2 / Jo i. The decay widths T. r <53) 1 3  dr cos 2 () (15. D.
35) N2 .40) VWY . (15. (15.\ \ * n = l.Jo dr' n M?) ~ 4 in sin sin JTCldrW2v{EV)j l \ 7T N 7W)sin = ( .32) we obtain to leading order u'{r) and therefore «'(0) = N N (1JU0) N (15.35) u T (•T=2TT/OJ dt cos 2 cot 2' (15.37) (15. & (15.38) N : sin V ^ Jr dr' 7T (15.l N n .2.e.1 )n . From Eq.316 In accordance with t h e relation CHAPTER 15. dE a—— / dn J0 so t h a t with Eq.10) with respect to n implies (which is permissible for small separations of neighbouring levels) 2/i ldEn rrc dr ^/2ii{En V) 2 dn J0 = irk.36) \i dEn irh2 dn According to Eqs. i. (15..rcu w X{r)dr = Kh .. (15.2) and (15..39) in / .4) we have for I = m = 0: ^(O) 2 = n'(O) 2 y O o(^0) 2 = ^ K ( O )  2 . A = (15. Power Potentials this averaging implies a numerical factor like 1/2.34b) VMEvy Variation or differentiation of Eq. so t h a t approximately l N2 / Jo k(r)dr . (15.
Thus only theoretical investigations performed immediately after the discovery of the particles \I/ and T are basically of an analytical nature* and therefore have not been performed largely with numerical methods and the fitting of as few parameters as possible. Bose and H. for instance. Hence Eq. W.37) we obtain: 1 (2M^)1/2 47r ft. J. and hence with Eq. W. E. . The case of arbitrary positive power potentials as above has been considered by R.43) or the results of corresponding calculations one finds the following dependence of (^(O)!2 on the quantum numbers n = 1. S. the expression (15. K.41) I*(0)2=_L^)^. § S. 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. 3 . MiillerKirsten [28]. .g.38) implies 317 (15.23) for the energy levels of the linear potential. (15. K.* Using the result (15. J.45) 'See the references of Quigg and Rosner cited at the beginning of this chapter. MiillerKirsten [147]. we have ^'(0) = ( . (b) for the logarithmic potential^ l*(0) 2 ex ^ v (15. W. Bose.44) (15. Hite and H. S. 'For one such investigation which includes the Coulomb potential in addition to a confining potential see e. J. 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). .3 The ThreeDimensional Wave Function Since we are considering potentials with V(0) = 0. for various potentials: (a) For a confinement potential of power v. MiillerKirsten [29]. (IM2) l*(o)r = M dEn dn irh 2 ~^WEn ~*T (15 43) ' Inserting here.1 ^ ( 2 ^ „ ) 1 / 4 . Itt^pocn2*"1^"^.l ) n . . (15.15.2. G. Kaushal and H.
g ^£+(<*Pl»riy = 0. Solution: Details can be found in the literature.1. ip = exp[(z — c)/2].L2 h4 = 4/3exp[(2a — /3)/f3]. 7 With the substitutions r = exp(z — c). i. K.5... . MiillerKirsten [29]. (11.133).I l n f ^ p . J. *=^f.j . in agreement with Eq.318 (c) and for the Coulomb potential^ CHAPTER 15. Power Potentials *(0) 2 oc . Example 15.^ +•••.c = —a//3. (15. rewrite the equation as ^ z + [ . = 1(1 + 1)." 1 11 L . D.— oo < z < oo. and there Eq.L 2 + U{z)]<f>.46) and the classical Kepler period of Eq. E' = E + gln(r0).. 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 . .e. Bose and H.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. We observed there in the case of the Coulomb potential that the period is proportional to n 3 . Eq = i q . Lifshitz [157]. M. (14.1: Regge trajectories of the logarithmic potential Consider the radial Schrodinger equation for the potential V(r) = g\n(r/ro).. = (I + 1/2) 2 .2. Landau and E. . (15. S. U(z) =  dz zf3e2a/?e2*. W.^ ( 3 ^ + l)55p^«(3^l) + . n = 0.75). q = 2n + l. {S = ^ > 0.
and in fact. This implies. It is clear that for ro — oo the > potential becomes of Coulomb type.Chapter 16 Screened Coulomb Potentials 16. that no matter how far the particle is scattered away from the source of the Coulomb potential. a charge as source of a Coulomb force leads to an effective polarization of the vacuum like that of a dielectric. in microscopic and hence quantum physics with the possibility of real and virtual creation of particles and hence a vacuum state. it is. This effect sounds unphysical. it will always notice the latter's presence. A very similar potential is the Yukawa potential V{r) = g2 — . Historically this potential arose with the realization 319 pVr . 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.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. 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.Phenomenologically the screened attractive Coulomb potential may be written V(r) = g2pr/ro . r where r is the distance from the source charge and the charges and other constants are collected in the coupling g2. In reality. The result is a screening of the Coulomb potential which thus attributes it a finite range ro.
x j 36 x 4 (2TT) k + p? 4 2 (2TT) J d x e^xl 4TTX ' We observe that the Fourier transform of the propagator is effectively the Yukawa potential. Screened Coulomb Potentials that the strong nuclear force is mediated by the exchange of mesons. Since. and that. i. in fact the parameter [i represents the mass of such a spinless meson. (16.1) We encountered Green's functions earlier.e.1) and is written in spacetimedimensional Minkowskian notation l/{jpvpu + /i 2 ).320 CHAPTER 16. the mathematical expression with an exponential is not so easy to handle analytically.3) where for the real potentials we consider here. (16. 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. one frequently resorts in calculations to the expansion of the potential in rising powers of r. from the relation Pi + P 2 + M2c2 = 0. all coefficients Mj are real and independent of the energy E = k2.* The relativistic equation of motion of such a meson when free is the KleinGordon equation which results from the quantization of the classical relativistic energy momentum relation of a particle of mass /x. .ii k .2 ) or (p = hk) 1 /" . Yukawa potentials and superpositions of such potentials play an important role in nuclear physics. Heisenberg [133]. In the literature reference is sometimes made to the socalled static limit of a relativistic propagator. A Green's function is effectively the inverse of a quantity called propagator which is intimately related to the expression in Eq. The realization that mesons and baryons are made up of quarks does not really change that picture at lower energies. who later — as is wellknown — deviated from this idea and invested his efforts into the study of nonlinear spinor field theory. 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 . Regge trajectories — which we encounter *The Smatrix theory of strong interactions was actually initiated by W. however. and in view of their relation to the exchange of virtual elementary particles have therefore been objects of intense study in elementary particle theory.
. Mandelstam [185]. A. Masson [180].§ 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.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. (16.3) have been considered by various authors. f H.r)=0. "These were first considered by C. Wu [47]. 2 . J. £+**PVV t/. however. Omnes and M. l+n+l = ^P. . i. Regge trajectories were realized to play an important role in the high energy behaviour of hadronic scattering amplitudes. Froissart [223].^ Regge trajectories arise as poles of the Smatrix in the plane of complex angular momentum. to be of the order of 1/k. Kinoshita [21]. Cheng and T. See also the other references below. Lovelace and D. C.Q. (16. 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. Potentials expandable as in Eq.1. 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.r). T.k. physical) values of angular momentum as functions of energy E. 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. . Thus in the present case it turns out that it is easiest to calculate first the expansion for the expression l + n+1. Squires [258]. Naturally it is easiest to familiarize oneself with these by studying solvable potential models. (16.k. i. Bethe and T.mo being the reduced mass of the system. only for the cases of n = 0.1 Introductory Remarks 321 in this context — are functions which interpolate integral (i. Frautschi [97] and E.e. $ In the following we consider screened Coulomb or Yukawa potentials of the type of Eq. i. S S.3) in the radial Schrodinger equation for the partial wave ip(l.1 in the above. S.4) where E = k2 and h = c = 1 = 2m.{l.16.e. for the Regge trajectories or Regge or /plane poles of the 5matrixll / = ln(K) 'Standard references are the monographs of R. . if>oo. bound state eigenenergies and the 5matrix.e. They are usually written I = an(E). (16. where n is an integer and we referred to these already in Chapter 11 in the simple case of the Coulomb potential.e.
Regge [31] and E. Squires [258]. W.9) the cut starts at E = k2 = 0.t Considering such superpositions of Yukawa potentials with /•oo / a(/j.** The energy is later obtained by reversion of the resulting series. for all n. W. M.6) 2K{M°~ An(K))x+ 2K^(^K) MiX ' (16J) **H. has only simple poles) in the plane of complex angular momentum and in the complex &plane (E = k2) cut along the imaginary axis. 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. < oo for all 0 < vr/2.19). o rV(r) I dpp\V{peie)\ regular at r = 0. one can assume an expansion of the potential V(r) in ascending powers of r. Proceeding now as in the case of the Coulomb potential we change the variable of Eq. Section 11. Miiller and K. W. tf H .z) Then x is a solution of the equation VaX= (16.4) to z = —IKv and set TP(l. 16. Schilcher [203]. (16. Muller [204]. J.2 Regge Trajectories Since our treatment here aims at obtaining an S'matrix as a generalization of that of the Coulomb potential. Longoni and T. *See in particular A. H. Screened Coulomb Potentials as a function of the energy. Bottino. .z) = ez2/2zl+1X(l. Muller [201] and [202]. in the present case at fc2 = 0 or k2 — Mi = 0.ndp.15) and (16. 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.)/j. (16. Under these conditions the S'matrix is meromorphic (i. A.322 CHAPTER 16.k. 'In the case of the Coulomb potential (cf. J.k. J.e. In the following we follow mainly the last two of these references. as may be seen from Eqs. The conditions for this have been investigated in the literature* and may be summarized as follows: /•oo V(r) = / d/xa(//)e^7r. < const. starting with the power r _ 1 of the Coulomb potential.
* Recurrence relations for coefficients of perturbation expansions for Yukawa.b. By a repeated application of the recurrence relation (16. . j)Smi(a. W.awhere (a. (a. which means that the hypergeometric series breaks off after a finite number of terms.b.a + + l). {0) X = Ha. (16.a+ l)$(a + 1) + (a. and all Sm(a.j)$(a + j). MiillerKirsten [249].16.z) = Ha). Sharma and H.j) for \j\ > m are zero.j) may be computed from a recurrence relation which follows from the coefficients (16.j) (16. (16. Eq.a + j)Smi(a.1).a + j)Smi(a.9) we obtain m l)$(a .2 Regge Trajectories where d2 Va = z^ and (cf. all other So(a.j +(a + j + l.j l) + (a + j. The function <&(a. b\ z) is seen to be a confluent hypergeometric function which for reasons of convenience we abreviate in the following as 3>(a).j) = (a + j l.L ^d {bz)a l+ l + ^ ^ l 2K =  n &ndb = 2l + 2 = 2n^4^ K (168) The right hand side of Eq.9) (16. a + 1) = a — b+ 1. (a.i 0) = 0. J. We also observe that the ansatz (16.0) = 1.10): Sm(a. (16. a) —b — 2a.z) is known to satisfy a recurrence relation which we write here for convenience in the form z$(a) = {a.5) is equivalent to a = —n.a)${a) + {a.10) zm$(a)= ^ j=m Sm(a. where $(o.11b) ^ The associated boundary conditions are: So(a.a — l) — a — l. anharmonic and cosine potentials have been derived in L.5)) a = + 323 . so that to leading order we have VaX{0) = 0. as in the case of the Coulomb problem.7) is seen to be of order 1/K. (16. K.11a) The coefficients Sm(a.
This follows from the fact that D a $(a) = 0.a\4. and hence Va$(a + n) = ra$(a + n). 1 . so that 1 1 2K— ' (2W[a'a]2+(2^F[a'a]3 1 r i . Evaluating the first few terms of the expansion (16.12b) The usefulness of this notation can now be seen in the ease with which it permits the calculation of any number of higherorder perturbation terms. This equation is seen to be 0 = 7n7L a ' a Jl + P a + n $ ( a + n) = 0. .5) the Regge trajectories I = ln(K).324 CHAPTER 16. oo _ . 0 < \j\ < i. i. #(a + ra) fjL$(a + n) V.o]i = M 0 . [ a .aj 2 (2K)A _ [a.a + j]i+l = MiSi(a. Any term /j. i R »] = 2K[a.7).aj 2 \ [a.[a>a + J]i+i*(a + 3). Mn The coefficient of the sum of all the remaining terms in 3? (a) is then set equal to zero and determines to that order of approximation the quantity An(K). Va+n = Va . (16. (16.7) and hence in Ra may be cancelled out by adding to x a contribution /x<I>(a + n)/n except. [a.13) one obtains the quantity An(K) and hence with Eq. of course. (16.l. a .[ a + l. Screened Coulomb Potentials Substituting the first approximation x^0* = 3>(a) into the right hand side of Eq. the latter can be written . (16.1] 2 .a + l ] 2 [ .n. [a. We first cite the final result . when n = 0. for the coefficients M^> of the expansion Details can be found in the references cited above. + (16.13) + ••• • One can now construct coefficients with their recurrence relations for the individual terms of this expansion. (16.e.j).An{K).12a) where [a.a]l^a) + ^2(2K)i+1 Y.<&(a + n) on the right hand side of Eq.' — .
( 2 1 ^ 6 1 ) [ 3 M 4 M 0 ( n 2 + n . is an asymptotic expansion for large values of the energy — is not very useful in the very interesting domain around energy zero. it is clear that one expects the trajectories to be finitely closed curves with asymptotes given by those of the Coulomb potential.1) + 3M3M02 + Mln{n + 1) +4M 2 MiM 0 ] + 0(K~8).1) +6M 2 Mm(n + 1) + 2M2M02 + SM^Mo] . However. Obviously the expansion — which. including an exploration of the domain of small energies E.1 by evaluating the first two terms. G.2. P.^ The plots confirm the expected behaviour for strongly attractive potentials but exhibit also a superficially unexpected departure into the lower half of the complex iplane in the case of the first few trajectories (counting in terms of the quantum number n) for weak coupling. With numerical methods one can achieve more. 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.l)n(n + l)(n + 2) + 2M 3 M 0 (3n z + 3n .§ This is an important observation which indicates the equivalence of the methods.^ p [ 3 M 4 ( n .2 Regge Trajectories 325 and then demonstrate the calculation in Example 16. . Inserting this expansion into Eq. Burke and C. Thus the expected appearance is that shown in Fig. Analytical Observe there the quadratic form of the centrifugal potential! "See in particular A. Ahmadzadeh.15) The same expansion may be derived by the WKB method from the BohrSommerfeldWilson integral for three space dimensions as explained in Example 16.e.l)n(n + l)(n + 2) + 2M 3 M 0 (3n 2 + 3n . i. (16.1) +6M 2 Min(n + 1) + 2M2M02 + SM^Mo] + g^s [3M 4 M 0 (n 2 + n . we obtain the expansion for the Regge poles. 16.1) + 3M3M02 + M22n(n + 1) (16. Tate [6]. The result is AnW = M0'[n(n + l)M2 + MlM0] <2» + ^ ^ 1 + — ^ [ 3 M 4 ( n .16. (16.14) +4M 2 MiM 0 ] + 0 ( K ^ 7 ) . of course.5). .1.
Frautschi [97].12b) we obtain [a. 114.^ .1 Typical Regge trajectory for a strongly attractive Yukawa potential.0) = (a. Similarly the conjugate variable to angular momentum is angle. .a) = b . (a.—" + In = — . Hence Si (o.11b) we obtain Si(a. and the lifetime At of the resonance satisfies the relation TAt ~ h. a] 3 = M 2 S 2 (a. a] 2 = M i Si (a. However. (16. ±1) = 0. Screened Coulomb Potentials expressions of the behaviour of a Regge trajectory in the immediate neighbourhood of E = 0 are practically unknown. 16. In fact. Example 16. . and the angle A#. through which the particle orbits during the course of the resonance. satisfies the relation ajAO ~ h. 0). 16. The conjugate variable to energy is time. just as a decay width T represents the width of the resonance in energy.a)So(a. p.1).13).O) = Other terms vanish since So (a. this is a particularly interesting domain since at the position soon after this point at integral I (as at I — 2 in Fig. where the imaginary part of I is very small. For a resonance with a long lifetime. C. From Eq. ai is small and A6 is large. one can expect a resonance with the lifetime determined by this imaginary part. 0) and Prom Eq.1: Evaluation of perturbation terms Use the above formulae to verify the first two terms in the expansion of A n for the Yukawa Regge trajectories. [a.2 a = In See S. Solution: We evaluate the first three terms of expansion (16. (16. so the imaginary part ai of I — an represents its width in angular momentum.0). For a bound state a j = 0 and the orbit becomes permanent." Re I 1 0 1 2 Fig.326 CHAPTER 16.
0) Si(a.a + l)S0(a.2 V~Az and z \ zz dr\ A 2B C dz V'A + 2Bz + Cz2 (6.1) + (a .2 Regge Trajectories 327 Analogously we evaluate S2(a.6 + 1 ) . (16.0) = (a . We obtain the quantities S i ( a . A" An An A" = = = (a.0)+ 0 + 0 = (a1). a)Si (a.2 a ) .^ 2 M " + ! ) M 2 + M i Mo] + • E x a m p l e 1 6 .(ra + 1) [n + (2K)3 A n+1+• n K K (M0 . (a.6 + l)a + A. using the Cauchy formula (6. One sets z = 1/r so that.0) + 0 + 0 = ( 6 .A „ ) H Mi 2An 2K if n + (2i^): r M 2 2K Hence 1 . 2 : E i g e n v a l u e a p p r o x i m a t i o n b y " p o l e s at infinity" Verify the dominant behaviour of the Regge trajectories of Yukawa potentials by contour integration of the BohrSommerfeldWilson integral in which 1(1 + 1) is replaced by (I + 1/2) 2 . 1) (a . .n ( n + l Inserting these expressions into expansion (16. Sec. 0) + (a + 1. 14.13) we obtain 0 (Mo .6)Si(a.0) + 0 + 0 = ( a . . .00 B + Cz 2ni VA + 2Bz + Cz2 B 2niI —= + vC with ambiguous signs of square roots I. 1). i ) again from Eq.16. a)Si (a. Thus (with terms which are 0): Si(a.4) with mo = 1/2 and around the classical orbit the integral in the following relation I h = <b ^J pdr: dr •K2 Mo {l + \)2h2 1/2 ^classical orbit The integral is most easily evaluated by the method of "poles at infinity". 0) + a S i ( a . a)Si (a.1 ) + (a.al)S0(a.65) 2m dz V'A + 2Bz + Cz2 • Z— 0.65) and observing that in z there is one pole at the origin and one in approaching infinity in view of the expansions VA + 2Bz + Cz2 _ VA y2 . 1) Hence 52(a.b)(a .1) Si(a.0) (a .1 ) + (6 .An) • [ M i M 0 + n(7i + l)M 2 ] 1 A n = M 0 .2o)Si(a.11b). (n+l)U+^pJ . Solution: Ignoring higher order terms in r. we have to evaluate (cf.1. (a.a)5o(a.
e. Thus we can write down the S'matrix for the present case by exploiting the limiting case of the Coulomb potential. (16.(iO mp~ ~ ' ' ' '+ l)M2 + M0Ml] (16. tSee Chapter 11 and V.3 The 5Matrix It is clear that if we now work through the usual procedure for the derivation of the S'matrix we pick up a logarithmic phase as in the case of the Coulomb potential.1) +61(1 + l)M 2 Mi + 3M2M02 + 3Mx2Mo] + 0(K~6).15). I. by expanding the right hand side in rising powers of v and evaluating the individual integrals. The turning points in this case are at r = 0 and r = oo.5) together with the expansion (16.. *J. Example 16. Here we do not perform this procedure. Singh [252]. however.* 16. i.2. n = 0. This shows that 1(1 + 1) has to appear in the BohrSommerfeld—Wilson integral as (/ + 1/2) 2 .l. Solution: For details of the solution we refer to papers of Boukema. In the second paper the second order WKB approximation is used and shown to yield complete agreement with the terms given in Eq.16) = n n=o 1 2 With this inversion we obtain = M0^^[l(l + ^ [ 3 ( Z . Screened Coulomb Potentials Mo . t?{l + \f M0 ±2K 1/2 2TT 2J 2V^2 (n + l + l)h = 2TT Mo h. (16. We know that corresponding to every Coulomb Regge trajectory we have a corresponding one in the Yukawa case. Boukema [34] and [35]. ±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. we can use Eq. 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 .1)1(1 + 1)(Z + 2)M 4 + 2M3M0(3Z2 + 3/ .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.t First.328 Thus here dr K' CHAPTER 16. .
W.l. Miiller [204].l)n(n + l)(n + 2) + 2M2M02 . J.. We observe also that the S'matrix is unitary as a consequence of the result (16. § See H. for instance. . Expanding the square root VfC2 + Mi for \Mi/K2\ < 1 one regains ln(K) with the expansion (16. the behaviour of the scattering phase in the domain of high energies.16. the following *See e.17).g. i.17) and is therefore real for real values of the potential coefficients Mj. such as.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\. Paralleling the case of the Coulomb potential in Chapter 11..4 The Energy Expansion We observe that Ai(K) has the property MK) = MK) 329 (16. 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 [211].0[(KZ + Mx)'6} (16.* 16. This has been done§ and one obtains . which are precisely the expressions yielding the Regge trajectories of above.19) we obtain the energy. H. J.18) of the Smatrix now to explore further aspects. n = 0.19) +2M 3 M 0 (3n 2 + 3 n . W.e.14). and could then have carried out the perturbation procedure not in inverse powers of K but in inverse powers of y/K2 + M\. Now reversing the expansion (16.l (two additional terms are given in the literature).2.. . 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. One can use the explicit expression (16.
W.l)n 2 (n + l) 2 (n + 2) 1 + • • • (16. E. MiiilerKirsten and N. Frautschi [97].1) + Af3Mff + 1) ^ lOM 6 M 0 2 (n .e.3 M 2 V ( n + l) 2 M(o I +2M 3 M 0 2 (3n 2 + 3n . Zauderer [286]. 107. Analogous expansions for specific energydependent Yukawa potentials have been investigated by A. the transformation was introduced in the form given below by G. Watson [280] in 1918 and later resurrected by A. J. p. p.1) + 2M2M.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 . Large coupling expansions derived in H.< 3 M 4 M 0 ( n .* We recall from Chapter 11 the definition of the scattering amplitude F(9. Screened Coulomb Potentials expansion. Sommerfeld [256]. C.11) + 2M22M03(9n2 + 9n .20) Extensive investigations of the energy eigenvalues for the Yukawa potential can be found in the literature.330 CHAPTER 16. 'According to S. B. i.l)n(n + l)(n + 2)(n + 3) +2M 5 M 0 3 {5n(n + l)(3n 2 + 3n . VahediFaridi [212] contain as special case the expansions of the first pair of authors. Iafrate and L.l)n(n + l)(n + 2) . k) as coefficient of the outgoing spherical wave in the asymptotic behaviour of the wave function tp(r).10) + 12} + 4M3M05 +2M 4 M 0 4 (6n 2 + 6n .2)(n .^ 16. in the following expression with an "See in particular G. N. Mendelsohn [135] and E. in which the leading term is the usual expression of the Balmer formula for the Coulomb potential: K2 = Mi+ 1 ° l4n(n + l ) ^ ( / + n+l)2 4(7 + n + l ) 2 Mio +4(2n + l ) ^ ( / + n + l ) 3 4(Z + n + 1)4 ( l) ^ — . . 282.5 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.f 24(2n + l)(Z + n + l ) 5 M4M0(nz M06 M$n(n ^ M9 + n . A. J. Warburton [279].
k) — e 2i5[(k) ~" p—iirl (16.fe)P . For this purpose we consider the following integral taken along the contour C shown in Fig.22) = \F(9..5 The SommerfeldWatson Transform 331 ingoing plane wave in the direction of z (here with h = 1.k) = ^T(2l + !)/(*.2. ..k)\2.iJ 11 —[S(l. Eq. c = 1. \x\ <C l .24) Si(k) being the phase shift.TOO= 1/2): ip(r] r ~*°° Jkz elkz + F{0.25) Setting Note the minus sign in the argument of the Legendre function. (cos 9). z=o in which (cf.26.kmcose). The scattering amplitude possesses the partial wave expansion F(0.1.23) as the contribution of the residue of a pole in the plane of complex I of some suitably constructed contour integral. 16.k) = 2zfc L " v "'' v .2 in the complex Zplane: *<«•*> = s / c c K <u(21 + 1) vf{i. I = n + x.184) to (11. n = 0.187). ~:' sin nl (16.k)~l] = ~ k" PiSiW sin 5i(k) and S(l. dn Iml 1 I c ( \ x o x 1 A 2 A 3 x4 X 5 Rel Fig.k) 0ikr (16.2 The integration contour C. Actually 5j lrr/2 as in Eqs.23) f(l. (11. (11.187)) (16. so that sin KI = sin7r(ra + x) ~ (—l)nKX = (—l)lir(l — n). 16..21) The scattering amplitude determines the experimentally measurable cross section o given by da (16.''' The idea is now to consider each term of the expansion (16..
Consider the reaction of a particle a{p) with fourmomentum p^ colliding with a particle b(q) having fourmomentum q^ and producing a particle c(p') and a particle d(q'). we obtain m® = ± I dl^±^f(l. 16.—) s = (p + q)2. Screened Coulomb Potentials Then integrating with the help of Cauchy's residue theorem.332 CHAPTER 16. 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. be elastic scattering of a off b. d(q') (16. 16. a + c —> b + d.26) P((+COS0) = 2 ( 2 n + l)/(n.3. Such a direct reaction and its "crossed channel reactions" are shown schematically in Fig.3 Reaction channels and their respective Mandelstam variables. in which case c = a and d = b.27) a(p) © b(q) Fig.28) . Thus the direct reaction may. for instance. u={pq')\ (16. t = (qq')2. as indicated in Fig.3. we have to digress a little and introduce a few simple ideas which played an important role in the development of particle physics.k) Zi Jc TT{L .—.n) oo {l)1 Pt{cosO) v ' (16. a + d — • c + b.fc)P„(cos0). 16. For the momenta indicated we set (with metric +. In order to see the relevance of Regge trajectories in a reaction of particles. n=0 which is the usual partial wave expansion.—.
which is an analytic function of the variables s.t.et) (16. the only kinematics we require in the following is given by the relations (e. Goldberger.31) i l Taking all particles to have the same mass mo. u. 16. B. . Khuri and S. (a) Mandelstam's hypothesis: All three processes (described by the s. u = .32) t = 4(q? + ml). A(s. etc. Blankenbecler.5 The SommerfeldWatson Transform 333 and for simplicity we assume here that all particles are spinless and have the same mass mo.u are known as Mandelstam variables. Ft = £ ( 2 J + l)F. A(s.u channels) are determined by one and the same relativistically invariant scattering amplitude A(s. Thus.9s)\2. s = 2qf(lcos0t)(16. In a Regge theory on the other hand.30) with partial wave expansions (observe the noninvariant factor k has been removed) Fs = ^ ( 2 2 + l)F z (g s )PKcos 98). Chew [48].^ (b) Chew's hypothesis: All (composite) particles lie on Regge trajectories. the variables t and u would describe momentum transfers in the crossed channels. (16. T See e.t) = y^Fs(s. if we had been considering the s reaction originally. F.t.c o s 0 s ) .0s).16.t) = VtFt(t.(<fe)^(cos 6t). e. Treiman [25]. if s describes the square of the total energy in the s channel.g. N.t. N.u) (if any of the external particles has nonzero spin. One now makes two hypotheses.t.4. In the Yukawa picture of a reaction the dynamics is described in terms of the exchange of mesons IT (called pions). (16. t = 2q£(i .u are not independent.t). the momentum k there would correspond to qs) s = ~4(ql + ml). Mandelstam [184] and R. S. except for poles and cuts which characterize the reactions in the three channels. § See G.2 ^ ( 1 + cos6 s ). there will be several such amplitudes which together describe all three processes). the quantities carrying these quantum numbers are the Regge trajectories.29) and with selfexplanatory meaning for the cross section of the reactions ~ = \Fs(s. one can write the amplitude as depending only on two. Symbolically we then have the situation shown in Fig.g. one then has A(s. M. L. the dynamics is described by the exchange of quantum numbers. For a Lorentzinvariant normalization of the initial and final states. t.§ Since the three variables s..g.The variables s.
.5 The Lehmann ellipse. So what can one do? 1 H. But for integral values of I.r 1+m 2 /2q 2 o ^t "" Re cost: Fig. However.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(^. 16.4 Yukawa versus Regge theory. Thus one is interested in establishing a representation of the amplitude in terms of ichannel Regge trajectories. But > > wave expansion X)(2Z + l)F. Lehmann [162].334 Yukawa CHAPTER 16. 16. in the domain of schannel physical values of the kinematical of the ichannel). the Legendre polynomial Pj(cos 9) does not possess a cut. 16.e. Screened Coulomb Potentials Regge oc(t) Fig. the partial wave expansion is not valid in the physical region of the (i. For s — oo : cos 9t = 1 + s/2q^ — oo.)e~ •fj.((fc)/Kcos0t) schannel ichannel schannel variables the partial converges only within the socalled Lehmann ellipse^ which is shown in Fig.
all poles of Ql > 0. 16. and there is only a finite number N + 1 in the domain of 9W = .16. k) have to the right potentials one can show t h a t the integral along the curved porA' of the circle at infinity tends to zero. 16.6.e. contour C / / D ^ ^ . to use the SommerfeldWatson transform. Again we have / Jc> • • • = 1ni y^ residues j3n „ For Yukawa tions A and f(l. „ i.26) but for a different choice of the contour.6 The integration contour C". Thus consider now the same integral but taken along the closed contour C shown in Fig.1 / 2 .t). Then / JDD' ••• + j JBB' • • • = 2TTI V ^ residues f5n. the contour representation (16. Moreover. i. B Iml 1 \ i All \ r\ 0 / closed . sin nl [16.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.33) .e.5 The SommerfeldWatson Transform 335 T h e answer is.^ B 1 A Fig. N F(6.
35) This result is valid for s — oo and t negative (cf. J. as we discussed briefly in the above. Potentials with a shortrange repulsion more singular than 1/r 2 have been considered by N. this behaviour has been confirmed in high energy hadronic reactions. the amplitude can be represented as a sum over ^channel Regge poles. at high schannel energies. In particular. The > result demonstrates that the high energy behaviour of an schannel reaction is determined by the leading Regge trajectory in the crossed ^channel. if ao(t) is the Regge pole with largest real part. ^M)^E i 2 a f: a „t ( t ) ^ m <^'F(s. (16. Omnes and M. 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. for high energies in • > the schannel). It follows that under these conditions. K. The slope of the Regge function an(k2) is an " Simple cases are the harmonic potential and the squarewell potential. i. we have P\/2+IR{Z) ~ 0(l/y/z) — 0 for s — oo (i. p.336 CHAPTER 16.32)) and finite. In subnuclear physics they played an important role before the advent of the quark idea and thus of quantum chromodynamics. Regge [229]. MiillerKirsten [29]. i. (IM4) This is the asymptotic Regge expansion of the invariant amplitude for s —>• oo. . 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. The logarithmic potential has been investigated in a way similar to the Yukawa potential above by S. Regge trajectories have. Predazzi and T.e. Limic [177] and E.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.e. W. we have (16. Foissart [223].t)&B(t)sa°V. of course.e. 42. Apart from more refined details. for plots and discussion see R. Bose and H. Eq.
W.g.** Whereas the former are related to absorptive properties of a reaction. For an overview see e. J. The way to do this is similar to calculations in Chapters 18 and 20. and also as socalled "fixed poles" . a composite particle being one with structure. Regge poles are not the only possible singularities of a scattering amplitude in the plane of complex angular momentum. . The small imaginary parts of eigenvalues or lifetimes of resonances have not yet been calculated. Singularities appear also in the form of cuts in the plane of complex angular momentum. 20.6 Concluding Remarks 337 important parameter in string theory. called "Regge cuts". like a meson or a baryon. and is composed of quarks. MiillerKirsten [209]. H.16.1. Related aspects are discussed in Sec. The above treatment of screened Coulomb potentials is incomplete. the latter are related to the distinction between "' elementary and "composite particles".
This page is intentionally left blank .
separation of Laplace's equation Ai/> = 0 in ellipsoidal coordinates leads to the Lame equation (no sn 4 z term). are not much harder to handle than trigonometric functions. 25. See F. The reason for these difficulties is. is effectively the Mathieu equation whose solution has for a long time been considered as being very difficult. called elliptic modulus. K being the complete elliptic integral of the first kind. The Schrodinger equation with this potential. p. Apart from the discontinuous KronigPenney potential consisting of a periodic repetition of rectangular barriers. that the Mathieu equation lies outside the scope of equations which can be reduced to hyper geometric type.Chapter 17 Periodic Potentials 17. however.or 27rperiodic trigonometric functions on the one hand and the nonperiodic hyperbolic functions on the other. the ellipsoidal wave equation reduces to the trigonometric form of the spheroidal wave equation. the most immediate candidate is a trigonometric form like that of the the cosine function.lP = 0 in ellipsoidal coordinates leads to the ellipsoidal wave equation with sn 2 z and sn 4 z terms.0 < k2 < 1.g.) which are functions that interpolate between the n. 2K or AK. example 14.* These elliptic equations involve Jacobian elliptic functions like sn(a. M. In essence these periodic Jacobian elliptic functions. 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. Thus Jacobian elliptic functions also pro*To be precise: Separation of the wave equation Aip\x2. and are themselves periodic with period e.1 Introductory Remarks Prom the beginning of applications of quantum mechanics periodic potentials of various types were immediately considered. since many analogous relations hold. 339 . depending on the parameter k. 19 and p. Taking in the ellipsoidal wave equation the limit k —> 1 (fc elliptic modulus) and putting tanhz = sin 6. Arscott [11]. motivated mainly by the regularities of the crystal structure of matter.
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. 8. cosine and tangent.1. W. (17. each with its own characteristic eigenvalue.1) where h2 is a parameter and — IT < z < IT. which is conveniently written ^ + [\2K2sn2z]y = 0. Here the Jacobian elliptic function sn z is a periodic function analogous to a sine function and has real period 2K} For n = 0. (17. This is a significant difference compared with the Mathieu equation. need not be afraid of them.e. k2 — 0 which means that in > > this limit the Lame polynomials degenerate into the periodic Mathieu functions. Both of these domains are important for a host of other considerations. 17. the Lame equation possesses 2 n + l polynomial solutions. Periodic Potentials vide periodic potentials like trigonometric functions. Dingle and H. and to relate these to the weak coupling bands or regions of stability. The finite heights of the periodic functions of the potentials permit tunneling from one well to another and thereby produce a splitting of the asymptotically degenerate harmonic oscillator levels.. c n z and d n z are handled with analogous formulas. For large coupling.7. B. those of integral order. these periodic potentials may be approximated by series of degenerate harmonic oscillator potentials. J. W. Muller [73]. or Sw&ve Schrodinger equation with elliptic potential. or 5wave Schrodinger equation with the cosine potential is conveniently written d2v —  + [A .2 . . We might mention already here the much less familiar but more general Lame equation. See also the discussion in Sec. J. In fact. such as sn 2 z + cn 2 z = 1 and double and half argument formulas etc. The Mathieu equation.340 CHAPTER 17.4. Miiller. 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.. Like the trigonometric functions sine. which one looks up in books like that of MilneThomson [196] or Tables when required. 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. Lame equation: H. which does not possess any polynomial solutions.2) where K2 = n(n + l)k2 and n real and > —1/2 and 0 < z < 2K.2h2 cos 2z]y = 0. i. The first of these conditions is already tMathieu equation: R. the three Jacobian elliptic functions s n z . [205].1. ''The reader who encounters these Jacobian elliptic functions here for the first time.
In classical mechanics stability is treated for instance in connection with planetary motion. i. the case of h2 small. i. 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. these concepts are rarely explained as such there. a bounded function. Correspondingly quantum mechanics — which requires the electrons of an atom to move around the nucleus — explains the stability of atoms. since the solution there is of the form cos vAz.2 Cosine Potential: Weak Coupling Solutions 341 seen to rule out polynomials.2. Vachaspati [242]. Salem and T. The orbits of planets are stable in the sense that deviations from the recurring elliptic orbits are small. we see that in a plot of h2 versus A.e. Although the mathematics literature on the subject uses the physical concepts of stability and instability. Thus a bounded trigonometric solution is indicative of stability and an unbounded exponential solution of instability.17. . Looking at the Mathieu equation for h? —> 0. The instability of an orbit would be evident either from an unbounded spiralling away to infinity or from a collapse into the centre. the semiaxis A > 0. y" + \y = 0(h2).^ 17. Adding the negative term — 2h2 cos 2z to A. we see that for sufficiently large values of h2 the periodicity or boundedness of the solution is destroyed and hence becomes one of instability.1.1 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.2 Cosine Potential: Weak Coupling Solutions We consider first the weak coupling case. 17. i. Our considerations below are deliberately made simple and detailed because their treatment in purely mathematical texts requires more time to become accustomed to.e.e.e. Here an important aspect is the determination of the domains of stability of the solutions and the boundaries of these domains. as in Fig. 17. h2 = 0 belongs to the domain of stability. 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). and one can imagine that a nontrivial parity factor exp(in7r) then turns into a complicated phase factor exp(iz/7r). i. we compare the For interesting related discussions see M.
e.2h2 cos 2z]y(z) = 0. i. A solution of the second order differential equation is determined completely only with specification of boundary conditions which determine the two integration constants. b: y(z) = A cos vAz + B sin \f\z and y+(z) = a cos yXz. Periodic Potentials equations ^ p + [A . (with Tn as translation operator) Tlxy{z) = y(z + TT) = cry{z). we obtain y(vr) y(2?r) a = — — a = —^^r.e.2/i2 cos 2(2 + vr)]y(z + vr) = 0. a = const. (17. y(0) Now suppose we write the equation for h2 small y" + Xy = Then solutions are y(z) oc e ± ^ [ l + G(h2)] or C S 0(h2). y+{z) .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.B and a. sin V Az or linear combinations.4) with (for convenience in connection with later equations) y'+{z) y'_(z) = avAsin\/Az = = bv\cos^f\z = —y(z). y(2vr) 2 a = —7—^ = 1. ° ^ [1 + 0(h2)}. 2/(0)' yfr) . (17. Setting z = 0. d2y(z + 7r + [A . y~(z) = 6sin vXz. We set (here and in the following frequently apart from contributions of 0{h2)) with constants A. — • . i.342 CHAPTER 17. d(z + TT) cos 2z We see that there are solutions which are proportional. TT.
This means.y_] = y+(0)y'_(0) — y_(0)y+(0) = ab\f\.aa][y'_{n) .3) we obtain y(z + ir) = a[ay+(z) + @y(z)].a2b2\ a± ~ 2a6\/A . the determinant of the matrix must vanish.y+(7r)y_(7r) = 0 or abVXa2 . 7 y\z) = ay'+(z) + (3y'_{z). y'_(Q) = bV\ with Wronskian W[y+.e. This expression is equal to the coefficient of a2 in the quadratic equation for a. and by'+(?r) = —ay/Xy{n). The roots a± are therefore obtained as 2bV\y+(ir) ± J4b2\yl(Tr)4. We can write therefore. we obtain V (TT) = ay+ (TT) + (3y _ (TT) = < [aa]. whereas the solutions y+ and y_ have here been chosen specifically as even and odd around z = 0 respectively. y{z) = aty+(z) + 0y(z). yV(0) = 0. y'(z + n) = a[ay'+(z) + py'_(z)}.{ay'_{ir) + bV\y+(ir)}a Now from Eq.y{ir)y'+{K) = abVX in agreement with its value at z = 0 above. \y+>y\ = y+{ir)y'{ir) . ay'~(ft) — bVXy+(n) and the Wronskian w + {y+(7r)y/_(7r) . (17. 2/_(0) = 0. Using Eq.oWX] . (17. with constants a and /3. 2/'(7r) = a ^ W + ^ .y+(7r)j/_(7r)} = 0. i.17.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.4). [y+(v) .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.
Eq. Setting cr+ = e . Prom this condition we determine later (cf.e.= — .6) We can derive various terms of this expansion perturbatively with the method of Sec. (17. 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. i. _ / ± v 7 ^ V .344 Setting / = 2by/\y+(Tr). This is the easiest application of t h a t method since only simple trigonometric expressions are involved. Periodic Potentials g= 2aby/\. where the constants are usually taken as unity from the beginning. <T__ + <T_ = 2y + (vr) . i.+ 2 2(i/ l) 32(i/2l)3(v24) ( 9 ^ + 58. we have for the sum of the roots a+ + <7_ = 2 cos TTV or" C O S T T I ^ ^ ^ C O S V A T T . (17.e. (5i/ 2 + 7)/i 8 vz + — . ** An important consequence of Eq. This calculation is demonstrated in Example 17. Thus this Floquet exponent is determined by the value at z = K of t h e solution which is even around z = 0 and is independent of the normalization constant a of t h e even solution.8) "Observe that for integral values of v (in lowest order of h2). (17. we have <7± — CHAPTER 17. this implies y+(Tr)/a = ± 1 . cr+ = 1/<T_. .2 + 2 9 ) ^ 2 + 64(1/2 _ 1 ) 5 ( z / 2 _ 4 ) ( l / 2 _ 9) + u ^ ^ ^ ..5) a The parameter i/ is known as Floquet exponent.0 This series may be reversed t o yield the Floquet exponent.5) is that for h2 = 0 we have v = A/A and hence more generally v2 = \ + 0{h2).7.1. 8. (17. .455A2 + 1291A . 2 _ hA 2(A1) (13A25)/i8 32(Al)3(A4) 12 ift (45A3 .1169)/i 12 64(Al) (A4) (A9) 5 2 + 0{hw) r 16 (17. T h e result for A is the expansion A = h4 2 . **This point is not immediately clear from mathematics literature.22)) the boundary conditions of periodic solutions. <7__cr_ = 1.
„„ x n. (17. (17.12) S e e J. W. we must have a2 — 1 and hence e 2iv/A7r = 1} i.2 how the expansions for small values of h2 may be obtained perturbatively. rr +h* 3 =7rsin7rvA 64(A1) (A4)AVA +Q(h12). (17.5) and (17. (17. 124.11) or smvz or e±ivz.9b) TT2 Next we observe that if we demand that the solutions y+ (z) and y_ (z) satisfy the condition (17. These cases therefore have to be dealt with separately.10) A = n2. (17. — 2/i 2 A and insert this into the Mathieu equation y" + [A — 2h2 cos 2z]y = 0. = J—^ + o(hr).17.9a) and is therefore not derived here in detail^ ri/ = VA + rCOS7Tf = COS7TVA + /l 4 h4 7rsin7rVA = 4\/A(A .1: The eigenvalue A for nonintegral v and h small Use the perturbation method of Sec. To lowest order the solution y is j / ° ) = yv = cos vz n where d2 Dv:=—~+u2.7 to obtain the eigenvalue A as a perturbation expansion for nonintegral values of v and b? small. p. as will be done later. which can then be rewritten as Dvy = 2fe 2 (A + cos 2z)y. (17.2 Cosine Potential: Weak Coupling Solutions and so 345 (8 . VA = ±n.e. . Example 17. cos TT\/A 32A(A1) 2 . With a perturbation theory ansatz for y+ around h2 = 0 as in Example 17. Meixner and F. n an integer.35A + 15A2)/l8 /.1 below one obtains the following expansion for cos7rz/ which is fairly obvious from Eqs.jl2.1) 15A2 .35A + 8 . Solution: We write the eigenvalue equation A = v2. It follows that in these cases i? = n2[l + 0(h2)]. Schiifke [193]. For these integral cases of v we demonstrate in Example 17.3). 8.9a) 4(1A)>/A 64(A4)(A1)3A^A One observes immediately that these expansions cannot hold for integral values of v or A.
Schafke [193].11) and so in R}?' may be cancelled out by adding to j / ' 0 ' the new contribution a(2v + a)' = 0. We observe that in these cases 2cos2z cosvz 2cos2z smvz 2cos2ze±il/z i. Meixner and F.v±2) = l. (i/. i. v) = 0. where except.2) — 2 "^ 2(2i/ + 2) " + 2 vanishes. 17.13) The first approximation y(°) = yv leaves unaccounted on the right hand side of Eq. Proceeding in this manner we obtain the solutions ce. i. = . .2) _ __ _ (y. se and me as* oo i y = „«>) + yW + i. The right hand side of Eq. The recurrence relation can also be obtained by substituting the right hand side of Eq. W. v .11) terms amounting to i?i 0 ) = = 2 h 2 ( A + cos2z)y„ = 2h2Ayu + h2(yl/+2 + y1y2) h2[(v.h2).„ + £ h2i £ P2i(2j)yv+j (17.2 — we can write down recurrence relations for the perturbation coefficients pa(2j) together with boundary conditions. v)yu + {v. Next we treat terms yv+a in j / 1 ' in a similar way. in each case 2cos2z2/„ = 3/„ +2 + j / „ _ 2 . i. or A = 0(/i 4 ).. h2).!/) = 2A We now observe that DvVv = 0.15) Hence a term fMy„+a on the right hand side of Eq.2) which applies when v is nonintegral.3 and 19.v + 2)yu+2]. v .14) therefore leads to the following nextto leading order contribution (v. since j / 0 ' = yv leaves unaccounted R\. siniy + 2)z + sin(i/ — 2)z. With our perturbation formalism — as demonstrated in cases considered in Sees. '.16) directly into the original equation in the form of Eq. so similarly yW leaves uncompensated / # > = ft2 ("• v ~ 2 ) R (o) .(2) + . Periodic Potentials The complete solutions of these cases are written respectively in selfevident notation cev(z.h2). e ± i ^ + 2 ' 2 + e±i("2)z.14) = = = cos(y+ 2)z + cos{y— 2)z. (17.16) *These smallh expansions are convergent. (17. me±v(z. h2(u. when a = 0 or 2v + a = 0. of course.11).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. and (v.e. (17. as shown. . for instance. We assume for the time being that a ^ 0 and 2v + a ^ 0 (the latter case requires separate consideration.3. (17. by J. (17. up to 0(h2) we have A = 0.e. (17. see Example 17.e. sev(z. (17.2)j/„_ 2 + {v. so that also Dv+ayv+a and hence Dv+a = Dv + a(2v + a) and Dvyv+a = a(2u + a)yv+a(17.2. .e.346 CHAPTER 17. ("• ^ + 2) (o) 2(2^ .
We consider the specific unperturbed solution yv = cosvz which is the dominant contribution j / ° ) of our solution y. the solutions are not yet normalized. Hence we obtain 0 = /i 2 (2A + l ) f h4 + ••• . Schafke [193].2—r——— (i/ + 2.. v) ' ( 2 v + 2) (17. In the case of integral values of i>.1 with A = v2 — 2h2A but with v = 1. 0 = h2(u. Solution: We start as in Example 17.. Following the first few arguments of Example 17."2)..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. .7).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].18) Again applying Eq.g. * Note.17. . however. . The higher order terms naturally require a little more algebra. (17. See e.17) Inserting 1 for the stepcoefficients of Eq.. (17. J.2 Cosine Potential: Weak Coupling Solutions 347 together with the equation determining A. (17.R{ . W. is set equal provided the coefficient of the sum of the contributions in 3/1 contained in R\ to zero. Meixner and F.v)\ .. + y~i) (17. . These expansions are actually convergent with a definite radius of convergence as explained in the mathematical literature.14). This completes the determination of the solutions for nonintegral values of v in the domain of small values of h2.u) + h4 (". one has to deal with each integral case separately as in the following example.2 ) v(y2. fo^ + 2) . „ .15) (but with v — 1 and a = 2) we obtain the next to leading contribution to the solution as (i) = h2 =_fe2 V {2(2v + 2)}„=1y3 8 y3' This contribution leaves uncompensated the terms amounting to R\ = ~ — 2h (A + cos2z)j/ 3 o = —[Ay3\ 4 \ 2 Hence the next contribution to the solution is 4 V 8 2(4)(2 + 4 ) / 4 V 8 48 Proceeding in this manner we obtain as the complete solution the sum y = y(0)+yW+y(2)+.e. For v = 1 we then have yi = cosz = y~\. we obtain immediately 0 = 2Ah2+ 2(^T)+' which verifies the term of order h4 in Eq. 121. i. p. thus determining the quantity A. Example 17. 2 ( 2 i / .
24c) below) . Meixner and F. h2). 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. 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.19a) This expression agrees with the result given in the literature (there the eigenvalue is called ai).n integers. We introduce a normalization constant c and rename the normalized solution y then yc. see below!) one obtains the additional terms given by Meixner and Schafke [193]. COS 6Z • 128 h2 COS 2 • COS 3 2  h / c o s 32 16 cos 5z 48 cos z 32 in agreement with Meixner and Schafke [193]. Schafke [193]. . 120.* The associated solution is ym+ym. 123) (there the solution is called cei(z. The reverse situation is later of importance. We consider this case again in an example. and v is to be found in ascending powers of h2. Jo dz cos z h2 cos 3z  iC2T It J* cos z H 64 cos 2 3z  64 It follows that 128 and the normalized solution is therefore h2 Vc . Eq. J.348 From this we obtain 2h2A = h2 CHAPTER 17. m. Periodic Potentials h4 r ••• .e. *" Jir Thus we have and uses / dz cos mz cos nz = — <5 mn . p. h4 (17. The specific normalization here is taken as I /*7r rn 1 = — / dzce2(z. Inserting this into A = v2 — 2h2A 8 for v = 1 we obtain (cf. the case of A an integer. W.19b) This expansion agrees with that given in by Meixner and Schafke [193] (p. i. With normalization (not to 1. (17. In the above we considered the case of v an integer and calculated the eigenvalue A.
. o 1• • • . H. J.3 1 A + 8) ( A . (17.4)Av A2v A ~ 32A(A .l ) 2 (A . Substituting these expressions into Eq.. J .2 Cosine Potential: Weak Coupling Solutions Example 17.3 1 x 4 + 8) 274233 (.('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 .Q . . 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 . W.9b) about A = 4. Hashimoto [124]. Liang and Yunbo Zhang [189].1) (A . S.1) .3: The Floquet exponent i> for A=4 and h2 small Show that the expansion of the Floquet exponent v around \f\ complex expansion: v = 2 349 = 2 is given by the following iVE/h 3 \2J 7% h (108^/5 \2J 8 1185H (h 31104^X2 Solution: We proceed as follows which demonstrates explicitly how the singular factors cancel out systematically. (11A2 .9b) and considering the approach A — 4 gives > (A4)2 + .4) 2 7T 2 /l47T2 + • 8 h87T2 8A(A .4 ) + . fe47T2(A4) 4%/A(A .3 5 A + 8)TT2(A4) _ 3 / / 64(A .= l 2vA sin27r + (A — 4)(cos vA7r)x = 4—= + ••• = j= h •• • . (17. .9b) v — 2 + <5.1)2/A TT2 +h8 (15A2 . MiillerKirsten.• 1 + 5n h 2 s 2932 Setting in cosm. We thus obtain the expansions cos"\/A7r sinvAvr = = cos27r + (A — 4 ) ( — s i n V / A 7 T ) A = 4 — = + . we obtain^ S 3 V 2 and therefore i/ = 2  iy/E ( h 3 \2J 7i . Manvelyan. ^S. Gubser and A.17. so that 2 2 cos nu = cos 7r(2 + S) = cos 27r cos TC8 = 1 and comparing with the above. on the left hand side of Eq. (17. We set A — 4 ~ 4e and expand the cosine and sine expressions appearing in Eq.
2*0 y~(z ± 7t) = 6 sin VA(z ± TT) = y_(z)cos VA7r ± 6 cos vXzsin v\ir.29.21a) and (17.e. concentrating again on the leading term in the even and odd solutions y±(z) for ease of understanding. bV\ y+(0) 2/40) 07.21b). i. in F. These are defined by specific boundary conditions. 28 . We verify these for the large/i 2 solutions in Example 17.10) applies. we have" y+{z ± 7r) = a cos vA(z ± TT) = y+(z) cos vXir = TV{Z) F sm v<\7r. may also be found. (17. 17. solutions for integral values of v. 17. We now derive these boundary conditions. for instance. "These conditions. Fig. (17. Taking in particular Eqs. Periodic Potentials Our next immediate aim is to specify the solutions for which Eq.W4W=UtWSta±I(. Arscott [11].4) into account.6.2.350 CHAPTER 17.w^. M. = and s+w 2tW±I. . pp.. as also conditions (17.2 Four t y p e s of periodic solutions The Mathieu equation allows four types of different solutions — briefly: Even and odd solutions of periods TT and 2ir.1 Boundaries of domains of stability.
we obtain y+(7r) _ i  2y_(f). f^\y+(j) /vr\y_(7r) and for the derivatives ^(i)^^'(D^ . . /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\. .2 Cosine Potential: Weak Coupling Solutions We select the equations with lower signs and put z — TT/2. ^\y±M /vr\yL(7r) . which after some rearrangement can be written y+(vr) _ . 1+ y+(f)^(f)y_(f)yV(f)' 7T Using the Wronskian (which we actually had above!) W[y+.21a) Since (with the help of the Wronskian) *W4)MM?)^ we also have y+(v0 _ .4(f) (17. /7r\j/!_(7r) y L l 7TJ .17. 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 . Then y+ y 351 U .
+ . of v is even or odd. .1). se2 = sin 2z 2 12 13824 ' h2 z —— sin 3z + • • • .21b) must be ± 1 .21a) and (17. in this order. The ordering of the eigenvalues which then results is a0 < bx < ai < b2 < • • • < bn < an < • • • for h2 > 0. (17.22) * a2n+i(h ). These may be subdivided into classes depending on whether the integral value n — 0. y > ce2n+i=qo.23) We can now see how the domains of stability and instability arise.2): hA 7h8 29h12 + Y l28^04.3.24d) v 12 ' **To avoid confusion we emphasize: In Sec. was treated in detail in Example 17.1 for the first few eigenvalues which are boundaries of domains of stability (as discussed in Sec. (17. These are defined by the following boundary conditions. (17. y > se2n+i=qo. (17.2. 27r. and whether the function is even or odd.' fh2 V2ceo = l .24b) 8 h2 z —— cos 3z + • • • . A > b2n+i(h2). IT.5) that the left hand sides of Eqs. There the second term on the right is nonzero.24c) 8 h2 sin 4z \ ..• a2n(h2). We see that there are four possibilities and hence four different types of functions. Periodic Potentials and hence** a afc\/A For integral values of v (in lowest order of h2) we know from Eq. the case of ai with solution cei (of period 27r). y > se 2n +2s 9o +i The functions so defined have respectively period 7r. Finally we cite here some expansions of the eigenvalues along with those of the associated solutions (an explicit example. 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.— — — • • • .. 17.. ce± = cos 8 64 h4 5h8 b2 = 4 1 .352 CHAPTER 17. where the notation for the corresponding eigenvalue is put alongside on the right: y+(f) = 0 with y+(f) = 0 y'{\) = 0 with y _ ( § ) = 0 with with A .2 we require this relation for nonintegral values of v. Mr7nA (17. 2K.1... A 2 y • ce2n=q0i.T + .. A *• b2n+2(h2). (17. (17. 19. . (17. This is shown schematically in Fig.24a) ao = hA h6 b\ = 1 — h —— + — — • • • . 17. sei = sin 8 64 /i4 h6 a\ = 1 + h2 — .
. 13824 .z3 ce2 = cos 2z . the walls are not infinitely high and hence tunneling occurs from one well into another. Thus in Fig.17.2 suggests that for very large fluctuations the potential can be approximated by a number of independent and infinitely high degenerate harmonic oscillator potentials. Dingle and H.> (17. Consequently one expects the eigenvalues in the large h2 domain to be given approximately by those of the harmonic oscillator. . 353 .3 Cosine Potential: Strong Coupling Solutions „ 5/i4 a2 = 4 H 12 763/i8 . of course.2 or by calculation that the potential 2h2 cos 2z has (harmonic oscillatorlike) We follow here R.24e) 17.3 17.17„. Since we have already a definite notation for the levels (boundaries of regions of stability) in the domain of small values of h2 we naturally want to be able to relate these correctly to those in the asymptotic laigeh2 domain.hz — 12 .1 that the boundaries of regions of stability merge to lines. 17.2cos4. . Miiller [73]. 17.1 Cosine Potential: Strong Coupling Solutions Preliminary remarks In the case of large couplings h2 we observe from Fig. The cosine potential cos 2z depicted in Fig. 17. In the case of the cosine potential.2 The cosine potential. B.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.3. J. The main objective of the present section is therefore the calculation of the tunneling effect in the form of a splitting of the otherwise asymptotically degenerate harmonic oscillator levels j t Fig. 17. W. We can see from the inverse of Fig. 17.
3) and is. Thus in this case the quantity on the left is not an exact odd integer..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.26) 17.3. This equation has normalizable solutions for X + 2h2 2h q0 = 2n + 1. We therefore wish to expand it about these points. and we set this equal to q. Periodic Potentials minima at z — ±ir/2. Ion (17. In the case of finite heights of the potential barriers. simply the onedimensional Schrodinger equation for the harmonic oscillator.. Taking the remaining terms of the cosine expansion into account. of course.25) This is a Weber (or parabolic cylinder) equation which we encountered earlier (cf.2.1. Sees. as in the case of the cosine potential... 8. we therefore set 9 A A = 2h2 +2hq + —.2/i cos z V = 0. Thus we rewrite the Mathieu equation as y" + A + 2hz cos 2 \z ± 7T y = 0 and take c o s 2 [ z ± .354 CHAPTER 17. 4 (17. n 0. there is tunneling from one barrier to the next.2 The solutions We insert Eq. (17.2 and 8. (17. but only approximately so.27) .26) into the original Mathieu equation and obtain y" + 2h q + rrxr .
29) is known.27) we )A = 0.3 Schematic picture of Mathieu equation eigenvalues. another solution. 16/i/ (17. (17. 2 (17. and substituting this into Eq. z). can be .17. the linearly independent one.30) D A): COS « = 4z + 1 iq sin 2. (17. (17.+2h': 2h level splitting oscillator 2nd excited state associated solutions A^9 h=4 oscillator 1 st excited sate ^1 oscillator ground state A^Oh 2 large h 2 =0 Fig.29) the equation can be written in the form D^A where 1 2 6 /i (16A" + 2 A ^ ) .28) A" ±4hcos zA' + 2h[ q =f sin z \ A.31) We make the important observation here t h a t if one solution of the type (17. Choosing y(z) = A(z)e2hsinz.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.
356 CHAPTER 17.35) not yet taken care of is set equal to zero. when i = 0 — the terms not involving Aq in (17.34) The leading approximation A^ = Aq therefore leaves uncompensated terms on the right hand side of Eq. Aq+4 2Jh . of course. Periodic Potentials obtained by changing the sign of z or alternatively the signs of both q and h throughout. (17. (17.32) cos^'+i) [\K . .4 ) ^ .35) But since and so a term iiAq+n on the right hand side of Eq.94) \ — Aq4 (17.(°) + A^ is the approximation of A to that order provided the coefficient of the term in (17. ^ ) A + ( ? . With some algebra one finds that 16^ + where (9.33) (17. D{q%Aq+4i = 0. 2[(92 + l) + A]._ 4 .e. (17.30) can be cancelled out by adding to A^> a new contribution — fiAq+4i/2i — except. 1 + 4 ) ^ + 4 + (^.9) = 0. 2 /i D^Ag = 0. (9.4 ] . 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 (17.4)A.9) (9.\z To lowest order the solution A is A^> = Aq.94) = (9 + l)(9 + 3).g + 4) .36) Then ^4.30) amounting to Ri ] q = .^ 6T [(«>« + 4 ) A ? + 4 + (Q. (9.35) can be cancelled out by adding to A^ the next order contribution A& = 1 (g. ? .Q)Ag + ( ? . 2 A A = («. (17. i.9 + 4) = (9. ? . (9l)(93).
q 8)Aq+8 (o)' ^±V(q 1 4. except those involving Aq.4) (q . h)..4. h) = A(z. (17. .40) . (q.35) and (17. A(z. Thus 0 = (g.q + 4)(q + 4.30) and hence y = Ae2hsmz to be a solution of the Mathieu equation.g) 26/i L_ 2 13 /i 2 ^±A{q + ^q) + —1 ^A{q^q) '17.8)A.q 4)Aq+4 q + 4 1 q + + + + + <^(. In its turn the contribution A^1' leaves uncompensated terms on the right hand side of Eq.30) amounting to RW 9 = fag+ — 2?h 1 ^ 2™h? +< 4 ) R(o) \ .g + 4)(g + 4.q + 4) (g.17. A.3 Cosine Potential: Strong Coupling Solutions 357 This equation determines the quantity A in the eigenvalue equation to the same order of approximation. q . the sum of the terms in Aq in Eqs. Thus in particular A(z.<J+4 Aq+& + 1 2 1 . q .)+ & ! = « ( .39) which is the equation from which the quantity A and hence the eigenvalue A is determined.37) and so on — left uncompensated — must vanish. . q .4) 1 1 (q. We note here again that if one solution y = A(z)e2hsmz 2hsmz is known. q.37) Here all the terms. can be taken care of by adding to A^ + A^ the next order contribution A^2\ where A (2) 2h u 2 (q.. q (17.g4) 4. Then for A = A(0)+A(D+A(2) + .4) (q ._ 8 (17. to satisfy Eq. ) ^ 4. Q .4. h) = A(z.39) is obtained by changing the sign of z or alternatively the signs of both q and h throughout.38) Aq8 + 1 2 Clearly we can continue in this way. an associated solution y — A(z)e~ with the same eigenvalue and thus the same expansion (17. (17. (q.4 . q. . h). (17. q. +4.g + .. q. (17.
e.) i ( Q . a solution Bq(w) of the equation Dq Bq(w) — 0 is Bq(w) oc He^^^iw).44) where D(B) 9 = dw2 $_ !)• (17.43) 9 2eh .45) {B) We recognize Dq as the differential operator of the Hermite equation.d2B w dw2 + w(l w dB_ dw d dw 1.41) so that the early successive contributions decrease in magnitude. Note different definitions in use.358 CHAPTER 17. (17." 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. (6. it is found that D(B) z+ — ]B = 0. (17. we show that two other pairs of solutions with a similar coefficient structure and the same eigenvalue expansion can be found which are valid in adjoining domains of the variable z.46) 2s(«. The differential equation there is Eq. 2 w2 + H* (17.28) and write the upper equation for a solution y = Bexp(2hsinz) in new notation B" + 4/icos zB' + 2h( qsin Changing now the independent variable to w{z) = iVhcOS I 7T + Z I .42) (17. and discussed there. this means the solutions are valid in particular around z — 0 since then l / \ / 2 3> l/Vh ~ 0. Periodic Potentials Finally we note t h a t the above trigonometric solutions of t h e Mathieu equation are valid in the domains cos 1 4 1 Z 7T ± » 1 2 VK' (17. We observe that for the large values of h we are considering here. We return to Eqs. Before we study the eigenvalue equation in more detail.67). where Hen{w) is a Hermite polynomial when n is an integer. . i.i ) We encountered Hermite polynomials earlier in Chapter 6.
49) Then the dominant order solution is C^°\w) = Cq{w) with and — again choosing the multiplicative factor suitably — Cq(w) = 2i(" +1 ) ^ . 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). the expansion for the quantity A being identical with that of the previous case.qi)Bq_4.3 ) \He* ! (9+1) H. i.sin z + by changing the independent variable to w(—z) = 4v/icos I 7r z (17.e. .4 (using the known recurrence relations of Hem{w)) the perturbation remainder can be linearized. Again a change in the sign of z throughout yields the associated Mathieu function y = i?exp(—2hsinz) with B[w(z)] = B[w(—z)] and domain of validity in particular around z = —ir/2.q + 4 ) £ g + 4 + (q.17.e. 1 COS I .51) . i.. (17. forms a rapidly decreasing expansion provided that \w (z)\ « Vh.3 Cosine Potential: Strong Coupling Solutions 359 because then as shown in Example 17. which is an amazing and unique property of the Mathieu solutions — cf. 16hJ (17. also the same coefficients in the case of functions C below. as in Eq.e. y — Cexp(2hsinz) and Cexp(—2/isinz). (17.e.50) C = 0. A{wz (17. i.47) where the coefficients are the same as before..7 T H 1 2 Z 4 <1.. q)Bq + (q. i. Finally we obtain a third pair of solutions.34). with the equation C" + 4h cos z C + 2h ( q .d2Bq + w{l dw2 "» 2 >^(» 2 + ^ = {q.48) This shows that this solution is valid in particular around z = ir/2. The solution with contributions B^°\B^>. (17.
We have thus determined three pairs of solutions of the Mathieu equation.A Hen .w2)He'n .4. show that the perturbation remainder of Eq. q + 4)C 9 + 4 + (q. 2' the remainder 72. Periodic Potentials since then (using the known recurrence relations of He^) ~Mw2^ +w(l+w2)^ + (w2 + o dwz dw V 2 (q. Example 17.(ql)2(q3)HeX (95) i(« 2 + i) + iA Hei (91) .44). ie 1 COS I .4: Hermite function linearization of perturbation Using the recurrence relations (6. We note that C(z.q. These domains of validity are indicated in Fig. Solution: Inserting the dominant Hermite function solution into the right hand sid of Eq. 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.(w2 + A)Hen. q)Cq + (q. (17. each of them associated with one and the same expansion for the eigenvalue. A closer look at that expansion is our next objective. 2 J where n=(ql).53) This solution is valid in particular around z = —7r/2.54) where the proportionality factor is complex. we have a remainder •R:=w2He„+w(l Using the recurrence relations wHen(w) = Hen+i(w) + nHeni{w).44) can be linearized with the same coefficients as in the case of the trigonometric solutions.n2(n l)Hen2 . (17. (17.q.52) and the coefficients are again the same as before. He'n(w) = wHen(w) Hen+i(w). (17.q. (9+3) (2n 2 + 2n + 1) + . . (17.360 CHAPTER 17.59d) of Hermite functions Hen(w).h) ocB{z.7 T 1 2 Z 4 <1. can be reexpressed as 72 = (n + 1 ~(q + l)Hen+2 l)He. Again the solutions have the same structure as in the previous two cases and the eigenvalue expansion remains unchanged.4)C g _ 4 .h). The solution y = C exp(2/i sin z) is a decreasing asymptotic expansion provided that w(z) ^ v^/i. 17.
q4) (17. in the following compact form with coefficients Pr(q.0 and —4. with (. 17. .56) (94. 2)Aq_8} + • (17.3 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). In addition (c) each move except the last has to be divided by the final displacement from q divided by 4. 2 + ^ M 3 + (17. q) = 2[(q2 + 1) + A]. The coefficients have thus been constructed in a way which now allows the formulation of a recurrence relation.' 361 in 2*f*1>[i(gl)]! = (q.57) . (q.q 4. r j=r.3).q + 4) (q + 4.4) = (q . (?) 1 1 + (9. e. ~l)Aq4} 2h 1 { P 2 ( < Z 2 + + P2(<? lAq+i ) 8 A ) ' " ' +P2(q. ^ 1 Hv" '^ ^' ^ ' .1i . We see that (a) every coefficient Mr results from a sequence of r moves from q back to q.g4)(q4.17.g) 1 1 and so on.l)(q .3. l)Ag4 + P2(q.3 Cosine Potential: Strong Coupling Solutions It follows t h a t with the definition of Bq (w) as Bq(w) that He K ..q + A){q + A. (q.j^0 e 2ft sin z A + ~ww 1 7 i + ^7ir{ p i(9' ^ ^ + 4 + Pi(q.l)[l(q _!)]. ) .q + 4)Bq+4(w) + (q. . We can write the solutions obtained previously. q)Bq(w) + (q. the solution y = exp(2/i sin 2)^(2).. (q. q .i ) ( w ) 2l(.g. q + 4) = (g + 1)(? + 3).55) M2 = M. no intermediate move to or from q being allowed. (b) The allowed moves are +4.)Bqi(w).j): y = = = e e2hsinzA{z) oo 2ftsinz r=0 V .
j). j) +(g + 4j + 4.l ) . B.P 2 ( g . Miiller [73]. j + 1) (17.j).362 Clearly CHAPTER 17.g + 4)(g + 4. (17. W.j)] = 2j]jPr2(g.g + 8)(g + 8. Dingle and H.j)Pr+1(g. the factor j removing a duplication of factors in the denominator at the junction. (17.g + 12) 1 2 3 (g.Each term in M2r can be considered as the product of the contribution from a sequence of r moves starting from q and ending at q + 4j.g + 4)(g + 4.j) = (l)rPr(q.g + 4)(g + 4.i (g. 2) 1 2 ' ' ^ ' ^ 1 (g. .g + 4) 2 1 (g. .j).4. M2 = P 2 ( g . x)_ (g._1) = Mzil.59) One can now write down formulas for M^r and M<ir+\ in terms of the coefficients Pr{q. g + 4 j ) P r .58) with P 0 (g. P2(q 2) Pl(g.g + 4)(g + 4.60) M2r+l = = *R.34) we deduce that Pr(q. g + 4j)P r _i (g. From the coefficients (17. Periodic Potentials PlM)=(M+i). Thus we have.3) ^3(9.j ^ 0) = 0.j).j). i. j .j)^2(9. as jP r 2 (g. By the above rules we can write down the recurrence relation for the evolution of a coefficient Pr by steps from the coefficients Pr\: jPr (g. p = (<?><? + 4 ) ( g + 4 >g + 8) .e. and in general* M2r = J]j[P2(g. l ) .g + 4 ) ( g + 4. and the contribution from a sequence of r moves from q + 4j back to g.g + 4) + 1 1 1 and so on. 0) = 1 and all other P0{q.g + 8)(g + 8.1) + (g + 4j. for instance.g + 4) ^3(9. g + 4j)P r _!(g. r ^JtPr^jOPr+lfejV^rfejO^r+l^j)] r 2^jPr(g. J. j) = (g + 4j .
62a) f R . *E.284g + 57) + (17. q is only known to be approximately an odd integer qo = 2n + 1. S. Miiller [73]. Ince [136]. L. 17. Its precise deviation from qo. This means that the in approaching such a domain the appropriate solutions must become proportional in view of their uniqueness. Up to > > this point.(5q 4 + l)^Jiq(q2 + 3) + 410g2 + 405) + 34q2 + 9) . a= (8/i)3(9x) [1(91)]! 1 .^ . It will be seen in Chapter 18 that in the case of anharmonic oscillators.61) +288161796g2 + 130610637) These are the terms given explicitly in the reference cited above. Goldstein [115]. the solutions and eigenvalue expansions again have the same type of symmetries. h — —h. [137].* We observe that the expansion remains unchanged under the replacements q — —q. remains to be determined next. . of course. t Terms up to and including those of order l//i 5 had been obtained by others and by different methods. [116].17. W. as pointed out earlier. Dingle and H. B.5 the following proportionalities there: B[w(z)] = aA(z). as a result of tunneling between wells. 17. In fact we demonstrate in Example 17.3 Cosine Potential: Strong Coupling Solutions In this way one obtains the eigenvalue expansion A = 363 _2h2 + 2hq^(q2 1 . These domains in the interval —7r/2 < z < 7r/2 are indicated in Fig. where solutions overlap.2 * +3) + ^ 2 (9g4 .( 3 ^ .4 The level splitting Above we obtained three pairs of solutions along with their domains of validity as decreasing asymptotic expansions.3. J.92q3 + 70g2 .^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. We observe that there are regions.4.
COS 7T 2!42 \4 2 COS /l 7T V4 Z H 1 2 Z H 1 2 .1 4 . The first term of the expansion of A(z) is the function Aq(z) given by Eq.364 and C[w(z)] =a~A(z).(17._ ^ l CHAPTER 17.1 ) ( 7T+ z 1 — COS 2 1 \l^+1> 2 Z C O S S ^ . Periodic Potentials + _(3<?+2<? + 3) + 2 15 /i 2 (9g 4 + 92g3 + 70q2 + 284g + 57) + • • • (17.7T H Aq(z) = = C O S ^ . Solution: We begin with the solution containing the function A{z).Z (g+1) 1 H 1!4 4 2 1 (g + l)(g + 5) . We write and expand this function as follows: .62b) *z domains of overlap Fig. 17. a .4 Domains of solutions and their domains of overlap. and determine the constants a and a. Example 17.32).1 ) ( 7T + .5: Proportionality of solutions in domains of overlap Show that in their common domain of validity B = aA and C = aA.
61) for the connection and Eq.g4)fa3) 1 1!4 / I V/i 2 = 1 (gl)(g3)2 2 9 /t Similarly substituting the asymptotic expansion of the Hermite function^ into the expression (17. with (8h)li''1) (3g2 . As we saw there. These are the conditions of Eqs. with e. \ql)/2(w) 2i(9D[i(.46). This is essentially the expansion of the parabolic cylinder function with the exponential factor removed — see Eq. 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 . the coefficient of the dominant factor c o s ( . (17. For further details and explicit expressions of higher order terms we refer to Dingle and Miiller [73].53) for the expansion.22). (6.!)(.i ) 1 (g. we have to impose the appropriate boundary conditions. these are conditions imposed on even and odd solutions at the point z = it/2._i)]! w*te» 2l(91 (.q) 2 7 /i 1 1 4(g + 3) 1 (g + l ) 2 ( 9 + 3) + 29ft ti(9l)]l i (8/1)4 ( 9 . we have as even and odd solutions y±ocA(z)eZflsmz±A(z)e 2/isin. s .6 at the end of this section we derive for these \axgeh2 solutions the conditions similar to those given in various equations in Sec. A*1) given by Eq.e. i. Since the solution with A is obtained from that with A by a change of sign of z. 17.36).63) In Example 17. (17.2. (17.g.3 Cosine Potential: Strong Coupling Solutions 365 Inserting this into A = A(°> + A' 1 ) + • • •. Hence we have to construct even and odd solutions and impose the boundary conditions at z = 7r/2.17. In order to link our asymptotic solutions with the Mathieu functions of integral order defined earlier in the consideration of small values of h2.2g + 3) 1 [£(91)]! 2?h 215h2 (9g 4 .2 for the case of small/i2 solutions and thus insure that in both cases the same boundary conditions are obeyed.g + 4) (q + 4. we obtain Bq[w(z)} H.3) 1!(8™2) )[i(gl)]!L Inserting this into the expansion of B[w(z)] = B^ + BW f • • • where w{z) — 4 / i 1 / 2 cos I IT + z J.i ) [(91)]! It follows therefore that over their common range of validity B = aA with a = b/a. (8. _ 1 " 2 ( 7 r / 4 + z/2) in the result is seen to be a = 1 2 /i 7 fa.92q 3 + 70q 2 .284g + 57) + • The other relation is obtained similarly.
He'v(0) = Hev+1{0). 80. The second of relations (17.62b). Periodic Potentials We can extend the solutions (17. Then y± oc ^ M e ^ i n . Here v is the not necessarily integral index of the Hermite function Hev{w) with Hermite equation ( jir w—+v)Hev(w) \ aw1 aw J = 0.62a) and (17. (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. i. Thus y+ = ce and y_ = se.66) then follows from the recurrence relation (6. i. With the help of the duplication and inversion formulas of factorials. He2n+1 (0) = 0. a ± C[w{z)}c_2hsinz^ a (17.e. The functions defined as in Eq.59d).g.366 CHAPTER 17.e. He'u{w) = wHev{w)He„+1{w).63) to the domain around z = ir/2 by using the proportionalities (17.22) and to correlate the solutions to those for small values of h2. p.66). Magnus and F. give the following expressions for polynomials with argument zero: ~19)"(^)!. and then replacing 2n by i>.66) ^E.22) are therefore given by the following boundary conditions: 1T y'+ v+fao I'=° y a a\dwjw=0 B[w = 0]c2h a a 1h dw\ ! ' . one obtains the first of relations (17. Oberhettinger [181]. .64) These are now the solutions around z — ir/2 which permit us to apply the boundary conditions (17.» a \dw _ C[w = 0}c_2h a w=0 = Q (17. W.
1) 7T .\4 q + m { 4« VJL ^[±(g3)]!sin{f(gl)} [J(ffl) "(^L^»{^': _ _q_ ¥h + g .68) Inserting here expansions (17. f^(16h)i/2e4h 2 [Uq1)]\ 3(g2 + l) + 2^h2 (9g4 .Qo) + 0[(q .67c). and similarly for C and its derivative.( g . Hence expanding the cotangent about these points.» 2 6 /i the condition reduces to the equation cot{ .11. the solutions would be given by q = go = 3.! ) ] ! [ . (17.67a) g 106g 87 2 14 /i 2 (17.87 4 2 ¥w (17. we have cot j ^ ( g ~ 1)} = ~\<(l .65) we have B[w = 0] _ a C[w = 0] ~ ~ ^ 4/i (17.82g2 .7.14) 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 .67b) <f ~ 82g^ .69) If the right hand side of this equation were exactly equal to zero.67a) and (17.67c) 4 2 \dw)w=Q C[w = 0] [\(q .106g . B[w = 0] 2TT _ _g_ ¥h+ g 24/* _ JL g4 .39 (17. and using the reflection formula (8. from which one obtains "1 1) 4<*3) 2(9!)/2 2 ^ \b.3 Cosine Potential: Strong Coupling Solutions 367 With these expressions we obtain by insertion into B and its derivative.67d) Considering now the second of conditions (17.40g3 + 18g2 . .136g + 9) + • • • .39 2~^h? (17.g0)3].17.
71) = 2 13 /i 2 fc{l6h)i/2e4h 1 3(g2 + 1) 2 [4(51)]! L (9g4 .40gg + 18gg ..72) This leads again to the result (17.5.368 Hence we obtain qq0 a [2(16h)i°/2e*h CHAPTER 17.40g3 + 18q2 .40gg + 18gg .65) we have (dB/dw)w=0 (dC/dz)w=Q Proceeding as above one obtains tan{(ql)\ 7T .408g0 + 1089) 2 /i +• (17.136g0 + 9) + • Clearly the last of conditions (17... we use \(q)~\(q0) + (qq0)(^J . i.120g£ + 467g .528gg + 3307Q .136q + 9) + .9.73) We now return to the eigenvalue X(q) and expand this around an odd integer qo.65) leads to the same result except for a change in sign. The remaining condition (17. Considering now the first of conditions (17. = _a a 6 _4h (17. Periodic Potentials ' " Vn [i(gol)]' 3(g02 + 1) 26/i .65) yields the same result except for a change in sign.e. .h qqo = * [£(*>!)]! 1 3(902 + 1) 26h +^^{9q^ .70) + ^ 2 (9?o .70) except that now qo = 1..(17. Thus we have 2 (16/i) qo/2 T2 4.136g0 + 9) 2 13 /i 2 1 4 2 19 3f(9gg . (17. .
M.3. just as any solvable problem is important for the insights it offers in a concrete and transparent form.3 we call this difference 26E„n. Goldstein [116].88q0 . Arscott [11].1. J. The tunneling effect.3 Cosine Potential: Strong Coupling Solutions Differentiating \(q) of Eq. . 120. results from the boundary conditions.. example 3.t From Eq. n ' r 1 / rrre 4/l ^odd(go) Seven (<Zo) Wfaqom (16/i)T n\V2n e~4h.^ ( 3 g 0 2 + 8go + 3) A ( Q 0 ) T (87r)V2[i ( g o i)]l L2 w . The dominant contributions were also found by S.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.e.3. 17. (17.. + ^ 2 (9<?o + 89o " 78g02 . n = 0. i. Dingle and H. made evident by the splitting of the asymptotically degenerate (harmonic) oscillator levels. p. (17. this becomes 369 = (16/i)^+1e4h l . i. B. (17.74b) The above results for the cosine potential are in many respects important. We see here that *R. 20. and that for the odd Mathieu functions se qo+ i and se9o (lower. In the literature one finds frequently the statement that the splitting is a nonperturbative effect.87) = Azpteo). Miiller [73]. See also F.e..74a) we obtain for the level splitting in dominant order* x r ^ \ / N 2(16/i)>+1 _4h A+(go) .A_(g0) ^ . quite apart from applications. *In Sec. minus sign). In the literature that we follow here* in all cases several more higher order terms have been given. (17. though approximate. In particular we saw that the perturbation expansion yielded only the degenerate eigenvalue expansion. . The complete.2. Effectively the degenerate eigenvalue expansion results from the anharmonic terms contained in the power expansion of the cosine potential. solution of the Schrodinger equation for the basic and nontrivial periodic potential is important for a thorough understanding of its quantum mechanics. and not the separation of these as a result of tunneling. plus sign) give the socalled level splitting AA(go) as a consequence of tunneling indicated schematically in Fig. W. 9o = 2n + 1.61).17.
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. and one can compare the methods.74b). The explicit derivation of the level splitting (17. ^ ± n ) = _ ( _ 1 ) T i(<JTl) A 9 ( 2 ) (the extra minus sign in the second equation coming from Aq(z ± 27r) = — Aq(z)). results from the imposition of boundary conditions. evident through the exponential factor exp(—4/i) in Eq.32) we obtain Aq(z ± TT) = (l)±i<9^'Aq(z). Naturally one expects the solutions there to be related to those of the periodic case considered here. The splitting can also be derived by the pathintegral method as will be shown in Chapter 26. The splitting can also be obtained with some methods of large order of perturbation theory. . the exponential factor we encountered in the semiclassical method where it has the structure of the factor exp(classical action/ft). 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 represents. T y'_(0) y'A±*) y{z). (17.2: y+(±ir) J4(±TT) y+{z±n) y(z±7r) = = ± —y+(z)± 2/+(0) . Thus we do not reproduce every step.2. in effect. Example 17. M±TT) . 17. We begin with the solution containing the function A(z). the Schrodinger equation with the cosine potential is a special case of several other more general cases. Periodic Potentials this nonperturbative effect.6: Translation of Solutions For the largeh 2 even and odd solutions (17. 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. 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. which one can then try to solve in order to obtain the behaviour of the coefficients of the late terms of the expansion. y^{z) = Aq(z)e~2hsin *. (17.74b) is also important for various other reasons. y+(0) 2/_(0) Solution: The derivation requires a cumbersome tracking of minus signs. _3/+(z) + ____2/_(z).§ The boundary conditions we imposed here are those of a selfadjoint problem which has real eigenvalues.370 CHAPTER 17. As mentioned at the beginning.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. yA{z) = Aq(z)e2hsin '. Then we obtain first y±(z) = yA(z)±y^(z). .
.[A'qM + 2hAq(n)] = iV2(q . Finally for the derivative of the odd solution we obtain y'_{ir) = K ( T T ) . 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+.3 The Ellipsoidal Potential Replacing VA.4ft) sin j ^n(q .( 2 ) .31). The Lame equation has not been a widely known equation of mathematical physics so . which is easiest by comparison with the foregoing treatment of the Mathieu equation. the operator having solution Thus ^ ( 0 ) = (q/2)Aq(0) = A~'q(0) and A'q(z = ±TT) = ~qAq(z qAq(z = ±7r).4.4ft) cos i TT(<J + 1 ) 1 .17. we obtain the conditions stated at the beginning. (l^ifr^i^O + Cl^ifo^j/aCO cos  . but also to enable a brief familiarization with the Lame equation.7 r ( g ^ l ) U + ( z ) q : i s i n { .g A ? ( z = ±TT).• Then y'+ (0) = 0. ( . y+(0)~2V2.z ) . since A q (0) = \/2 = Aq(0): y+(±n) = ±isin^(9=Fl)J3/+(0).7 r ( q = F l ) } j / .q).( ± 7 r ) = c ° s  . A' (z = ±ir) = (.1 Elliptic and Ellipsoidal Potentials Introduction Our intention here is partly to deal with other periodic potentials.1 )  . where Aq(0) = Aq(0) = V2. Similarly we find y(z±ir) = = and by putting z = 0: V.%/2(g . — j / + ( 2 ± 7 r ) = ± i s i n  ^ ( g = F l ) U + ( z ) .c o s  ~(<1 T 1) f y .2hAq(n)] . y'_ (0) = \/2(4h . = ±ir). (17.VA DV 371 1 V±. Replacing sines and cosines by the functional expressions. we obtain — with Aq{z + 7r) = (1)^ ^ 2 A .4ft) ~ 8(q . and y+(z = ±TT) = V2(q .y]z=„ = [y+yL . Aq(z) The derivatives can be handled with the help of Eq.yV+\z^ = [j/+(0)] 2 (<? .T ( ? =F 1) p + (0). 17.) cos < 7r(q 1) [ = .4ft.Ah) ~ 32ft. A ?q( Z = ±TT) = .4 17. etc.( z ) From this we deduce for z = 0.
*A. one arrives at three equations of which one is the ellipsoidal wave equation pt + [A . Ince [138].. the nonoccurrence of the Floquet exponent.0.. however. MullerKirsten and D. This line of investigation. (17.. and was continued by Erdelyi. for general n the solution of Lame's equation is not singlevalued owing to the singularities in the finite part of the uplane.75. 194) remarks to parallels with the Mathieu equation:". so that Floquet's theorem cannot be immediately applied . W.5 The potential sn2(z." Fig. L.372 CHAPTER 17. .98. H. Liang. 17.0 and .. inverted double well and cosine potentials (J. The most conspicuous difference compared with the Mathieu equation is. k) for k = 0.. f E . Arscott [11].5. Periodic Potentials far. In particular the Lame equation was recognized to arise as the equation of small fluctuations about instanton solutions for practically all basic potentials. 194) remarks. Tchrakian [165]).* and the work of both prepared the ground for presentday investigations. Arscott [11] (p. but also cubic potentials.* As Arscott [11] (p. p. as alluded to at the beginning.. 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 . If one separates the wave equation § V 2 * + J2* = 0 in ellipsoidal coordinates (which we do not need to consider here). 19. J.K2sn2u .fi2A..0.5 < z < 5.1.75) ' T h i s means double well. § F . But more recently it has been observed to arise in various contexts. Erdelyi [85].4sn4u]y = 0. M. H. the new stage of the development of the investigation of the Lame equation was really initiated by Ince^ around 1940.Q.0.
2k.75) reduces to Lame's equation and one writes K. we consider mainly the Lame equation. the usual factor —h2/2mo. is the elliptic modulus of the Jacobian elliptic functions snu.2 = n(n + l)k2. for very > high barriers (harmonic oscillator approximation around a minimum of the potential). (17. If we put fl = 0. . . in front of the second derivative has to be kept in mind (mo being the mass).e. Although the results have been calculated for the ellipsoidal wave equation. 17. For barriers of finite height the parameter q is only approximately an odd integer qo in view of tunneling effects. W. where n is real and > —1/2 (and n is an integer in the case of solutions called Lame polynomials) where k.4. In order to distinguish the above equation from that with the Lame potential. i. MiillerKirsten. Eq. (17.3 The Ellipsoidal Potential 373 Here K2 and A are separation constants and U2 = l2(a2 — c2). the equation ^  + [A .77) into Eq. (17.17. K being the complete elliptic integral of the first kind.77) where q —> go = 2iV + 1. cmx and dnu. The range of the independent variable u is 0 < u < 2K.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. we refer to the potential consisting of the two terms with sn 2 n and sn 4 n as the ellipsoidal potential. in the case n2 — oo.a > b > c being related to the lengths of the three axes of the ellipsoid in a Cartesian coordinate system.K) = qK+^±. 2 .e. J.78) We follow in this brief recapitulation the description in H. In comparisons with the Schrodinger equation. 17. .76) and to write the solution y = A(«)exp  f Ksnudul = A(u)[f{u)]K. The first step is to write the eigenvalue A as A(q. (17.1. . N = 0.5 for several values of fc.K2sx?u}y = 0.2 Solutions and eigenvalues In the following we sketch the main points of the method of deriving asymptotic expansions of the eigenvalues and eigenfunctions.\k\ < 1. Jianzu Zhang and Yunbo Zhang [206]. (17. The function sn2u is plotted in Fig. and of the derivation of the level splitting'. i. The second step is to insert (17.
q. determine their proportionality factors) in domains of overlap (their extreme regions of validity). A. A second solution is written y = A(u) exp < / nsnudu >. one pair in terms of Hermite functions of a real variable.81) Since these expansions are not valid at the extrema of the potential (where the boundary conditions are to be imposed).Es(ii). more precisly for dnu = cnu F 1 » . /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. . / z J . i. but solutions B.77). B and C.e. ll^ ^A(u)[f(u)]^k±A(u)[f(u)}^k.q.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. by transforming the equations for A. dnu =F cnu . the other in terms of those of an imaginary variable. i.e. one has to derive new sets of solutions there and match these to the former (i. ^ (17. \/8K (dmz = cn« \ ' F . Periodic Potentials . C replacing yl. A into equations in terms of the variables z(u) = ^—{k' . (17. dnu± cnu K Thus one can construct solutions Ec(u). C which are valid for dnu±cnu <1.82 . A(u. which are respectively even in u (or snu) or odd.80) The domain of validity of these solutions is that away from an extremum of the potential. one obtains again the same expansion for A. (17.374 where CHAPTER 17.K). Thus in the third step two more pairs of solutions B. A are derived. . _ Solving the resulting equations iteratively as before. k' = Vl~ k\ \dmi±cnu/ 17.K)=A(u..K — > A(u)=A(u + 2K).e. (17.
^ ^ { ( 1 + k2)5(63q6 + 1260g4 +2943g2 + 486) 8k2{l + k2)3(49q6 + 1010g4 + 1493g2 + 432) +16fc4(l + fc2)(35g6 + 760g4 + 2043g2 + 378) 64fi 2 fc 4 (l + fc2)2(5g4 + 34g2 + 9) + 256f22£." A = qK±(l 1 + r {(l k2)(q2 + l)^{(l + ifc2)3(5g4 + 3 V + 9) + k2)2(q2 + 3)4k2(q2 + 5)} 2WK2 Ak2{l + k2)(5qA + Uq2 + 9) .4. Erdelyi. one sets at it = 0 and u = IK altogether** Ec(2A") = Ec(0) = 0. F. C = a~A. 64. a of _ _ B = aA. p. . (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.17.85) The first three terms of this expansion were first given by Ince [138].3 The Ellipsoidal Potential 375 In their regions of overlap one can determine the proportionality factors a. (17. Es(2f0 = Es(0) = 0. J. Oberhettinger and F. W. the ellipsoidal equation. Tricomi [87].75).384n2k4(q2 + 1)} q r {(l + A. (17.e. *A. i. 17. as well as \duJ2K 11 (17.3 T h e level splitting In the fourth and final step one applies the appropriate boundary conditions on these solutions. W.6(5g4 . Magnus.38g2 .63)} + ••• .2)4(33g4 + 410g2 + 405) 24/c 2 (l + k2)2(7q4 + 90q2 + 95) + 16fc4(994 + 130g2 + 173) +512f22fc4(l + k2)(q2 + 11)} . G.e. Miiller [205]. who obtained this expansion for the eigenvalues of Lame's equation (i. This expansion is found to be in the more general case of Eq. £2 = 0).86) \duj0 o h r 2K > \duJ =[sr 0 =0\duj (1787) H.
go being an odd integer. 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 . 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.j 26K2 (1 85) = A(g0) + (q .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)*. Periodic Potentials These conditions define respectively functions EcX°.go)« qo(l + k2 22K {3(1 + k2)2(q2 + 1) .136g0 + 9) n 3. J. from which the difference q .89) one obtains the eigenvalues from which the level splitting can be deduced.:J {3(1 + k2)2 (9q* . 2K and 4K respectively. 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).g 0 ) ( ^ . Finally expanding A(g) ~ A(g0) + ( g .2 /c +256fc2g0(g2 + 5)} (17.90) + 3.1 / 2 (2TT) 1 /22« l + (l_fe)^«( i .i Here the upper sign refers to Ec*> l or Ec£°.4k2(q2 + 2g0 + 5)} + • • • ]. 2K.g0 is obtained by expansion around zeros.376 CHAPTER 17.91) tt H.Ec£ 0_1 and Es2° of periods 4K. Muller [205]. .78g^ .Es^ 0+1 .l n 2^+ i J + 0(«i) (17. + fc )(3g + 8g0 + 3) k2)2(9q% + 8g^ . (17.88g0 .87) (17. and the lower to Es*>+1 or Es*5.40g03 + 18g02 . W.
that calculations with the Schrodinger equation are not only easier.* Apart from the choice of notation. V. h . Hence the conditions fi = 0. L.4. Thus in the case of the dominant contribution of Eq. 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./2i and hence A^ A/ A AU I2 . (17.17. (17.92) reduce the periodic ellipsoidal wave functions and their eigenvalues to corresponding Mathieu functions and their eigenvalues.3 The Ellipsoidal Potential 377 This result agrees with a result of Dunne and Rao** who calculated this expression using instanton methods. *H. K = 2/i. this equation becomes y" + {A .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. J. it = x ± ^ . Thus if k — 0 and n —> oo in such a way that > K2 = n(n + l)k2 ~ finite. Ince [138].86).f ln[(l + fc)/(lfc)] _ e 4h) ( 8" \ ^^ __ (16h)«.93) (17. 17.90) one has in this limit: /l±±\ ~K/U = e .e. + E. .87) agree with those of Ince. Miiller [206]. A(g)>A(g0)T4/i\/e ^ [Hqo ~ !)]•' in agreement with Eq. Replacing u by x ± IT/2.* M G .2h2 cos 2x}y = 0. Dunne and K.4 Reduction to Mathieu functions Under certain limiting conditions the ellipsoidal wave equation reduces to the Mathieu equation. however. . but also more easily generalizable. (17. W. (17.4 h ( i e M 9 0 7 2 \. One can see. then snit — sin u and Lame's equation (f2 = 0) becomes > y" + {A . in which they dubbed the classical configurations Lame instantons.Ah2 sin2 u}y = 0. A = 0.74a). the conditions (17.2h2 . K2 = h2 (say). Rao [79]. i.J1 O.2h2 = A. A .
**Y. all basic potentials. de Veigy and S.e. Tchrakian [165]. In particular we shall encounter the Lame equation as the equation of small fluctuations around classical configurations associated with cosine. Rana [287]. M. This is an interesting problem. H.Q . W. A generalized associated Lame potential has been considered by A. Khare and U. for instance. MiillerKirsten and J. Thus. A. Cho. H.§ This is therefore a very important equation which in the limit of infinite period becomes a PoschlTeller equation.378 CHAPTER 17. W. thus revealing an unexpected significance of this not so wellknown equation of mathematical physics. H. also because the role played by the Floquet exponent in this problem is not yet well understood Jl Another recent appearance of the Mathieu equation is in the study of the mass spectrum of a scalar field in a world with latticized and circular continuum space. See e. Ouvry [276] and Jianzu Zhang. D. Gu and S. Shiraishi [49]. Ganguly [103]. i. Sukhatme [148]. J. See Chapter 25. S. N. A.g. § .** The associated Mathieu equation appears also in string theory in connection with fluctuations about a L>3brane (see Chapter 19). MiillerKirsten and D. 1 1 See Z. inverted double well and cubic potentials. in the problem of two parallel solenoids the lines of constant electromagnetic vector potential  A are elliptic with the Hamiltonian separating into a Mathieu equation and an associated Mathieu equation. A host of related elliptic equations has recently been discovered and studied. double well. Liang. Quian [123].W.5 Concluding Remarks The potentials we considered above will reappear in later chapters. .^ The equations considered above also appear in diverse new problems of physics. Periodic Potentials 17. J .Y. J. Kan and K.
f C . This seemingly simple problem revealed extremely rich internal structure . [19]. Bender and T. D. J.Q. in nearly a thousand of physics articles the problem of the anharmonic oscillator was touched in one way or another. H. MiillerKirsten and A. MiillerKirsten [163]. T. In the cases treated most frequently in the literature the anharmonic *We follow here largely P. W.. M. both of which make the calculation more difficult.* 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. W. Weinstein [281]..Chapter 18 Anharmonic Oscillator Potentials 18. ' T h u s A. Wu [18]. 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. Turbiner [273] remarks: "It can not be an exaggeration to say that after the seminal papers by C. Bender and T. demonstrates that derivations of such a quantity are much less familiar than calculations of discrete bound state eigenenergies in quantum mechanical problems. There is no end to this: An entirely new approach to anharmonic oscillators was recently developed by M. J. In particular the investigations of Bender and Wu. The recent work of R.1 Introductory Remarks The anharmonic (quartic) oscillator* has repeatedly been the subject of detailed investigations related to perturbation theory. Wu.". T.t which related analyticity considerations to perturbation theory and hence to the large order behaviour of the eigenvalue expansion attracted widespread interest. Priedberg and T. Achuthan. Wiedemann [4] and a revised version of parts of this reference by J. Lee [98] referred to it as a "long standing difficult problem of a quartic potential with symmetric minimd'1.
Case (1) is obviously the simplest with the anharmonic term implying simply a shift of the discrete harmonic oscillator eigenvalues with similarly . Anharmonic Oscillator Potentials oscillator potential is defined by the sum of an harmonic oscillator potential and a quartic contribution. 18. 18. *~z (1) (2) (3) Fig.1. which are nonetheless linked as a consequence of their common origin which is for all one and the same basic differential equation. V(z) = \\h*\z* + \\<?\z*. To avoid confusion we specify first the potential V(z) in the Schrodinger equation dz2 + [EV(z)]y(z) =0 for the different cases which are possible and illustrate these in Fig. described as the case of the double well potential. with the potential described as an inverted double well potential. (3) Complex eigenvalues with tunneling: In this case.1 The three different types of anharmonic potentials.380 CHAPTER 18. V(z) = \\h*\z2 I 2i 4 \cr\z . 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. and thus lead to very different physical situations. These contributions may be given different signs.
i. This type of potential allows tunneling through the barriers and hence a passage out to infinity so that a current can be defined. except for a change of sign of c 2 . (18. however with decay as a consequence of tunneling.1 Introductory Remarks 381 normalisable wave functions.18. We do not dwell on Case (1) since this is effectively included in the first part of Case (3). as is sometimes hoped.86) below. In this case our aim is to obtain the aforementioned complex eigenvalue. that the general applicability of the method becomes evident. Case (2) is also seen to allow only discrete eigenvalues (the potential rising to infinity on either side).e. The eigenvalues of this case are given by Eq. Case (3) is seen to be very different from the first two cases. The level splitting will be rederived from path integrals in Chapter 26. The imaginary part will be rederived from path integrals in Chapter 26. The result is an expansion in descending powers of h?. The shift of the eigenvalues is best calculated with straightforward perturbation theory. 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. and we present a fairly complete treatment of the case of large values of h2 along lines parallel to those in our treatment of the cosine potential in Chapter 17.175) below. however the central hump with troughs on either side permits tunneling and hence (if the hump is sufficiently high) a splitting of the asymptotically degenerate eigenvalues in the wells on either side which vanishes in the limit of an infinitely high central hump. In the case of the double well potential our aim is to obtain the separation of harmonic oscillator eigenvalues as a result of tunneling between the two wells. We begin with the latter. We therefore consider in this chapter and in Chapter 20 in detail some prominent examples and in such a way. If the barriers are sufficiently high we expect the states in the trough to approximate those of an harmonic oscillator. since the potential decreases without limit on either side of the centre. 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 eigenvalues of this case are given by Eq. 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. The boundary conditions are nonselfadjoint and hence the eigenvalues are complex. . (18. it will not be possible to obtain the exponentially small contributions with convergent expansions. Thus we are mainly concerned with the double well potential and its inverted form. and this behaviour is that of asymptotic expansions our treatment largely terminates this muchdiscussed topic. Calculations of complex eigenvalues (imaginary parts of eigenenergies) are rare in texts on quantum mechanics.
h 4 2 2h1> ~1/2 2c?.1 and more specifically in Fig.z Fig.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. 18.2 The inverted double well potential with (hatched) oscillator potential. 18.2. v(z) = ^h4z2 + ^c2z\ (18. and the Schrodinger equation to be considered is C L^ + [E + v(z)]y = 0. . 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 . We take here h = 1 and the mass rriQ of the particle = 1/2.1 refer to the two cases. The potential in this case is given by V{z) = v(z). we can pass from one case to the other by making the replacements: 4 E1/2 = 2Ei.3) "1/2 harmonic oscillator h8/25c2 . (18.2) We adopt the following conventions which it is essential to state in order to assist comparison with other literature. (18.1) for h4 and c 2 real and positive. 18. If suffixes 1/2.2 18. a point which has to be kept in mind in comparisons.382 CHAPTER 18.2. Anharmonic Oscillator Potentials 18.
5) becomes .1/7  Vg(w)y(w) = o(J^j.2.4) we can rewrite Eq. /i). (18.8) h8 v"{z±) = h and V(z±) = 5 2 2 c ' 2 2 Thus for c > 0 and relatively small. 4 . then to specify the necessary boundary conditions and finally to exploit the latter for the derivation of the complex eigenvalue. The result will be that derived originally by Bender and Wu. y{w) oc Di. (18.2. (18.2 The Inverted Double Well Potential 383 Introducing a parameter q and a quantity A = A(qr. (18.7b) The perturbation expansion in descending powers of h suggested by the above considerations is therefore an expansion around the central minimum of V(z) at z = 0.e.. and a variable w defined by setting E= b* 1 + A A and w = hz. 18.. _i\(w) and q = qo = 2n + l.7a) The problem then reduces to that of the pure harmonic oscillator with y(w) a parabolic cylinder function Dn(w). the harmonic part of the > potential dominates over the quartic contribution and Eq.. n = 0. (18.2) as Vq(u >)y(w) = with T)Jin\ 1 (A + c2w )y(w 0 (18. The positions z± of the maxima of V(z) on either side of z = 0 in the case c2 > 0 are obtained from h2 v'(z±) = 0 as z± = ± — with (18.. The problem here is to obtain the solutions in various domains of the variable. although our method of matched asymptotic expansions here (which parallels that used in the case of the cosine potential) is different. i.5) (18 6) "^'"dw* ' * 2' 2 In the domain of w finite.1.d2 .18. to match these in domains of overlap. \h?\ — oo and c finite. and h large the eigenvalues are essentially perturbatively shifted eigenvalues of the harmonic oscillator as is evident from Fig.
2 A (18.11) V = 0+ dz2 + 2qh + W The solutions in terms of parabolic cylinder functions are valid around and extend up to z ~ 0 ( l / / i 2 ) . Inserting (18. We observe — before touching the square roots in Eq. be and A = A(q. these solutions are not valid around z = 0. h) is obtained from the perturbation expansion of the eigenvalue.9) z =0 these Thus these c2z4 1/2 + • 2 (18.9) we obtain 1 h4z2 c2z4 A L2 (18.14a) 1/2 VA( ) = A(z)exp (18. This observation allows us to define the pair of solutions yA(z) = .10) E= qh2 ^ j. 1 d r A!{z)±iA{z)±{ 0. (18.4(2. as encountered and explained earlier. •A~2 hrz" 2^\ c^z V2 —— + (18. as we shall see.10) into (18.2.384 CHAPTER 18.+ ^ .13) + 2qh u2 A + W A{z) Later we will be interested in the construction of wave functions which are even or odd around z = 0.14b) .11) y(z) = A(z)exp ± z dz h4z2 4 + E hAz2 + cV y(z) = 0 (18. Before we return to solutions we derive a new pair which is valid in the adjoining domains.12) Then A(z) is found to satisfy the following equation r /iV c V V/2 A"(z)±2<^. (18. This construction is simplified by the consideration of symmetry properties of our solutions which arise at this point.2 We are concerned with the equation d2y(z dz2 where ^ 1 . In order to arrive at solutions we set in Eq.) exp Z ifdz[ifdz[ h4z2 4 h"z2 + + cV cV 1/2' (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.13) — that one equation (of the two alternatives) follows from the other by changing the sign of z throughout.
(18. 1 (18. 385 (18. One finds that these solutions are associated with the same asymptotic expansion for A and hence E (given by Eq. i.4 .. (18. /H 4 Z h +~C Z 21.20b) . we do not pursue their calculation.13) TzA'(z) T ^A(z) + qA{z) = O We define Aq{z) as the solution of the equation zA'(z)(ql)Aq(z)=0.2 The Inverted Double Well Potential with A(z) = A{z) and yA(z) = yA(z).16) For large h2 we can write the Eqs. / c z Aq{z)+0 1/2 .34) below) — to be derived in detail in Example 18.19) We see that one solution follows from the other by replacing z by — z. Clearly Aq(z).13) and A(z) that of the lower of these equations.e. (18.18.Aq(z) approximate the solutions of Eqs. Since these higher order contributions are of little interest for our present considerations. Thus we now have the pair of solutions yA(z) = exp yA(z) = exp dz I^4 + l„2„4 1/2.zh2 =()*2"(18.15) where A{z) is the solution of the upper of Eqs.1 and as a verification again in connection with the solution y# — as the other solutions. (18.18) M*) zf(9+l) A_ g (s) = (^2)1/4 exp ?/" dz (z ) / 2 1 2 (18.13) 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. (18. We take the square root by setting z2h^1/2 +.20a) These expansions are valid as decreasing asymptotic expansions in the domain \z\ >0 h. 1 2 4 2 Aq(z) + 0 (18.
ft = .v ' = o (« .. to obtain in leading order the eigenvalues E. 1 . 2 . c*i = 1 1 . (18.20a) and (18. W i t h proper care in selecting signs of square roots we can use the solutions (18. procedure. 128 8 16 A = i. „f 1 \ S o l u t i o n : We rewrite the upper of Eqs. Vq+ii = Vq + 2i. Anharmonic Oscillator Potentials i. 03 = . .^A(z) °° /1r2z2\i + J2 { fir ) \**A'(*) 1 + jA^Wl. A = . one observes that with £>. „ . we write > V±(z) = . h2.20b) to construct solutions y±{z) which are respectively even and odd under the parity transformation z —> — z (or equivalently q —• —q. a2 = 2 1 .! ) ^ .1) 2 2 2 A?(Z) " ^^^^^ = liq~ 1){q ~ VA"^ The lowest order solution A^ 0 ' = Aq (z) therefore leaves uncompensated on the right hand side of the equation for A the terms amounting to i ^ / 2r2z2 2z\ y 1 °° (2c 2 l i=3 Clearly one now uses the relation z2iAq{z) = Aq+ii(z). A A _ 3 .21) Example 18.2 . * = — . .e. away from the central minimum. h2. (18.13) in the following form with power expansion of the square root quantities and division by h2: 1 + (q .e. z)). X » / A q + 4 i = Mg+4i .f .. (18.. i.l)A(z) 1 = tfA"(z) A .1: Calculation of eigenvalues along with solutions of type A Use the solutions of type A. 04 = 5 .18) we obtain A » = H^rL^z) 2 z .2 ( * ) 2 and from this or separately A"(Z) : 1 (q . zA'(z) where the expansion coefficients are given by ao = l.:=*£ +i(5l).. (18.e. h2 —> —h2). z) ± yA(q. (18.386 CHAPTER 18.• Using Eqs. As always in the In this way Rq is expressed as a linear combination of functions Aq+4i(z).20b). i.17) and (18.[yA(q.20a).
22) are parabolic cylinder functions D i .5). Hence in the present case In its turn A^ leaves uncompensated terms amounting to A „ /2c 2 \ 2 9 „ M 2ft4 2 { 4/i2^ .—77 [i(9l)]. give an equation from which A is determined. Hence to the order we 0: {^)^) [{ 1){ 2 ^ ^ ^3)^ 0 +1 ^ + 2^ + 3)]+0 {h) 2ft6' A = 24 1 2 ^ + 1>2^+°G9A= ..(q + 3)(q + l)Aqvri(q '^ ' . (18. (18.w — > > ±iw) or functions R B w q( ) .Aq+4i in Rq 387 can be taken care of by adding to A(°> the contribution — _ ^ 4 ' except. and the coefficient of terms with i = 0.e..22) is invariant under the combined substitutions q — —q. i.. w = hz.2 The Inverted Double Well Potential Thus a term /j. of course.'2H9i) ^faDfr™) and . those in Aq(z). We return to Eq.. The solutions yq(z) of the equation Vq(w)yq(w) = 0. (18. when i = 0. r ( CJw) = ^ . Rq '. (18. c V + It follows that l) + 0 (i The same result is obtained below in connection with the solution of type B. 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. +1J±iw) (observe that Eq. In this way we obtain the next order contribution A^1' to A^°>. [1(9 + 1)]! .18. _1J±w) and D_\. = •>— r.23) . 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). ^f( g + i)(±H2^ + 1 > \ — 1.
27) Now.25) where in the case i = 2: SA{q. (18.q + 4j] = c2SA{q. Oberhettinger and F. Tricomi [86]. See Example 18. (18.26) The first approximation y(w) = y(°>(w) = yq(w) = Bq(w) therefore leaves uncompensated terms amounting to 1 4°) = where [q. (18. (q±2)(q±S). Erdelyi. pp. Vq+4j = Vq + 4j. As an alternative to Bq(w) in Eq. and for j ± 0 : [q. S2i(q.388 CHAPTER 18.q + 4j}yq+4j.28) (A + c w )yq(w) E 2 4 1  2 ^ [q.24) The extra factors in Eq. .±4) S4(9.3. j=i (18. l(q2 + l). (18.±8) 5 4 (g.Aj)(18.0). and so Vqyq+Aj(w) = Ajyq+Aj(w). Actually these factors can also be extracted from W K B solutions for large values of q.4j)yq+Aj.29) ' A . Magnus.0) = = = ±(q±3)(q±7).123. (18.( < ? .l)2/?4. we also have Vq+^yq+tj = 0. Comparison with our notation is easier if this reference is used.^ For higher even powers of w we write i w2l V<i = Y. F.23) have been inserted to make this recurrence relation assume this particularly symmetric and appealing form.q}=A + c2S4(q.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 + . 115 . since Vqyq = 0. 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 .3)y g _ 4 . W. (18. G.
Equation (18.e.^ g ( 4 g + 29) + O(^ 2 1 2 Qr2 2 4 2 / 1 \ J. Thus the next order contribution to yq is y H = ie 2^ —z.—v*+*j ( 18 . h ) = qh . and even powers of q in combination with even powers of 1/h2. (18.(!) («.35 ) .1. so that the entire expansion is invariant under the interchanges q . and c 2  replaced by — c 2 . (18. i.e.10) we obtain E(q. . p i(9»9 + 4 J'WK7». We can now write the solution y(w) in the form y{w)=yq{w) + Y.) + j/ 2 ) (W) + • • • with the corresponding equation from which A can be obtained.[rp ) Y.18.) ty^^w) to be a solution to that order we must also have to that order [q.) + j.• q.32) Evaluating this expansion and inserting the result for A into Eq. .2 The Inverted Double Well Potential 389 Hence a term (iyq+4j on the right hand side of Eq. ( 18 .—A(q + 1) . (18.q]=0. i. In Case (3) the parameter q is only approximately an odd integer in view of tunneling. (18. n — 0.3 °) For the sum y{w) = y(°)(«. .34) is the expansion of the eigenenergies E of Case (1) with q — qo = In + 1. (18. A = ^(q2 + l)c 2 + o(J^. This type of invariance is a property of a very large class of eigenvalue problems. 2 . h2 —> h2.34) We observe that odd powers of q arise in combination with odd powers of 1/h2.27) can be removed by adding to T/°) the contribution (—^/4j)y<j+4j.31) Proceeding in this way we obtain the solution y = yW („.
o. ? T 4 ] [ g T 4 . Our third pair of solutions is obtained from the parabolic cylinder functions of complex argument.37) For further details concerning these coefficients.11) is invariant under a change of sign of z. we may infer that given one solution y(z).38) [yB(z)]argz=7T = [?/s(2. and for j ^ 0 all other P0(q. H. (18. 2 4tPi{q. / l \ 1i i( ) + J2 ( ^e ) J2 P ^q + 4j)Bq+lj{w) iu=/iz. Wiedemann [4] . We observed earlier that these are obtained by making the replacements q —> —q.argz=0 (18.i 1 + 4j) in complete analogy to other applications of the method. > "p. 5 + 4j) = 0 for \j\ > 2% or \j\ > 2i + 1. q + At] (18.)]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.g±4] ±4 ±4 [g. We thus have the following pair of solutions VB(Z) VB(Z) = B w . Achuthan.e. q + tt)= J2 i=2 p il(?> Q + ±3 + 4*)[q + 4j + 4t.—. J. W. w — ±iw. their recurrence relations and the solutions of the latter we refer to the literature. ^VO(<L<?) ^i(?. there is another solution y(—z). Anharmonic Oscillator Potentials ±4 ±4 ±8 ±4 P2{q.q±A) = MM!i±MM+[M±8][g±8. Again we can write down a recurrence relation for the coefficients Pi{Q. q + 4j) = 0. q) = = 1.36) with the boundary conditions p o(q." Since our starting equation (18. g ± 4 ] T4 ±4 and so on. MiillerKirsten and A. i.390 where for instance v J CHAPTER 18.
J. Thus in the transition region some become proportional. like w of Eqs. with normalization since the normalization constants are also asymptotic expansions.40) "Variables like those we use here for expansion about the minimum of a potential (e.The solutions yc. (18. h2 — — h2 as long as corrections » resulting from boundary conditions are ignored.2 The Inverted Double Well Potential These solutions are therefore defined by the following substitutions: VC(Z) = [yB(z)]q^q. 18. in fact.h^ih 391 (18.Vc suitable asymptotically decreasing expansions in one of the domains \z\ < O 7T argz We emphasize again that all three pairs of solutions are associated with the same expansion of the eigenvalue E(q.2 _4 A } 2cVl1/2 h4 8c 2 1 in the solutions of type A which are not valid around z = 0. one has to stretch each by appropriate expansion to the limit of its domain of validity.z2 .18.2. In order to be able to extract the proportionality factor between two solutions.5) are known in some mathematical literature as "stretching variables" and are there discussed in connection with matching principles.h^ih.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. see e.4)).g. 4 . (18. In this bordering domain the adjoining branches of the overall solution then differ by a proportionality constant. Aq(z)) does not appear in any of the higher order terms. we obtain: exp 8c2 3 < 2cVl3/2 /i4 h2z2 „(zA exp 12c2 (18.** First we deal with the exponential factor 1/2 exp exp dz z z h* + c z J2c z 2 2 . . We add parenthetically that all our solutions here are unnormalized as is clear from the fact that the function in the dominant term (e.q + 4j) as in ys. the solutions of type A being valid away from the minimum. Such contributions arise.g.39) are with the same coefficients Pi(q. h2) in which odd powers of q are associated with odd powers of h?. Vc(Z) = [V!B(Z)}q^~q.g. so that the eigenvalue expansion remains unaffected by the interchanges q —> —q. Integrating and expanding as follows since h? is assumed to be large. Mauss [192].
Anharmonic Oscillator Potentials Considering the pair of solutions ju(.) ( ^2 „22)\ z £(</!) e 4re z 4 4 = hz: [i( 9 l)]!24(«D l+O h? (18..44) In the solution y~g(z).392 CHAPTER 18. eh* /12c* e\z*h? (18.7 r .42) we obtain for the solution ys(z).l ) ] ! ( . Comparing the solution VA{Z) of Eq. y^(z) we see that in the direction of z = 0 (of course. Thus from the literature [86] we obtain Di_{q_1){w) = w*{ql)e\w* but D Wi) ^=.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. with z in ys(z) replaced by — z.41) with the solution ys(z) of Eq.z). ^Jw) has a similarly complicated expansion for 1 5 — — > argw > . (18. . We do not require this at present. > n.ir_I + M + 1)]! (2w*y U i\[~W • i 5 with 7r > argw.42) H W2 (ql)e$W I „ .w yB(z) ~ B. The function Di.45) .43) (since z — — 2 implies > argz = ±7r). we would have to substitute correspondingly the expression (18.4 i . (18. not around that point) VA(Z) VA{Z) = ehVl2c 2 e\z*h* z^) + o(± zfr+i)+o(±\\.43) W + 1)]! w^+V A . 4/T 4 Prom (18. 2 ) i ' ir! i iHim 1 3 j argw < 7T.(«.0x ! [ £ ( g . 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. (18. 2 ^i![i(g4il)]!(2«.2 « .
2.2 — 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. We proceed similarly with the solutions yc(z)..5.1 ) e 12c 393 Kgl)]^"1) l + O h? (18. i.9.18. 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.11. hence we have to . (18.46) However.41). — For instance qo = 1 (or n = 0) implies a ground state wave function with the shape of a Gauss curve above z = 0. (18. Looking at the potential we are considering here — as depicted in Fig..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. the ratio of yA(z)iVB(z) i s n ° t a constant. we see that at the origin we have to demand the conditions yV(0) = 0 and y_(0) = 0 (18.yc{z)pansion (18. (18. At z = 0 the solutions of type A are invalid..e. we see that in their common domain of validity yA(z) where a = =Vc(z). ^^+i)Ah^ze* [_i( g + l)]!24(9+D Inserting the ex l +O h2 (18.50) will be seen to imply qo = 2n + 1 = 1. 18.50) and y+(0) ^ 0.2 The Inverted Double Well Potential with a 2 (K ^ ( 9 . Recalling the solutions y±{z) which we defined with Eq. large probability for the particle to be found thereabouts.7.y'_(0) ^ 0.21) as even and odd about z = 0.42) into y~c(z) w e obtain Vc(z) = (h*.49) d yc{z) Again there is no such simple relation between yA(z) 18.47) yA(z) Comparing this behaviour of the solution yc(z) with that of solution of Eq. and the second qo = 3.48) (h2y 4(«+i) [i((? + l)]!24(^i) l +O /i2 an (18. The first of the conditions (18.
[ .54) dio q(w) .u=o follow with the help of the "circuit relation" of parabolic cylinder functions given in the literature as (q r — ) Di(.1)('») + ^ r + l)]! ^.(0) = . with the reflection formula (—z)\{z — 1)! = IT/ simrz.F ? . Then imposing the above boundary conditions we obtain 0 = y'40) = limJ[y'A(z)+y'A(z)] = i»i..394 CHAPTER 18. y'B(0). y c ( 0 ) .^{(9+1)}.(«. [C. we obtain i V7T24( g .s i n { .2. + 1)]!' D ' j .(0) = and htoi)(°) ^—'.58) diu [  ( g . one finds that [ 1 IT (q + 1)1! = .i ) . (18.(0) + ^ ( 0 ) l z—>0 2. IT 4 dio In fact. 1 [7(93)]! n «(°) = . A. e.( g + 3)}. M.5 ( 9 + !!)(*"')• i ( 1) . ..) [i(«l)]![i(?+l)]! 2TT V5F[i(gl)J! V (18.55) Solution: Prom the literature.i)H = e.fc(0) Thus we obtain the equations (18. Stegun [1].. Anharmonic Oscillator Potentials use the proportionalities just derived in order to match these to the solutions valid around the origin.SxW) = \ i»fl(0) .i ( g + 3)]l (18.g.(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)}.(0) (18. Example 18. We leave the detailed calculations to Example 18.51) and 0 = V(0) = Jim i»xM .56) D 4(. ( 9 + 3)]! s i n { .oi(.l ) ] ! [ .53) yc(0) a' Clearly we now have to evaluate the solutions involved and their derivatives at the origin.( g + 3)}..(iu)].! ( . 4W >S [ I ( ( ? _ 3 ) ] ! s i n { f ( g + 1)} and hence =^= 1.57) [i( 9 _l)] ! 2 i(9i) B.52) # 1 = ^ y' c (0) a and 2/B(0) = a (18. = = V^ >—= = — i i ^ i [1(93)]! ^_ sin{_(q + i)} (18. Abramowitz and I./o^Tpif (q+i) . Expressions for C 9 (0). C."*(«!) ^ ' ~ [ .' '.2: Evaluation of yB(0).l)(°) Thus with the help of the reflection formula cited above: D B. Z (18. — C.
53) in dominant order and insert the appropriate expressions for a and a from Eqs.+i)(°) [ _ I ( .1/2)!.(0) = *>i(„+i)(0> 2 .56)._l) From this we derive C. + i)]!2i(«+ ) 1 [(«3)]! V5F[i(9l)]! sin{(q + l)}. (18.62) . (18. 7T 4 V 7T 4 (B) Evaluation of the boundary conditions We now evaluate Eqs.sin{^(g+ 1)} = J .18.60) (1861) [C>)]o = iV2i [1(9 + l)]![i(9 " 3)]! (18. we obtain 2 i(»+i)[_i(. ~ ^ 1 5(9+1) 2i(''. W .46). 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) _^_ .)[i(g3)]! and ^i(.i .3)}. + i)] V* H(?+l)]'[£(9l)]l Similarly we obtain [°'i(.6 ) } ^ — i t a n < — (q + 3) >. (18.2 The Inverted Double Well Potential From this relation we obtain 395 ^^^i^w^vy ^(9i)(°) (18 59)  Inserting from (18.49).sin{^( 9 . _ l)]!2l(91)(/J2)(9+D We rewrite the left hand side as 1 sin{f(g + 3)} r„.+1iH]«=o = . 4 r T (18. isin{f(g + 3 . using the above reflection formula and the duplication formula ^/TT(2Z)\ = 22zz\(z . Starting with the derivative expression we obtain (apart from contributions of order 1/h2) lsin{f(g + 3)} i Sin{(g .3)} (h?)faV[\(q [Uq U + l)]\2*(q+V _£ e 6c^.i ( « + i ) [ _ i ( .
.1).. this is the case of the purely discrete spectrum (the differential operator being selfadjoint for the appropriate boundary conditions. . equivalently.1.5._ e =*. we again obtain (18. (18. (18.64) In each of Eqs.. . This is Case (1) of Fig. ^ .63) and (18. Anharmonic Oscillator Potentials Then the derivative relation of Eqs. by the rotations E > einE = E.63) Proceeding similarly with the second of relations (18.^ >^_>m e J*..9. . .. With a Taylor expansion about go the left hand side of (18.53). (18. (18.. 7 . In fact the left hand side of (18.9o) ^ cos  . z2 > z2. and the left hand side of (18.5 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.go) ^ (1) It follows that we obtain for the even function with q — go = 1.5.9. (18. 1 1 .65) Expanding similarly the left hand side of Eq. Our next task is to extend the solution all the way to the region beyond the shoulders of the inverted double well potential and to impose the necessary boundary conditions there. i.64) for q = g0 = 3..64) the right hand side is an exponentially small quantity..63) becomes (? .L ^ i . The analytic continuation of one case to the other is accomplished by replacing ±c 2 by =Fc2 or. (18.2.64) about go — 3.i .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.. Thus we have to determine these conditions first.. we obtain cos((g + 3)) = .53) becomes 8 m(I(.11.e. 7.^ . the energy E is real. 18. Recall > the original Schrodinger equation (18.396 CHAPTER 18. z^ ein/2z. + 3)} = . 18. the vanishing of the wave functions at infinity).( g 0 + 3)  + • • • ~ (g . (gg 0 )^±^ y e5?. ..2) with potential (18.63) vanishes for q = go = 1. For c 2 < 0 and the solution y(z) square integrable in —oo < ?Rz < oo.
6 6 In the case of the inverted double well potential under consideration here (i.cos 36.e. 1/2. 2 c2>0. or — < 6 < . we therefore demand that for $lz — +00 and • —7r/3 < arg z < 0 the wave functions have decreasing phase. i.68) . r / c 2x 1/2^3 = exp{ ± i — —\. + i s i n 3 6 0 i y ) ~3~^ c2\l/2\z\3 oc exp <M — J — .e. It is then necessary to insure that when one rotates to the case of the purely discrete spectrum without tunneling.sin 3(9 This expression vanishes for \z\ —• +00 if the angle 0 lies in the range — ir < > 36 < 0. (18.67) Rotating z by vr/2.18. (18. Thus exp exp (<£ff} [+<£ )%} sin * 0 for $lz —s.2 The Inverted Double Well Potential 397 One can therefore retain c2 as it is and perform these rotations. Thus in our case here the behaviour of the solutions at infinity has to be chosen such that this condition is satisfied. the case of complex E). i. replacing sin 30 by " 0 + ^) 1 IT = .66) z3 = \z\3em. y{z)~exv\i[) 7T dV'V. 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.e. we set z w r = \z\eie. Now. if TT/3 < 6 < 0. i. in the domain — TT/2 < 36 < > 7r/2. Then c 2 y/vi ^p^My) yf = exp _ r HvcV2!*!3 cos36.e. —}. the resulting wave functions vanish at infinity and thus are square integrable. we see that the solution with the exponential factor is exponentially decreasing for \z\ — 00 provided that cos 39 > 0.+00 in arg z € (!•" * 0 for 5ftz — —ex) in arg z € » N> (18.e.
we match the WKB solutions .e.Z\.3 The inverted double well potential with turning points ZQ.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.21) to + infinity and to demand that they satisfy the condition (18. We do this extension with the help of WKB solutions. 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. (B) Evaluation of the boundary conditions The following considerations (usually for real z) require some algebraic steps which could obscure the basic procedure.68) for c 2 > 0. and there impose the boundary condition (18. i.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. Eq. 18.68) on y±(z) (by demanding that the coefficient of the solutions with other behaviour be zero).21)) were defined in terms of solutions of type A which have a wide domain of validity. This is the boundary condition also used by Bender and Wu [18]. 18. This is not the asymptotic behaviour of a wave function of the simple harmonic oscillator. Anharmonic Oscillator Potentials V(z) Fig. Our procedure now is to continue the even and odd solutions (18. (18. We therefore explain our procedure first. Our even and odd solutions y±(z) (cf. Looking at Fig. for z > Too.Z\. For c 2 < 0 we have correspondingly y(z) ~ exp I ± ( — 1 — y c2 < 0. We have to remember that we have various branches of the solutions y(z) in different domains of z. Equating to zero the coefficient of the term with sign opposite to that in the exponential of Eq. (18.398 CHAPTER 18.
70) where z < z\. 291. t To the right of the turning point at z\ these solutions match on to (r. Dingle [70]. f R .z\) ( \ y^NKQyz) qh2 z 4 dz h H—c z 2 x sin ^ . Vol. z\ ~ .=c z 4 dz 1 L2 9 + 2 4 1/2 x exp . Dingle [70].*l) Z 1 2L4 1 qh? + zzh* 1/4 _2i4 VWKB\ ) . In using these expressions it has to be remembered t h a t the moduli of the integrals have to be taken. z2h4/A > c2z4/2.e.4.71) We now come to the algebra of evaluating the integrals in the above solutions. equations (21).B. (22) or A.^ 4 1/2 + icV 7T Jz + (18.70). *R.zi)/ \ 1/4 ywKB\ ) z — g/i* .4 h + c z 2 1/4 < . (18. I [195]. i. The distant turning point at z\ as indicated in Fig.e. 18.69) The W K B solutions have been discussed in Chapter 14.2. 291. and then use the W K B procedure (called "linear matching" across the turning point) to obtain the dominant W K B solutions beyond z\. p. (18.^ T rz\ 1 + yh* . 4 2 T ' i. 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. B.4) for E and ignoring nondominant terms) t o .2 The Inverted Double Well Potential 399 to the left of z\ to the solutions of type A.V 2 iL44 + <?z / „2 4 X COS < f / il/2 G?Z g/l 2 1 2L4 z 4 M 1 2i..3 is given by (using Eq. Messiah. A J24 1 ? 2 • "^ •z h \—c z ~ — q h . p.18. Sec. We begin with the exponential factors occurring in Eqs. h2 (18. 1 2i4 1 2 4 / ~ 2 4 ~ 2 C * (18. 6. ^ 2 4 •K + —(r. ^2? 1 2qc2 /2? . .
2.i c V 8c T 2 1/2* ti>_(2 ~ exp 8c \3 2 r< 2cV 2c2 z2 i24£Y'2u h* 3/2 >. (18. Thus the above factor yields '2c2\l/2 h4 so that (cf. We obtain this factor by expanding the expression in powers of zjz\ (since in the integral z < z\). (18.. Eq.40)) E± = exp _/^_2 f ± In A 2c2 . (18.2 \ 1/2 .2.4 In 2c 2 J + 1/2 ^1 2c2 1/2 21 and hence (with use of Eq. In the remaining factor we have (looking up Tables of Integrals) 21 2 dz h 4 2 2c. CHAPTER 18.69)) rzi J<z z dz Z[&Z*]W = + ^Y /2 m '"U4 1/2 2d2 + 2c2 Since we are interested in determining the proportionality of two solutions in their common domain of validity we require only the dominant 2dependence contained in this expression.^ I 2 2 1/2 zV = F 8c2 Jz [i . Anharmonic Oscillator Potentials E± = exp = exp .73) . zi 2q 1 c 2 [1 .2^3/2 2_(h^\ll2 2czz 4 N 1/2 8?3\ 2 ~hf ' =F9/2 ^ 2 ^ .e.72) Here the first part is the exponential factor contained in 2/A(Z)) J / A W respectively (cf.40)).\qh2 + \z2h± .211/2 1 \ /j.±9/2 = ^ V M z^l2 21 2c2 exp ± i dz 1 4 1 1 V2 2 (18. 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 . (18. .^ i ^ ^ c 2 .400 i.
\z2h4 + J Z\ nl/2 \M fZ .20a) and (18.71) and these into (18.77) Now in the domain 2 — oo we have > \qh2 .21) we now have V±(z) 1 2 \yA{z) ± yA(z)} = \W^(z) WI ± _(J. we see t h a t in their common domain of validity VA (z) = /3ywKB 0)> where .18.78) Inserting this into the solutions (18.1 ) 9 / 2 _(2c 2 ) 1 /2_ 1 (18.76a) P P r 2/1 2 " 9 ( . In these expressions we have chosen the signs of square roots of h4 so that the conversion symmetry under replacements q — —q.79) +S_(±)exp . > » Returning to the even and odd solutions defined by Eqs.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 ) ] .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) . h2 — — h2 is maintained.75) 9/2 1/2 0= or '(2c )V2 2 and (3 = — 2 2 (fr2) (2c 2 ) 1 /2_ 9/2 (18. (18.21 1/2 .*l) (3y^(z) (18.Zl) ( \ yWKBV^J \z h* 2 1/4 and ywKB(^) \z h* 2 1/4" (18. CZ2 JZ1 cz 3^2 V2 h6 12c 2 ' Ac2z2 cz" 3y/2_ (18.*l) (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.74) Comparing these solutions now with the solutions (18.(J*i) VA 0 ) = PVwKB (Z C.20b).
65). i.h2) + {qqo)(^pj = E(q0..80) Imposing the boundary condition that the even and odd solutions have the asymptotic behaviour given by Eq.76a) for (3 and /3. (18.e.85) + . this equation can be rewritten as or as the replacement + / u6 \ 90/2 (/i 2 )W 2 =^(_) 2®"1 f — J . (18. i. Eq.82b)) /j6 \ 9o/2 («*) = W? mam ''*• with qo — 1./ ? = 0. h2) = \qh2 . (18. Inserting this into the latter equation we obtain (the factor "i" arising from the minus sign on the left of Eq.84) Expanding about q = qo we obtain £(g.e.h2) + (qqQ)h + (18. E(q./i2) = E{qQ.81) Inserting expressions (18.2..^ j . (18..^(q2 + 1) .6 The complex eigenvalues (1883) We now return to the expansion of the eigenvalues..68). (18.34).5. (18. (18.82b) This is our second condition along with Eq. Anharmonic Oscillator Potentials where 5+(±) 5_(±) = = Q / ? ± ^ e x p ( ^ ^ T ^ ) e x p ( . (18. 18.3. we see that we have to demand that 5 + ( ± ) = 0.. i / 3 ± .^ § ( 4 g 2 + 29) + O ( ^ ) .402 CHAPTER 18. .
3.4 1 2 4 1/2 ft + i c V 2 1/4 + f}' z.86) will be dealt with in Chapter 20.~p{q. ^ 2 4 y qh 2H ""{f' z h H—c z 4 2 1/2 1 ^ 4 + IC2Z4 dz qh2 4 2 .g.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 . !*WKB(Z) ~ 2 .36) of their Ref. 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.h6/6c2 E = E(q0. [19]) for ft = 1 and in their notation 90 = 2 ^ + 1. zo} means "to the right of the turning point ZQ" . In the above we did not require the matching of WKB solutions at the lower turning point ZQ to the solutions yA(Z).18.2 . Since this is not without interest (e. S W K B ( 2 ) = PVA(z). where the superscript {r.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.yA{z). above) (1. in comparison with the work of Bender and Wu [19]) we deal with this in Example 18. The large order behaviour of the eigenvalue expansion (of Case (1)) which Bender and Wu derived from their result (18. Example 18.l ) ] ! > + = ^ 2 .ZQ) . i. h2) of the relations KWKB (Z) =PVA(Z).h2) The final result is therefore E () i ( 2 7 T ) V 2 [ l ( g o . o 2 + l)^(4. * _ qh2 1/4 *• 2 j _ 4 i 1 2 4 z h H—c z dz qh2 2 1 2L4 •{/.g ( . h2). ft6 ^ = e.e. Solution: The turning point at ZQ close to the local minimum is given by qh2 2* 4 z2hA c z ~ 0.3: Matching of WKB solutions to others at ZQ Determine the proportionality constants p(q.
1/2 h? \z{z zlf' \z \n\z 2 2 2 + {z2z2Y/2\ 2q h2 4 • In (V2zh V s/q Then to the same order of approximation ." / 4 hj (ft2 2 2)( 9 +l)/4 J'W'KB (Z) 2\1/2(fe222)^1)/4 h) exp(±h2z2) ^2ey/4 * Solutions of this type (which are asymptotic forms for q — 00) have been investigated in t h e > literature. and we have 1/2 dz h?_ 2 J /i2 qh2 + iz2h4 _ ^ 4 J dz zn qh2 + z2^ 1/2 . 2 z 2 ) / 2 e \ .404 CHAPTER 18.. Schwind [247]. 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. f x Zn 4) h2 2. 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. .] { hz combined with the particular constants contained in y ^ ' ^ g . J ^ K B match on to the W K B solutions with exponential behaviour to the right of 20. Anharmonic Oscillator Potentials where zo > z.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. N.zo)/ \ qh2 2 ZO + z2hA 4 2 .zo) % ' K B (Z> 1 / 4 2 l / 2 h z V Q / 2 7i/2 4 2 e x p ( ^ e . cf./ ^ 2 X 1 / 2 exp(ifi. the (approximate) solutions* 1/2 exp • ± i ( .ic224 2 4 qh2 + i^fe 4 . Expanding the square root occurring in ^ W K B W ' ^ W K B ^ ) ' Igft2_I22h4+lc2z4 2 4 2 and we see that 1/2 = 7 ^ 1/2 2.O.
as in the case of the Mathieu equation. (j . (18. J/J Hence ^ and h so that \]q] \2"'2 ~ ( 2 7 r ) 1 / 2 e .^ 1 .z ± ) = .41) for yA(Z). in fact. in Eq. (18. ~p requested at the beginning are given by /1y/2[j(gi)]i2^c^ h 6 /12e3 /ly^ih+iwi"" w 18._ .g.18.13) of Bender and Wu [19].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.(r.23)) appear quite naturally.3. We consider the following equation ^f& with double well potential V(z) = v{z) = jz2h4 + ^c2z4 for c2 > 0..DfJ»v we see Comparing now the expressions (18.5 "l _ o f .zo).3 18. „ / M 1 / 2 i .kyil(.1)! = ( 2 7 r ) 1 / 2 e . Then .J ' i J . We can reexpress these relations with the help of Stirling's formula.87) The minima of V(z) on either side of the central maximum at z — 0 are located at h2 h8 z± = ± with y(. h4 > 0.3 The Double Well Potential 405 It may be observed that 3/Y/KB ( z ) corresponds to solution (3. (18.88) + [EV(z))y{z)=0 (18.89) ." / 4 ( ? « / 4 . 4" ) (h?z2)^q+1) for q j£ odd integers. i. we shall employ basically the same procedure as above or. The last expression can be made acceptable for q = odd integers by applying Stirling's formula in the denominator of the second of the previous pair of expressions.e. But there are significant differences. In this form the WKB solutions reveal their similarity with the solutions VB i VB > Vc i Vc a n c ' demonstrate that the factors which we inserted (e.yA{z) with expressions 27WKB i J'WKB ' that the factors p.
94) (18.95a) . (18.406 CHAPTER 18.z±).91) becomes T>q±(w±)y(w±) = o(^)y.90) V(i)(^)=0.93) w± = h±(z .91) (18. the previous equation (18.92) (thus we sometimes use h4 and sometimes Ii±) and EV{z±)=lq±hl With the further substitution + ^. . 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 .4 The double well potential.z Fig. (18.V(z±) l{z z±fhA + 0[(z . 18. (18. Anharmonic Oscillator Potentials and V{2\z±) V(4)(z±) = = h\ 12c2.z±) y = 0. vW(z±) = ±Qch2. (18.i>5.
where U{z) = and near a minimum at z± 4_ 4 jp[V(z)V(z±)].97a) U(z) = (zz±)2 + 0[(zz±)3}.96) dz2 + 5 « 4 +  .100) . Inserting the expression (18. we obtain another by replacing z by —z. (18.98) Evaluating the exponential we can define these as the pair yA{z) VA(Z) = A(z) eyip = A(z)exp + —z V2\3 4c'' Ac' (18.2 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 (18.99) 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. given one solution... (18.95b) with the equation of parabolic cylinder functions u(z) = D„(z).96).97b) Our basic equation.95a) and (18.3 The Double Well Potential where ^ 407 d2 •« 2 2  d8..93) for E into Eq. (18.1. d2u{z dz2 + v+ 2~r u(z) = 0. i.95b) By comparison of Eqs. Eq. (18. qo = 2n + l.18..3.e.i 4 f W » = o. (18. we obtain d*y (18. Thus again. (18.87). we conclude that in the dominant approximation q± is an odd integer. n = 0.2. 18. is again seen to be invariant under a change of sign of z.
z) may be obtained from the solution yA{q. z) = yA{q. ~h2.1 6 ) . We leave the calculation to Example 18.105) "8 + 4 ^ .yA(z) are associated with one and the same expansion for A and hence E.^ A"{z) + 2c V2 *A(z) (18.101a) {z2+z2)A\z){qz+ + z)A{z) = ^ A"{z) + £A{z) (18.103) Looking at Eqs. z). (18. but the solution is a different one. (18.101a). Anharmonic Oscillator Potentials and selecting z+ (z2+z2)A'(z) h z+ = —.h2) .4.101b). The result is given by A and E(q.c W + 1) . Vq = {z2+z2)— + (qz+z). yA(q.102) yields the following expression M*) I z2 _ z2 11/2 *+ z + z+ q/2 z+ 1(91) \z + z+ 1(9+1)' (18. (18. h2 h4 2c' with g+ = q.106) . (18. i. Aq(z) = Aq(—z) = Aq(z).e.102) We observe that a change of sign of z in this equation is equivalent to a change of sign of q. these equations can be rewritten as + (qz+z)A{z) = . n± = V2h2. z) = yA{~q.104) Both solutions yA{z). (18. h2.h2.^q(l7q2 6 4/i + 19) + (18. i. The dominant approximation to A is then the function Aq given by the solution of the first order differential equation VqAq(z)=0. we observe that the solution y^iq. h2.408 Since CHAPTER 18.z) by either changing the sign of z throughout or — alternatively — the signs of both q and h2 (and/or c).c W + I ) 2C yfi 2h 5 2 A V2c* g(17g2 + 19) + 0 ( / t . Integration of Eq. 8hw (18. h2.e.101b) To a first approximation for large h2 we can neglect the right hand sides.
First. . . .106).18.3 The Double Well Potential 409 Example 18.114a) . + Aq+2 + Aq Aq+2 . .109) We observe some properties of the function Aq{z) given by Eq.4: Calculation of eigenvalues along with solutions of type A Show in conjunction with the derivation of solution y^ that the leading terms of A and hence E are given by Eqs.102). <4q+2 — z — z+ Aq+2iAq+2j (18. Similarly we obtain _z+_ Z _ Aq + 2 — Aq Aq+2 + Aq 12 ^g+2 A ± ± _2A 2 ± 6 + (18. (18. (18. as a linear combination of terms Aq+2i. we reexpress A'q in terms of functions of z multiplied by Aq. Solution: The structure of the solution (18. Thus it is natural to explore analogous steps.. The first such step would be to reexpress the right hand side of Eq.101a) with A replaced by A . J = 0.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 . such as periodic potentials. (18. ± 1 . for instance by componendo z z et dividendo.z\) + 2z(z .105) and (18.(l + + 2  z y* . however. (18.111) {Aq+2  Aq2)2 (4zz+)2 {z2  2 .{z2 .113) 2 .4zz+q+ z\{q2 + 1)].qz+)2 . z • .+ 2 ^ ± l + 2^±i + .103): . We know the first derivative of Aq from Eq.108) We wish to rewrite this expression as a sum y coefficientiv4. i.2 z\) • Aq$4 £Aq + Aq—4.112) From these we obtain.110) Aq+2+Aq2 9+2 A.*l) (* ~ "+> ) A (z)exp q /2 \ 3 4c J (18. = Aq+2i+2jAq.z2 ) N t.qz+) [2z2 . . (18. A Differentiation yields » = ^^w (18.103) for Aq{z) is very similar to that of the corresponding solution in considerations of other potentials. ± 2 . (18.Aq 1 + 2^± 2 Aq+2 (^M^f (18. 92 ZZJf (18. .e.107) KV) ~ Aq(z) {z*z\y Aq{z) (z 2 z\f (z .
Eqs. . In particular the dominant approximation of A is obtained by setting (q. (18.3 ) A .115) (z*z%) 2 i2 (z 2 .112) CHAPTER 18.2A g + A g _ 4 ] .l ) ( g . (17.<? +{q.aAg+S 12+ 16 (18. Since VqAq = 0.z\Y Z2 + {~)\Aq+42Aq 1 224 Aq. ' T h e reader may observe the similarity with the corresponding coefficients in the simpler case of the cosine potential.117) 4(q + l)2Aq+2 Here the first approximation of A. leaves uncompensated on the right hand side of Eq.e.2\2 (z* . A(°~) = Aq. (q.9=F4) = (g=Fl)(?=F3). <?) = 6(<j2 + 1) + z: 2 2A +72"' (18.4)A„_ 4 + (q. •••].101a) the contribution R . (q. we obtain 5 K = ( ^ f ) [ ( 9 . q + 4)Aq+4 + (q.4A ? _2 + 6Aq .116) into Eq. (18.119) It is now clear how the calculation of higher order contributions proceeds in our standard way.l ) 2 A .33) and (17.q + 2)Aq+2 (18.2 ^9+2 „^9+4 + \4z \4z+J z+ Finally we have also [A q _4 — 2Aq2 + 2Aq+2 — Aq+4 (18.116) [Ag4 .108).2Aq + Aq4] 1. where the lowest coefficients have been determined above as (9. Anharmonic Oscillator Potentials A 1 . q) = 0.4 ( g .112). (18. (18. cf. QA?+4 _ 10 A + 6 .z%) 1 n z+z 2 \4z+J [Aq+4 [Aq+4 . _ 2 + 6 ( 9 2 + l)A ( ] + (q + l)(q + 3)Aq+4.4 4+* + . (18.115) and (18. Vq+2iAq+2i=0 and Vq+2i = Vq + 2iz+.+6 + • • • ] . (18. .4]j.q2)Aq2 + (q.118) 2s/2c ~^j[(?.q + 2) = 4{q + l ) 2 . _ 4 .q + 6)A<. i.(0) V2 ' 2c K + h4+Ao .410 and from this = Hence with Eq. 1_4^±^+8^±±12^±^ +.q)Aq + (q.114b) .4Aq+2 + Aq+A Inserting (18.34).
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. of course.105).(0).0) = y^(q.4 ) ( 0 ) ( g . (i) _ / V2c\ [ ( g . in the domains away from the minima. g) 8z 4 .g) 4z+ {q q)+ 2r (9. i. y'+(q.h2. (g.92) ig4 ' H ~ 23/X4 iz+ An2 H + (9.0) = ^ .g) Az.Aqjr2i in this may therefore be eliminated by adding to A^0' the contribution A^1' given by (9.124) . The first approximation A(°> = Aq leaves uncompensated on the right hand side of Eq.121) This coefficient of Aq set equal to zero yields to that order the following equation 0 £) ' {^) (g.{^ h2.123) ^[yA(q. (18.0)=0. for i = 0.h2. (18. g + 4)(g + 4. h2. g .h2.g) 8z+ (g. \ZZ±\ > 0 ( y We can define solutions which are even or odd about z = 0 as y±(z) = = £ fe^te' h2> z) ± y A.2 ) (0) (18. 9 + 4) ^+48z+ (18.z)±yA(q. thus yielding the next approximation of A as given in Eq.yA(z) derived above are valid around z — 0. h2 2y^ Age? h4 (18. A' (0) = q/z\) y+(q.101a) the contribution Rq . where Rq is the sum of terms left uncompensated by AW.28. y'_(q. *)] (18.g ( 1 7 g 2 + 19). 9 + 2) Aq+2Az+ (9.3 The Double Well Potential we have T>o 411 2iz+ J Aq+2i except.g2)(g2.h2.94) 8z+ (9. i.h2. which reduces to 2 \/2c 2 0 = 2(3g 2 + i) + — A + ^ g .122) The solutions yA(z).e.0)=0. g .e. z)\ = 2^A^ h2 '>z)± VA(Q.94)(g4.z) Considering only the leading approximations considered explicitly above we have (since Aq{0) = l/z+ = A. h2.g + 2)(g + 2. (18. Terms jj.
Evaluating the exponential factor contained in VA{Z) of Eq. we have w±(—z±) = —2h±z±.We draw attention to two additional points. Inserting (18. Since w±(—z) = —h±(z + z±). (18.125a) = VB{~Z) = Bq[w±(z)] + 0(h±2) = Bq[w±(z)] + 0(h±2) (18.^ % [i(?l)]!24(«.3 M a t c h i n g of solutions Next we consider the proportionality of solutions yA{z) and VB(Z).412 CHAPTER 18.yc(z) with complex variables and Cq(w) given by Eq. (18. in this case we use the Schrodinger equation with the potential V(z) expanded about z± as in Eq. Moreover.95b) that the solution there is of parabolic cylinder type.yB(z) is obtained around a minimum of the potential.95a) and (18. yB(z).95b) — 1 T>q(w±)y(w±) = j£ Thus we write the first solution yB(z) = Bq[w±(z)] + 0(h±2). (18.91). 18. Anharmonic Oscillator Potentials Our second pair of solutions. but w±(z±) = 0.125b) It is clear that correspondingly we have solutions yc(z). the equation is — with differential operator T>q as defined by Eq. we > . and another VB(Z) ± 25^ch3wl + c2wi .126b) These are solutions again around a minimum and with yB(z). (18.3.23) with appropriate change of parameters to those of the present case. 2(9+) .2A y(w±).93) and setting w± = h±(z — z±).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)). in view of the factor "i" in the argument of Cq the solutions VAJVC have the same exponential behaviour near a minimum.99) for z — z±. Thus yc(z) Cq[w±(z)} = = Cq[w±(z)]+0(h±2). (18. We see already from Eqs.1 ) D (18. Bq[w±{ ^ . + 0(hl2). (18. This means.
(18.^/w + ± 1 . (18. . Consequently the above integral from z = 0 to z± differs from that from z = 0 to a turning point (as in expression /2(C)) of Eq. (18. Thus for h very large these turning points are very close to the minimum at z+. and comparing with Eq.frW.« + ) 2 (9+1) e (18. Allowing z to approach z+ in the solution IJA(Z).147)). (18. 24(9D[i(gi)]! hj (18. Eqs.z±)dz .146).3 The Double Well Potential (cf. Later we calculate the coordinates of turning points ZQ. (18. We observe that for z — z± the approximation yields exp[±/i? t z 2 t /4]. (18. ein+z+e in+(z 22+(<?+i) .172)) we add here comments on this approximation.• • " .172) obtained later. 2 An+z+e 4w+ (18.97b) ~ exp .164)) only in a nonleading contribution and hence implies the equivalence of the exponentials in the relation (18.97b)) exp 413 '\hlJQ'uV\z)dz = exp = exp (18.129) .18..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 ( « .^(±>(^*4 ^\h\{zz±f exp 7>2 r 2 ^n±z± For later reference (after Eq. .ZI and find that these are given by z+ + 0(l/h) for finite q (cf.127) Recalling that around arg w± ~ 0. Here z± = ±h?/2c are the positions of the minima of the potential.128) Similarly we obtain in approaching z+: VA(Z)  ( 2 ^) (9 " 1} e. a= ^ . the dominant term in the power expansion of the parabolic cylinder function is given by Dv(w±) ~ wv±ew±>4. . .• )2 (^)itoi) ( 2 z + ) 5 ( 9 + i )e ' yA(z) z+> lft2 2 _ I .
The next aspect to be considered is that of boundary conditions. 18. S.130) Therefore in their common domain of validity a a = dte+vM' {2z+)^. de Deus [66] and W. K. S. The solutions in terms of parabolic cylinder functions have a wide range of validity.131) We have thus found three pairs of solutions: The two solutions of type A are valid in regions away from the minima. 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. Bhattacharya and A. R.\h\)^+l\\{q l + l)]\ 1+ 0 J_ . K.(18. The involvement of these WKB solutions leads to problems.414 and CHAPTER 18. The pair of solutions of type B is defined around argz = 0. they assume large quantum numbers. D. Thus 1 E . we naturally expect the wave function there to be similar to that of the harmonic oscillator.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.z+)]fr+V ' (18. P. and this means at both minima. J. Anharmonic Oscillator Potentials D_h{q+l)(iw+(z))2i(o+V Vc( ) z 2i(g+1)ei^(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. Thus it is unavoidable to appeal to other methods such as the WKB method to apply the necessary boundary conditions at that point.3. but it is clear that none of the above solutions can be used at the top of the central barrier. H. Bhattacharya [23]. Since it is more probable to find a particle in the region of a minimum than elsewhere. even above the turning points. since. g . basically.7r and the solutions of type C around argz = ±7r/2. Rau [22]. Various investigations^ therefore struggle to overcome this to a good approximation. We have to impose boundary conditions at the minima and at the origin. Furry [102] .+)2 [1(1 + I)]' [i(q+ l)}\[ih+(z .
135) Hence the conditions (18. However. 18. (18. as indicated there.5. y(z±)=0.132) Fig.133) imply yB(z±) yc(z±) « a' and y'B(z±) y'c(z±) a a (18.132). At dtz — ±oo > we require the wave functions to vanish so that they are square integrable. as indicated in Fig. (18. The odd wave function then exhibits a correspondingly opposite behaviour. ^(*±)#0 (18. 18. and y+(z±) = 0.134) y'+{z±) ± 0. We have y±(z±) = ^VA{z)±yA{z)]z^. an even wave function can also pass through zero at these points. 18. and y'±(z±) V'B(Z±) ± =y'c(z±) (18.Zii yB{z±) a ± a yciz±) (18.3 The Double Well Potential 415 the most basic solution would be even with maxima at z±. as indicated in Fig. We have therefore the following two sets of boundary conditions at the local minima of the doublewell potential: y'+(z±) = 0.5 Behaviour of fundamental wave functions. y'_(z±) = 0. y{z±) £ 0.18.136) .5.133) y+(z±)^0.
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 . _ 8)} s .«*{=(. Eqs.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. (18. also (18.„*{!<. (18.5. .138) Using formulae derived in Example 18.2 we can rewrite the second of (18. Anharmonic Oscillator Potentials (B) Evaluation of the boundary conditions Inserting into the first of Eqs. 1 1 . (1. (18. .m Using Eq.142) (<Z9o)J(l)^°+1)(l)1/2 . .54) +l j 2^[I(Ql)]![I(g3)]!sin{f(g+l)} ^^ ^ r i r ( T(y+me'm' (18 137)  where we used first the reflection formula and then the duplication formula. 7 .416 CHAPTER 18.9.136) the dominant approximations we obtain (cf. +1)} = 4 m.136) as  M ... (18.
z\.5 B o u n d a r y c o n d i t i o n s at t h e origin (A) Formulation of the boundary conditions Since our even and odd solutions are defined to be even or odd with respect to the origin.144) Thus we require the extension of our solutions to the region around the local maximum at z = 0. y'+(0) = 0.3. y+(0)^0. 18. V(z) — • z E=V Fig.» ^ B ^ ' . We emphasize: We needed the solutions of type A for the definition of even and odd solutions. Hence we have to impose at z — 0 the conditions j/_(0) = 0.18. 18. y'_(0)^0. We do this with the help of WKB solutions.3 The Double Well Potential Thus altogether we obtain 417 ^ • . 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. Since the type A solutions are not valid at the minima. we must also demand this behaviour here along with a nonvanishing Wronskian.6 Turning points z$. (18. We deduce .1 " <813 1 4)  In Example 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.
148) ("f * "The superscript (I. As a consequence. the dominant terms of the WKB solutions are* 1 yWKB\ ) Z 1 1 ~l/A 1/2 ~ qh\ x exp + h%U{z) . to the left of ZQ where V > E.145) \qh\ + \zW Using z+ = h2 /2c. (18. one finds that .e.146) and Iqh* 1/2 z\ Ac" + +o ^ (2g2)V4 2c + /i ' (18.144).418 CHAPTER 18. in leading order the integrals to be studied below from ZQ to z = 0 are equal to those from z+ to z = 0.^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. i.24 C Z O 1/2 ZQ = Ac2 + + °(r A_ 2c 1 2 2q\ 1/2 2 / cq 5 + h± + 0(h~ ) h%) (18. we see that the turning points ZQ and z\ are very close to a minimum for reasonable values of q ~ go> the latter being an odd integer.96) that the two turning points at ZQ and z\ to the right of the origin are given by \qh\ i.e. We also note that the height of the potential at the turning points is h8 qh V(z) 20.\qh\ + \h%U{z) (18. + £\h%U(z)=Q. ZQ) means "to the left of 20" .147) Since the minima are at z± — h2 /2c with /i 2 very large. Anharmonic Oscillator Potentials from Eq. (B) Evaluation of the boundary conditions We now proceed to evaluate the boundary conditions (18. In the domain 0 < z < ZQ. h8 25c2 (18.
we have to match yA(z). Thus we consider in the domain \z — z+\ > {2q/h2+)l/2\ h 1. Hence we have for ywKB the exponential factor exp ("f * Z+ \*l*e\*(\h% 1 1/2 \qh\ 9/4 + \h\U{z) e\{zz+?h\_ (18152a) . .^ e therefore have to consider the exponential factors occurring in (18.y^(z) to 2/VVKB(Z) anc ^ ^ W K B ^ ) .148) and (18. 20 dz \qh\ + \h%U{z) .149) In order to be able to extend the even and odd solutions to z = 0.151) In identifying the WKB exponentials we recall that y^KB is exponentially increasing and ywKB is exponentially decreasing.^ + ) 2 + igln2(zz+).(29/^)1/2 1 ^(zz+){(zz+) 1/2 2 2q_\ 1/2 hJ 2 l l q+ q  ^ ( ^ . (18.i (2q ln (18. (18.3 The Double Well Potential and 1/4 419 ywKB\ ) z — \qh\ x exp I / + \h\U{z) 1/2 I. )+^*+)2lr} rnD h\z'2)}z Q )\ = .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 >'^.18.150) Evaluating this we have ±h ~ h2 GMln 1 In 2g q In +4 1 (z .149) and consider both types of solutions in a domain approaching but not reaching the minimum of the potential at z+.^ j O e .* 4+ .
Thus using the Stirling formula z\ ~ e~(z+1\z + l)z+2y/2~rr. 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.ZQ) . 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 see therefore. using Stirling's formula (and not the inversion relation) — we have —(1. In a corresponding manner — i.e. since there is no way to obtain an exact leading order approximation with Stirling's formula for small values of q.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. This is the aspect investigated by Furry [102]. \ yWKB\ ) z 1 fl(g3)]! V*(h%)V*\zz+\h(*+V\2 q/A h + e* \(zz+)W+t (18.125a) and (18. l/2 ~h\U{z \(zz+?h\ (18. of course. The only way to relate these solutions is with the help of the Stirling formula which converts t h e product or ratio of such factors into factorials such as those inserted from the beginning into the unperturbed solutions (18. t h a t the Stirling approximation is amazingly good even for small values of the argument).126a).125a). However. the results necessarily require adjustment or normalization there in the q—dependence.420 CHAPTER 18.154) . Anharmonic Oscillator Potentials We observe here t h a t the W K B solution involves unavoidably the quantum number dependent factors exp(g/4) and (q/A)qj/4 which do not appear as such in the perturbation solutions. we can write the exponential as exp (27T) ("f 1/2 I dz Z+ 1 qh\ + q/A 1.153) This expression is valid to the left of the turning point at ZQ above the minimum at z+.
e. 7 of t h e matching relations y^NKB\z) i. as the results also seem to support.n('>2o) 3/WKB(Z) =1VA{Z)I (18. (18. . this result is somewhat imprecise.153). (18.103) and (18.144) we obtain fi(0) Tin the present case as 1 (91) 1 — 5 (18. multiplied by (1 + 0(\/h\))) the proportionality constants 7 .3 The Double Well Potential 421 where [\q}\ was written as [\(q — 3)]! for q large. linearly matched) W K B solutions. (18.159) 7 i(«i) r(9^3) ^2~ j(9l) . [\{q — 3)]! are really correct replacements of [\q}\ only for q large.155) (2z.143) which was obtained with our perturbation solutions from the boundary conditions at the minimum.18. it is our philosophy here that the factorials with factors occurring in the perturbation solutions are the more natural and hence correct expressions.K?+1). Comparing for z —• z+ the W K B solutions (18. \hl.143) T^^*)) 27T (18.157) is used in Example 18. (18. The relation (18.e. 27T1/2 9/4 1VA{Z).5 for the calculation of the tunneling deviation q — go by using the usual (i. we obtain in leading order (i.157) Since the factorials [^(q — 1)]!. and is shown to reproduce correctly the result (18.123) we have V±(z) 1 [yA(z)±yA(z)} = ^y^l(z) ± ^j#&(*) (18. However.e.154) with the typeA solutions (18. Returning to the even and odd solutions (18.156a) and 7 _ i\(Qm •K •/2(M )l/4 ^ 2 n« q/4 (2*+) •(9D e ^^4.99).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 (18.104).158) Applying the boundary conditions (18. .
161) we can rewrite the integral as r I2{z) = = dz^{a2 .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. G2 K [i 4.z2)(b2  z2).k'/2. (18. We have 1/2 dz Jz \qh\ + \h\U{z) dz Jz 4 ^? lfh 2\22c dz ¥+ A4 l 4 3 1/2 1/2 dz Jz — cz h4 CZ qh\ dz4 qh\ <Wc h+ 1/2 Jz zo 2V*\Vc h 4 ) T Jz 2\/h ~~2 /2(^2 —pz I V2 dz ^?^ h+ Z 4? + ^f ^2<«2 (18. Byrd and M.iV2 ! i . 60 and 361.3 evaluated at z = 0 is then given by = y/T+u[E(k) uK{k)} (18. n + uY 2c „ ^6 8V2c2 (18. 213.162) T h e integral appearing here is an elliptic integral which can be looked up in Tables... p. formulae 220. p.19. .148). Anharmonic Oscillator Potentials Thus we have to consider the behaviour of the integrals occurring in the solutions (18.164) / 2 (0) z ca [(1 + k2)E(k) . F. Friedman [40]. (18.05. 3G2 12V2c ^See P.163a) (18..422 CHAPTER 18. D.163b) The integral hi?) . K(k)} 3 ^ 2 y/T+u[E{k) uK(k)\.160) Setting 6 =4?^T"z°' rb Ac2 + VQ—> (18. .149) near z = 0.
. 1.1)]' Gfo..e T ^ 5 U W ( 2 „ ) i / 2 ( ^ i W4C • (18 170) (18 1?0)  ^The expansion of I2 used in Ref.2.3 )l ! .1. (G\ . 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 . . .2 A( L ) 4 s / and z dz yWKB\ J and d^WKBW 2=0 (*?) a ( .3 The Double Well Potential 423 where K(k) and E{k) are the complete elliptic integrals of the first and second kinds respectively. ^ ^ . n. Here we are interested in the behaviour of the integral in the domain of large /i 6 /c 2 .18. .) = 2 \.r1)/10//2 ( f g 1^ yS» /L ( ^ ) l^4 . where the result is shown to be (with q ~ q0 = 2 n + l . n = 0.± ' 3G 2 ' T ( ^ (18.( ft 7.§ Since the integral is to be positive (as required in the WKB solutions) we have to take the upper signs. ± . .165) in agreement with Ref. which implies small G2.166 ) ^^iffil/^(2*)1/affiii£ Correspondingly we find ^ (18168) S W I 2 = o .1 " e " O T (18 169)  Y^V / 4 tfo.( 2V + l)ln V 4 / 4( 2 n + l ) l nV 2 n .. /2n + l 2 ± — . [167]. The nontrivial expansions are derived in Example 5.. [4] misses an n—dependent power of 2 in the result.
157).143) the replacement: {hlY'2{2z+fe^zl + 2 9 ( ^ y Z e"A.3. Thus our second condition in the present case of the double well potential turns out to be an identity which confirms our discussion at the beginning of Sec. (18. 2 .3 concerning the exponentials.. (18..^ W ^ / 4 t o .1. 18. . Anharmonic Oscillator Potentials With these expressions we obtain now from Eqs.172) into Eq. with the help of Eq.172) A corrresponding relation holds for h+ and z+ replaced by /i_ and z.174) .173).i ) ] ! e ~ s A ' ..n = 0. (18. Expanding this around an odd integer qo we have ( 8E\ h? —J Inserting here the result (18.159) — on using once again the duplication formula in the same form as above — 2"/ 2 (2h% 1 V/2 7 Comparing this result with that of Eq.106). 18.»W.143) with qo — 2n + l. and obtain ( * . we see that the relation is really an identity (the preexponential factors on the left and on the right being equal) with the exponential on the left being an approximation of the exponential on the right (which contains the full action of the instanton). . (18. .h2) + (qq0)^=. Inserting the expressions for h+ and z+ in terms of h. (18. (18. (i*™) We obtain the energy E(q. (18.h2) and hence the splitting of asymptotically degenerate energy levels.3.175) ~E0(q0. . we can therefore impose the boundary conditions at z = 0 by making in Eq.6 Eigenvalues and level splitting We now insert the replacement (18. h2) the split expressions 29o+1/i2(—W2 h6 (18.424 CHAPTER 18. we obtain for E(q0..
176) is described by Bhattacharya [23] as a "modified well and harried result AjfWB.e.h2)E+(qQ. the level splitting.fr2) is given by Eq. 1 7 425 2 .^ 2^ h* . 2h4' i.n = 0.. i n U n / ! Thus AE(q0. i. Combining Eqs. which evidently supplies some correction terms.^ e 21/^6c2 fe6 ( 1 mass mn = (1 8 . i.7 =( e N n+1 .2.h2) / h 6 \ —K />6\n+l/2 q o / 2 2«>+2h2 — —^= m 1.157).. the pure WKB result of Banerjee and Bhatnagar [14] (i.2 = 2 .e.1 7 6 ) (2n+9)/4^/ft V7T«! V C / \ eiI72^. in the WKB restricted sense) defined by ^( n +2^KB(0) 2 Here the derivative dE/dn corresponds to the usual oscillator frequency. (18. 1 u2 c 2 (3 9 2 + l) ^ « b . T \ *n ! 1 ' (18..1.106).178) "In S.174) and (18.V2^\ \n+ i . . for finite h2. Bhattacharya [23] the "usual WKB approximation" of this expression is — with replacements h4/4 <» k. can be given by In the work of Bhattacharya [23] the WKB level splitting is effectively (i..e. the following expression for large q ~ 2n + 1.e.143) with (18. c 2 /2 <> A — given as E0 = 4\ h(2fc) 1 / 2 o y J H q2 4/c > ^— + q ^ 2s c2 V2 . 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 / » : = \ .18.h2) = E_(q0. (18. The result (18.3 The Double Well Potential where £0(<7o.e. K.q2 7. the difference between the eigenenergies of even and odd states with (here) qo = 2n+l.
H. Then E^h2) and AE becomes 2^+U 2 (^)W 2 h6 = Eo{q0.180) h8 E0(qo. Then the splitting A ^ W B of Eq. that if this symmetry is taken into account from the very beginning. The Furry factor represents effectively a correction factor to the normalization constants of WKB wave functions (which are normally for mo = 1/2. h = 1. we would have obtained the result with E. (24) there is k 2 (i0n+i3)/4 //c3/2\n+1/2 V2fc3/2 6XP n\ ft \/nn\ in agreement with AE(q0. h4 and c2 replaced by 2E. these deriva> tions do not exploit the symmetry of the original equation under the interchanges q <> —q.426 CHAPTER 18. and A there with c 2 / 2 here.^ 2 + ^h2 If in addition the potential is written in the form A / . This. Eq.. we identity the parameter k there with ft. The factor set equal to 1 represents a somewhat "mutilated" Stirling approximation. (66). Anharmonic Oscillator Potentials which is unity for n — oo with Stirling's formula. 2/z4 and 2c2 respectively (see Eq. ** Comparing the present work with that of S.2 / 7. K. supplies the bridge to the perturbation theory results and removes the unnatural appearance of factors e and nn. A>0. however. (18.176) together with Eq. as we do here. Bhattacharya [23]. (18.h2 <> — h2.4/4 here." Of course.6) to yield an improvement of WKB results for small values of n. We see therefore. 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.2X2 V(z) = \z2tL) . the Furry factor corrected WKB result follows automatically. h2)\mo=i = .6 \9o/2 e~h&l^\ with (m 0 = l) (18. (18.3)).178).181) "W. (18. as is also explained by Bhattacharya [23]. Furry [102]. 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.h )/2 2 2 ( — ' 3 6\/2c A h6 n+l/2 h6\ 2{2n+5)/if \ exp 2 \ c / of Eq.h2) * ^ 2 q o / ^ q l _ m ^  (18179) l~o o». .** Had we taken the mass mo of the particle in the symmetric double well potential equal to 1 (instead of 1/2).
Lee [98]. and g2 is given as g2 = r]2/m3. Bhatnagar [14]. M.—Q. We considered above only the symmetric twominima potential.3.11). (18. D. carried out along the lines of the corresponding calculations for the cosine potential and thus of the wellestablished Mathieu equation. In principle one could also consider the case of small values of h2 and obtain convergent instead of asymptotic expansions. Eqs.c2 = A/2. Equation (18. . (18. Jentschura [293]. ZinnJustin and U.3 The Double Well Potential 427 a form frequently used in field theoretic applications. these are presumably not of much interest in physics (for reasons explained in Chapter 20).15).180) agrees similarly also with the result for arbitrary levels obtained with the use of periodic instantons* and with the results of multiinstanton methods.h>) * 2V2^^^y/2e^3/3A. P. W. first paper. Gildener and A.W. however. formula (4. so that by comparison with Eq. §J. J.182) implies AlE(l. Patrasciociu [110]. MiillerKirsten [186]. Banerjee and S.^ Thus for the case of the ground state Eq.88) h4 = 2/x2.§ 18. (2.7 General Remarks In the above we have attempted a fairly complete treatment of the largeh2 case of the quartic anharmonic oscillator. ^To help the comparison note that in J. Liang and H.34) and (E. m 2 <• p 2 / 2 . Friedberg and T. D. 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. H. J. ZinnJustin and U. Miiller—Kirsten [167] the potential is written as V2 VW = \ for mo = 1. See R. The asymmetric case can presumably also be dealt with in a similar way since various references point out that the asymmetric case can be transformed into a symmetric one.^ *See E. Mansour and H. provided their result is multiplied by a factor of 2 resulting from a corresponding inclusion of antipseudoparticle (antiinstanton) contributions. J. The comparison with Eq. (18. K. 'This will become evident when the perturbation theory result of Chapter 17 is compared with the path integral result in Chapter 26.* A similar correspondence is also found in the case of the cosine potential. Jentschura [293]. M.181) therefore implies the correspondence rj1 <> A/2.18. D.
as some relevant references with their view we cite papers of Harrell and Simon [129]. 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. Pradhan [182]. Anharmonic Oscillator Potentials Every now and then literature appears which purports to overcome the allegedly illnatured "divergent perturbation series" of the anharmonic oscillator problem. and numerical studies are presented to support this claim J It will become clear in Chapter 20 — as is also demonstrated by the work of Bender and Wu [18]. The wide publicity given to the work of Bender and Wu made pure mathematicians aware of the subject. in both symmetric and asymmetric form. There is no reason to view the anharmonic oscillator differently.g. can be found in an article by Coleman [54]. though incomplete list of references in this direction has been given by Garg [105] beginning with the wellknown though nonexplicit (and hence not really useful) ground state formula in the book of Landau and Lifshitz [156]. A reasonable. [19] — that the expansions considered above are asymptotic. Very illuminating discussions of double wells and periodic potentials. as he discusses. P. in his recent paper on simple uniform approximation of the logarithmic derivative of the ground state wave function of anharmonic oscillator potentials. for instance. In fact.[131]. Mahapatra.428 CHAPTER 18. though not exclusively numerical. As references in this direction. Tables of properties of Special Functions are filled with such expansions derived from differential equations for all the wellknown and less wellknown Special Functions. Perturbation theoretical aspects are "See e. The immense amount of literature meanwhile accumulated for instance in the case of the Mathieu equation can indicate what else can be achieved along parallel lines in the case of special types of Schrodinger equations. Santi and N. Sophisticated mathematics — like that of the extensive investigations of Turbiner [273] over several decades — seems to approach the problem from a different angle. Simon and Wightman [178]. B. like those for anharmonic oscillators. There is no way to obtain the exponentially small (real or imaginary) nonperturbative contributions derived above with some convergent expansion. The ground state splitting of the symmetric double well potential has been considered in a countless number of investigations. mostly in connection with instantons. Martin. . The double well potential. has also been the subject of numerous numerical studies. N. Ashbaugh and Harrell [12] and Harrell [130]. we cite papers of the Uppsala group of Froman and Froman [100]. [132] and Loeffel. since — as we shall see in Chapter 20 — these nonperturbative contributions are directly related to the largeorder behaviour of the perturbation expansion and determine this as asymptotic. B.
3 The Double Well Potential 429 employed in papers of Banerjee and Bhatnagar [14]. We have 4 ~k' 2u(l .u)1'2 = V2u{ 1 .183) S o l u t i o n : We have k2=1^.192) 'F 12 16 . Example 18.188) T h e following e x p r e s s i o n a p p e a r s f r e q u e n t l y in t h e e x p a n s i o n s of elliptic i n t e g r a l s . 1+ u : Gy/2q.\G\ + • and 1k2 =2Gv/2q4.190) Hence ln]=ln : l n ' 7 2 ^ + 2 (18. Bender.5: Evaluation of WKB exponential with elliptic integrals Show that (l + u)^2[E(k)uK(k)}'. 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. We obtain the expansions from Ref.^ + • (18. (18. q ~ 2n + 1. Thus k' and hence k' = v ^ ( l . + (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. [122] as 13 1. fc ~ 1 . Srivastava and Varma [24] and Bhatacharya and Rau [23].G2q + .f + • • •) 2u\ .184) Hence we obtain for G close to zero: 1Gy^g l + Gy/2q~ and 1 .185) (1 + u)^2 = (1 + GVW/2 = 1 + GJ±. T h e r e f o r e it is convenient t o deal w i t h this here./4 E{k) = 1 + • l n In (18.191) Our next objective is the evaluation of the elliptic integrals E(k) and K(k) by expanding these in ascending powers of k' . which is assumed to be small.189) =2u 2u2 (18..2G^2q + 4G2q . k! (18. Datta. Biswas. (18. Cooper and Strottman [221]. Gutschick. X +2 + ' ^ (18.186) (18. Saxena.18.187) We now reexpress various quantities in terms of u. 2 a.
u) + 2}] 1 H 1 o u(u .3u(6u . 1 3 « 2 = .a = (l«)(jl 2u/ 2 ) .u)} u{u(l .195) we obtain E(k) 1 + ln uK(k) . 8 + 9u 2 .3u(6u .3u(6« .u) + 2} + i u ( u . (18.3u(6u .u) + 2u} + i u 2 [ 2 + (1 .18u 2 + 39M} + 12u 3 (u .195) With (18.193) 2u(l . Anharmonic Oscillator Potentials *« = M £ H Consider first E(k): E{k) = 1 + .u){4 + 3u(l .u ( l .13)} + 12u 3 (u . 4 \ u In —= I + V W 2 1 + ln x(l u) 31/ +.3 y .{ M 2 ( 1 .1)(1 .«) + 2} + V [ .194) ln '^J + 2 In i/2u )+H l + 2u(l . (18.6 in the denominator and in the last step pick out the lowest order terms in u (i.197) . ) J 4 (18.13)}.13)} + ±u3(u .4..1)(1 .u ) (^){« + 5^ 1 )} + 5.u) 2 {24 .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){24 .^ L ) H[(i _ U ){4 + 3u(l .u ( l . 8 .2{«(1 .3) . (18.3).— ) .« 2 + 3u] In [ .196) Now consider the last line here without the factor 2.430 and CHAPTER 18.3) .e.u 3 ( u .3 u .2)1 4 uA"(fc) In ( ^ = ) ^{«(1 .u)} .3 .u){24 .u){24 .{l u ( l u) / W 2 + 2} . up to and including u2): u(u .13)} + ~u3(u .« ) ^ { 2 4 .3) = = ~ .u) + uf\ 2 + ju2(l 16 + (u W 2 12 4u2(lu)2 +  l ) u ( l .u 13 \ 1 ln(± k> fc'2 +  (18.u) 2 {8 .3 « .194) and (18. . 8 + 6 u 2 .1)(1 .u)} + — u ( u .3 u ( l .1)(1 .3) 1 + In ^ .u)(u .6w2 + 13u} + 12u 3 (« .
155)) <*VA{Z) = VB{Z) a n d VWKB(Z) = 1VA{Z).e.2 l_ln(1 JL\ V {2. Inserting here from the above expansions the contributions up to and including those of order u2.. this becomes _2 2 u 2 / j M 2 / 3^2s 2 ~ 3G 2 2G2 H \ / 2 H / .uK(fe) ~ 1 + In I — = Mu2(3u. where w±(z) — h±(z — z±). (18.it)} . / to + 3GH~ \ 8 .156a) we obtain a and 7 and hence the ratio (always in the dominant order) T2~ ) w i t h h += ^ h • Since (cf.«){4 + 3u(l . we have..2 (18.18.^HfHGH Example 18.128) and (18.199) We now return t o the expression on the left of Eq. we obtain: h ^ —^ 1 + 3G 2 V 2 8 2 /„ u 1 . 2 ..e. ' 8G 2 q = 4 2 q ln f 26/2 \ o .3) .2 /o 2 431 (18.200) Remembering that u = Gy/2q. VB(z) = y\hTB(z) with VB(Z) ~ Bq[w±(z?  ^ [±(q  J l)}\2^V .J? dz^\E V(z)\ \ 87T2 / lEVtz)]1/4 Solution: From Eqs.3).196) and (18.197) we now obtain 4 1 • E(fc) .zo) HWKB normalized _ ( h* y^eM. .u H I .3 The Double Well Potential Now consider the bracket [. (18 201) = Thus 3ffl« (2i/aGi/a2i/V/V"4 3ffl2 (Gii7»J4  '• . (18. V^LJ 4 2 16 ^ . i.4u . From (18.4u .128) and (18. [(1 .196).198) 9 «. (18.173) t o be evaluated.195) with (18. Eqs. + iu^Zu2)^~u+ — +0(u3) l .6: Normalization of WKB solutions Show that the normalized form of the WKB solution (18. l2 = ^ ( l + uf/2[E(k)uK(k)}. (18.148) (for particle mass 1/2 and in dominant order) is given by H.2{w(l u) + 2}] = u(3u2 .] in (18. i.
204) .146) and (18. 93. 1 7(l. 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+.3)]!23<«D [5(91)]' >A. (18. W.7: Recalculation of tunneling deviation using W K B solutions Determine the tunneling deviation q — go of q from an odd integer go. Eq.Aw) = s/2i 2("!) V (41) and [(9 .143). (18. Anharmonic Oscillator Potentials With the standard normalization of parabolic cylinder functions.35)). by using the periodic WKB solutions below the turning points. y±(z) = IVA(Z) ± yA(z)\ = — i M O O ± 1 Ji. zo) meaning t o the right of ZQ and note the asymmetric factor of 2) 11/4 y±(z) ^ W K B (z) ± ^ = % K B ( Z ) : \qh\ 1/2 \h\U{z) ±— cos 7 **See e.*o)/ ^WWKBOO (18. 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 [ . i. pp. Oberhettinger [181]. We start from Eq./ . Magnus and F. Solution: The turning points at zo and z\ on either side of the minimum at 24.** oo / dwD\.e. and obtained as Eq. normalized = We see that the normalization constant is independent of a quantum number. (18.hA/2 = h^/4 this implies . Then ((r.432 CHAPTER 18. oo . The constant obtained here will arise in Chapter 26 in the evaluation of the normalization constant of the WKB wave function (cf. are given by Eqs. ° \Edzy/\EV(z) V(z)V4 VWKB.g. (18.147). i. 82.e. we obtain (for q reasonably large) 1 (91) and hence Joo Hence with h+/V4n = (h4/Sn2)1^ W2TT SI {[UqIWH"1)}2 2TT h2+\1/2 ~VWKB \ ~^T f 4^ \V4ir ^g^ J VWKB .f hi\ VWKB.158).l)]![i(0 .zo). (26.
.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 .207) (18.18. We could.Jz0 sin ^o J LJzo 4 Jzo 2 y •••4y ••• = . z{) we expect tt (18. Messiah [195].209) "27 In the present considerations we approach the minimum of the potential at z+ by coming from the left.132) and (18.. 6.g.210) Choosing the point z to be z+. ^(7(z) r t J z0 •I Zi I V2 dz \qh\ 1.212) tt See e. i.3.2.g.. from ZQ..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. this implies sin cos sin cos Thus e.208) and cot L />G + 1/2 ^(ig4^+^))"" + J 1/2 2qh%h%U{z)\ 27 +  :± (18. JV = 1.i ) N c o s / •••+! provided the BohrSommerfeldWilson fZ1 dz^/E J zn quantization 4 condition holds.206) !/2 ' + J*+dz(^qh2+±h%U(z) 1/2 7T' 4 1 =p — sin 7 f*+dz(^qh%h\U(z) ^ TV + 4 (18.2.J^Wj +(18. . \h\U{z) 1/2 . of course. Sec.iK h\U{z) + / 4 (18. Then at any point z € (zo.. i.211) 7 z l .( .e..6.r + ^=co S [r. A.V(z) = r J zo I ZQ dz[\qh\ \ Z h%U(z) (27V + 1 ) .205) We now apply the boundary conditions (18. 1/2 . approach the minimum also from the right. from z\.e.^r + + . (18.e. 7T\ + ^ 4 7r\ dz . i.
90 — =F— for q0 = 1. 7 . . and cot< (q + 1) '!}» for q = 50 = 1. M. P. where u> is the oscillator frequency of the wells. .210) assume a form as in our perturbation theory. (18. (18. 1 1 . + l ) j } ^ 1_ 27' (18. the level splitting is given by AE(q0. W "Cf.209) and (18. Lifshitz [156].215) we can insert for 7 / 7 the expression given by Eqs.5. 2IO/K.146)). 5. K.27 9 . The latter give 7 7 1/2 exp dz dz \qh\ \qh\ + + \h%U{z) 1/2 exp I — / \h%U(z) (18. We note here incidentally that this agreement demonstrates the significance of the factor of 2 in (18.215) which results from the factor of 2 in front of the sine in the WKB formula (18. S. and L. they become c o t { ( . K. (18. (18. 7r"y in agreement with Eq. The prefactor 2\/2h2/n is.434 CHAPTER 18.e. (18. In Eq.exp ( \/2h2T V2 7r 7 dz l / qh\ + \h\U{z) (18. Landau and E. Anharmonic Oscillator Potentials Similarly under the same condition (l)Jvsin J Zn where it is understood that qh2/2 ~ Ey(z±).215) we can expand the left hand sides about these points and thus obtain . Bhattacharya [23].216) Inserting this into Eq. (18.213) The integral on the left can be approximated by 1. One may note that the exponential factor here is not squared as in decay probabilities (squares of transmission coefficients).h2) = = h2 h2 1 y 1/2 2{EEo)=2^{q~q0)=2^±l V2 V2h2 . . (18.. Bhatnagar [14].205)._ £/>(£<•>•) tan{(g + l Since tan ) ^ ^ . Banerjee and S.215). The formula thus agrees with that in the literature. D. 9.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 (18.217) 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.148) and (18.214) {«+"i} 0 for q = go = 3.143). ..Zi / r.2 . in fact (with our use of qo). We now see that Eqs.3. ' /^i .149). 2Sm (2 g /ft2)i/2 in agreement with the right hand side (the last step following from Eq. i.
J. W. 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. W. also in P. 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.Chapter 19 Singular Potentials 19. f K . Feshbach [198]. 435 . M. are not as troublesome as one might expect.* However. H. Giittinger and H. However. The centrifugal potential oc r~2 is generally considered as exceptional and is treated in detail in wellknown texts on quantum mechanics. 'For a review from this perspective with numerous references see H. In particular Case pointed out that for a repulsive singular potential the study of scattering is mathematically welldefined and useful. n > 2.* The physical analogy between singular field theoretic interactions and singular potential scattering of course breaks down at short distances. 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. Case [45]. Morse and H. since no probabilistic interpretation is available for the field theory matrix elements in virtue of creation and annihilation processes during the interaction.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. Aly. Some decades ago — before the discovery of W and Z mesons which mediate weak interactions — and before the advent of quantum chromodynamics. in an early investigation Case^ showed that potentials of the form r~~n. M. MiillerKirsten [9].
J. Singular potentials arise in various other contexts. J. for instance. R. In view of the unavoidable use of Mathieu functions expanded in series of Special Functions. MiillerKirsten and Jianzu Zhang [226]. roughly speaking a brane is the higher dimensional equivalent of a membrane visualized in two dimensions from which the DS brane derives its name. D. Manvelyan. H. This property singles this case out from many others and attributes it the role of one of very few explicitly solvable cases. N. In fact.2 19. It is therefore natural that we study this case here in detail.436 CHAPTER 19. In the concluding section we refer to additional related literature and applications. Callan and J. (19.Q. Singular Potentials theories in mind. Tamaryan. MiillerKirsten. H. . S. some also permitting further research. G. This chapter therefore gives the first complete solution of a Schrodinger equation with a highly singular potential.e. Park. K. § 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). the attractive singular potential 1/r4 was found to arise in the study of fluctuations about a "brane" (the D3 brane.2. Hashimoto [124].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. We shall see that not only can the radial Schrodinger equation be related to the modified or associated Mathieu equation (i. 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. Liang and Yunbo Zhang [189]. but also that the Smatrix can be calculated explicitly in both the weak and strong coupling domains. W. J. D referring to Dirichlet boundary conditions) in 10dimensional string theory. Gubser and A.2) § C . Maldacena [41]. 19. W. S. that with the hyperbolic cosine replacing the trigonometric cosine). our presentation below is deliberately made elementary and detailed so that the reader does not shun away from it. Recently. this particular singular potential plays an exceptional role in view of its relation to the Mathieu equation. S.
and so =4 5 and n ez = r = ~ = e^4fir. r = 7 e z . MiillerKirsten. Eq. 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.Q. It is clear that we draw analogies to our earlier considerations of the periodic potential. (19. W.4) dz2 + In the literature this equation is known as the modified Mathieu equation or associated Mathieu equation. h2 = ikg. (19. Liang and Yunbo Zhang [189].4) first for small values of h2 and then for large values of h2. (19. In the following we study the solutions of Eq. It may be noticed that since I is an integer. This is rarely possible and therefore this case deserves particular attention. H. We begin with the derivation of various types of weakcoupling or smallh2 solutions which we construct again perturbatively.7 that we employed also in previous chapters. Our ultimate aim is the derivation of the /Smatrix for scattering off the singular 1/r4 potential.19. and as we have in mind in Sec. h = ei^y/k^.28) below).( I + 0. This has advantages compared with higher dimensional cases. \[\ in the correspondence is nonintegral. ig h \j g (19. using the method of Dingle and Miiller of Sec. (19. R. [l+2) ~ * A> ~ the equation converts into the periodic Mathieu equation of Chapter 17. Manvelyan. The radial Schrodinger equation then assumes the form 2h2 cosh 2z . 8.3 below) g2 is negative and hence h? is real for E > 0. J. J. We observe that with the replacements p2 z^iz. Subsequently we derive the same 5matrix from the consideration of largeh expansions and calculate the absorptivity in the case of the attractive potential. 19.3a) In the case of the attractive potential (as in the string theory context referred to above.2 The Potential 1/r4 — Case of Small h2 437 where E = k2. . The following substitutions are advantageous: y = r1/2cf>.^ where \f\ can be an integer and thus leads to singularities in the expansion of v (see e.mo = 1/2 and h = c = 1.g.3b) 7 = % j.
First we make the additional substitution w = 2/icosh z. w 2 . (19.^% dw1 1 1 —— = dw = ][Zv22Zv 4 w Zv + Zv\ = (Zv\ 2 + Zv+2]. — Zv+i). Proceeding along the lines of the perturbation method of Sec. J.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.2 CHAPTER 19..I .7) We shall see that this parameter is given by the same expansion as in the case of the periodic equation. 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 = 0.7. and H„ (w).i l . MullerKirsten and N.e. Eq. The solutions Zv(w) are written Jv(w). . (19. (19.6) terms amounting to 1 „ 2 d2Zu A V (19. W. Neumann or Bessel function of the second kind and Hankel functions of the first and second kinds respectively.g) This equation is wellknown as Bessel's equation or as the equation of cylindrical functions. dwA w dw { wA ) (lg.6) becomes to zeroth order </>(°) = <j>v of <f> in h2 Dvcf)v = 0. i.2 A / i 2 .1. where £>„ : = —_ + i _ + J l . H.9) leaves uncompensated on the right hand side of Eq.Nv(w). —Zu — (Z„_i + Zu+i). VahediFaridi [10]. H. (19. (19.4) becomes dw2 w2 w2 \ dw2 w dw \ j Next we define a parameter v by the relation V2 v2 = (/ + . where these are the Bessel function of the first kind. (19. 8.11) We follow here largely H.438 19.5) so that Eq.2.Hv (w). The zerothorder approximation = Zv[w) (19. Aly.
13).15) (19.u + 2)*Z„ +2 ]. wz The expression (19._2 + {v. (i/.13) therefore lead to the first order contribution jfU = h2[{v.2.6). therefore 4>^ leaves uncompensated i#> = / i 2 [ ( i / ^ .20) 2)*ZV_2 + {y. where We observe that DUZV = Q.10) is now particularly simple: i?i°) = h2[(v.2 ) * i ^ 2 + (!/. (19.i/4)*Z 1 / _ 4 +(u. = §.. v . v + 2)Gv+2].13) (19. However. Du+aZu+a but Dv+a — Dv so that DuZv+a = a(2l/ (19. it is more convenient to use these relations in order to rewrite Eq. where the starred coefficients are defined by (19.2)*Zi. I / ± 2 ) = 1.i/) = 2A.16) ( I / .21) +(i/. ~a(2v + a)~ w2 * a> Zv+a = a(2u + a)Gv+a.u (19.!/+ 2)*4°J 2 ].14) (19.19. We assume in the following that 2u + a ^ 0 (the case of 0 has to be treated separately). The next contribution to (^°) + t^1) therefore becomes 0(2) = /i4[(z.2 The Potential 1/r4 — Case of Small h? 439 we can rewrite the expression (19.10) in terms of functions G„ defined by Gv+a = —nZ v+a . (19. v + 2)*Z„ +2 (19. can be cancelled out by adding to </>(°) the new contribution liZu+aja{2v + a) except.2 ) * ( z / . v)Gv + (y.17) Thus a term fiGu+a on the right hand side of Eq. .10) as a linear combination of various Zv. v2)*{y2. when a or 2v + a = 0.v + 2)*{u + 2.2)G„_ 2 + (y. . of course. (19. v .18) Now </>(°) = Zv left uncompensated Rv . z . The terms (19. and this means in (19. v + 2)*(v + 2.12) (19.u + 4fZu+A}.
(19.25)/i 8 32(al)3(a4) .25) Finally we have to consider the terms in Gv which were left unaccounted for in Rv .24) P2i2(2j + 2){v + 2j + 2. Ri. i/ ± 2)* (i/ ± 2.26) iy — 2. Singular Potentials Proceeding in this manner we obtain the solution ^ = 0(0) + 0 (1) + ^(2) + .v) + hA[{v.17)).1169)/i 12 . Evaluating t h e first few terms. + 2(y2\) — + 32(i/ 2 l) 3 (^ 2 4) —^ (9.4 + 58. . in Eq.v + 2y{v + 2.22) where p2(±2) p4(±4) = = (i/.I/±2)*(I/±2.l)(z/ 2 .4)(»/2 .440 CHAPTER 19. v + 2)*{v + 2.I/±4)*. v + 2j) (19. . (I/.. Zv + Y^h2i i=l J2 p2i{2j)Zv+2j. . p 0 (0) = 1.9) 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.v)] (19. p4(±2) = (i/. and p 0 ( 2 j ^ 0) = 0. . . i/ + 2j). i/ + 2j) + p 2 i _ 2 ( 2 j ) ( ^ + 2j. etc. Reversing the expansion and setting a = (I + 1/2) 2 (not to be confused with a e.v + 2)*(i/ + 2. ^ 64(a_1)5(a_4)4(a_9) + 0 ^ )• (1928) 1 6 .455a + 1291a . i/ ± 2)*.2 + 2 9 ) ^ 64(^ 2 .2.v) + {v.v2)*{v2.23) These coefficients may also b e obtained from t h e recurrence relation P2i(2j) = p2i2(2j . .g.j^0 (19. subject to the boundary conditions P2i(2j) = 0 for  j  > i. v) +(i/. (19. we obtain 9 h4 = a 2(al) 3 2 (13a . we obtain 0 = h2{v. Adding these terms and setting the coefficient of Gv equal t o zero. . (19.2)(v + 2j .i/±2)*. we obtain t h e expansion . P2*o(0) = 0. j=i. v ( 4 5 a . i/)] + • • • . v — 2)*(v — 2. 1 1 \ 2 + 2) ~ ) s X 2 A = v 2 hA (5v2 + 7)h8 v .
19.2. Substituting (19. ±{ v+ z e = sinh(v + 2)z + sinh(z/ — 2)z. = 0. (19.30) £>„&.h2[(j)u+2 + 4>v_2] h2 [{u.An(u + n). The solutions 4> of the modified Mathieu equation are now completely determined.cosh 2z)4>.^ + e±(v~2)z.34) . v)(f>v + {v.33) = = cosh(^ + 2)z + cosh(^ — 2)z.2 The Potential 1/r4 — Case of Small h? 441 We note that this is an expansion in ascending powers of h4. and is therefore real for both cases g2 > 0 and g2 < 0. we may say that the first approximation <^°) leaves uncompensated terms amounting to R^ = 2ft2 (A .29) sinh vz or e±uz.19.31) It follows that Dv+2n4>v+2n = 0. apart from a normalization factor which we have chosen (so far) such that the coefficient of Zu in <f) is 1. Thus v is real for small values of  h21. v . we can rewrite the latter as j ~ . (19. (19.2)<t>„2 + (v. . 2 . — We return to this question later but mention here that this problem has already been dealt with in Chapter 17.v2(j> = 2/i 2 (A . Du+2n so that D„(j)v+2n = ^n(u + n)<\>v+2nSince 2 cosh 2z cosh vz 2 cosh 2z sinh vz 2 cosh 2z e ±vz = Dv . ±2.v + 2)<t>v+2]. so that J2 (19. £>„ = — .cosh 2z)4>v = = 2h2A(/>„ . (19. (19.4). We are still left with the question of what will happen if 2v + a = 0 or v — ± 1 .32) (19.3 Small h2 solutions in t e r m s of hyperbolic functions For later purposes we require yet another type of solutions.7) into Eq. Thus to O(0) in h we have (jr ' = 4>v — cosh vz.i .
v ± 2) etc. Denning (19 36) (">" + <*) =o^7Ta)> we now have the solution  p(z.I/±2)*. The solutions in terms of cylindrical functions are written *J. • (19. In order to avoid confusion arising from the use of different equations. Schafke [193].J7t0 p2i(2j)<f>u+2j. In fact we could have obtained the same Ru by starting with the modified Mathieu equation for h2 replaced by — h2. Singular Potentials (I/.h). (19.J/±2) = . 8.1 . > (f> with </v = sinh vz — Seu(z.* The solutions of the modified Mathieu equation which we are considering here are written with a first capital letter.38) (f> with (\>v = exp(vz) — Meu(z. > These solutions correspond in the periodic case respectively to the solutions and me^. 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.h) = <(>„ +Y. F. Arscott [11].I/) = 2 A .37) where p M (±2) = (I/. W. (19.h). .442 where (I/.7 — we refer to Meixner and Schafke [193]. For rigorous convergence and validity discussions of any of these solutions — our's here differ only in the method of derivation with the perturbation method of Sec. we prefer to adhere to one equation with different solutions. The use of the symbols (v. CHAPTER 19.4 Notation and properties of solutions We now introduce standard notation as in established literature. etc.h2i *=1 J2 j=i. M.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.h). Meixner and F.2.
h) = exp{vniri)Me„(z. h).44) (19. Mi. Oberhettinger [181].h)±Sey{z.19.41) i=l We see immediately that Meu(z + niri. h). this normalized solution possesses contributions exp(i/z) in higher order terms.h). <f)v = N„(2hcosh z) MJ?\z. (19.2 The Potential 1/r4 — Case of Small h? similarly: <fi 4> <f> (p with with with with <j>v = J v (2/icosh z) . 4>v = Hv (2h cosh 2) * M^\z. 4>v = Hi. Schafke [193].h).h2) oo = Y. (19. as Mev{z. Magnus and F. i. h) are therefore proportional to each other as a comparison of Eqs. see this reference p.h) = Jv{2hcosh z) + ^h 2i J^ j=iJjLQ p2i(2j)Ju+2j(2hcosh z). — oo cv2r{h2)e^+2r>. (19. which.h = £ p2i(2j)(l)j exp[(^ + 2j)z] (19. (19. 17.42) The solutions Meu(z. For the relation of Hankel functions used below. Also from the expansion of Jv{z) in rising powers of z. we obtain$ Ju(2hcosh(z + inir)) — Jl/(2hcosh z exp(m7r)) — exp(inuTr)Jv. as we remarked earlier.e. 443 (19. W.40) Cev{z. whereas the — as yet — unnormalized solution (19.h) = exp(vz) + Y. (19.44) implies. h) = av{h)M^ (z. 'See e. As emphasized in Chapter 8.42). h). i.g. . h). Meixner and F.45) T Below we frequently write the normalized solution as in J. Me„(z.h).40) does not. 16. W. (19.43) and hence similarly M^\z + inir. introduces the dominant order function into higher order contributions''' — oo i 2i Meu(z. p.h) and oo i M^\z.e.h). h) = exp(inwn)MP(z. ' (z.39) Writing two solutions out explicitly we have for example — apart from an overall normalization constant.(2hcosh z).MP(z. {2h cosh z) • MJ?\z.
H™(z) = e^HfXz).e. These points are important in our derivation of the S'matrix below. we use the notation of Meixner and Schafke [193] and so write the expansion of the function Mev(z. Singular Potentials where clearly MeJQ.43). 178) to be convergent for  cosh z\ > 1 but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z. (19. ^ . p. we have the following for a change from v to — v. Meixner and Schafke [193]. oiv{h) = (19.46) H^z) = e^H^(z).47) With this equation we obtain for nonintegral values of i/:§ ±isin WK MJ)3A\z. Mi 3 . h) = M™(z.50) E. Meixner and Schafke [193]. i.49) Inserting this expansion into the modified Mathieu equation (19. The functions Me±v{z.48) The series expansions of the associated Mathieu functions M„ (z.g. MJ>>(0.e.e. h).h) on the other hand can be shown (cf. s (19. Equating the coefficients to zero we obtain [{y + 2r)2 + \]cv2r = h2(cv2r+2 + c£ r _ 2 ).h2) = ] T cv2r(h2)e{y+2r)z. r=—oo (19. h) . p.444 CHAPTER 19.h) . As mentioned earlier. 130) to converge uniformly for all finite complex values of z. 2M™ = eiuvMi3) + e~ivlrM^\ .h) Using further properties like (19. For later essential requirements (i.h) can be shown (cf. i.h2): oo Mev(z. the explicit evaluation of the S'matrix element) we elaborate here a little on the computation of coefficients of normalized Mathieu functions. 4) = exp(±^7r)i\4 3 ' 4 \ M^3'4) = M™ ± iMJ?\ (19. .8.exp(TM"r)MW(.4) we obtain oo £ r=—oo [(i/ + 2r) 2 4.\c\r + h2(c^2 + 4 r + 2 ) ] e ^ + 2 ^ = 0. the first relation is obtained as follows: 2M<£1 = M^l+M^l.
51) we have 2r (19.2 The Potential 1/r4 — Case of Small h? which we can rewrite as fc2^±2 c 445 = [A _ („ + 2r) 2 ] .^ C 2r+2 (19.2 [ A . Meixner and F. h~2[\ . + 2 r ) 2 ] .53) /i2[A(^ + 2 r ) ] . 117. \ .^ p 2 2 ^ or c 2r+2 2r2 1 (19. .^ c 2r 2r 2T~2 or 2r ^2 /l2[A_(I/ + 2r)2]£2l± 2 .(v + 2r)2} / z . 2r (19>51) 2r For ease of reading we give the steps in rewriting this. Schafke [193]. + o2„r 2 l2 ]  C 2r 1 2r c 2r2 or 1 c J =^[A(.(/^.^ + 2r + 2 ) 2 ] . Thus now c 2r .52) in agreement with Meixner and Schafke [193].54) 2 C2r 4 ^ h.[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. Alternatively taking the inverse of Eq./ i 2 C ^ . (39). .* For 'Actually.T22r[\ A . see J. 122. Eq. eg = c~v = 1. (19. W. \) /.19. p. p.
(19. 8 4 ( i / + 1) 128(i/ + l)2(v + 2)(vl) + 5 10 32(i/ + ^ l)(i/ + 2) + 768(^ „„ . Aly.27) and truncate the continued fraction after the second step.5 D e r i v a t i o n of t h e S . J.52) and obtain c% 1 eg fc"[Afr + 2)»)h_a[A_(.8 ^^> 1 hi(v2+4i>+7) 32(^+2)(i/l)(^+l)2 4 ( i / + 1) /l 2 4 ( i / + 1) which is the result expected. J.( " + 2)2] ~ h2 h. H.1 for the evaluation of a typical case): hb + 0(hw). MiillerKirsten and N. M. W. Spector [257].55 ' Example 19. but follow our earlier reference H.2.12) ^ 384(^ + l)(z/ + 2)(z/ + 3)  " ^rW^hy. W.446 CHAPTER 19. Singular Potentials nonintegral values of v examples obtained in this way are (see Example 19. t For this purpose *We do this in the manner of the original derivation of R. + 0(/. .C + l) 3 (i/ . VahediFaridi [10] (this paper contains several misprints which we correct here). Solution: We put r = 1 in Eq. Liang and Yunbo Zhang [189]. Then di 2 1 t" + 2(^1) + • • • .55). Here we insert for A the expansion (19. H.1: Evaluation of a coefficient Evaluate explicitly the first few terms of the coefficient Cj/cfJ given in Eq.. (19. 128(i/+ l) 2 (i/ + 2 ) ( ! / ..+0^"> < 19 .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.Q . . H. MiillerKirsten. + i). J .[v^. Manvelyan._{u+m h2 M ( " + D + 55^iy] + i ^ 5 j /i 2 4(. i/2 + 41/ + 7 7i6 + 0 ( / i 1 0 ) .+4)!ll_.^ ^ + 2)(i/ + o> l)(i/ 3) 0(^10). and also R.l ) 19.
.58) The behaviour of y reg near r sa 0 is therefore given by 2 1/2 2/reg "\ 5 exp 1\TT r\ —r exp i .3a) and (19. and then the continuation of this to infinity. 22. (19. 2 ) 2 1/2 ex Z/7T IT H=i —J TTW P l+O 10 (19.5) w — 2h cosh z = [ kr \ r Thus r —> 0 implies \w\ —> oo..19. (19.2 The Potential 1/r4 — Case of Small h2 447 we require first the regular solution yreg of the radial Schrodinger equation at the origin. In the case of the repulsive singular potential here. I / + 2 2 l\TT +E ^ i\ E j=i.57) ^ 1 . Magnus and F.j7L0 P2i(2j)exp i[v + 2j + 2 2 . we have W r = ip + ut — 7r/4' . \w\ S> 1 and — IT < arg w < IT is known to be given by* (19. The asymptotic behaviour of the Hankel functions H„ ' (w) for \w\ ^> \u\. Then y reg = = r rV2M®(z.56) where by Eqs. this wave function near the origin is the wave transmitted into this region (as distinct from the reflected and ingoing waves).59) which tends to zero with r.h) 1/2 w) (19. Oberhettinger [181]. and considering the propagation of this wavefront. W . We obtain the regular solution by choosing Zv(w) = H„ (w) for 5ft(z) < 0. p. The timedependent wave function with this asymptotic behaviour is proportional to (here with ui — k) —iuit—g/r+iTr/4 Fixing the wavefront by setting <p = —lot + ig/r + TT/4 = const.
17.j^Q p2.A1"—2 z '+"2 A~~ n or rA .62) Also with Eq. p.448 CHAPTER 19. 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. Singular Potentials so that when t — oo : r — 0. h) through the entire range of !&{z). This means that the origin of coordinates acts • > as a sink. 178) to be convergent for  cosh z\ > 1. Meixner and Schafke [193].A14A.3b) ln • 7T Q"*4' r fg ° Vk (19.57) the condition  cosh z\ > 1 implies I cosh z\ kr + ig/r >1.61) In fact. The series M^ (z.h) = exp(Mr^)il4 4 ) (z. but uniformly convergent only when  cosh z\ > 1 for otherwise complex values of z. . p.h) oo j 2 — rV flp^ + ^fc* £ i=l j=i. 4%r g >0 4r^r ' rg See W. We require therefore > > the continuation of Mi. AV . In a similar manner we can define solutions y± by setting for 9fJ(z) > 0: y± = rl'2Ml?>*>(z./l ~ 2 ~ 2z + " 4 > 0. we can derive y_ from y+ since one can show from the circuit relations of Hankel functions^ that M^\z + m. ^ 1 ) (e i 7 r z) = Hi2!>(2) e^H^iz).60) Using the above asymptotic expressions for the Hankel functions. '2"4 .V + § 9 2 > 4hh* . Magnus and F. (19. We now require the continuation of the regular solution y reg to solutions behaving like y± at r — oo.3a) and (19. 2h and the square of this expression implies 2 2 ~ ' A.(2j)<gH • (19.j¥=o Ki(2j)exp(Tmj) (19. '(z. (19./i). Oberhettinger [181]. Now from Eqs.#.h) can be shown (cf. Prom the relation r = (ig/h)ez we see that r — 0 > corresponds to $l(z) — —oo and r — oo to $l(z) —> oo.
130).1 The domains of solutions.rl)(r2 . z = In h i— r0 4 and 7T z = In • ro { 4 for ro < r < oo. the domain r_ < r < r+ — as illustrated in Fig. Thus we choose . . A suitable set is the pair of fundamental solutions Me±v defined by (19.r2_) > 0 with r£ = rg(2 ± >/3). and in order to maintain this symmetry. 449 This condition. (r — r+)(r + r+)(r — r_)(r + r_) > 0.38) or (19. which has to be bridged by using another set of solutions. Since the real part of z changes sign at r = ro. (19. Originally we chose the sign of m / 4 as in Eqs. This interchange in the > solution My interchanges y reg with y+.3a). i.e.62).3b) and (19. (19.ijt/4 Rez . 19.19. i Im z z = In r/r0+i7t/4 .e. p.49).2 The Potential 1/r4 — Case of Small h2 Hence (r2 .1. r 7r for 0 < r < ro. 19.ft/4 _Z z = In r/rm/4 0 r 0 =(9/k) 1/2 Fig. is satisfied only for r < r_ < r__ or r_ < r + < r. These solutions converge uniformly for all finite complex values of z (cf. Meixner and Schafke [193]. (19.63) Thus there is a gap between the two regions of validity — i. The one further point to observe is that the variable w = 2/icosh z = (kr + ig/r) is even under the interchange z — —z. we have to assign different signs to the imaginary part of z in the two distant regions.
450 CHAPTER 19. i. Hence we match the right hand sides of Eqs. h) = Me^u(—z. Thus we set rl/2M^\r) = = rll2[aMev(r) + (5Mev{r)}. a4[r1/2Mev(z)] dr a'^lr^MeAz)} It follows that (19. + ^[r'^e^z)] dr z=—iw/4 + (31 ^[rll2Me. we express in the domain r_ < r < r+ the regular solution as a linear combination of solutions Me±v with coefficients a and /3.)](1964) The right hand side of the first equation now represents the regular solution in the domain r_ < r < r+. these relations can be reexpressed for one and the same point z = —iir/A: rl/2[aMe„(z) rl/2[a'Meu(z) + + /3Mev(z)]z=_in/A p'Me^(z)]z=_i7r/4. (19.z=—in/4.B^0. 131).v{z)\ . at this junction we require also the imaginary part of z to change sign.65) in/A a>d[rl/2Mei/{z)]+f3>d[rl/2Me_Az)] Since Mev(z.67) Next we have to continue the solution beyond the point r+ to a linear combination of solutions y+ and y_.e. Singular Potentials Then starting from the region r ~ 0. i. to r^2[AM^ + BM^\ A.v(z)\ —m/4) + ^[r^Me. .(2)]z=w/4 + 0Me. (3 = a . p. (3) (r)] aA[rl/2Mei/(r)]+/3i:[r.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_.^)] dr z=in/4 dr (19. and we determine these coefficients from this equality and the additional equality obtained from the derivatives.l/2Me_i/(T. Thus. as just explained.66) a = (3 . (19. h) (as one can check or look up in Meixner and Schafke [193].[ r l/2 M. At r = ro the real part of z vanishes and then switches from negative to positive.
3a) and (19. (19.69)) cancel out in view of the nonderivative relations.19.2Mei/{r)]+d[rl/2M] dr dr = = rxl2[AM^\r) + Ad[ri/2M(z)(r)] dr +B^[rl'2M^{r)].g]:=f(z)*&g(z)WZ) dz dz Thus.71) We now determine the coefficients a. (19.68) \j3Me„(z) + aMev(z)] P±[Me„(z)}+afz[Me„(z)] = = [AM^\z) + BM^(z)}. (19.70) by Me'_u (the prime meaning derivative). (3. and we are left with relations expressible only in terms of z.2 The Potential 1/r4 — Case of Small h2 451 This solution can be continued into the domain below r = r+ by matching to the right hand side of Eq.3b) we infer that (z = l n r + const. Then. and the second of Eqs. Thus we obtain from Eqs. (19. A and B.64) the relations M&Hz) jz[M^\z)] and from Eqs. we use the notation W\f. ^ ] = ^ + ^ ^ .68) the derivatives with respect to r by this relation.67): rl/2[(5Meu{r) + aMe_u(r)} pd[ri. For the Wronskian W of two independent solutions f(z) and g(z) which is constant and can be evaluated at any point. (19. (19. the nonderivative parts (from the first term on the right of Eq.65).70) by MeU and subtracting .70) = a±[Meu(z)}+P^{Me_„(z)} A^[M^{z)] +B^[M^(z)}. (19. (19. = aMev(z) + 0Mev(z).) d_ _ 1 d dr r dz Hence for any of the functions Mv:  [ .65) and (19. (19. dr BM^(r)]. (19. (19.69) Thus if we replace in the derivative relations of Eqs. with the replacements of Eq. for instance. (19. multiplying the first of Eqs.68) From Eqs. (19.
. and Wronskians W[M^\MIJ)] = [i.Mev]' P l ' From Eqs. we obtain immediately W[Z.u.47) we obtain (av defined by Eq.4] = . (19. (19.48) (or cf.^ 47 9? MJ>4)] MeV] [3.46). (19. or obtainable by substituting e. i..73) We now use Eq. h)/M^\o.58) and considering large values of \z\. W[M^\Mlf)]W[Meu. With these expressions the quantities A and B are found to be A= f au ^ 2isin VK \ a_u 1 a.75) With these relations and Eq.M^]= IT e~iv*av. 170/171).M^] J™aV. (19.u}W[Me. . B = J 2 L _ e . Me. 7T [1.A] = W[Mi3\z). p.. using Eqs.M^] W[Mev.* « * r r z H . Meixner and Schafke [193].\ \ ( av 2iwKav ^ v .Mevy W[M?\Mev] W[Mev.Mev] = = = av. 7T W[Mev. and the second is obtained similarly: W[M?\Mev\ W[Mev.g.e.71) we obtain in a similar way A B = = W\MP\ Me. (19. = IT <*„. i. M^4)] + W[MJ>3).Me^} (19.74) Moreover.57) and (19.46)) W[Mev.452 CHAPTER 19. (19. av{h) = Me. (19. 169) M0) = e«^Ma) _ isin U1X M(A) _ (19.(0.j] given in Meixner and Schafke [193] (p. we use the following circuit relation which can be derived like Eq. Singular Potentials the second resulting equation from the first resulting equation. 7T (19.76) .MiA\z)} = 2(i)(2/Trw)(dw/dz) = (4i/Tr)(2hsinhz/2hcoshz) ~ 4i/>.77) av I 2ism VK \ ctv av ^Thus. h).M^} W[Me„. .4] = .^ ^ a „ Q _ . TV W[Mev.3] = [1.] W[Mev. (19.e. we obtain the first of the following two equations. the leading terms of the respective cylindrical functions. W[MJ?\ M^] W[Mev.
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 .85a) . Z ) e . Z ) e ^ ] .h)] ( m / 9 \ 1/2 {L\ {Aeikrei(V+l/2)*/2+Beikrei{v+l/2)*/2^ g) We set fl = ^ = MM).79) where the second expression follows from Eq. in Chapter 16). Then f' .(ite™)" 1 { 8 J ' f(k.f c .(l)le^}Pi(cos 9).82) The S'matrix is defined in the following way in the partial wave expansion of the scattering amplitude /(#).83) The Smatrix element is therefore given by ifc*"* .46) together with the relation Mev(z. (19. (19._ [ / ( f c . (19.g.2 The Potential 1/r4 — Case of Small h? The regular solution in this way continued to r ~ oo is then y reg ~ ~ 453 Tl'2[AMiz\z. (19./) = (l)1/2R2^2i^)ei^)U^.h).78) we obtain Ay reg « _ . eii/7r/2 (19. B=tfD'"". f c r .80) 2^fc (1981) where \7r/ i t sin VK /(*.19.h) = Me_u(z. (19./ ( .h) + BM^\z. 2zitsm z/7T zzitsin I/TT Inserting these expressions into Eq. as we recall from earlier considerations (e.l) ' We note already here that with the substitution R = ei7r7. (19.
we can write this (multiplied > by 2ismvTT.86) .85b) We shall obtain the 5matrix in this form in the case of large values of h2 later.454 we can rewrite the .58)) in the limit r — 0. We can rewrite Eq. and recalling that yreg is proportional to a function Mc> {z. we can define respectively as amplitudes of the incident wave.81) in another form from which we can deduce the amplitudes of reflected and transmitted waves. Fig. (19. (19. (19. (19.56) to (19. Thus in terms of the (3) variable z. Singular Potentials r(i+k) sin 7r(7 + u) (19. the reflected wave and the transmitted wave the quantities: A Ar At = = Re1' R 2zsin 7r(7 + u). (19.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.2 The repulsive potential and the various waves. and with the left hand side following from Eqs.56). 19. cf.Smatrix in the form s i n 7T7 Si = CHAPTER 19. R — — = 2i sin ivy. h) (see Eqs. 9?. R 2i sin TXV. V(r)=g2/r Fig.2. 19.58)) 1/2 2r e~ 2hir cosh z.
" \R~Ji\2 1  tO _  2 i s i n VK\2 [  . / i ) = Y. (19. (19.6 E v a l u a t i o n of t h e S .Q. i. 178). S. (p.e. J. " S e e S.19. MullerKirsten. therefore we cite it from Meixner and Schafke [193].( i J e ^ ) " 1 ! 2 = R2 + ^ . Manvelyan.h) = 1 Me V\ 5 / r J2 __ 0 O c»2r(h2)J„+2r(2hcoShz).39)) the associated Mathieu function expanded in terms of Bessel functions of the first kind.88) the power expansion of the Bessel function one realizes soon that the expansion is inconvenient in view of its slow convergence. but in the 10dimensional string theory context. if R becomes a pure phase factor and eil/7r a real exponential. Gubser and A. A derivation is beyond the scope of our objectives here. 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. Eq. however not here. W.„„„X (1 Q Q'T'l l \Reiun e~iuw/R\2 ~ \Reiv* .89) .88) where the factor Me„(0. This implies basically the evaluation of the expression R of Eq. (19. i.h) The function Mo (0.m a t r i x Our next task is to evaluate the expression (19. The expansion is: 4 r ( / i 2 ) M S ( z . i. (19.±)2 + (2sinj/7r) 2 .e. this expansion is given as 1 oo MJ> Hz. in which v is complex.2COS2VTT = (R.e.79). unitarity is preserved.28)) and R = ey real.** 19. By inserting in Eq. i.e. 180). y We observe that this relation remains valid if the real quantity R = e and the pure phase factor eW7r exchange their roles.In Meixner and Schafke [193] (p.2.. (^)l4lu{h2)Jir(hez)J±u+i+r(hez).h) is (cf. Liang and Yunbo Zhang [189]. H. (19. unity minus reflection probability = transmission probability. h) serves the use of the same coefficients as in the other expansions. Jv+2r. Hashimoto [124] and R.84) of the ^matrix. Eqs. (19.e™*/R\2 ' ' . and then exploit it for the evaluation of the quantity R. J.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. (105) to (108). The latter is what happens in the S'wave case of the attractive potential. l=—oo 1 1 lite*"* . M^l)(0.
49) from which we obtain for z = 0: Me„(0.94) . (19. setting r = 0 in Eq.91) Inserting here the coefficients given in (19.90) and choose r = 0 (one can choose e.3 9 ^ .. ^ (19. e.h) v 2 = ^2 (l)lc%ih2)Ji~r(h)JM+r{h). (19. we obtain M&(0.55). + 2 u l\2j 2 fh\2 "2+2 2 MX4 {v l)(v 4)\2 2 2(^ + 4 ^ . (19. by allocating different values to r. /=—oo (19. one obtains *. Here in the case of the repulsive potential. (19.g. u»^.90).90) is in some sense amazing: It permits the evaluation of one and the same quantity M±J (0.g. i. Eq. r = 0 and 2.e.6 2 ) m 6 (^_i)3(z. (19./>) = Co> 2 ) £ 7 " = — OO j ± > . As a matter of introduction we recall the definition of the coefficients with Eq. (19. r = 2 as a check). 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.2_4)(l/2_9)iv2y) + <AA. Next we evaluate M^1}(0.456 so that in particular for z = 0 oo CHAPTER 19. The formula (19.28)).49).h) = Jo(h)J±Ah)^^Ji(h)J±„+i(h) c0 {n ) ^ ' C^r^J2{h)J±l/2{h) cr(h2) + •••. h2 is complex but v2 is real (cf. h) in many different ways. Singular Potentials 4r (h )M$(0. /n ^ .90) Here the coefficients c2"{h2) are the same as those we introduced earlier for the normalized modified Mathieu functions in Eq./i) with the help of Eq.
1 ) 2 ( I / + 1 ) ( V 2 .93) and the coefficient expansions (19.7) F " (i/±l)2(^^l)(i/24)V2.9) V2 1 fh ±v 1+ 2 + • (19.97) + This expansion will be used below in the low order approximation of the absorptivity of partial waves.4)(i/ .95) These expansions imply for the quantity R: I/! 2v ri1 i .55) and collecting terms in ascending powers of h? (here for the case of nonintegral values of v).l) 5 (19.32i/ .96) With this explicit expansion we can evaluate the S'matrix.96) we extract for later reference the relation sin7rz/ R IT h h 2\ 2 Av [v\(viy.2 The Potential 1/r4 — Case of Small h2 457 Inserting the power expansion (19.... In this procedure our attention is focussed particularly on the quantity called absorptivity in the case of the attractive potential. We observe that this approximation involves v and h4 and hence is real.!' or with the help of the inversion formula of factorials R = v\{y — 1)! smKv f h 7T 2v (^ . . + # "+" ( I / .2^ 2 ± 59^ .2 _ ^ 4 ( ^ 2 _ 4)2 (19./I) h u . 2 (h\2 ^ l/2_1l2^ 2(i/ 2 +3f~7) /feu . From expansion (19.4 ) V 2 > 1 (I/)!V2/ [1+ 2_Mx2 2(i/23«/7) (h)4 (iy+l)2(i/l)(^24)' . _2_(h\2 .23) (hxQ 2 3 2 2 V ± l) (i/ T 1) (^ . one obtains M£!(O.111) / / i 5 4 3 1+ 4i/ 2 + +' (z.19.] 2 \v i + {v2 .12i/2 + 64u .1) 2u(4u + 15i/ .IV 2 2 2(v =F 3i/ . 4(z/ T H ^ 3 .
(19.100) Here we set 2^/3 = 1 + j / ? e 2™ = l + gt (19.9 / . _ i?* 2 2+ g # 2 i?* 2 2 3/(i + s)(4 + i?*2). Thus we set v = n + i(ot + i(3) = (n — 0) + ia.102) (19.2.101) where / is complex and g is real. + ( i + s w (E 4 .99) which can be rewritten as SiST „2ira 1 1 1 R2)V 1 1 R*2 16 1 R2 Ana „2na + i 2 leZ7rQcos27r/3 R* +1 )e 2 ™sin27r/? }]"' (19. (e iun e~iv*^/R*2)' (19. In this case R — R* and so 1/i?2 ~ Ofji4). where n is an integer and a and (3 are real and of 0(h4) We have with Eq.458 19.28)). Eq.101) is > zero. (19.R*2 ^ * #2 \ 11 tf2 R* (19.28) that v is real.104) We now consider the case of a — 0 implying that g of Eq. 9 / « i(2yr/3)2 ss i(sin 2vr/3)2 ~ 2 sin2 TT/3. and sin 2TT/3 = ft/.84): Si ST 1 (11/R2)(11/R*2) .^ i 4 2+^ # 1 / •. (19.eiun/R2)(eiu*n ~ (19. (19. This is the case of real . 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. (19. Then cos 2TT/? = 1 . 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.103) Then SiSt (1+5) i .98) (cf.7 CHAPTER 19.
61n2 .2 The Potential 1/r4 — Case of Small h2 parameters v = n — (3.* )  14(^V = 1 T h e absorptivity A is therefore given by A = 1 .28). J. MullerKirsten. p. Gubser and A. We can also evaluate now the amplitudes (19.107c) Here h2 is actually Vh4. With t h e help of Eq.1)\} 2 h \2 iv 1 + Au {y . "This is the result effectively contained in R.SiSf » 4 sin7rz/ R (19. . At = 2i[l + 0(h2)}. Using the derivative of t h e gamma function expressed as t h e psi function ip(z) = T'(z)/T(z).Q. (19. (19.1 ] Oih* For S waves this implies an absorptivity given by ((—1/2)! = 1/7F) (19.19.3. = 1 . 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. i. S. H. (19. W. 2 /l4 '•i Al=0 2[(l + 1 / 2 ) 2 .86) and obtain in leading order for I — 0: Ai = *[l + 0(h% Ar = ±[l + 0(h2)]. Hence with Eq.6C . ( I l l ) ) in agreement with a result in S. Here in t h e case of the attractive potential (as remarked after Eq.107b) = Ah2[l + 0(h4)}.106) and evidently violates the unitarity of S. Manvelyan. J. Hashimoto [124].107a) In this result* we now have to insert the expansion of the parameter v given by Eq.^ ) ~ 2 2 459 1+ 9 / • & ( ! . Liang and Yunbo Zhang [189] (there Eq.3b)) h2 is real for E = k2 > 0.l ) 2 2 \) +0{h«) (19.61n/i} + ~9~ 0(h8) (19. one can calculate the next term of the expansion so t h a t At=Q = 4 ^ 4h4 {7 .577.97) this can be written A 4TT2 [v\{v . (19.e. Magnus and F. 1 29/ ^ (1 . f W .102) we obtain Sis. and t h e values of ip(\/2) and ip{2>/2) as given in Tables^ with Euler's constant C ~ 0.105) sin7r/3 (19. Oberhettinger [181].
completely different solutions.108) For the large values of h2 that we wish to consider here. Aly. Park.460 CHAPTER 19. J. 2. H. . MiillerKirsten and Jianzu Zhang [226]. i. K. J.1. H. Tamaryan.4) which — for some distinction from the smallh2 treatment — we rewrite here as 2 + [2h cosh 2z . 19. (19. Miiller and N.h) (19. W. where the part of interest here is based on H.4 Fig.3.3 The function q(l.109) ^We follow here mainly D.3 19.H. 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. This is therefore a highly interesting case which found its complete solution only recently. Wannier [278]. 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. S. W.e. H. Singular Potentials 19. we again make use of the replacement (17.$ q(l. VahediFaridi [8] and G. h) for large h and upwards I = 0.3. 2 2 a = [l + 2 (19.h) 10.a }tp = 0.26) and hence set a1 = 2hl + 2hq + A{q. We are again concerned with the associated Mathieu equation given by Eq.
which therefore cancels out or can be taken to be 1.113) 'We are considering nonintegral values of v. 19.112) In Chapter 17 the unnormalized even and odd large/i2 solutions of the Mathieu equation were given by Eq. (19.§ a W[y+.3.2 T h e F l o q u e t e x p o n e n t for large h2 In our treatment of the periodic Mathieu equation in Chapter 17. (19.109) in this way enables us to obtain asymptotic expansions of solutions very analogous to those of the periodic case. y± oc A{z)e2h sm * ± A(z)e2n sm z . We also found the boundary condition (17. Therefore the second term on the right is nonzero. and a is its normalization constant.y} where VF[y+. i. See also Eq.l + 2y+(l.3 The Potential 1/r4 — Case of Large h2 461 where q is a parameter to be determined as the solution of this equation.21b).3. with an essential mathematical step. The replacement (19. since it is clear that large h2 considerations require a knowledge of the large/i2 behaviour of the Floquet exponent v. and A / 8 is the remainder of the laigeh2 expansion as determined perturbatively. to which we have to return in this subsection.19. (17. We begin. y(0) = 0.yJ\ — 1 and cos TTV . With the boundary conditions y+(0) = l.e.h*y_(l.h*y (19. however. .110) a where y+(7r) is the solution evaluated at z = TV.e. J4(0)=0. which is even about z = 0. i.21b).5) that this is given by the relation cos iris = ^ M .111) we have W[y+.y^] = ab\f\ is the Wronskian (in leading order) of the solutions which are even and odd about z = 0 respectively. (17. we defined the Floquet exponent v and observed with Eq. The behaviour of q as a function of h for some values of / is shown in Fig. (19. y'_(0) = 1. 19.63). (17.
K [ (z)}2hsinz (19.114) at z = 7r/2. (19.32). We now obtain the expressions needed in Eq. (19. Hence we require (obtained from Eq. .39 2 14 /i 2 + (19. W. Kin B y± oc y+(z) = N0 A(z)e2h A(z)e2h sin z + A{z)e~2h . which is \/2 at z = 0. Dingle and H. Miiller [73]. 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 . (17. See R. and hence with Eqs. W.e. (17. 1 1 Or see R. This paper contains the normalization constants of large/i 2 periodic Mathieu functions. (19. Miiller [73]. in the second expression 2h is obtained in addition from differentiation of the exponential factor). Dingle and H.118a) "We emphasize again: For integral values of v the normalization constants are different.111) we have^ iVo = 2 3 /2 1 + °{k and K = ¥frh l +O (19. (17. B.462 CHAPTER 19.117) a a Referring back to Eq.115) Setting z = 0 we obtain in leading order y+(0) = 2iV0A(0) and y'_(Q) = 4hN0A{0).43) we see that z = 7r/2 implies W(TT/2) = 0.114) a a Our first step is therefore the determination of the normalization constants No.A(z)e2h sin 2 sin z sin z (19. (19. i.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 .116) (in the first expression we have A(z) ~ Aq(z) with Aq(z) given by Eq. Singular Potentials and their extension to around z — TT/2 by the relations C[w{z)\ ^_2h sir Mz)}„2h. J.112) from evaluation of Eqs. B. J.
119a) This formula — derived and rediscovered in Park.8 g + 3) +0 ^2 (8h)^[~l(q + l)]\[l(q + 3)}\ 2eh ' ~ (19. (17.62b) (see also the explicit calculation in Example 17.[_ l ( g 2h and hence 5 + 3)]! 'All Ah 1+2 (3g2 .112) we obtain COS TTU + 1 7T '2Wa[\(ql)]l[\(q+l)]\ 02h 1 ¥h + 2TT ' 2^ha [\(q . p.119b) COS TTV + l ire This result gives the leading contribution of the lavgeh2 expansion determining the Floquet exponent v.87 2 14 /i 2 [ i ( g .62a) and (17.! ) ] ! [ . (19.h .3). a in Eq. Tamaryan. Presumably they extracted it from "between the lines" of a sophisticated paper of Langer's [159] which they cite together with others. .3 The Potential 1/r4 — Case of Large h2 and 463 as dz )z=n/2 dw (dw\ o\dzJz=n/2 1C%2 . (19.2g + 3) 27/i ( 3 g 2 .118b) The factors a. Thus in particular (8h) i(?i) a (3q22q + S) 27h + l\(qW " ' Inserting these various expressions now into Eq.117) are known from the matching of solutions in Eqs.1Q11QM cos irv + 1 ~ •==— e .\ { q + 3)]! 7re 4/i 24/i + (2/i 1 /2) ( 8 / l ) 9 / 2 [ _ l ( g + !)].i ( g +3)]! 1 24/i {2hl>2). With the help of the duplication and inversion formulas of factorials the result can be reexpressed as cos( g 7r/2)[£(gl)]! . (19.19. 210. (19. MiillerKirsten and Zhang [226] — is cited without proof by Meixner and Schafke [193].! ) ] ! [ .
h. —z).109) into the equation (19.3.3 Following the procedure of Sec. Vcosh z \ 1 .120) The resulting equation for the function A(q.123) We again make the important observation by looking at Eqs. dA 1.z). .z) = A(q.109) remaining unchanged. z) are TP(q.2 we insert the expression (19. of course. i = 1. . In the following we require solutions H e ^ ( z ) . —h.z) := = exp[i7rg/4] j===^Aq{z) exp[2/wsinh z i(77r/4 V2ih k(q.108) and set ip{q.z)exp[±2hismh z\. the expression (19. h(q.2. Then straightforward integration yields > inq 4 „.h.z). —h. Singular Potentials C o n s t r u c t i o n of largefa 2 solutions 19. h. h. each with a specific asymptotic behaviour.s i n n z ± iq)A = ± K qj dz 2 Aih d2A dz2 (19.. z) = ip(—q.h. for h — oo).e.3.120) to (19.? s m h z) Vcosh z (19. h. z). —z). tp(q.z) or as tp(q. .124) Since this function differs from a solution ip by a factor k(q.h)iP(q.4. we can obtain the linearly independent one either as ip(—q. h.h) =?==.464 CHAPTER 19.h. h.121) We let Aq(z) be the solution of this equation when the right hand side is replaced by zero (i. it is still a solution but not with the symmetry property ip(q. (19. 1 / l + isinhz\T9/45R^oo e^ / Ag(z) = ===[r^— ~ .z) i/>(q.h).122) Correspondingly the various solutions ip(q.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. We define these solutions in terms of the function Ke(q.h. With the solutions as they stand. 17. (19. „ 1 A „ —A cosh z——H . V— 2in (19. h.h.h. z) can then be written as .122) that given one solution ip(q.3. z) = ip(q.
<5±1) (19 125)  (the expression on the far right in leading order for h2 large). [ \ .45). It follows t h a t S e t t i n g s i n h 77 = t a n 0 .19. we have to match a solution valid at Kz = — oo to a combination of solutions valid at $tz = oo.fi2) can be reexpressed in terms of Hankel functions. we h a v e 77 = 0—> 6 = 0. 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. This is.e. achieved with the help of the Floquet solutions Me±v(z. In order to be able to obtain the Smatrix. : 7r/2.e. we h a v e ( u s i n g cosh 2rj = 2 c o s h 2 77 — 1) ry Jo ' = dr. (0. the Bessel functions contained in the expansion of the associated Mathieu function M{.^ s i n h 7?)]g. (19./i 2 ) M (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]. Me„(z.WJ " C o n v e r t i n g t o n o t a t i o n of W a n n i e r [278].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 . . / 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 performing this cycle of replacements the function picks up a factor. 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 . as we saw earlier. i.3 The Potential 1/r4 — Case of Large h2 465 Instead. One can show that the quantity < o of Wannier* (see above) is related to q by $o = & iqir/2 + OO.e. (19. . (z./h). 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 . av{h2) = ^(0J ^ Mi i .127) sin z T /'1 M TT\ — <>{ 4 . i. * ** = > rojKe( ". i. We observed that these satisfy the relation (19. or — equivalently — by the appropriate expansion of the Bessel functions as given in Tables of Functions. Ju(z) TTZ VK 7T cos z T~ 4 1 + 0 ^V.h2) = OvMPfrh2). 2j<.h2).126) For large values of the argument 2/icosh 2. (44) of W a n n i e r .fc 2lqk A + 1/2 + 4fc2 c o s h 2 77 A / 8 P . >• * F $ 0 * e".
exp[—ikr — in/4\ kr Re{2)(z.130a) Re^(z. He (z.129) The solutions so defined have the following asymptotic behaviour (where e(z) = (2/icosh z)~ll2 and h2 = ikg): H. HeW(z. (19.q.4 T h e c o n n e c t i o n formulas We now return to Eq.h) = e{z) exp ihez + iexp[ikr + i7r/4] kr 7T . (19. (19.q.g./i).3. z): Re(2\z.q.h) (3) = = = = Ke(q.eW(z.h) = e(z) exp ihez .q. (19.h )~ \ ot±v cos(2/icosh z = vir/2 — 7r/4) F V2/tcosh z (19.h) He (z.g.h). (19.vr .128) We now define the following set of solutions of the associated Mathieu equation in terms of the function Ke(q.h.q.q. h.130b).466 CHAPTER 19. In this way we obtain in the domain . 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.130b) r_0 rll2exp\g/r (ig) ' 19.q. 3te»0.z).h) = e(z)exp exp ihe^ 4 + i\ 12 i4 3^cO.128) and consider the cosine there as composed of two exponentials whose asymptotic behaviour we identify with that of solutions of Eqs. 3te»0.h).h) Be^(z. (2) (19. KeW(z.130a)./i) Be^(z.q.
. h.133) In a similar way one obtains the connection formulas — sin nv HeW(z.19.h).h) is asymptotically small (cf. g . Changing the sign of z. z) and He (3) (z. h) = Ke(q.134) in the region where the function He^(z. e x ^ r / 2 H e ( l ) ( 2 > g> />) _ e .3 The Potential 1/r 4 — Case of Large h2 fc>0 the relations: = 2TT 467 Msv{z. and thus can be matched in this region.h2) a. (19. q.« * r / 2 H e ( 2 ) ( ^ g> ^ ft) _ e W 2 H e ( 2 ) ^ ^ ^ = a_ 2TT e .132) These relations are now valid over the entire range of z.q.q.(z.q. z) are proportional there./i) — sin 7T7 Re^(z. (19.135) sin irv 'This means. t With Eq. Substituting the Eqs./i 2 ) = Me_j. (19. h) = sin 7r(7 + v) He (1) (z.129)) He (2) (z. (19.130b))./i). and He(*> exponentially decreasing.li 2 ) Oil. (19. q.g. e W 2 H e ( 3 ) ( ^ ? j ^ _ e M/*/2jj e (4) ( ^ q j ^ e . q. (19. these are exponentially increasing there.134) Considering now the first of relations (19.131) and eliminating He") and setting again exp[m7] := we obtain the first connection formula ozwr — sin irv He (4) (z.i ^ / 2 H e ( l ) (z> ?) (19. Eqs.(z.h2) = 2vr i a_ 2TT />) _ e ^ / 2 H e ( 4 ) ( Z j ^ Q (19. h) = Ke(q.h2) Me„(z. h) — sin 7T7 He^(z. (19. (19.131) where the second relation was obtained by changing the sign of v in the first.g.130a). we see that the solutions (compare with Eqs. / i ) + sin7r(7^)He(4)(2./i).* ^ / 2 H e ( 3 ) ( Z ) g? Mev(z.132) into Eqs. — sin 777 H e ( 3 ) ( 2 .h) sin irv = sin7r(7 + ^)He (3) (2.125) the proportionality factor can be seen to imply the relation exp •ffa + 1) sin7r(7 + "} (19. we obtain in the domain tfcz <C 0 the relations: Me. q. h.g.
(19. where 8i is the phase shift.136) ^ jrfl+l/2) (19:I35) sin 7r(7 + v) sm WKljrtla/2) S m TXV (19.139) ''Equation (19.3. sin TXJ. Thus here the S'matrix is defined by (using Eq. and therefore remarks after his Eq.5 D e r i v a t i o n of t h e ^ . (19. (19.m a t r i x From Eq. The latter is defined by the following larger behaviour of the solution chosen at r = 0. (60) in the work of G. cos TXV + sin TXV.135). Squaring Eq.1 .l ) ] ! c o s ( W 2 ) 2 2 (8h)i/ 2TT29/ 2 +0 (19. ' cos TXV ± v 1 + elin sin2 TXV =p v'cos 2 TXV . (19. is only the dominant term for large h2} 19. exp 2^(9 + 1) and solving for sin TXJ one obtains the expression sm 7T7 sm TXV = ~iei7rq/2sin le iqK • COS 7T7. (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". which in our case is the solution He^4^. Singular Potentials The factor on the left.138) TXVICOS TXV We obtain the behaviour of the Floquet exponent for large values of h2 from Eqs.e.119a) or (19.119b). Wannier did not have the relation (19. q.137) We see that this expression agrees with that of Eq.(19.468 CHAPTER 19. of course. H.l ) (l„—ikr e From this we can deduce that Si (19.135). The quantity 7 is now to be determined from Eq.Sin TXV.133) we can now deduce the 5matrix Si — e2t51. Wannier [278]. i.( .85b) obtained earlier. (19.133)) — sin TXV He^4' (z.l ) ' s i n 7 T 7 ^VK /2 ikr sin 7r(7 + v) _l)leikr i5i„—ilir/2 SxelKr . (19.135) is the equation corresponding to Eq.119b) as „4/i 1 + COS TXV ~ rrl ( g .119a) and (19. h) ~ — sin TXV (1) sin 7r(7 + v)e kr ikr in 4 rl/2eg/r+m/A / (ig)1/2 ( .e'i7ri 1. .
A{l. 9/2 (91) with the approximation q ~ h obtainable from Eq. (19. This is different from the behaviour for small values of \h2\. for the attractive potential.109).138) we obtain. the real part of v must be an integer.139) could actually be neglected) implying  cos m>\ = cosh nvi. irrespective of what the value of I is) by cos™ + l = c o s ( T j^J y/e1. (19. . imaginary part uj (so that 1 on the left hand side of Eq. i. Using Stirling's formula in the form z\ ~ e~zzz+1/2V2^. 19.8h COS From Eq.e. since cos TVU is large. h) of the Zth partial wave.3 The Potential 1/r 4 — Case of Large h2 469 Since the right hand side is real for real h.137) together with (19.141) COS TTU irvj — i s i n WR s i n h TTUJ. we can approximate the equation for q ~ h (i.e. Si xe ~ te ilir COS KV V cos2 irv — 1 — e (COS^Q COS ITU ig7r 1 + e'iqn ielln cos TXV 2 2 C O S TTU (lp/2)n (19. (19.e.28) shows. as a comparison with Eq. T h i s is r e a l for UR a n integer.h):=l\Stf ^cos rr(uji + ivi) = cos Tn/jicosh COS ^7TQ N 2 1 (19.140) From this we obtain the absorptivity A(l.4 The absorptivity of the S wave (attractive case).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. i. (19.
**R. Challifour and R. Pope and T. Shabad [248].4 (there are tiny fluctuations in the rapid approach to 1). Jones [78]. 11 J. J.' Rudimentary aspects of a singular potential with a general power n have also been considered previously. H. Fubini and G.^ A highly mathematical study of the potential ~ r~2 as an emitting or absorbing centre has been given recently.139) 2Tr(16h)q /32\ f e Thus with Eqs. H. 19. D. N. Apart from the papers already referred to. Y. S. W Some of these sources can be helpful in further investigations. H.470 CHAPTER 19. Eden [46]. Masson [191]. Pao and J. (19.e. Treiman [271]. one may expect corresponding results for other singular power potentials. YuanBen [285]. . Lii. Dombey and R. Esposito [88].k) in the approach to unity are too small to become evident here. Furlan [20]. 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. Tiktopoulos and S. Bertocchi. Cvetic.107b) and (19. Limic [177].** 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. and. N. The diminishing fluctuations of A(l. Cvetic.4 . (19. S . Rosendorf [225]. it seems it is not possible to guess from the above the result for the absorptivity of a singular potential with an arbitrary negative integral power. **A. small h2) case have been considered by some other authors. In essence. I r a n [61] and M. Paliov and S. N. Partial or other aspects of the weak coupling (i. This is sketched schematically in Fig. and G." as well as other aspects. E. C. D. B. Apparently this is one of the very rare cases which permits such complete treatment. such as phase shifts. A. A. Singular Potentials with near asymptotic behaviour (for h2 — oo) > Afi . see also M. F. Lew [128]. "G.P. VazquezPoritz [61]. in further contexts beyond those already mentioned. Handelsman. in fact. Lii and J.142) we can see the behaviour of the S'wave absorptivity as a function of h. A. L.4 Concluding Remarks In the above we have considered the solution of the Schrodinger equation for the potential r .
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. Muller [74]. Dingle and H.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. f R . Dingle [71]. T. the subject is much older and had been explored earlier in great detail in particular by Dingle^ with the subsequent investigation of the behaviour of the late terms of asymptotic expansions of the eigenvalues of the Mathieu equation and others. *R.Chapter 20 Large Order Behaviour of Perturbation Expansions 20. 471 . W. Wu [18]. 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. how the large order behaviour of the expansion of the eigenvalue is related to the discontinuity across the latter's cut. [19]. Bender and T. also convergent expansions are known and a lot of literature exists on the equation.g. the asymptotic solutions in different domains all have the same coefficients. J. We shall also see explicitly. the cosine potential is more suitable for such studies than the anharmonic oscillator since its case is simpler.* However. M. B. e.* In fact.
20. 19.2 The function ur = r\/pr vs.1 and cases in Sec. r for p = 4. Since the method is also applicable to convergent expansions.472 CHAPTER 20. Fig. 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. 8. Presumably this has never been done so far.8.6.2 and below. r for p = 4. 10 12 14 16 18 20 Fig. This approach leads . 20. 5.2. 8.1 The function ur = pr jr\ vs.2. Large Order Behaviour of Perturbation Expansions expansions using in particular the recurrence relation of the perturbation coefficients obtainable with the perturbation method of Sec. one could naturally also explore the large order behaviour of these in a similar way and expect the behaviour discussed in Sec.7. as evident from Example 17.6.
20. the function E(g2) is given only approximately by Yln=o En92n.2. and the integral E(g2) is called the Borel sum. .1) is an asymptotic series if for any N lim 92^o 1 g 2N N E(g )^2En92n 71=0 2 = 0 (20. The behaviour of the rth term (r large) of the convergent expansion is generally of the form of a power divided by a factorial (as.3) n=0 Thus for any given value of g .J2 2n h E (20. for instance.4) Jo Jo its Laplace transform F(z) is called the Borel transform of E. whereas in the case of the asymptotic expansion it is the opposite: a factorial in r divided by a power. and the expansion is said to be Borel summable. 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.e. E(g2) = f2 Eng2n = Eo + J2 n=0 n=0 E 2(ra+l) ^9 (20. The formal perturbation expansion of a quantity E(g ) . It can be seen from these. 20. 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].e. If the function E(g2) can be written as the Laplace transform E{g2)= dze'z/g F{z)=g2 dte^FigH). In some cases the answer is affirmative. i. equivalently for g = l/h. this term represents the size of the error.1 and 20. that the maximum of the absolute value of terms in a convergent expansion and the minimum of the absolute value of terms in an asymptotic expansion are reached approximately when the value of the variable is approximately equal to the number of the term.2) or. The question arises: What is the information about F(z) that can be obtained from the asymptotic expansion of E.1 Introductory Remarks 473 automatically to asymptotic series. i. as we discussed earlier in Chapter 8. The behaviour of such terms is illustrated in Figs. (20. in the power expansion of the exponential function). T V lim h2N 2 n EQ?) . even if each En is known.
Large Order Behaviour of Perturbation Expansions Eq. . (20. In any case. In general this is not so easy. .6) The power series of F(z) converges inside the unit circle about the origin at z = 0. We can therefore write down a dispersion relation representation choosing the cut from 0 to +oo: . J we obtain Now if />oo dteHz\ °° oo ' 0 dteH*\ TEn+1g^V . 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. In fact.1)? Asymptotic expansions originate from integrating a series (i. We observe that a function E(h) with asymptotic series ^ r=\ A has an essential singularity at h — 0. . n=0 (20. n=0 (20.n _ En+\ n! En+1 ^ (!)"«!. In this ideal case the integral of Eq. . . that of F(z)) over a domain that is larger than its region of convergence (see Chapter 8 and Dingle [70]).e. many asymptotic series do not even alternate in sign (as in the case of the series in Eq. 1 we obtain oo F(z)£(*)" = — . (20. El . (and E0 = 0). Thus if CO F(z) = aiS(z) + J2 anzn.474 CHAPTER 20. (20. it is important to know the large order behaviour of the asymptotic series so that the question of its (exact or only approximate) Borel summability can be decided.4) can be evaluated and E(g2) thus recovered from the asymptotic expansion by Borel summation.4). . Q.n E(9 )~ 2 n=0 ^ E0 +1) = Ea^2( °° oo /»oo = J2ang2^ „n 'O n=Q dte~H\ a_i = E0.6)).5) then with the integral representation of the gamma function F(z) = (z — 1)!. (20.
if the potential is V(r) = +grN. Comparing the last two equations we obtain (by expanding in the latter the denominator in powers of h'/h) I jQO Ar = 7T JO tir~l%E{ti)dti.3 The potential V(r) gr». N = l. M...10) is considered in detail in this work. (20. and with g = 1/h and it is found that§ En = 4 n ) ~ \lis).1 Introductory Remarks 475 except for possible subtractions. (20.g. Weinberg [228].. Popov and V.20. e. § where 7(ff) = To exp  . (20. 20.2. S.. if the potential or a boundary condition is complex such that the problem is nonselfadjoint..2. This is the case if the real potential has a hump of some kind. 20.../ \p\dr\.. we have a quantum mechanical problem where probability can leak away. and E represents an eigenenergy. ? When ?s E ^ 0. i.10) r as shown in Fig.N = l.e.. The potential (20. for < ^ 0.3 and g < 0. or can we determine Ar independently? Fig. .11) V. $s E can be calculated by the WKB procedure.9) The question is therefore: How can we determine 9 E(h).
20. Eletsky. Solution: The solution can be found in a paper of Eletsky. Thus in the following we apply these methods to our typical examples. of course.11).134)). Example 20. Weinberg [83]. the BreitWigner formula (10. 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 .73) in the W K B or semiclassical approximation. . S. (20. 8.11) is. (14. Popov and Weinberg.1. as was (and still is) not the case for t h e equation 1 V .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.^ The logarithmic factor in Er is a novel feature of this case. L. since this is effectively the Mathieu equation for which — for comparison purposes — extensive literat u r e was already available. Popov and V.76) (see also the comment on the comparison with the Kepler problem before Eq. V. Evaluating the integral show that 00 r=0 v ' where n is the principal quantum number of the Coulomb problem as in Eq.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. An interesting application is provided by Example 20. (20. (11. Straightforward RayleighSchrodinger perturbation theory usually does not enable this in view of the unwieldy form of coefficients after very few iterations. the eigenvalue expansions of the cosine potential and of anharmonic oscillators. as we expect on the basis of Eq. The most direct way to approach the problem of determining the behaviour of the late terms of an asymptotic expansion is to consider the equation determining these coefficients.9) for Ar.2 Cosine Potential: Large Order Behaviour It is instructive to investigate various methods for obtaining the large order behaviour of an asymptotic expansion. M. Equation (20.113)). the cosine potential being t h e most completely investigated will also be seen in the following. In the first such investigation it was sensible to consider the Schrodinger equation with the cosine potential. However. Here n is the principal quantum number of the Coulomb problem (see Eq.476 CHAPTER 20. (11. the perturbation method of Sec.
55)) X = E(q. Dingle and H.Bq. 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. .3 ) P i _ 1 ( g .e. Eqs. i. B.2 A = Mx + ~M2 + 7^WM3 + •• • . Eqs.M2i+1 for large i. (17. (cf.Cq.l(q. The recurrence relation (20.58) and (17.j) +(q + 4j + l)(q + 4j + 3)i^_ x (g.13) In the special case j = i the boundary conditions stated after Eq.13) represents effectively a partial difference equation. Eqs.l ) +2[(q + Aj)2 + l + A]Pl. Thus in the case of the Schrodinger equation with periodic cosine potential we saw (cf. Miiller [74]. (20. W.j) of functions ipq+4j. 3)? . are given by i .j + 1).2 Cosine Potential: Large Order Behaviour 477 with an anharmonic oscillator potential.34)) that the coefficients Pi(q.12) (20.14) Thus these coefficients can be obtained once the coefficients Pi(q. (20. % = Aq.j) are known. With some simplifications for large values of i this equation can be solved approximately and the large order behaviour of the coefficients Pi(q.60). (17.l ) ( g + 4 j . (17.26) and (17. J.j) a n d M2i can be deduced.61)) that the coefficients M2i of the expansion of the eigenvalue A = E(q.l ) / 2 ] ! i![(gl)/2]! • We have seen earlier (cf.58) eliminate the last two contributions and one obtains the exact expression ^ ' Z j _ 4<[2» + ( g .h) = 2h2 + 2hq+j. (17. or equivalently of the quantity A.20.J) = (g + 4 j . This has been done" and it was found that (we cite the results here since we rederive them below in detail by other methods. obey the recurrence relation jPi(q. M2i = 2j2 J[Pi(q. h). arising in the ith order of the perturbation expansion of the wave function oo 1 i ip — const. j . (17.
l (20.17) We see that assuming i is very large compared with m is here equivalent to assuming m is approximately zero. Large Order Behaviour of Perturbation Expansions where the bracketed expression at the end is the binomial coefficient.16a) (20.478 CHAPTER 20.91 l)]![i(g3)]!} 2 { [ .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 7.19) ^nOW =2(9"1) We observe that successive terms do not alternate in sign. Hence exact Borel summation is not possible.•91 {[l(? + i)]![(? + 3)]!}2 (20. .l ( 9 + l)]![(Q + 3)]!}2 i!^" 1 1 or (with the help of the duplication formula.16b) With the help of Stirling's formula one can derive the following relation: i large 1 (i + m)\ \o (20. also written lCj.*.1 22 the result becomes i t h term (i + 2n)!2 2 " 1 where n 1 (20.18. and from this for i even or odd it is found that Mi ~ (16)^! 1 2g2 + 3 1i .th term iW~^q 1 {[i(gl)]!} 2 2 7 r(8/ l ) i . In the following we concentrate on the cosine potential and rederive the large order behaviour obtained above by other methods. (20. see also below) . Using this relation and the duplication formula lw('\W±^. In the first method we do this by first obtaining the imaginary part of the eigenvalue with the help of nonselfadjoint boundary conditions.
Reeve [262].2 solutions obtained earlier. 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. but with different boundary conditions. We retain the previous boundary condition at one minimum. Previously. . t To enable comparison with the work of Stone and Reeve we shift the cosine potential considered previously by 7r/2. the difference q — q$. J. What changes as a result of different boundary conditions is the deviation of q from the odd integer qo. we emphasize a few points *M. The potential we consider here then has the shape shown in Fig. MiillerKirsten [200]. when this is g0 = 2n + l . This is a very instructive exercise which shows how the eigenvalues change with a change in boundary conditions. f H . this difference was found to be real. in the calculation of the level splitting as a result of tunneling.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/?.5 A i 1 1 r f \ \ M\ \ \ / ! \ \ 1 \ \ I 3 \ \\ \ /41 i \ \ 0. 20. . but now change the boundary condition at the neighbouring minimum in such a way that the difference q — qo becomes imaginary.4 The cosine potential shifted by TT/2.3. Stone and J. 20.e.3 Cosine Potential: Complex Eigenvalues 479 20.4. \ 1 \ \ \ \ \ 0.5 \ \ / \A Fig. remains unchanged.* However. i. . At the risk of repeating what we explained earlier. we follow here a formulation in the context of our earlier considerations of the Mathieu equation.3 20. . From our previous considerations it is clear that the formal perturbation expansion of the eigenvalue X(q) in terms of the parameter g. A means to achieve this was found by Stone and Reeve. .20. 2 . W. n = 0 . l .
a boundary condition which at the face looks somewhat abstract. Large Order Behaviour of Perturbation Expansions for reasons of clarity (since some considerations here may otherwise not be sufficiently clear). which means replacing z by z±ir/2. and thus obtain ip"{z) + [A + 2h2 cos 1z\i>{z) = 0 (20. 8 the equation can be approximated by d2ip(w) (20. The physical wave functions there are therefore the solutions which are real and exponentially decreasing. Thus around z = 0 the potential behaves like t h a t of an oscillator well with minimum there. as we found these for this potential. i. To be more precise we recall the Mathieu equation we had earlier: ip"{z) + [A .Ah2z2 • • • }^{z) = 0. 20.20a) As stated. Around z ~ 0 w e can expand and obtain i)"{z) + [A + 2h2 . Here we choose t h e second way. X + 2h2 = 2hq+—. Previously this situation was the same at the neighbouring minimum.4. Effectively we consider the tunneling from the left well in Fig. and correspondingly we assume alternately even and odd oscillator eigenfunctions there. . the degeneracy of eigenvalues of neighbouring oscillators being lifted by the finite height of the barrier separating them. We can distort the potential there or alternatively impose a different behaviour there on the wave function.21) <u> + ( i « .2h2 cos 2z]ip{z) = 0. but t h a t with complex argument there.T K ^ a D_i_{q+1){±iw). 20. (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. (20.480 CHAPTER 20. but achieves exactly what we want.20b) with potential V(z) = — 2h2 cos 2z as shown in Fig. Setting w = 2h1/2z. Now we change this situation at the minimum on the right. Thus one chooses not the solution with real argument at the minimum. we shift the argument by n/2.e.4 (around z — 0) to the one on the right (around z ~ ir). and allows its imaginary part to tend to infinity.4. 20.
2. 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. Thus. i. 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. Eq. 20. we naturally constructed the even and odd solutions with these (cf. We consider here explicitly only the case of the ground state around z = 0.4. . considering around z = 0 the ground state of the harmonic oscillator for which go = 15 and replacing go by g. (20. that at z = 7r. by a different method. 3 . . . 17.24) Considering t m O i n Eq. . we see that ips ~ cos \j2qhz for z ~ 0. 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 = 1.23) . This somewhat abstract looking boundary condition is effectively simply one imposed on a complex solution of the equation giving the oscillator eigenfunctions. such that this deviation becomes imaginary.3 Cosine Potential: Complex Eigenvalues 481 As explained above. (where for the ground state q = go — 1) D_i{q+1)(±i2h1/2z) ' ± zoo (20. A (which we can loosely dub solutions of WKBtype). .> 0 for z > ±ioo. whereas here we start off with a solution even or odd about a minimum.20. Thus we write the boundary condition at Kz = ir: tp(±iw) — 0 for > w — 2h}/2z K z— > i. however. we now choose an harmonicoscillator type of real solution around the left minimum of Fig.22). A derivation is given subsequently. whereas here we construct these from oscillatortype solutions. so that the deviation q — qo results from the finite height of the barrier taken into account with boundary conditions at the next minimum.e. Since the large/i2 solutions valid away from the minima were the solutions of types A. (20. the even or symmetric solution about z = 0 can be taken as *• = liDfrriVhWz) +Dh{q_qo)(2h^z)}.e.63). 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.
Oberhettinger [181]. . 17. Magnus and F. (20.30) 'See e.29) Equation (20. Setting v = 0 in T(—v) = (—v— 1)! gives infinity. W.2 that one parabolic cylinder function is exponentially increasing in h and the other exponentially decreasing.•w) + 2ir T{u) <"+!)% Dvi(iw).482 CHAPTER 20. we know from the solutions of types B and C in Sec.27) requires the differentiation of this expression with respect to the index v. The differentiated function Y{y) is in Tables of Functions expressed in terms of the socalled psi function.26) Thus here we can approximate ips by expanding it about E2 i)s~DQ{2hll2z) + E2 ~{DE2/4h(2h ^ z) 1 2 + DE2/.g.3. Thus the only nonvanishing contribution comes from differentiation of Y[—v). The arguments of the functions appearing in Eq. or logarithmic derivative of T(^) for which:§ ^(1/) r» (20. and then evaluation at u = 0. However. As before we set A = 2hz + 2hqA where in the present case Ei = 2hz + 2hq0 + — and A & = Ex + E2 (20. Oberhettinger [181].27) The differentiation of the parabolic cylinder functions with respect to their indices is nontrivial. W. p. From Tables of Special Functions'.(. p. tp(v). §See e.w) = e T Dv{w) 2TT n») AvViDv^iw). Magnus and F.24) are linearly independent around z = 0 for q ^ qo. This is the feature which gives rise to an exponentially increasing contribution in the neighbourhood of z — 0 and hence provides the dominant contribution in going away from the origin. 2. (20. (20.25) qo) E2 0: (20.g. Large Order Behaviour of Perturbation Expansions Although the two parabolic cylinder functions in Eq. We see this as follows.we obtain the following circuit formula of parabolic cylinder functions Dv(w): Du{w) T DV{. 92. (20. this can be handled in a few lines.28) Extracting Du{—w) we have A.27) differ by an angle IT.h(2h1/2 J £2=0 (20.
31).29) v=0 (in + In w)D0(w) + v2ire (m + lnu.+ v ^ v k^x k ^ + ^ where C = Euler's constant = 0 .34) This is the behaviour of the symmetric ground state solution away from the origin towards the limit of its domain of validity.4.29).27) exponentially decreasing contributions. which are valid around z ~ 7r/2. "See e.e. dropping from the derivative part of Eq.31) dT(u) r*(v) dv 1 i/>(v) ^ o l. Differentiation of this equation and subsequently setting v — 0 yields. Magnus and F.33) = IV2~TT. Oberhettinger [181]. W. . i. and "Prom W. we obtain 4>s D0(2h^z) D0(2h1/2. (20.//n and ij>{y) = C . . 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.g. . r(i/) We now return to Eq. we obtain the formulas (X) / \ 1 OO I » J ] (l+)e.3 Cosine Potential: Complex Eigenvalues From formulas given in the literature we deduce the limiting behaviour^ 1.e. This result is » not easily generalized to <jo > 1.20. Thus calculations for qo ^ 1 are not given by M. Reeve [262]. including the derivative of DE2ufl(2h1/2z)). p..32) which we can use here since all other problems have been resolved: d D {w) dv v (20. pp. Stone and J. those of type A. 3 . For u — 0 one obtains the result (20.)e" w /4 2 l ^Dx(iw) (20. . i/^0 483 T(v) It follows that d dv [T(u)\ (20. 1E2 + ± 0(e~hz2) 2 Ah Ihir ehz* E2V'< 16/i2 V2^ehz2~ 2hM2z (20. 5 7 7 . We now go to the WKBtype solutions. (20. (20. 92.+w2/i IW Inserting this result into ips and retaining only the dominant contributions (i. Magnus and F.
where we can then match these to the solution ips. (17.38) in its limiting domain towards 2 = 0. near z = n the boundary condition (20. x . Thus in the present case and with q ~ qo = 1 we obtain „2ft.39) Now.34) and (20. (20.e. Large Order Behaviour of Perturbation Expansions extend these to their limit of validity to the left.40). Then V'WKB becomes (with cos z ~ — 1 + z' /2.23) demands (with constant A) ip = AD_1(±i2h1/2z') > 0 for z' + ±ioo.sm I We now want to match this solution to that of Eq. We now return to the WKBlike solution (20. To this end we set z' = z — ix.e. cos z/2 ~ (20.36) we can identify the dominant large/i2 behaviour of the constant C\ as c\ = e~2h. i.36) V'WKB cie~2hehz' 9h  .484 CHAPTER 20. 92) we take the formula: Dv{w) = T { y ^ \eiv^l2D^v^{iw) + e'^D^iw)] .41) 2TT . i. We observe that in the matching domain (i.— 721 . 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. smf (20.29) and (17. c2 the WKBlike linear combination e2hcos ^ W K B = ClipA + C2lpA ~ Ci z _—2/icos z — + C2 rT (20.37) Comparing Eqs. (20. (20. that in approaching z = ir. Recalling that we have shifted the potential by 7r/2.35) We construct with constants c\.34). y COS c\ J (20. (20. we obtain the required solutions from the earlier ones. 2 c2e2hehz UJ2 . (20.36) cos  . with expansion of cos z in the exponential) D0{2hll2z) = e~hz ~ e~2h / „2/i cos z \ — ).36) and consider this at the other end of its domain of validity. by there replacing z by z + ir/2..cos z — 2/icos z ipA^ Tcos  and i n A —. Eqs.e.40) From Magnus and Oberhettinger [181] (p.
(20. ^'mliw <2a47) For qo = 1 we recover the result (20. (20. (20.40) can be written ^ ~ AJ^e~hz'2 + A~(i2hl^2zTlehz'2. (20. J Hence Eq.34) we obtain E2 = i%\ = ±ih22Qe~8h. with c\ = exp(—2h) from above. 1 2 4 1 2 8h ^^2 2 (20.20.46).38) can be written TAWKB (20. (20.^.e.43) we obtain. We derive the general result in the next subsection using our own method.45) Comparing this result with tps of Eq.2/. for both even and odd harmonic oscillator wave functions around z — 0 with qo > 1) is nontrivial.e.46) Repeating these calculations for the general case (i. i. The general result stated by Stone and Reeve [262] is therefore presumably guessed.1/V)_le^_ Z With this result Eq. . (20.3 Cosine Potential: Complex Eigenvalues Thus for v = 0 we have DQ(w) = —j=[D\(iw) V 27T 485 + Di(iw)] = real.complex conjugate] (2Q42) (2032) y /  Z e _ ^ + 2(. A = ±i4/l1/2e4h ^ t c2 = T i2(2/ivr) 1 /2 e 6/ l . 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 .39) and (20.44) = DQ{2h / z) ± (2h7r) / e.43) Comparing now Eqs.
20. W. we impose the usual type of boundary conditions as for the minimum of a simple harmonic oscillator there. •_(:)*<>.z. At the minimum to the right of the barrier shown in Fig.) + 7T V> = o.20a). i. (20. at z = TT/2. Considering the original equation.*. and expanding cos 2z around the point z = — TT/2. •+(=)* V(f) 0.46) we consider a half period of the cosine potential as illustrated in Fig.5 The cosine potential from — TT/2 to n/2.Ah2 ( z + .5.486 20.5. harmonic osc 71/2 E>V / E<V z+ O \ E>V TC/2 \ \ .e. the result is formula (334) there.e.2 CHAPTER 20. 20.E curre it \ j. 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. V(z) . (20. H.48) These conditions provide the required quantum number qo = 2n + 1. MiillerKirsten and A. 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 . 20.e. (20. i. J. Wiedemann [4]. Eq.()#0. Large Order Behaviour of Perturbation Decaying excited states Expansions For our derivation of the generalization** of Eq. Achuthan. We have to impose two sets of boundary conditions. Fig.3. . i.
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.' 4 / 1 1 / 2 COS ( Z + 7T \2 4 . Proceeding as in our previous cases and substituting the expression (20. (20. as shown earlier. (A) Boundary conditions at the right minimum We begin with the evaluation of the boundary conditions at z = 7r/2.r/2 ~ e i(2hq) V 2 .Hz^+Ol^ Bq[w(z)}+0 h1'2 (20. we have IPA(Z) „2hsinz Aq{z) + O il>A(q.z)=tpA(q.50b) which are valid as asymptotically decreasing expansions in the domain away from a minimum.3 Cosine Potential: Complex Eigenvalues 487 (using Eq.21)). B and C.h. for COS I Z ± 7T 1 1 2 4 >o^\ The second pair of solutions is ipB(z) J2hsmz B.21) into the Mathieu equation we obtain.51a) where H Bq[w(z)} w(z) = h(«v (w) Bq[w(z)} = Bq[w(z)}. (20. (20. i.z). three pairs of solutions of types named A.51b) 2i(«1)[i(9l)]'.20.49) This is a complex boundary condition which requires the coefficient of the other exponential in ip±(z) (continued to and beyond — 7r/2) to vanish. we demand that ^(*)L<.h.e.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. Recapitulating these from Chapter 17. (20. For the continuation we use again the WKB formalism.
we obtain 1>±W = \ a a (20. (20.488 CHAPTER 20. ^ W H .^ i .57) where go = 1) 3.52a) = Cq[w(z)}. Finally the third pair of solutions is Cq[w(z)}+0 Cq[w(z)}+0 Cq[w(z)} where Cq[w(z)} = 22(9+1) •((73) 1 h1/2 1 hW> (20. (20. Large Order Behaviour of Perturbation Expansions the function H!/{w) being a Hermite function. and the second around z = 7r/2.52b) The first solution is valid asymptotically around z = —vr/2.=F2 2\1/2{28h2y/4e~Ah l + O (20.54) The even and odd solutions are defined as iP±(z) = tyA(z)±iPA(z)}. These solutions are valid asymptotically with the first in the domain around z = ir/2 and the second in the domain around z = —ir/2.qo . (20.53) where a (8M («!)]« 1 + 0 { \ a = 1 (8/i) K9+ ) l + O (20. We return to this condition after derivation of the .55) Extending these solutions to the domain around the minimum at z = 7r/2. Thus we have IPB(Z) = aipA(z) and VcO*) = O ^ A O 2 ) . Here the upper sign refers to the even states and the lower to the odd states. i. we obtain our first transcendental equation. \vw). We saw in Chapter 17 that some of these solutions can be matched in regions of common validity.e.48) and proceeding as in Chapter 17.56) Applying the boundary conditions (20. 5. the relation q .
49) and we can apply our boundary condition. formula 280. 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. p. 163 — but here we are interested in the WKB solution in a domain where it is proportional to a solution of type A.Ah2 cos2 z ~ 0.2hq] / dz) We know that the solutions I/JA{Z). Some nontrivial manipulations are again required to uncover the generation of important factorials.[ [Ah cos z . I^A(Z) overlap parts of the domains of validity of these solutions. and we obtain the proportionality by appropriate expansion.58) Ignoring the correction A / 8 . i. r ywKBlZJ ~ ~ — [4/i2cos2 2/i ]i/4 2 9 z 2 2 e x p ( £ [Ah2 cos2 z  2hq]ll2dz) ' 1 2 .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 .e. (B) Boundary condition at the left minimum The solutions of type A have been matched to the oscillatortype solutions at the right minimum. (20. C O S 1/2 ^" (l) ± ' 2<2+<2 (2 °.61) [Ah2 cos2 z2hq]1/2dz. These then provide complex exponentials like (20. The solutions of type A are valid to the right as well as to the left of the maximum of the barrier.Ah2 cos2 z ip = 0. F. Byrd and M. Friedman [40].01. Our procedure now is to match these to exponential WKB solutions. e x p ( .67). Thereafter we extend the latter across the turning point down to periodic WKB solutions. Jz+ ' T In principle the integral could be evaluated exactly — see P.20.3 Cosine Potential: Complex Eigenvalues 489 second transcendental equation from the boundary condition at the other minimum. as we shall see with the result of Eq. . Inserting the approximate form of the eigenvalue into the Schrodinger equation we obtain dz2 + 2hq+ — . (20. we see that the turning points at z± are given by 2hq . D.
59) 2h COS Z+  ~ 0.) +0 z+ IT Inserting this into the integral I. 2/i In the same spirit we can set h[z + l 2h 1 — cosi I z + — 7r 2/i[l + sin z\.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. using the relation 1 + sin z = 2sin2(2i/2 + 7r/4). we write cos2 z = l .» _ i y v .sin2 z = l .490 CHAPTER 20.64) .cos2 ( z+^) = (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 (20.63) We handle the logarithmic terms in Eq.62) where we set — in accordance with our expansion *+ + 2 IT 1/2 <? + (20. We have for the following expression contained in (20. (20. (20. we obtain f ~ I = = 2 1/2 Ah [z + 1 2hq + o\h \z+ 2 dz J Z+ 2h I /MH)' dz 1/2 1/2 2h 2/i /•«"(.62) as follows. Large Order Behaviour of Perturbation Expansions Since cos(z + n/2) = — sin z.
64) into Eq.50a) and (20.67) (27r) /22«+5(4/i2)V4V^y The last expression is again obtained with the help of Stirling's approximation. . 1/2 (20.62). <$&?(*) e2he2hsinzei/4(q/2h)i/4[tan(% + l)]'q/2 2?(4/i 2 ) 1 /4{2sin(f + f)cos(f + f ) } 1 / 2 e2heq/A(q/4y/4 /2\q/4 ./ * [Ah2 cos2 z IZ+ 2hq\l/2dz) + f )]«/ 2 (4/i 2 COS2 z)V4 e2he2hsm zeq/4(q/2h)i/4[tan(% /2^~Q/4 (4/i 2 )V429{2sin(f + f ) c o s ( f + f)}V2 ^27*291 (4/l )V4 e ?/4( g / 4 ) 9 /4 ^ J e" ^^(2TT) /2 2 1 2 ^ ) / 4 /2\"  . 2*4(4tf)V« W ^ e[l( V D]! 1 /2y^ (20.3 Cosine Potential: Complex Eigenvalues Inserting the relations (20. .^ Proceeding similarly with the other WKB solution we obtain rWKB (z) ~ e x p ( . but mention that for q an odd integer: j(9D l + O and 1 r(93) 4(9+1) ! = ±7I . we obtain I ~ 2h + 2hsm z Hence exp 491 V^m 2h\~'I z TT 4 — tan .65) [Z[4h 2cos2z2hq]1/2d< e2he2hsin ^eg/4(g/2/t)g/4 2<?[tan(§ + f )}i/2 (20. (4^2)i/4 [ i( g _3)] ! V/ l (20.+ qj V2 4 A. We ignore this here.50b).63) and (20. .68) ' T h e use of Stirling's formula here and below — essential to obtain factorials — assumes q to be large so that corresponding corrections would have to be calculated.20.66) The WKB solution therefore becomes (approximating the denominator of (20.60) by (4h2 cos2 z)1!4). (20. on using in the second step Eqs. (20.
U ^K^. (20.—^ (2^)^(16^2)1/4^ ^ _ _ [T+(±)exp{i(2/ig)1/2(z+_z)} 2(2/ig)V4 + r _ ( ± ) e x p { .. S1H (2/ KZ ) / (z f 2 (2^).z+)/ VWKBW 2\2j (2^72 l(8^/4e2fe(27r)i/22V2(4fr2)V4 2 < nz+) ^WKBW [\{ql)]\ 1 (8/i). i 27r(8^/ 4 2/t /4 ' r(±) . we obtain T _ ( ± ) = 0. Thus with IPA(Z) and TPA(Z) taken from Eqs. (14. (14.KW4^8 4 .™) .2 f e (27r) 1 / 2 (16/t 2 ) 1 / 4 2 (2/wz)V4[i(9_i)]! cos (2M1/2(^^)+4 7T .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)]! .492 CHAPTER 20. we obtain l(8/i)g/ 4 e.i ( 2 / i g ) 1 / 2 ( z + . Large Order Behaviour of Perturbation Expansions again using the Stirling formula.+) + B^) HAh^/\ir.49).2 [ ? ) Imposing the boundary condition (20.l^l !T ^^* T2 7 r ( ^^ /. Eqs. where _ 1 + [±) *) + £ (20.67) and (20.e 2[J(«l)]l[j(?3)]! i l[(g3)]!e (20.49).50)). i. 1 2 1 ±2 . We now return to the even and odd wave functions defined by Eq.z)}].69) l(8M9/4e2^ \2 [ i ( g . (20.?/ e^2 /2(4/ l 2)i/4 [ i ±2 (^)V^ 4 1 (g _ 3)]! (/]Z+) ^WKBW Inserting here the oscillatory W K B expressions (cf.1 ) ] ! 1 (8/Qg/ e4 2fe l[(g3)]!e^ .T 7 77 r )V2 /4(2 .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+.
75) . (20. (20. (20.57) and (20. Ex = 2h2 + 2hq0 + A 1 (<? .l)j /2 e4fe(162/t2)^4_±l or ±i 7r\ 1/2 (16 2 /i 2 )9/ 4 e.21) expanded about q = qo.h) + (qq0)( BE — 90 E{qQ. for q an integer both sides are imaginary.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. Equation (20.e. i.72). 20.26). we obtain E(q. In both cases we wrote (cf.47).72) we obtain qqo q0fiSh 2(16/i)*>e {[!(<?<)I)]!} 2 1+ 0 (20. (17.qo) El Ah' (20.4h (16 2 /» 2 )9/ 4 (91)]! = (16 2 /i 2 )«/ 4 = (16/i) 9 / 2 . Combining these equations by replacing (28/i2)<?/4 in Eq.h) + i l + O '4{[^ol)]!}2 E(q0. (20. the result (20.25)) E = A = Ei + E2. Eqs. h) of Eq.3 ) I = }(»« !7rl/22I(.20. We now have the two equations which have to be satisfied.3.h) = = E(q0.h)+2h(qq0) (16h)i°+le'Sh E(qQ. who derived it for qo = 1 and guessed the form for general values of qo. (20.h)+i%Ego.74) This result agrees with that of Stone and Reeve [262].3 Cosine Potential: Complex Eigenvalues Using the duplication formula \{1l) this becomes 7rV 493 \ ^ .57) by the left hand side of Eq. Eqs.73) Inserting this expression into A = E(q. (20. in fact.72) is our second transcendental equation.72) We observe that with the square root as on the right hand side both sides are complex. (20.
3. * It would be interesting to see a derivation of the formula — reminiscent of an optical theorem — from first principles.W [£(901)]! < (20.494 CHAPTER 20. Fateyev [26]. Large Order Behaviour of Perturbation Expansions Referring back to the calculation of the level splitting with selfadjoint boundary conditions. f E . The same formula can be derived in the case of the double well potential.) {I6h)™/2e4h 1 + . 25Eqo=8h[ 2\1/2 . W. its tunneling properties and the large order behaviour. . we recall that we obtained there for the difference 25Eqo between levels with the same oscillator quantum number qo. (20. i. A = .76) he Thus SEqo is the deviation of one of these levels from the harmonic oscillator of level. cf.e.I g  ± l ( 2 „. Bogomol'nyi and V.77) One can say that this simple relation summarizes the intricate connections between the discontinuity of the eigenenergy across its cut. MiillerKirsten and 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. the discontinuity across its cut from zero to infinity. B. Wiedemann [4] and Example 20. J.2 below. A.* The existence of a formula of this type was conjectured without an explicit derivation like that given here and only for the ground state.74b). (20. Setting ng AEqo:=2iSfEqo. i=l We can calculate Ai for large i from the relation Ai = * Jo 1 P°° ti^QE^dti (20. For definiteness we recall the appropriate expansions together with the definition of coefficients Aj. Achuthan. (17. 7 8 ) i=l V . by others. we can obtain the imaginary part calculated above from the formula 4 (H) A ^ o = 27ri(<5£90)2.80) *P.+  . Eq. We had A= or _2ft2 + 2ft. H. we can recalculate with the help of Eq.9) the coefficients Ai of its perturbation expansion. 20.
81) We observe that this result agrees with the result of Eq. W.l ) +2[(q + 4j)2 + l + A}Pi_1(q.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.14). We sketch this method here slightly modified and extend it to the final formula for comparison with our previous results.j + 1).4 Cosine Potential: Another Method of Calculation 495 with $$E(h) given by Eq. J). For convenience we write Eq. The integral is recognized to be of the type of the integral representation of the gamma function. i ^ = ^ + E?27MT E ^ifoj^W (2082) The coefficients Pi(q. so that an exact Borel summation is not possible. the integration is seen to give (20. (20. (20.l ) ( < ? + 4 j . (20. and the method can be applied in many other cases. B. *R.83) One now sets ftH = ( 2 % P * ( 9 . MullerKirsten [200].t We consider again the periodic cosine potential with (unnormalized) eigenfunction expansion oo .74).1). We also observe that these terms do not alternate in sign. J.19) for the large values of i under consideration here.13) and the coefficients M2i or M2i+i of the eigenvalue expansion can be expressed in terms of these as in Eq. (20.j) + {q + 4j + l)(q + 4j + 3)Pi1(q. 20.. j . f H .j) satisfy the recurrence relation (20. (20.3 ) P i _ 1 ( c ? . Dingle [74].20. a = J + \<1> (2084) .13) out again: jPiiQj) = (g + 4 j . there is nothing special about this case. But as Dingle remarks. Using again n = (go .
85).496 CHAPTER 20.a R . g < 4 j .89) Replacing here K by a and \x by 1/2.e. (with a = j + g/2 in the coefficients) 8/? —Pi(a) = P i _x(a .ll{z)= L . a.e.K + ^ ] L + K^)WK.85) Considering now the generating function of the perturbation coefficients.88) satisfy among other relations the following recurrence relation which is of significance in the present contextt {2Kz)WK>li{z) + WK+1. i. i. (20.4. Pi(a) i.z)WaA/2(z) Setting V.86) one can convince oneself that Eq.4 ) ^ .(a) oc .^(Z). A.h) 2+ 8h a p(a. (q + 4j) 2 ~ 4j(4j + 2g).1) + 2 P i _i(a) + Pi_i(a + 1). z2 (20.83) can be written j .l/2l Wa.h) = ^2pi(a) with p.e. Abramowitz and I. (20. which are solutions of the differential equation known as Whittaker's equation. formula 13. 0 4j(4j + 2g) pi_i(a . Taking dominant terms. nx pi_i(a + l).h). d2WKh dz2 + 1 4 K z zM2n WKtll = 0. Now. 4j(4j + 2g + 4) + / . we obtain (2a .91) + Wa+lil/2(z) = (1 . p(a. (20. the recurrence relation (20. (20.31.27h pi(a) a = 4j(4j + 2 g .87) Taking terms of 0(h~l+1) of this equation. . (20.85) implies p(a . e. Stegun [1].h) +p(a + l.1) + 2 (a — 1) a .g. we regain Eq. Large Order Behaviour of Perturbation Expansions and considers large values of j for large values of i.1/2(z) (a1)! (20. 1  1 / 2 (4 (20. Whittaker functions WKjfl. (20.90) See M.l.
e..Tl ss\{a)\{al)\ (z) s=0 s — a)\(s — a — 1)! for z — oo. a (20. (20. such that p(a. i.i/2(z). (20. > (20. In the present case this means the replacements q — —q. Dingle discovered the solutions^ p(a.20.(« .\)V . h) is expressed in descending powers of z = 8/i.g(h) are functions of h which in view of Eq.84)) {8h)q/2( Ah for j < 0. Magnus and F .86) have to be chosen such that p(a.87).88) which reminds us immediately of a radial Schrodinger equation with Coulomb potential. p. ^Observe that if one solution of Whittaker's equation (20. (20. h) is expressed in descending powers of h. h —> —h.(« 1) 2 ]. Oberhettinger [181].92) with Eq. g(h)(20.j — —j. (20. The more precise form obtained from Tables of Special Functions' is w*A* z 2 e / z" n=l [f .h) = \~\Pi(&) with pi oc h l.i/2(z) + Va+lA/2(z) = 497 2+ Va. 89. another solution is obtained from this by changing the signs of K and z since this leaves the equation unchanged.(« .h) ~/(fe) ' (a1)! .93) (a1)! ' where f(h).„ +1)2] 7 n\z \n which with K = a and /z = 1/2 is: WAz)~e^z«Y. v . . We can infer the approximate behaviour of the Whittaker function for large values of z from the differential equation (20. Thus the approximate largez asymptotic behaviour is WKtp(z) ~ exp •FH V dz z e ^' 2 exp(/clnz).94) Hence to insure that p(a. we have to remove a factor (recalling a = j + (7/2 of Eq. > > ^See W.88) is known.92) Comparing Eq.4 Cosine Potential: Another Method of Calculation we obtain Vai.
g — —q. (20.95b) Then for j > 0 we obtain from Eq..j _ g / 2  1 / 2 ( .96) and a similar expression for j < 0 with /i — —/i.94) with s — s — j : > x = J + y u ( 1)J ~ f (*+&!(*+*?pi i_ Using the reflection formula (for q = 2n + 1. x 4i[2» + i(« — 1)1! il! [ ^ ( 9 . this becomes / 1 \ (1)5'" ^ U + iq)<(s + iq1)\ 1 Picking out the term in hrl. ^ .94) here we have a series in descending powers of h..+ i f or 3 < 0 (20. (20.. we obtain the result of Dingle: (20. Comparing > > > this result now with Eq.95a) Inserting (20.84) 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. J.). Muller [74]. n = 0. i). (20.1/2(8^) ( . Dingle and H.l ) ] ! 11 R.1. W. Large Order Behaviour of Perturbation Expansions and write / i \ pl<x = j+ 2<7>h) « (8h)~q/2e4h v ^+9/2.498 CHAPTER 20. _ i s. Correspondingly we set for j > 0 / 1 A pla = j + q. . B. 2.8 / t ) for j > 0. h\ oc (8h)q/2e~4h ( _ .83) and is given by 1 1 . This expression can be obtained from the recurrence relation (20. j — —j.
i. J.14). (20.20. (20.j) (i + lg)[2i + i ( g . formula 0. E* J'=I i+M 2 [2i + 9 . A special case of the formula can be found in I.99) In the next step we use for both factorials [2i + ^(q — 1)]! in Eq.3)]!2 2 '+^. formula (30). p. S. is M2l = 2j2j[Pi(<l.l ] ! (20.98) the duplication formula 22zl / y P*l>'7J<*l)'(»j»and obtain Mi% + \{q^)W + \{q.l ) ] ! 2 M [i + \q)\ i\\l{ql)]\(J + \q) [j + lq]\[ij]\ (20.157(4).+i?_l]!}2 (20. note the misprint there: (r — s) ought to be (r — s)\.4 Cosine Potential: Another Method of Calculation 499 We observe that Po(q. Ryzhik [122]. M.{j + \<l). 5. Dingle and H. .0) = 1 as desired (although not really enforcible for large values of i).100) **R.e.l) (i+y) 2*%(ql)}\ [2i + §(gl)]l i\[i+\ql]\ ' Inserting this expression into Eq. of Eq. Gradshteyn and I.97a) we obtain (observe that as required this expression vanishes for j > i) Pi(q. Thus ° 9 (i + \q) 2«(i+Igl)! p . B. W. (20.98) We evaluate the sum by approximating this to a formula given in the literature ** r \ f R\ __{r + Rl)\ J (rl)\(Rl)\' E j=0 i.l ] ! {[.e.J)} 2 i=i 2 [2i+l(ql)]!22il2 i![i(gl)]! i (20.Then [2i + ±(ql)}\2* Pi{qJ) [i + \q]\ (20.97d) Then the coefficient M<n of the eigenvalue expansion.1 )/ 2 2 2i ' ^i\{\{ql)]\ [2i + g .97c) We assume that in a leading approximation we can cancel the factors (i + \q). l{Q. Miiller [74].
.500 CHAPTER 20.5. i.]\.81) for q — qo = 2 n + l .Am i .e. the result of Bender and Wu [18].1. i. We refer back to Eq.i ) ] ! [ ^ +  ( g .86) and set (cf. The imaginary part we obtained is interpreted as effectively the discontinuity of the eigenvalue across its cut from h2 = 0 to infinity..3 ) ] ! _ 1 Hence M or.e.5 20. i\ M i large 1 . Large Order Behaviour of Perturbation Expansions In the following step we use the approximate formula (20.. 20. .e.2. replacing 2i by i: 2 ^2 8 ^ % + *"*}' (20.102) Thus the result is M m 1 2«Fi + g . (18. [19].101) This formula allows us to cancel all factorials [i.4)) £ = ^ = ^+ w and y= ^ > ( 20 .. i. n = 0. (18.e.17). Eq.l.l l ! with the large i behaviour in agreement with the result (20.e.105 ) the latter because we see from the eigenvalue expansion (18. With this we can now derive the behaviour of the late terms in the largeh2 perturbation expansion of the eigenvalue of the Schrodinger equation for the proper quartic anharmonic oscillator.86). (18.' 1 i • (20. Eq. i.34) that £ is an expansion in ascending powers of h 6 /c 2 .1 Anharmonic Oscillators The inverted double well In Chapter 18 we obtained the imaginary part of the eigenenergy E of the Schrodinger equation for the inverted double well potential. i. [^ + i ( g .o i • .. since \{q — l) + \{q — 3) = \q..
20. in terms of K where go = 2ftT + 1.e. one obtains the result Mr 2io+hr+1+f (l)r[r + f ] ! 7T3/2[1((?01)]! (20. i.5. and then the relation (20.108) Proceeding in this way.109) or. 20.e.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. (18.86). in fact as one with Borel summability in view of the factor ( .5 Anharmonic Oscillator The Borel sum can now be formally written 501 (20. This can then be compared with the level splitting we obtained for this potential.77) can be verified.110) This is the result in agreement with the large order formula (4. and hence inserting here the imaginary part from Eq.107) we obtain (i)r r 71" JO yirZ£{y')dy'.(20. h2 _ h222K+llHr+K+ll2{l)r+1 r+ K .l ) r . Once this has been obtained one can insert the imaginary part into the Borel transform and obtain the behaviour of the late terms in the eigenvalue expansion. (with E = h2£/2) my') = 290+1 (y')9o/2 e 2/'/3 (27r)V2[i(go_i)]!' and integrating with the help of the definition of the factorial or gamma function.106) With £= E r=0 Mr (20. (20. the discontinuity across its cut. We leave this to the following Example.4) of Bender and Wu [19]. i. The result establishes the series unequivocally as an asymptotic series. .
Paldus and H. V.Q . J. Solution: For the solution we refer to the literature. Graffi. J. Nakai.77) and its possible generality has been referred to by Brezin and ZinnJustin [36]. J. Cizek. W. J. " P . H. . MiillerKirsten and A. *R. Harrell. J. J. Grecchi. 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. S. Liang and H. T. A relation similar t o (20.77) was conjectured without explicit derivation and only for the ground state by Bogomol'nyi and Fateyev [26]. K. 1982) [139]. The relation (20. J. of International Workshop on Perturbation Theory at Large Order (Florida. K.* A more recent status evaluation is provided by the book of LeGuillou and ZinnJustin [161]. J. G. Its usefulness has also been demonstrated there. in t h a t when the quantity on one side is known. S. (357) to (371). Large Order Behaviour of Perturbation Expansions Example 20. Wiedemann [4]. Wu [18]. W. + 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. Bender and T. MiillerKirsten [164]. as in later chapters.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. Harris. MiillerKirsten and A. M. R. Damburg. V. Gram. J. Paldus. [19] can be traced back from articles in the Proc. M. Wiedemann [4]. *Some of the literature in this field following the work of C. . Silverstone [62].6 General Remarks The study of the large order behaviour of perturbation theory has become an individual direction of research. H. and is there shown to have a deep meaning (cancellation of the imaginary part of the Borel sum by explicitly imaginary terms in the perturbation expansion). J.502 CHAPTER 20. An approach to arrive at the relation via instanton considerations has been examined by ZinnJustin [292].77) has been obtained in the Schrodinger theory of the molecular hydrogen ion H^ . E. Damburg. W. Cizek. Harrell. J.^ 20. R. Propin and H. t p . M. 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. Silverstone [52]. it can be used to obtain the other. Eqs. Achuthan. R. J . Grecchi. t The relation (20. E. Achuthan. Propin. S. J.
Coulomb problem) we refer to existing literature. Hibbs [95]. A brief introduction is also contained in R. ^ Our interest here focusses on quantum mechanics. Chapters 2 and 5. since in such cases the probability of this system choosing a path far away may be considered to be small. Feynman [93]. our main interest in path integrals and their uses will focus thereafter on their evaluation about solutions of classical equations. The path integral is. P. A standard reference is the book of R. Probably the most readable introduction. 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. in general. which explains also difficulties. Kleinert [151].Chapter 21 The P a t h Integral Formalism 21. A further wellknown reference is the book of L. Feynman and A. S. Felsager [91]. This formalism is particularly useful for evaluating scattering or transition amplitudes. The formalism is not so useful for boundstate problems (in this case the initial wave function is. 'See in particular H. However.e.1 Introductory Remarks In this chapter we introduce the path integral formalism developed by Feynman. 503 . However. We illustrate the method here by rederiving the Rutherford scattering formula with this method.* This formalism deals with an ensemble of paths {x(t)} rather than with wave functions and constitutes an alternative to canonical quantization in quantizing a theory. *R. P. For the application of path integrals to the hydrogen atom (i. Feynman [94]. however. R. 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. Thus we are in particular interested in rederiving the level splitting formulas obtained in earlier chapters with this method. can be found in the book of B. Schulman [245]. particularly useful in cases where the behaviour of a system is to be investigated close to its classical path. P. unknown).
tf) = / dxK(xf.4) so that the timesliced action S becomes rtf «+i S = / 'ti n+1 dtL = d£L = n—»oo.tf\x. 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).1) Let the wave function i/> at time t = ti. (21. =J dx(xf. the initial time.t). ti).x.2) where K is a Green's function.x fc _i) 2 2e2 V{xk) Thus: Given xk(tk).ti). Obviously K(xf.e—»C lim V e m0xk fc=l  V(xk) = lim n—>oo. (21.tf\ip) . The Feynman Path Integral Formalism 21. we wish to know* ip(xf.5) Sn = X / fc=l mo (zfe .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. 2mo dx2 (21.504 CHAPTER 21. */ = U + (n + l)e.ti\tp). U.3) We divide the time interval tf — ti into n + 1 equal infinitesimal elements e such that £() ti = = hi ti + e. (21.e.ti){x.e—»0 Sn.ti)ip(x. We are interested to know the wave function tp at a final time tf > ti and position coordinate x = Xf. (21.x).tf. we can compute the action S1. *In Dirac's notation we write this equation later (xf. i. be ip{x. ti) = 5(xf .
t).6a) . Here and in the following we need one or the other of the following integrals (cf. 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). 21.ti) to its final position at (xf. 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. The usefulness of the method therefore depends on the fact that some paths are more probable than others.21. 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.1 A path in (x. Thus there is a countless number of possible paths that the system may choose in propagating from its initial position at (xi.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. 21.2 The Path Integral and Green's Functions 505 Above we considered time intervals of length e. as will be shown to be possible below.28)): f dr\ exp ia a) ' (21. of which one is illustrated in Fig. the Fresnel integral (13.
Xi.* W i t h the first of these integrals we obtain (compare the result with the Green's function (7.506 CHAPTER 21.m0 .6c) In the evaluation of these integrals we replace a by a + ifi. J—oo dxneiS/h — • 0. ie f me\ 2a\a~) 1/2 (21.x f c _ i ) 5 \m0 J (21. One therefore removes the troublesome factor by defining (Sn being the sum in Eq. one has j ^ dr1eifa+i^ /£ = .mQf Xk L ~2e^ ^ .8) Note t h a t in spite of the n integrations.s f c _i) + (xk+i . 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. The Feynman ia Path Integral Formalism f J—c J dr) r\ exp v = 0.z f c _i) 2 } } . for example.ti) '•= f g m ) (N(€\]n+i dxL.xk_i) dx exp i — { 2 a . with f™ dze"™ 2 * 2 / 2 = s/2ir/w2.6b) —c ia o drjr] e x p—V € (21.xk+i) exp . 2 + 2a.xk+x oo —oo oo .0 ^7 (x f c + i .Zfci) 2 } I +{xk+i x*.5)): G(xf..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 dx2.2 + (x + x f c + 1 . (21. (21..37)) dxk exp d(xk .ir 1/2 £dxeXp[i^{2(3 dxexp\i—<2lx+ 72 iV(e) exp Z7TT. We then integrate and finally let /x go to zero. dxneiSn/n. tf > U.t ~Xk+l> + xk+i 2 ) +{xk .(x fc+ i . n > 0. dx exp z—{a.tf..
we arrive at a path integral without the measure factor N(e) above.e. A mathematically rigorous definition of the path integral is difficult and beyond the scope of our present aims.2). (21.U) = / JXi=x(ti) = K(xf. (21.10) can be looked at as the product of a succession of free particle Green's functions (or propagators). The additional factor is needed later to combine with the number n contained in the extra factor y/n + 1 (in Eq.g. B. (21. However.26)) that — ignoring the potential — the multiple integral (21. (21. in fact. p. the Green's function K of Eq. 21. i.g.tf. (21.Xi.9) T>{x{t)}eiS/h = ^ww^IdxlIdXnelSn/h(2L10) This is the formula originally given by Feynman. and will therefore not be attempted here.Xi.2 The Path Integral and Green's Functions 507 Eq. § See e. These integrations are indicated by the horizontal arrows in Fig.Xi.tf. Then one integrates over X2 from —oo to oo and obtains an exponential involving (x^ — XQ)2.' These factors N(e) were introduced to cancel corresponding factors arising from the Fresnel (or Gaussian) integration. summation over all possible paths. . as in our discretization in Eq.7) for n — 1) to give a time interval. however prefixed with a new factor.1. B. and so on. We shall see later (with Eq.t In the following we demonstrate that the function constructed in this way is. 183. 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.5).24).ti) for tf > U. Felsager [91]. 11 For discussions see e. nor the limit of integration of some variable Xi. Felsager [91]. i.tf. In the expression rx" / ?){x}eiI{^ Jx the quantity x" is only a symbol to indicate the endpoint of the path. 183.5). G(xf. p. Thus one first integrates over x\ from — oo to oo and obtains an exponential involving (x2 — xo)2. Different constructions of these families of paths usually lead to different measures. See the square root in front of the exponential in Eq.e. (21. It is neither a variable of integration in T){x} —> Ylndxn.§ The factor [iV(e)]_(n+1) associated with the multiple integration is described as its measure.21. which all vary from —oo to oo.ti) For tf > ti we now write* G(xf. it may be noted that with a different philosophy concerning the summation over paths. (21.
Consider a time t = tf + e for position Xf = xn+2. (21.tf + e. We have G(xn+2. 187 . One way to avoid this problem with the integration measure is to consider the ratio of two path integrals.Xi.Then we can construct a recurrence relation as follows.11) For reasons of transparency in the next few steps we replace xn+2 by x and suppress temporarily tf.ti. 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.ti) = 1 j^p: f°° cte n +iexp } • fm0. 16 x2 Sn+l) \ 7T2~(Xn+2 V(xn+2) Setting now xn+\ — xn+2 = n.U) imo 2 G(xn+i W)Idr. Felsager [91].3 in the evaluation of I a specific path integral (see Eqs. pp.72) to (23. Then the measure drops out. the other with that of the action of the free particle SQ.1 and allowing n to go to infinity.tf + e. .xi.508 CHAPTER 21. Expanding this in a Taylor series around x. the discussion in B.ti). we have G(xn+2. we have See. The Feynman Path Integral Formalism The factors N(e) can be avoided by instead summing over piecewise linear paths. (23.74)).tf.Xi. we obtain the same as before except that there is no factor N(e). so that the piecewise linear paths fill the entire space of paths. It is for these reasons that — unless required — the measure is frequently ignored. I We employ this method later in Example 23.exp 2e n ieV(xn+2) G(xn+2 + V. 21.188.xi. We next determine the equation satisfied by G in order to verify that this is indeed the Green's function for tf > t{. Each of these linear pieces is completely characterized by its endpoints. however. one with the full action S. and one is calculating the Feynman amplitude or kernel relative to that of the free theory. Then we have on the right the factor G{x+rj). Performing the sum now in the sense of summing and hence integrating over the vertex coordinates in Fig.
ti) 2 —l€ — ^2 + V(x)G(x.— p G ( j .tf.xi.(21.x.ti) = j^rr dnexp I ^rj1 .Xi.16) .ti) +V(x)G(x. U) = e^G + 0(e 2 ) dtf 1 d2G(x.15) We see that G(x.tf. U) + .U + 0(e ).tf. tf.6b) and (21.Xi.ti)\ . tf.tf. tf. . (21.U) +••• \ 509 G(x. U) .ti) for tf > ti (21. ti) satisfies the timedependent Schrodinger equation.13) We integrate with the help of the integrals (21.G(x.ie 1 d2 2nV0dx^G{X.ti) +\if(^G{x.Xi.tf + e.xi.ieV(x) + ri< —G(x.tf + e.Xi.xi.tf''Xi.xi.xi.6c).t Thus + 0(e2 (21. tf.21.U).Xi. d and this means d i—G(x.tf + e. Xi. U) + Q(e2 = (lieV(x) + 0(e2)) G{x.6a).Xi.Xi.ti) .tf.tf)= dxG(xf.Xi. 2mo dx2 (21.U ieV(x) 2i7re\ 1 / 2 m0 ie d2 G(x.ti) N(e) or G(x. It follows that ip(xf.ti) = G(x. tf + e.ti) = 1 d2 + V(x) G(x. we obtain: G(x.tf.xi. Retaining only terms up to those linear in e.xi.ti)ip(x.14) = G(x.tf.tf.xi.tf.tf.2 The Path Integral and Green's Functions so that G(x.Xi.tf.
now called So.x.3 The Green's Function for Potential V=0 We now calculate the Green's function K for potential V = 0. as this is also called. as may be verified by inserting this into Eq.tf. It follows therefore that {{W ~V+2m~0~^) K{Xh t/. The Feynman Path Integral Formalism also satisfies the timedependent Schrodinger equation. is the Fourier transform of the nonrelativistic propagator.*<)' ( 2 L 1 8 ) 0(tf ~ li)QfG + 5(Xf ~ X)5^f since G(xf.U) = / T){x} eiS\ [fdt \mx2. The result.ti) = 5(xj — x). (21.17) Differentiating K with respect to tf we obtain —K(xf. The function G which satisfies the timedependent Schrodinger equation (21. (21. (21.1)).x.U)6(tf . Eq.x. (21. We have therefore established that the Feynman integral (21.15) is a wave function (cf.1 Configuration space representation In the case of V = 0 we have from the previous section for the action.U. In a problem where V is small it is convenient to calculate K with a perturbation method. or K(xf.3.Xi.tf.tf.U)G0. K0 = 9(tf .510 CHAPTER 21. 21.20) . Hence our original function K is the same as G.ti) = = 9{tfti)—G + G5(tfti) . and the Green's function KQ\ S0= Here (cf.x.10) is indeed to be identified with the timedependent Green's function. Eq. 21. Thus we consider here the zeroth order of the latter case and begin with the configuration space representation of this free particle Green's function.tf. X U) = i6{Xf ' ~ X)6{tf " U)' (2L19) Thus K is the Green's function for the problem of the Schrodinger equation.U) = G(xf. denoted by KQ.U).8)) Go(xf.16). (21.
tf. (7.Xi.ti) = = G0(x.23) for Go and obtain Go(xf.Xi.ti) lim n^oo lim n>oo m0\(?1+1)/V27rie\"/2 1 z —— exp 2e(n + l )(a.oo /*oo / oo cfcci / J—oo cfcc2 • • • / J—oo + (x.t) is the configuration space representation of the free particle (V — 0) nonrelativistic Green's function.Xi) exp \2Kie{n + 1) 2e(n + 1) Note the square root factor in front which originates as discussed after Eq.tf.t) K0(x./ .23) We now insert our expression for In into Eq.24) ( s / .tf. we have Go(xf. (21. (21.x 0 )" n + 1 (in agreement with Eq.3 The Green's Function for Potential and hence Go{xf.7) for here n = 1 (n+l)/2 lim ( —^~ K2meJ m /n A = 0 V 2e (21.ti) Ko(xf.Xi.tf.1 we show t h a t oo /.Xi. But t = tf — ti = (n + l)e.21) fe=l and t = (n + l)e = tf — U and h = 1..37) — by explicit solution of the differential equation which the Green's function satisfies.i ) 2 . 2t Go(xf.22) • \ n/2 . n + i .8).tf. (21.ti)= iX (o.Z i ) \2KieJ \m0 J y/n + 1 1/2 m0 «mo (21.Xi.z / c .ti)9(t).. We obtained the same expression previously — see Eq. exp where Xj = xo and Xf = xn+\ and there k = 1).t) = ( m0 \2mt 1/2 irriQ and writing x = Xf — Xi. TJ+1 dxn exp[i\{(xi 2 — XQ) + • • ^n) }] (21.tf. (21.21.25) T h e quantity Ko(x. f c .Xi. oo dxi 71+1 / J—c /_ dxn exp oo irriQ oo ~27 ^](a. exp 2 (21. . In Example 21. Then Go(xf.ti) = V = 0 511 m0 lim n—>oo \ 2irie oo (n+l)/2 /oo oo /*o / dx2.
2y(xn+2 .. Solution: The result can be obtained by cumbersome integration.ti. (7.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.25) we can rewrite the Feynman path integral or kernel (21. as / JXi f r D{x{t)}eiS°lh .xi)2 = 21 xi ^2o(„.tn)••• K0(xi. Eq.. the latter being a Green's function with interaction like that for the harmonic oscillator. ••• 2 / \ 1/2 \2iriheJ dxldx2•••dxnK0(xn+l.tn+l•xn.512 CHAPTER 21.54)._*R±HY+l_{xo_X2f J —! r dXieiM(xlxo)2 + (x2~xi)2 ei^(x2x0) and using Eq. n+l = lim / n ^{a^jexp fc=i ^°° Jxi = lim / • • • / dx\dxi • • • dxn I —1 nKx J J \2mneJ / \ 1/2 . (n + l)A"_ We set y := x n +i — Xi = x n +i — XQ and (x n + 2 .^(Xn+1~X0^2+iX(Xn+2~Xn+1^'i.6a). ^ (xi . (21.26) with n integrations and n + l factors KQ.)!} J — OO . Then oo 2 / (21. (21. we have instead of 5*0 and KQ expressions S and K.)+(1"+2.xo)2 + (x2 . For n = 1 (see below) h This result can be verified like Eq. If the potential V(x) of Eq.t0) (21.f_^y e.10) in a product form.22).e. (21. or else by induction. (21. i.x0.x „ + i ) 2 = ( x n + 2 Xi + Xi .7) by direct integration with ^ . In our proof of induction we assume the result is correct for n. 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.x ^ 2 + y2 .x „ + i ) 2 = (x„ + 2 .22) oo Indxn+leiX<x"+*x»+^ 1/2 f°° dxn+ie^:. Example 21. e 2he ^{Xn+lXn)2 = \2iriheJ lim ••• n^ooJ J x(^*) e^() .5) is carried along.1: Evaluation of a path integral Verify Eq.
(21. ) H — (xn+2Xi) n+2 n+l n+2 y2 .Q .21.Xi) + ( x n + 2 Xi)2. n+l T h u s as r e q u i r e d 1/2 Jn+l ^ e J — oo 1/2 .27) '4a in can be verified by completing the square in the argument of the exponential in the integrand to ia(p + x/2a)2 and using Eq. it .25): Ko{xJ) = = (S) e{t)(^X'2 y J .i) H —3/ . i.2y(xn+2 . A(n + 2) 21.27) we have (with a = t/2m0) . s 2 n + 2 2 „ / ^ n + 1 / \2 (x„+2a. (21. This quantity is also described as the nonrelativistic free particle propagator. (21. and + 2 2 l n + 2 ( x n + 2 .30a) —6(t) / dp exp ipx — to Joo 1 i—t 2m.28) exp (— V 2irimo J —( dp exp ipx — i p* 2mo (21.2.i j ) H n +— z 1 .a . n + 2 . iir(n + 1) 1/2 jn+1 n+l L(n + 2)A'n + l I 1/2 4 (n + l ) A .e.2 y ( a .3. e > 0 . 21. The first. (n+l)A n n j(a.. Using Eq. rc+2 rl 1 / s2 513 so that n dz = dy = d i n + i . We pass over to momentum space with the help of the following two integrals.max exp i 2t 6(t) 1/2 \2mtJ f°° 2KimQ J dp exp J—c ipx P t 2mo (21.3 The Green's Function for Potential V = 0 Next we set z : re+1 (x )2 — Xj).29) Hence with KQ from Eq.x2rriQ 1/2 toij_. a dp exp [ipx + iap2] = exp — i(21. The second integral is an integral representation of the step function.i) . (21.2 M o m e n t u m space represenation Our next task is to derive the momentum space representation of the free particle Green's function KQ.6a).n+2a. 0(t) Ti: i r Art alr This integral can be verified by applying Cauchy's residue theorem in the plane of complex r and using one or the other of the contours shown in Fig.
The Feynman Path Integral Formalism Im t \t>0 Rex Fig.t) i(2 h?L L E iv 00 dr exnlipx T '2mo t + itr] — te ~U2^LdPJ ^E+£. The generalization to three space dimensions (required below in Eq. i. 21.e. the second order two factors of V and so on. Replacing 9(t) by its integral representation (21. an expansion in rising powers of the coupling constant (contained in the potential). 21. ( 3 b) ° The expression E+ V — it 2m.*' where 2mo — r.Q (21.51)) is fairly selfevident. (21.4 Including V in First Order Perturbation We now proceed to calculate the first order correction in a perturbation expansion. The first order correction contains one factor of the potential V. This is the type of expansion most people describe as perturbation theory.28) this becomes K0(x.2 Contours of integration. Thus in first .31) is the nonrelativistic propagator which describes the propagation of the free particle (V — 0) wave function.514 CHAPTER 21.
^ J_00 V(xk.xi. .)} (21. ^ = k ) = lim (S^ A H*\ nKX) *—J \2irieJ r "+1 />oo X / dsi / dx2.t)] n+l . (21.xfc_ OO J— OO J—OO L J.32) Since — we recall this for convenience G(xf.tf.t/..i V e / D { a . \2meJ n+l J.21) this can b e written as Coo oo /•oo />oo / oo dt / J —oo ^Go(cc/.Xi.20)) and with elS = exp i S i o / hi dtV(x.tk) J_( x exp fc=l = ie(—2(xkxk 1>{x(t)}exp iSoi dtV(x.t) JSo dtV(x. .21. } e °° f l f •/ oo /"oo F ( a ._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 . dxr.5) a n d (21.i.U).(xfc . x (n+l)/2 \ lim . .t) the first order correction is (G\ ~ G — Go) Gi(xf.^.4 Including V to First Order: Perturbation Expansion h=l.ti)= n+l ri_> f V{x(t)}eiSo „ iSo f ' dt[iV(x. li f f 515 order (cf.ti. (21.Xi.33) xG0(xi.U) = — lim fv{x(t)}el —— js (n+l)/2 I dxi dx2. (21. . Eqs. dxn exp i ^ —.^){iV(a.t)\.+i 6XP L — {xk XkiY {iV(xhti)} 1/2 X Em fc=l — 0 (xkXki) W i t h Eq.. dxi ( m0 V V27rie/ m0 \^He \ exp fe=..tf.i..
(21. The factors of "V" originate from the series expansion of the exponential exp ftf . The first two terms lead to identical contributions and therefore cancel the factor 1/2! in the expansion.i / dtV(x.tf.35) We also observe here that the problem of timeordering arises only in perturbation theory.e.xi.ti)K0(x.xi.37) Jti The last two contributions do not describe the time evolution from ti to tf.t) (21. dx2Ko(xf.t2)V(x2. t') + • • • .x2.t.t2) xKo(x2.t2)V(x2.tf.I dtV(x. in writing down the expansion exp i dtV(x.tf.U) xV(xi.36) {if •j.34) We observe that we can extend the ^integration to — oo since KQ contains the step function 9{t).t2.x. The Feynman Path Integral Formalism and hence correspondingly for the first order contribution Ky.516 CHAPTER 21.ti).ti) t2)K0(x2. OO /"OO / oo dt / J—oo dxKo(xf. (21. / 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'. Xi.t)V(x. i. This is cancelled by n! different time orderings. t) Ju Thus the term containing n times V contains a factor 1/n!. (21.x2. It can now be surmized that the next order contribution to KQ + K\ is K2 = (~i)2 / dti / dt2 I dx\ I ti.e. = i dt2 / dx2Ki(xf.ti).t)K0(x. They must therefore be deleted. Jti Consider the term of second order.t) =li dtV(x. i.xi. t2. .t) ! Jti r*f dt'V(x.
i ] P I dt J dxKn(xf.t)V(x.Xi.40) The rules for the present case are given in Fig.t)Ko(x. Here. E) which is given by (in onetime plus onespace dimensions) V(x.t) is V(p.t)K0(x.tf. The Fourier transform of V(x.ti). .3 Feynman rules and representation of the Green's function.t. (21.ti) = K0(xf. 21.t)= f dpdEeipxiEtV(p.39).tf. (21. U) (21. of course.t) ) K K K0 'V X K o Fig.ti) (21. The diagrams represent mathematical quantities and hence are designed on the basis of rules.Xi. For various reasons — such as illustration. tf. 21. we are concerned with quantum mechanics.Xi. transparency or brevity — it is useful to introduce a diagramatic representation of individual perturbation terms.Xi.Xi.t)V(x.t2 K o (x 2'l2 .x.t. called Feynman rules. This can also be written in the form of an integral equation which when iterated yields this expansion: K(xj.xi.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.ti).tf.E). or in momentum space as mostly in field theory.3 along with the diagramatic representation of Eq. and hence the diagrams representing perturbation terms here are Feynman diagrams corresponding to those in field theory. Such diagrams in the context of field theory are called Feynman diagrams.tf. x r l i 'V(x.21. x"^ X x"2.ti) i dt = K0(xf.39) dxK(xf.38) . These may be formulated in configuration space as below.tf.
^(x. (21. for the allowed values of k: kx = 2imx/L. to one particle in all of space. i.ti). mo V since ip is normalized to one particle in volume V.t) = . 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). * 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. The Feynman Path Integral Formalism 21. here with E — k 2 /2mo. in onespace plus onetime dimensions by A= dxf dxiil)*f{xf. T = duration of time for which potential is switched on. (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.= l. with p = Kk. i. The transition amplitude A describing this process is defined by the Green's function sandwiched between the inand outstate wave functions.e. The plane wave function normalized to one over a box of volume V = L3 is. 2 The expression A 2 /T can be related to firstorder timedependent perturbation theory in quantum mechanics.42) VV dn dp Jv We are interested in the total cross section defined by transition probability per second incident flux / Here (1) dn/dp = density of (plane wave) states in the box (of volume V — L 3 ) per unit threemomentum interval** = V/(2irh)3.ti)il)i(xi.k = 2wn/L.tj)K{xf. (21.e. We recall for convenience from **Note that from the wave function we obtain. and dn = dnxdnydnz. .7 = e i k ' x _ i : l i with f i/}*i/jdx.ipf have to be properly normalized. dn = (L/27r) 3 dk.41) We have to pass over to three space dimensions.Xi. (3) The transition probability per second is = \A\2 .518 CHAPTER 21. There the wave functions ipi.tf.
V(x.T)\2 f • (2L46) We take the following expression for the timeregularized Coulomb potential.44) = u(Ek.k. Eq.t) = e~^lT\ r where r = Ixl. P(*) = j r (2145) (cf. (21.r)eiE^\ H'ki(t) = J dv u*k(r)H' Ui(r) Thus the amplitude ak is effectively the spacetime Fourier transform of the interaction or potential XH' contained in the Hamiltonian.5 Rederivation of the Rutherford Formula 519 Chapter 10 the Schrodinger equation with the subdivision of its Hamiltonian into an unperturbed part and a perturbation part..= 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. Hence (h = 1) atot = f J j2^~k dp V Vm0\A(p. (10.117) and Example 10.21. H = H0 + \H'. Then H'ki can be taken out of the integrand in Eq. and with these the following perturbation ansatz: ih^* with an(t) = HV. i. The transition probability per unit time is . (21.47) . Then for stationary states M^t) and itm^ = f HUt'V^t'dt'. «= — .— . u = ^\H'kfp(k).4).. in which T is a large but finite time. Proceeding in this manner one obtains Fermi's "golden rule". # = ^an(i)VnM) n = a^(t) + \a^(t) + .44) and f_oo can be replaced by fQ. 47T 1 3 7 (21.e.
we have to replace the onedimensional quantities dxf etc.xuti) = ^ J°° d p e i p x _ i ^ * ..50) Here x — xj — Xi and t = tf — ti.f.>/>.tf. so that 0(t) can be dropped) K0(xf. dt dxip*f(xf.tf. i. x. The threedimensional generalization is (taking tf » T and ti <C —T.ti).49) i.30a) K0(xf.51) Inserting this into Eq.Xi. (21.ti). The Feynman Path Integral Formalism In calculating the Rutherford formula with the help of our previous formulas.tf)Ki(x. £. is now in three space dimensions A = Ai= / dxf / d^ty*f{'x. The lowest order approximation of A. (21. (21.t) (21. (21. (x*. we obtain A1 = clS(ti tf) / dxf(i)V(xf.e. T a large but finite time. (21. This now has the form of the amplitude (21.tf.49) together with the following wave functions (of asymptotically free particles) 1>i(*i.* / ..44) (here in the scattering problem with initial and final energies equal).i x (i)V(x.48) where from Eq. t)K0(x.x. (21.t) = ^ J*ie*<*'*><<*'*>.tf)ip*(xf. In our previous onedimensional considerations we had with Eq. (21. Thus with the ansatz Ko(x. t.f.520 CHAPTER 21.39) OO /"00 / dt / OO J —OO dxK0(xf. also called Born approximation. Ai = dxf / dx. by threedimensional dxf and so on.tf)K0(xf. t. (21.44) by contracting the propagators to instantaneous point interactions.x. We can relate this amplitude to the amplitude ark of Eq.tf.t)K0(x. Xj.tfx.tf)ipi(xf.tf).).52) .t.^ / .t)V(x.U) = ~ e ^ l ^ \ ^(*/>*/) = _ L e * P . ti) = co5(t — U)5(x — Xj) etc.ti)'ipi(xi.yLi.
fdxe^rt* dxe< p) x fdte1 ' (4) 2m0 T2 e "^^J2" 16 .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 . £). k2 ipx + i—t + ikx — i 1 2mo 2mo (21. 2 „ x v~ m o (21. .k.p ) .54) Now inserting the expression for the potential V(x. At2 dteiat~^ a2T2 f°° =e " J dre^lT2 p2k2 2mo = (~)l'2ea2T'' / 1 6 .57) we have A(p.T) = JdtJdx(^\V(x. P2 . Then integrating over q and q' we obtain .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 . which is evaluated in Example 21. Then we require the following integral. (21.).56) (21.p 2.53) The integrations with respect to Xf and Xj (including in each case a factor (27r)3) yield delta functions 5(q — p) and <5(k — q') respectively.t)exp A(p. (21.k.T) = ^JdXJdt%*t2/T2exp i ( k .5 Rederivation of the Rutherford Formula we have (for ingoing momentum k and outgoing momentum p A(p.k.k.59) .2: piqx / Y iukawa •" dx e~Mr 47T q + M2' 2 (21.58) In doing the integration over x we introduce a temporary cutoff r = 1/M by inserting the factor e~Mr in the integrand.iaT2/8: L With 0 0 • . this becomes A(p.T) = x exp L z/ q/2 k2 521 f dx. (21.r) = = y .21.i ( k 2 .~ .
51 — 2 /w2 y ~ —. e 8{k) = lim 7=*—^ (21. Using the result (21. where M has the meaning of mass (in the case of the Yukawa potential the mass of the exchanged meson). otot contains the factor \A(p. y 2 [ ( k .3.61) We can now Hence in our case \A\2/T has to be replaced by \A\2/(nT2/8)1/2.522 CHAPTER 21. 66) o ." + M ] p) ' "•"• ^2 .„ 2. T ) ~ .62) We want to consider the limit T — oo. / / P 2 _ .T2f 8 V/ 2 . The Feynman Path Integral Formalism This means: The result of Fourier transforming the Yukawa potential ~ exp(—Mr)/r is the propagator (21.60) Now.63) we have: Probability per unit time = 77777.59).e. { I i _ F T l } .T)2/T. (21.t o t= y dp ( 2 v r ) 3 "V 3 I W: 2mo y m0V k 32^a2S(^) y ( k y 22 [[ ( k _ p ))2 + M 2 ] 2 ' kP 2 (21. ( 21 .„. e (21. .a ( — ) ^ .P ) 2 + M2]2 4 VvrT2. write down the probability per unit time which is ( 8 \1'2 _a2 16^ . (21.\ 2?7lo / 1 (21. 2 with 2mo J 2ir a2 Ct 16TT2 LU7I F [(k . we obtain . dE = PjV = ^ E ± .^ .63) We establish this result for T — 1/e in Example 21. In our case this has not been done. For this we require the following • representation of the delta function: —k2T2 rp —k2T2/8 S(k) = lim T . i.< ^ £ \M VTTT2. k . Using Eq.64) Inserting the result now into our expression for the total cross section.59) we can reexpress A as I /7rT2\1/2 e (P 2 k 2 )T 2 f Ajr ~) p * P . k2 . This expression assumes that the integration over time t is normalized to 1. <rtot. mo m (21 .65) But E=*.k. integrating over the square of the ^dependent factor in V we have /"°° dt(e4*/"1*}2 = (^p) ' 2 instead of T.
^ Jo 47T q 1 M — iq 47T i q2 + M 2 Jo G— 9 Example 21.e .j .^ .p ) 2 = 2 k 2 ( l . (21. and inserting the expression thus obtained into the final result of Example 10.65) with the cutoff M > 0.2: Fourier transform of the Yukawa potential Show that [ j / j eiqx dx—e .. so t h a t (21. Then (x = r) foo „2jj.j . esfn (21. giving the potential a coupling constant.70) .67) and (21. Example 21.= e " f c a ^ o 2v^fa /4a = lim £^o —. setting M2 = 0. (21.69) This is the wellknown Rutherford formula.e.4. with k = y 2moE / we obtain the differential o?m\ 4fc sin 4 (0/2)' 4 cross section (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 — .21. (21.67) T h e delta function here implies energy conservation.^ . 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. 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 . —Mr 4w = .68) into Eq. /Yukawa = ) Solution: We choose the 2axis along the vector q. i.e.M £ M * = / m0kdn = m0^2m0E (k .68) Inserting (21.cos 9) = 4k 2 sin 2  . 2mo / dft.59) by replacing there q 2 by ( k — p ) 2 . p 2 = k 2 .5 Rederivation so t h a t of the Rutherford Formula 523 M^f) . This result may also be obtained from lyukawa of Eq.* 9 ' d r e .
We consider again nonrelativistic quantum mechanics. The Feynman Path Integral Formalism o° 2 / oo dxe~ax eikx. We have x> = I J ^ l p X p l x ) .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. a. 7 r°° d 9 xdxe~ax elkx = — dx — (e~ax oo 2a Joo dx Partial integration of the right hand side implies oo /. Direct integration /4a .75) .70) and (21. (yx) = 6(y . Differentiation of (21..71) with respect to k yields the same as °° 2 . (21. The two representations are related to one another as we saw in Chapter 4 by the following expressions or Fourier integrals: p> =  d x  x ) ( x  p ) .524 The function required is CHAPTER 21.71) we obtain a representation of the delta function: <5(fc) = lim — / 1 /"oo a ^ O 27T J _ 0 0 dxeax 2 eikx = lim —g(k) o ^ 0 2?r ^ O 2K 1 i 1 = lim k 9 .71) Solution: In general one uses for the evaluation of such an integral the method of contour integration.q). (21. Va we obtain g(k) = Je~k V a = 0.x) (21. a simpler method suffices.73) 5{p) = (W / dxe"P'x' 5{x) = J^rI dpePx '' (2L74) The normalizations (qp) = (27r)3<5(p . co oo dx co (eikx)e~ax dx J = — / ikelkxe~ax 2a 7 _ 0 0 dx = ia^ Thus g(fc) satisfies the first order differential equation g'{k) + (k/2a)g(k) yields g(k) = Ce~k2/4a with the constant C = g(0). In configuration space the state of the particle is written x).^0 a^C 2^/TTO.oo J / )eikx. (21. We write the state vector of a particle of threemomentum p in the momentum space representation: p). In the present case. '<*> = . ' i a = lim e —0 pk 2 /e2 e^TT 21.72) co From (21. Since dxe" 1 1 1 = « / .* . however.
) H = (x</>o)s.t).We now introduce the Dirac bra and ket formalism. i.e.21.The time development in the Schrodinger picture is given by \tpt)s = e~lHt\4>o)s.82) (21.p). so thattt d — x.6 Path Integrals in Dirac's Notation therefore imply 525 (xp) = e ipoc . Then ( X  ^ .t'. (21. The time development is given by the Hamilton operator H. so that \if>t=o)s = H>)H. Ryder [241]. (21. (21.t'x.^ l x ) . . pp.81) Comments on signs in Eq. x.t)} constitute a "moving reference frame" in the sense that ip{x.i'x.t). Then the states {x. The relation between Heisenberg picture states and Schrodinger picture states is \ipt)s = e~lHt\ip)H. See also L.t\.82): The general Schrodinger equation has the state vector to the right. at initial time ti to its final position xy at time tf.t). (21.x.e.80) (21. i.77) (21.i) = K(x'..e.e. (21. H.£) = ( x V ^ ' . 4) is to satisfy the Schrodinger equation.t\ = iH{x. i.If '/'(x. t) = (x. i. we must have d/dt{x.t).We define x.76) These relations imply the completeness relations 1 = J dxx><x. Similarly we have a position operator X with Xx) = xx). (px) = e~ipx. We consider first an infinitesimal transition. d/dt\x. Hence for the infinitesimal transition (x'.t) = +iH\x.t> =iH\x. Thus (qPp) = p(qp) = p(27r)3<5(q . as had to be shown. (with ft = 1) d/dt\ip) = —iH\tp).79) Feynman's principle is expressed in terms of the propagator or transition function (x'. t\ipo)s. 159 . t' — t. M^ip(x.t) = e* H t x).78) Pp) = pp) is to be interpreted to say: The quantity p is the eigenvalue of operator P with eigenvector p).i) = e iH4 x.e.t) = —iHip(x. which describes the propagation of the particle from its initial position x.. i.0) = e""x).83) (21. x' — x infinitesimal.160. The equation 1 = J ^p)(p.
"  » We suppose that for a single particle of mass mo moving in a potential V(x) the Hamiltonian is (21.P V<P e l )  P > e P X <2184 = /(w/(^ ' ' *" '"'' *=£Uv«. (21.526 CHAPTER 21.p'). (2187) (21. The Feynman Path Integral Formalism We can rewrite this relation as <X t M ' ' = /(^/(l]3 < X ' I P ' > < P '  e ""'"'' )  P ) < P  X > e. 2mo Since (21.* .x = V(p .i p '.x') + V{x)5(x' .91) . i..y<&(p'ix')(x'ivix)<Xp) = f dxe.85) the amplitude or expectation value or position space representation (x'#x) = .90) (P'IVIP) = y<&.88) Thus P has the operator representation P= /dxx)V(x.89) Using this and previous relations we obtain for the Hamiltonian (21.V 2 < 5 ( x .x y(x)e i p .x).q) = qq>We also have by mapping onto configuration space Pp) = = P /*dxx)(xp) = P /"dxx)e i p x = p /"dxx)e i p x f dx\x)—eipx= /"dxx)V(xp).85) The momentum operator P satisfies the relation P = /(2^3PP)(P' (2L86) so that it projects out the momentum q of a momentum state q).e. (21. Pq) = J 7^3Pp)(pq) = J dppp>*(p .
p').tj+i\x.t') \^(27rf5(p 2mo +V(pp' where we used (21. tf subdivided into infinitesimal transitions (xj+i.j.6 Path Integrals in Dirac's Notation we have in momentum space the representation (p'ff p> = ^(27r) 3 <5(p .94) In the above we have chosen the signs such that p = —iV and H = i— ot (21.82)).tj). Thus with •"•/ — ^ n > *•/ — ^ni ^ 0 — Xj. t). (x'. We have <x'.t'f / w? eip''x'^(p ~p/) = / l r ip''x' / "x"e~JP'x"v{x")e" = = Hence + y(x') 2771Q d /dx"<5(x'x")F(x"KP'x' y(x')e i p ' x '. (21.p) + i{t . ti to x j . i. But e'ipx + 0[{t . the signs are reversed when these operators are applied to the ket vector \x. 527 (21. We now consider the transition of the particle from Xj.21.t> d f dp' P' r_dp f dV rivf*!ini\rmtf) J (27T)3y (27T p e p) e ipx +i(tt'){p'\H\p) + }eipx .92) We now return to the expression (21.93) + ipx P r ip(x'x) iH(t't) (2TT) 3 (21.e.95) when applied to the bra vector (x. for the amplitude of an infinitesimal transition.p') + V(p .fx.i'x. i.p') • • • j W < 5 ( p ' .t) (cf.84).92). (21. EQ — Hi .
a — —e/2mo and divide b o t h > sides by 2TT. d x n .. .ti. dx.ti) = x 0=Xi n^n fc=1 ? <^Pi (2TT)3_ exp / ^fa . .t 0 ) + H(p2.e.tixo.. i.2.tn respectively.x 0 ) + p 2 • (x 2 . (21.tf\Xi.e. x i .97) H hH(pn. The resulting relation is m0 2irieJ 1/2 m0x exp ze ) " / 2TT exp Pk 2 mo 1 (21. d x i . x n which the particle passes at times tQ. d x i ( x n .xic) by fc — Xfej H(pk.. .528 CHAPTER 21. . The Feynman Path Integral Formalism we obtain (it will be seen below why we do not introduce normalization factors here) (x/.X n _i)] h) (21. then corresponds to the sum over all paths.98) tk — *fcl # ( p f c .i/xj. . T h e integration over the n—1 intermediate positions.tfei) fe=i . x f c ) = ^ .27) in which we make the replacements x —> eifc.x i ) H h pn • ( x n . zmo T h e n t h e integration over all p ^ can be performed by using Eq.+ V(x f c ).i /Xj=x(*i) f J [ ^ 1 dp2 3 3 J (2TT)3 (2TT) . (Xf.dpnl dpn (2TT) 3 (2TT) 3 exp [i{pi • (xi .xn)(tn in_i)]. t n  x n _ i . .. i. . x i ) ( t i .tf\Xi. We now insert for each of the infinitesimal intermediate transitions the amplitude (21. t n _ i ) t„2><xi.tj) = / ••• / d x „ _ i d x n _ 2 .i { f l " ( p i . x exp [ . ....x2)(t2 We can rearrange the factors here in the following form c„=x/ n—1 (Xf.t0>(2196) We define a p a t h in configuration space by a succession of n + 1 points x o . = 1 X X<jPfc We now replace H(pfc.99) . d x „ _ i .94) and thus obtain a phase space expression.ti) rx/=x(i/) d x i dx2 .xk (21..
102) We see t h a t the factor (with h = 1) m0 2m(tk 3/2 3/2 [N(e)}< = tki) \2irie) is precisely the normalization factor inserted earlier for the onedimensional case in Eq.H(x.103) is another form of the pathintegral for which the integration with the help of Eq. (21.98) can be rewritten as 3n/2 2^") x / n dxfe exp * ^o** ~tfcV rnl <i 2 m ° x fc ~ ^ (21.98).] ex H ^I (. Smolyanov. (21. > Thus the phase space representation of the pathintegral* (21. Truman [254]. i. / {xf.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. i.21. .tfci) fe=i £(**'**){ p***^} —tki m0±2k (21.8) as N(e) so t h a t the integral does not vanish for e — 0. G..e.101) Then Eq. G. (21. m& ze ) J (2vr)3 exp it p fc • xfe IP* 2 mo (21.ti) = J D{x(t)}D{p(t)}expi [' dt\pi.100) Hence with all tk — ifci = e we obtain /n dpi i _(2vr)3 exp •. to / m0 \ 3 / 2 las. Tokarev and A.tf\xi.fc=i 3n/2 r n m0 exp 2_^i{tk 2m(tk .100) over V{p}= L J ^ rwooll V (27T)3 1=1 ' lim T T T ^ *The reader interested in a rigorous treatment of representations of solutions of Schrodinger equations in terms of Hamiltonian Feynman path integrals may like to consult O.e. (21. A.p)]\.
105) D{x} n ^ o o [7V(e)]3n Udx kk=\ In this notation the measure factor is contained in ©{x}. The Feynman Path Integral Formalism yields automatically the correct normalization factor or metric in the configuration space path integral^ in agreement with Eq. £ J ) with no operator in between.tf\xi. We check below that this is given by x = / ( ^  p > i  ( p   <2L107) See also C.104) where for the particularly simple Lagrangian L under consideration here m0±l . operator (21.ti) where nl D{x(£)}exp '. xfc).V(xk) = m0±l . 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.H(xfe.ti) = ^lim^ N(e)3n / * JJcbc fc exp i / fc=i * dtL(x. .106) In order to be able to deal with this expression we first require the momentum operator representation of the operator x(t). We have now obtained the path integral representation of the matrix element ( X / .530 CHAPTER 21.tf\xi. £ /  X J . t to x. (21. Clearly for a better understanding one also wants to explore the case of sandwiched operators in the path integral representation.x) (21.e. Thus finally we have Feynman's principle: (Xf. nl (xf.iJ(x fc . Garrod [106]. pfc) = Pfc • xfc . We know from classical mechanics that this relation is a Legendre transform which transforms from variables x. p .8). x.t'}.8). (21. pk) = L(xk. this agrees with the corresponding discrepancy in the onedimensional case of Eq. I hi dtL(x. t.x) . i. (21.
(21.108) (x 4 p)^^(px 5 )(x 5 e^ t 'x 6 )(x 6 p / )e.v ' x ' and thus {x". Thus <xV'xx'. so that we obtain a cnumber position coordinate which we can shifted across other cnumbers. With the wave functions (21. With contractions and integrations over delta functions the expression for the matrix element then becomes <xVxx'.i p '• x .5 l^ t 'lp / > e . (px).*') = / I F / (2?FeiP"'x" / ^ 5x5 (p // K^> 5 )(.t'} = = fdx5x5(x"\eim"\x5)(x5\eim'\x'} f dx5x5{x". .76) we can rewrite the factor (x4p)^(px5) as (x 4 p)x 5 (px 5 ). x fdxG  ^ P <p"x3)<x3e^V) x (21.21.* * f dp = j (2^ p) ^ d .107).* * . ••• y (2^ x= f dp p) ^(p d We now consider an expectation value in which we replace the operator x (which is sandwiched in between exponentials containing Hamilton operators) by the expression (21.t"\x\x'./ ^ I < « .109) . .t'}. .0 = J'^ j'^e^''(p"\e^''xe^'\p')e^'.t"\x5){x5\x'. (21.77) and thus have 531 *!*') = *M = * 7 ^ I P > « .6 Path Integrals in Dirac's Notation For the verification we use Eq.
(21.i ( t " . ( x " .tj+i\x\xj. (21. £ n _ixx n _ 2 . .t') (xV'xx'.i(t" . (21. This is the expression corresponding to that of Eq. we obtain (here and in the following the limit n —> oo at the end being understood) (x/.iit!' 1') 2mo + vtf) + <5(x"x').t') dp . Inserting here the integral representation (21. Thus with x^ = xn. we obtain with (21.t n _i) x ( x n _ i .110) Recalling Eq.£ n xx n _i. (21.tj) at equal time intervals e. The Feynman Path Integral Formalism In the second last expression here we expand the exponentials and obtain (xV'xK.i t " ( x "  #  x 5 ) ( x V ) +.i{t" .0 dp x„ / (27T): 1 .113) Inserting for each of the infinitesimal intermediate transitions the amplitude .90) we can rewrite this expression as (x". x / / e .74) of the delta function.ti = to as before.f ) i f e i p . t0).tf subdivided into n transitions (xj+i.x') . tixx 0 . .*/<x"x5)(x5tfx'> + .x') . = x"6{x" .111) 1 .85) (xVxx'.532 CHAPTER 21.x ' ) (2^ i^+vM + D «p(x"x') (21.tf — tn and x$ = XQ.112) where x" is a cnumber.£j to Xf.0 = and hence /dx 5 x 5 [(x"x 5 )(x 5 x') .94) for the case with no operator in between.it / / x / (x"ffx / ) + it'x"{x"\H\x>) + • • • . ] .£j) = / ••• / dx n _idx n _2 .. We now consider the transition of the expectation value of operator x from Xj.t'){x"\H\x') = X + •••} (21. tn2) • • • (xi.t"\x\x'. We now proceed from there along parallel lines.£'> = x"[«5(x" .• dx1 (x n .£/xxj.
n—>oo *•»• 21. (21.t') = and con 5x"Vx„{x".t') .21.117) "•"One may observe here that if x = x^ = x„ in the integrand of Eq...\Xi. (21. tj \ xt. tj  xj1 x».ti) 533 rxf=x(tf) = / • • • / dXidx.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...x2)(t2 ~ h) + • • • + H(p ••• + p „ ..t"\x'.xi) + f f dpi dp2 dp n _i dp n The momentum integrations can be carried out as before and we obtain finally (x/. dyi l xi x 2 .t') (21. For simplicity we consider the case of one spatial dimension.. Before we can consider canonical quantization in the context of path integrals we have to investigate in more detail the role played by momenta.x . .x n _ i ) ] .7 Canonical Quantization from Path Integrals (21.t"\x'. Xn _i X„ xi dx. U). x n _ i x n reduces to the single factor x^.115) where (there is no integration over x n ) nl 0){x(t)}= l i m [ i V ( e ) ] ..2 .t"\x'.t"\x'.89) is consistent with the conjugate expression dL (2L116 * .{x".tj) = / JXi 2){x(i)}x(i) exp i I I Jti dtL{. .x 0 ) + p 2 • (x 2 .114). .3 " n d x f c .y (27F (27F •" (2^(2^)3 exp[..112). We demonstrate first that the operator representation of P given by Eq.t') = (x" + 5x". x (21.10) or (21. (21.x 1 )(t 1 Z & ) +H(p2. x n _ i x n i JXi=x(ti) /Xj=x(tj) 7 . the string of factors xi X2 .2 .114) x exp [i{pi • (xi . we obtain* (xf.t') sider the variation S(x".tf\x.t"\x'.( x „ .iyxxj.and hence may be put in front of the integrals to yield (xf.105). dxnn _i Xi X2 .« • > We reconsider therefore the transition amplitude (x".{^( P i. tj) = xf (x/.
t'): 1 fx" 6(x".t"\x'. Then. (21. .t"\x'. we separate the metric factors from T){x} in Eq.t') =  6— / 1 V°{x}t . t". For the purpose of enabling some formulations below.t i_(dL_ x't' h\dx J t„ (21.i/(x)/fi ) j ^ ) dtL(x.120) However: In our case here 8x ^ 0 for t — t" (although 5x — 0 at t = t'). V°{x}[i6I{x)/h]e* Jx' The following steps are familiar from classical mechanics: yV 51{x) S /1 dtL(x. Hence w « = (t). (21.122a) = 17) 5x"Vx/. (21.118) — = (21 5x"(x". Thus in this case of classical mechanics dL_ d fdV.x).l&cdi. dL r /•*" .t'). with N = N(t" . dL d5x / oft—5x + / dt dx dt Jt' ox Jt t' and hence <9L —oxdt + ox ox nt t.tt).10) and set D{x} = D°{x}/JV(i/ . " M{x)/h &(t ) = fa (L2) 2 11 and — using Eq. The Feynman Path Integral Formalism with 8x" = 5x(t") ^ 0 and 5x' = 8x(t') = 0. jt/ 37 ( £T.534 CHAPTER 21. dx dt \ dx _ (21.119) The classical equation of motion follows from Hamilton's principle for which 5x = 0 at t — t'. <9L' dt —dx + — 5 x ox ox f . " (t). and we demand the validity of the equation of motion. dt\Ox w(x) = r dt d^_dL(dL dx dt \ dx °L 8x + ^ T!5 X 1 ox (21. x) Jt Jt dL.(x".118) JI(x)/h .
(21.t"\p(t")\x'. I(x) = / JA dtL(x.89) we see that Having explored the appearance of canonical momentum in path integrals we can proceed to extract the canonical quantization.125c) Corresponding to our relation (21. § The evolution of the system is expressed by the Feynman functional integral (xf.115) in three space dimensions.U) = T777TT / &{x}eaW\ IXi=x(ti) ) iV(A) JXi=x(U (21.1 (A / ). F. x' (21.x). Streater and A.122b) By comparison with Eq. (21. Wightman [264].Xf = x(tf).125a) (2) the completeness relation holds. / Jxi V°{x} / Jx' V°{x}. (21.t')(x'. S. The following properties are assumed to hold: (1) / Jti V°{x} = ^ x. 535 (21.126) R.124) where with A = tf — ti.t').1 (A)iV. ^2\x'..t'\ = l.125b) and the measure property (3) N~\A + A') = A^. .tf\xi. Here f*f D°{x} indicates that the integral is to be taken over all x with boundary conditions Xi = x(ti). we now have in one spatial dimension {x^tflxlxuU) = J^ff'DQ{x}x{t)eiIWh. 6 (21.1>) = ^{x".21.7 Canonical Quantization from Path Integrals or JL(x»y\x. We consider this here in quantum mechanics in one spatial dimension but parallel to field theory.
127) We can now write down an expression for a timeordered product of operators. Taking the functional derivative 5/iSx" of this equation we obtain 'ilx7'^' '^"^x"\x ''*') = (21.131) = ih(x".t'){x\t'\x2\xi. (21. . (21. Thus (xf.t\x'.t"\x"\x'.ti).tf\T(xix2)\xi.tf\xi\x'. (21.t>).x](x.t"\x'.t') (21.ti) = x' 'Y^{xf. Thus if T denotes such ordering.129) = ^ y With Eq.t')=x"(x". But for t = ti we have (and equivalently for t = tf) as we can see from Eq.t"\x'. we have the relation (xf. (21... (2) and (3) above.tf\xi. Then the relation follows with the help of the properties (1).536 CHAPTER 21. tn lie correspondingly in consecutive order between U and tf.t"\x'.t\x'.ti) = Xi(xf.132) The generalization to higher spatial dimensions proceeds along parallel lines.tf\x(ti)\xi.t') = ih(x. (21.t').127)) (x". (21. (21.130) ~i5x7jX"{x".128) where tf is later than ti and t\. We consider the case n = 2.t') \ i ox" i ox"J or in commutator form [P.t"\X'.122b) we can deduce the nontrivial canonical commutation relation arising here.t') + x"^J^{x".. (21.t"\x'.114) (xf. We have (cf.t"\x'.ti) f ' V0{x}xlX2eu^lh. Eq. We can rewrite this expression as the commutator relation (ITU*" x"^](x".t').t').133) = ih{x". The Feynman Path Integral Formalism The time t is in general a time between ti and tf..tf\T(Xl • • • xn)\xi.ti) = —!— p V0{x}xx • • • xneu^h.
Watson [283]. Such an expansion parameter is usually known to be or assumed to be small in some sense. it was known from quantum mechanics that RayleighSchrodinger perturbation theory by itself does not yield e.e. N.Chapter 22 Classical Field Configurations 22.e. > 537 . although the first few terms indicate a convergent behaviour. E. i. the exponentially small level splitting which occurs in "The expansion involves both positive and negative powers of the deviation.e.g. Such expansions are strictly speaking. in the context of quantum mechanics are not convergent. for sufficiently large r. integrated over) beyond its radius of convergence. z = a + i/3 with a — oo or f3 —> oo or both. such as the fine structure constant of quantum electrodynamics. in the sense of rigorous convergence tests. i.g. The vast majority of perturbation expansions discussed e.g. Expansions of this type arise. 102.e. see e.1 Introductory Remarks In field theory a perturbation expansion is generally understood to be an expansion in ascending powers of a coupling constant. T. Whittaker and G. In field theory it has been known for a long time that expansions in ascending powers of a large coupling constant are fairly meaningless. An essential singularity is really an ambiguity rather than an infinity. the series converges. divergent. It was therefore natural to search for alternative methods of expansion. 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. Consider e. nonzero radius of convergence in the domain of the coupling parameter. W = e x p ( l / z ) . i. the modulus of the ratio of the (r + l)th term to the rth term is larger than one. but asymptotic. as we have seen in Chapter 8. If the expansion is mathematically well defined. it has some finite. p. In physical applications such cases are rare. so that successive contributions to the first approximation do not invalidate this approximation.g. Besides.
such as soliton theory. One therefore faces the problem of quantizing a system which is subject to constraints. The classical. and the corresponding expansion is called the "loop expansion". There. i. In particular it enabled nonlinear problems to be studied and led to a consideration of topological properties. A method which achieves this is in particular Feynman's path integral procedure. such > a series begins with a contribution proportional to (e. However. i.. cnumber.e..h2...e.538 CHAPTER 22. The study of expansions of this type has turned out to be extremely fruitful. it is an expansion in rising powers of a semiclassical expansion parameter which plays the role of Planck's constant h in the quantum mechanical WKB approximation.g. The procedure requires a change of variables from the original ones to collective and fluctuation variables (in analogy to centre of mass and relative coordinates). These methods are often described as methods of collective coordinates. nature of the dominant approximation does not imply that this describes classical motion. and has led to insights which previously seemed unimaginable. and in general this change is accompanied by certain constraints. the consideration of conditions which insure the existence of Green's functions of this new type of expansion procedure (which again leads to asymptotic expansions) and so on. the procedure discussed thus far) to systems with constraints was developed by Dirac [76] and is introduced in Chapter 27. Classical Field Configurations the case of the symmetric double well potential or in the case of a periodic potential. In the following we consider various typical examples. higher order corrections are of orders ±. the latter's topological properties if any.h°. 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.h3. On the contrary. A further challenging task was the development of methods of quantization of theories which incorporate classical. the dominant contribution is singular (has an essential singularity) for vanishing semiclassical expansion parameter (h — 0).h. although this terminology is not precise. cnumber first approximations. This type of expansion is decsribed as "semiclassicar. 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.e. Many of these aspects are interesting by themselves . In searching for other means of expansion. specific boundary conditions have to be implemented in order to yield these effects which are therefore frequently termed "nonperturbative".) l/h2. The fundamental extension of the method of canonical quantization (i.
The important aspect we want to draw attention to here is that of "spontaneous symmetry breaking".2 The Constant Classical Field 539 and in any case deserve detailed study. 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>. It is evident that symmetries and their violation by the classical approximation play a significant role in the entire consideration. Tong [272]. 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. modified by going to imaginary time). 22. 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). d^] = d^d^ . V{</>) = m§0V + A0(<^)2.2 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. (22. to keep this case in mind. 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.g.1) ' T h e reader who wants a highly advanced overview of instantons. t The classical first approximation in the procedure outlined above may be simply a constant (time and space independent) quantity. or even a solution to a modified field equation (e. However. . vortices and kinks. Thus references to field theory will be exceptional and hopefully do not irritate the reader. This means theories with field densities which are therefore infinitedimensional. may like to consult the article of D. We shall not be concerned so much with the case of a constant first approximation.* assuming that this is acceptable to a reader with some familiarity of the basics of the more complicated field theory of electrodynamics.V(4>). a static (time independent but spacedependent) solution of the classical equations of motion. and also to see these in a somewhat broader context (by comparison with higher dimensional cases). monopoles.22. it is important for a better understanding of the considerations which follow.
540 CHAPTER 22.2) We ask: Are there classical (i. i.3) See e. For UIQ > 0 the potential V((f>) has a single well with minimum at </> = 0.e. p. The EulerLagrange equation is d„ dC d(d. 22. then after cooling with a phase transition one considers expectation values 0 7^ 0 corresponding to minima at \<$>\ ^ 0. To insure that the Hamiltonian H = f dxH with Hamiltonian density 7Y(0(x. .e. m l + \04>*4> )$ = <). B. 433. as illustrated in Fig. § (22. for m§ < 0 the potential is a double well potential with two minima at \<f)\ ^ 0. Classical Field Configurations where d^ — (—d/d(ct). £)) is bounded from below we must have Ao > 0. (In models of the early universe one often considers V(0) with m\ > 0 at the early stage. V ) . 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). Felsager [91]. and hence here [d^ + m20}(l) = Xo(f (22. of the equation of motion? For these we must have all derivatives of <j) zero.g.S£/<90(x.£).1 Different potentials for different signs of m§. V[<t>] m 2>o o Fig. cnumber) solutions <j)(x) = 4>c = const.1. 22.
whereas the field (bc becomes (a) A oi(0+<x) V~2 + Every new phase defines a different solution. We can convince ourselves that <fic is the field associated with a state of minimized energy. 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. aq> For (b = const.5) . i. 22.2 The Constant Classical Field For Ao > 0.4) Here (3 is a spontaneously chosen phase like a stick held upright along H in Fig.ir]=7r<fiC. the only such solution is the case we wish to consider. mg < 0. Thus the classical solution 4>c violates the U{\) symmetry of C and of the equation of motion. H = —C = V. not the solution x(t)). w A iP 3ml 5* "XT" (22.e.22. The Hamiltonian density H is defined by the Legendre transform H[<f>. 22. The U(l) phase transformation <b — exp(ia)((> leaves both £ and the EulerLagrange equa> tion unaffected. the equation has this invariance. TTIQ > 0. TT = —J. But for Ao > 0. (22. 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>. we have 541 0.2 The spontaneously chosen phase. and so ft[<M=m6:(^* 1 + Ao(0*0)2. with 0 = 0.</>2)plane.e. Fig.2 when allowed to fall chooses an unpredictable phase parallel to a radius in the (0>i.
and we wish to investigate the behaviour of the field cf> in its neighbourhood.6) Inserting this into Eq.e. >° for m 0 < °> .7a) (22. in a transverse direction.7b) In Eq. 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. i.3). This is the condition (22.e. the one for (3 = 0.e.3ml = 2ml = ml + TA 0 A 2 = 0.Then m\ m\ = ml + A0A2 = m2. 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). We identify the coefficients of the linear terms on the left hand sides with those of the masses of the fields ipi. (22. Suppose we choose one such configuration. (22. V?) (22. (22. . i.e. we set cf>(x) = <f)c + ri(x).3) obtained above. 22. where (f>c = — T](x) = —= [Vl (x) + ilp2 (x)} • (22. i. Clearly these are the field configurations which trace out the circular bottom of the trough of the double well potential.2.7b) the constant on the left and the coefficient of A on the right of Eq.tp2.2) and separating real and imaginary parts.g. e. and partly by climbing up the parabolic wall on either side. For every value of the phase (3 we have a different constant cnumber field configuration 4>c. along the trough of V (which we can call a longitudinal direction).542 CHAPTER 22. ^ ) o + m 2 + 1A 0 A 2 ) ^1 = A (± A0A2 + m 2 ) + 0 ( ^ 2 . &H/d(fi = 0) if [m20 + ^>]</> = 0.7a) vanish on account of Eq. (22. as indicated in Fig. Classical Field Configurations the density Ti is minimized (i. one obtains the equations (d^ and (d^ + ml + ^ A 0 A 2 V = 0 ( ^ 2 .
t'). tangential to the classical path. t. Goldstone's theorem applies to fully relativistic field theories. (3=0 Fig. has to be removed in order to allow a well defined perturbation procedure (i. r'. This wave function is in the quantum mechanics constructed about the classical configuration (j)c at this point or collective coordinate of the classical path a vector in the Hilbert space pointing in the longitudinal direction.e. We have here an example of the Goldstone theorem which says: If the solutions of the equation of motion do not possess a continuous symmetry of the Lagrangian. though not identical phenomenon.3 The spontaneously chosen phase seen from above. There the wave function with an associated vanishing eigenvalue is called a "zero mode". whereas ipi(x) is y ' " " " " " ^ . the existence of individual perturbation contributions). i. 22. (22. of course. Thus the Green's function is similar to that in electrodynamics. then a massless boson exists. 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. irrespective of the question whether the perturbation series as such converges or not.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. ^ = h eip / \ ^2 l . which.r')S(t . Like the Goldstone mode in the above Higgs model it leads to a divergence in the Green's function of the theory.7b). The appropriate Green's function G is the inhomogeneous solution of the the equation [8^ + m20]G(r. the component which climbs up radially outward along the profile of the potential implying a tp2(x) term with the potential V(</>) and so in the Lagrangian density £. We can see the problem here by looking at Eq.22.e. t') = d(r . The elimination of the . In our later discussion of theories with nonconstant classical field configurations we shall encounter a similar.
spinless field $(x.g. 7711 = — 1 ( 1 6 Minkowski manifold). C. instead we shall consider static and timevarying solutions in important models of one spatial dimension. MullerKirsten [291].d^} = ~d^d^V[^) = ^2^(^j V[$]. we shall need only at a later stage.544 CHAPTER 22. . Zimmerschied and H. The Lagrangian density is taken to be C[^. A. W. *S. t Potentials of this type have been discussed by M. We see therefore that the classical cnumber field configuration and the attempt to develop a perturbation series in its neighbourhood leads to intricate connections between symmetry properties.e. Khare [17]. zero mass configurations and constraints even before the question of quantization of the fluctuation field T){x) can be considered.8) where V[<&] is a selfinteraction or potential of the field (to be specified later). Later we wish to develop a perturbation series about a classical configuration of $. The principal aspects will always be very similar so that it really pays to study simple models in considerable detail. N.£) in one spatial dimension. however. J.* We consider the theory (later: quantum theory) of a real. and we choose the metric ?7oo = + 1 . To insure that the energy (see below) is positive definite we require V[$] > 0. We make one more assumption concerning V which. see e. so that we are interested to have a parameter *For a collection of many informative papers on solitons etc. i. In the following we will not be concerned with classical cnumber solutions of the EulerLagrange equations which are simply constants. equates to zero the coefficient of this wouldbedivergence. 22. Rebbi and G. Lohe [179]. Behera and A. 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.* Potentials of higher polynomial order in $ than $ 4 can also be considered. F.3 Soliton Theories in One Spatial Dimension We consider here the two basic soliton models known as <J>4 and sineGordon theories. Soliani [237].t In particular the sextic potential can be considered along parallel lines. (22.
10) n*]  m4 r 25 i m2 and the sineGordon theory with cosine potential by ±V[g$} = \v[g$.«{x) = V'(<p). For this purpose we assume that V[$] depends on a scaling parameter g such that Vm = V[t>. the socalled ^theory = ^V[g$. Since $(x. The field $(x. ^ ^ * + V'[*]=0. such that J > — — — = 0.11) (22. 9 9 The EulerLagrange equation is seen to be (with c = 1) D * + V'[*] = 0. (22. . (22. The other condition which we impose (and study in detail) is that of stability. but in such solutions which are subject to (ignoring the dimensional aspect) reasonable physical conditions. In a quantum theory the field $ is an operator defined on a space of states. In fact to begin with we restrict ourselves further and consider cnumber fields <> —• (f) which are static.l].§ Unstable classical configurations are also important and will be considered in later chapters. From Eq. 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). the reason for this condition being that we visualize the classical solution as representative of a lump of energy.12) we see that they satisfy the Newtonianlike equation <j.9) with quartic potential is defined by ^V\g$\ = ^V\g*. i. (22.22. at least in the present case. 1]. (22. — m$ 1 — cos Im 1 I ^ $ (22.t) has an explicit timedependence it is a Heisenbergpicture operator.e.e.12) where the prime denotes differentiation with respect to $ .g] = \ v m For instance. (22. Before we consider quantum aspects we study classical cnumber fields which satisfy the EulerLagrange equation.3 Soliton Theories in One Spatial Dimension 545 which serves as the expansion parameter of the series. i. The first such condition to be imposed is that the energy of the solution of interest must be finite.l}.13b) We are not interested in just any solution of this equation.13a) We write such fields <j>(x).
2\dx 2 ' 1 fd& + V[$].e.16) In the static limit we obtain therefore and write this as Em = H[<t>\ + E(4>) = fdx 1 2 \dx 2 + V(</>) (22. HI tanh m(x>k) 9 Fig.4 The wellknown soliton of $4theory. the classical solution will therefore have to be such that this integral is finite. d.17) is reminiscent of classical mechanics if we interpret x as time and V(<f>) as the potential energy of a particle at location d>. The energymomentum tensor T^ and the conservation law it satisfies are given by the equations dC T^ = n^C + Q^^d^. (22. (22. <9$ i.T = <I>. We observe here already that the integrand of the integral in Eq. i. (22. (22.546 CHAPTER 22. 22.14) where the overdot implies differentiation with respect to time t.7r) where IT is the momentum conjugate to $ defined by dC ? r = .e. dC • Too = r?oo£ + =r$ = C + vr$ = H. Classical Field Configurations The energy JE?[$] of the system is defined to be the spatial integral of the Hamiltonian density 7f($.c = <£z + . . n = & .T^ = 0.17) Since we have to integrate over all space.15) The zerozero component is equivalent to the Hamiltonian density H.
3 Soliton Theories in One Spatial Dimension 547 As we shall see below. < = — tan [e >  ° ] 9 2 rrm/g — Fig. i.1 . In fact. in the case of the first. around <j> = 0 the sineGordon potential behaves like the KleinGordon potential proportional to 2 4> . (22.18) Here XQ is an integration constant. one obtains by straightforward integration (but see also below) the soliton configuration m 4>c{x) = ± — t a n h 777(2. the $*theory.21) This sineGordon configuration is depicted in Fig.10) and (22. "The name sineGordon is derived from the analogous KleinGordon equation or theory. m ( x .e. See J. Rubinstein [240].777. — XQ). 22. it is not difficult to find classical cnumber solutions 4>c to Eq.5. (22.18) is depicted in Fig. (22.20) (22. In the sineGordon theory^ the classical or soliton configuration is (see below) found to be (x) = ±4— tan" 1 with energy 777 . The energy is correspondingly obtained — as will be shown — as 4 m & 9 The typical shape of the configuration given by Eq. Again XQ is an integration constant. 22. O±m(xxo) (22.22.4. . The sineGordon theory differs from the KleinGordon theory in that it possesses invariance under the shift rf> — <f> + 2n". 4m . > .x n ).13b) for the specific potentials (22.11). 22.5 The soliton of sineGordon theory.
Here the constant is zero for V(<p) and <j>' zero at x = ± o o .20) are obtained from Eq.18) and (22. Classical Field Configurations The solutions (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. and hence 4/71 9<t>{x.^^! m ^m <p(x0) m 4>(x) = ±—tanhm(x — XQ). 0(x) = ±4—tan" 1 j e * " 1 ^ " * 0 ) ! . Thus J<t>(xa) \/2V(6) JXOJxn U(x0) y/2V{</>) dx = x • XQ.e. With we obtain g(/>(a. <^>(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 .548 CHAPTER 22.10) we obtain f<t>{x) x — XQ i 9 ! 1 = ± / J<j>(x0) dtf> 1 . 4m In tan tan 5 l/e±m(*xo)tan^£^)U. ± 1[tanh. (22.9 2 4>{x) = i. Inserting the potential (22. In the case of the sineGordon theory defined by the potential (22.o) =m7r.* ( m' m 2 r)] ^2" .1 "»JtKxo) 9 1 .13b) which we can write d \W? = !"(•». . or \(<P'? = V(4>)+ const.
the discussion of Bogomol'nyi equations) in a simpler way.19). The kink solutions are also frequently described as "domain walls" in view of their analogy with the domain separating upward spins from downward spins as. the curves rising monotonically from negative to positive values are known as kinks. Jackiw [141].23) See for instance R. in the onedimensional Ising model. However. Then we have to demand that the functional derivative of the energy E(<p) be zero. A state of stability is therefore associated with minimized action and/or energy. instead of doing this now.4 Stability of Classical Configurations The concept of stability can be complex.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) . i.17). Naively stability means that a system does not deviate appreciably from a state of stability (or equilibrium) if it is allowed to fluctuate between neighbouring states. (22. . Suppose now that we demand stability in the sense of minimized energy. (22. 22. for instance. however. The expressions (22.22) and that its second variational derivative be positive semidefinite.e. (22. This is called more precisely "classical stability J The EulerLagrange equation is obtained by extremizing the action or Lagrangian. this does not require the action to be minimized. the others as antikinks. (22. The relation (22.22. SE((f>) 5cb(y) 0. we obtain these expressions later (cf.21) for the energies can also be obtained from Eq.4 Stability of Classical Configurations 549 Prom their monotonic behaviour the solutions <f)c in these cases derive their name as "kinks".
This is. The Newtonlike equation (22.25) Before we begin to study this equation in detail for specific potentials we can make a very important observation on very general grounds. in fact. its eigenvalues w\ > 0. with eigenvalue w2. dx (22. the function 4>c(x + a).e.28) .27) shows that the zero mode results from application of the generator of translations to the classical cnumber field configuration.26) Comparing this equation with Eq.e. = 0. and the EulerLagrange equation retains this property.13b) retains the invariance although the solution (f>c does not. i.a / 0 is not the same. A theorem on zero modes Let S[U] be the action of some field U defined on Minkowski space. Thus the vanishing of the first variational derivative yields again Eq.13b). Eq.550 CHAPTER 22. The eigenfunction 4>'c{x) is for this reason called a "zero mode".A. if we apply the generator of spatial translations d/dx to the classical equation. Classical Field Configurations Since y ^ ±00. we obtain 4'(x) = v"(4>c ^ or d2 dx2 4>'c(x) = 0. i. (22. the integrated contribution vanishes. dx2 + V"{4>) (22. Thus we have to investigate the eigenvalue spectrum of the Schrodingerlike equation d2 + v"{<t>c ipk(x) = wkipk(x) dx2 (22. (22. in fact ip0(x) const. Since it arises from the violation of translational invariance by 4>c. The original Lagrangian was.13b) with classical solution 4>c.24) For socalled "classical stability" the differential operator of this expression has to be positive semidefinite. we see that the latter has one eigenfunction. In fact. it is also called a "translational mode". of course. written down in Lorentzinvariant form. i. namely (p'c. (22. a very general phenomenon which can be established as follows. 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. Assume that S[U] is invariant under the transformations of some symmetry group G with elements g = exp(i\T) € G.e. (22. in particular the invariance under spatial translations. if 4>c{x) is a solution. (22.25). Proceeding to the second variation we obtain at 0 = <fic the relation 52E((j>) 5<j){x)5(t)(y) S(xy).
i. The right hand side can be interpreted as meaning: The application of the second functional derivative of the action (at U = Uc) to (TUC) is zero.13b) as ~d~4> > ' The first integral of this expression is \(tfc? = V(<l>c) + const. (22. it is an expression like (22. We see therefore that the occurrence of these zero modes is a very general phenomenon. and we have seen that Eqs. In the onedimensional case under discussion we can rewrite Eq.29) Suppose now the EulerLagrange equation possesses the classical cnumber solution Uc.e. We return to our considerations of stability. 0 = S'[UC] = S'[gUc}.4 Stability of Classical Configurations 551 where T is a generator and A a parameter.22.20) are such configurations in two particular models.30) Differentiating this with respect to A (applying —id/dX) and setting A = 0. Invariance of S[U] under transformations of this group implies S[U} = S[gU].] = 0 or x—>±oo lim V{(/)c) = 0 for const.e. (22. we obtain 0 = i^S'[exp(iXT)Uc]\x=o = S"[UC](TUC).33) x—>±oo . 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. and that S[UC] is finite (or equivalently the energy). As stated above.17) we obtain oo dx[V(4>c) + const. We can choose the potential V such that the constant is zero. (22. in fact. = 0.32) = V'(</>). (22. i. lim [V(<f>c) + const. Then the vanishing of the first functional variation implies that 5S = 0. Thus TUC is a zero mode. (22.18) and (22. / oo Thus for E(4>c) to be finite. (22.31) Inserting this into Eq. (22.].27). (22. we are interested in classical cnumber field configurations of finite energy. this integral has to be finite.
In the sineGordon theory > • defined by Eq. V{4>) = 0 at different values of </>). 22. .11) we must have correspondingly for x — ±oo: » 771 9 2vrn. (22. In the case of the 3>4theory we can define a conserved current k^ by (22. in view of the antisymmetry of eM^. Thus the classical energy or vacuum configuration is not unique.6 The minima of V in $4theory at x — ±oo. The charge Q is therefore given by dxk° = / dxe01di(bc = [ ^ ( s ) ] ^ = ±—. This number is defined like a charge in field theory by the spatial integral of the time component of a current. For a configuration to have finite energy.35) and it is seen that in this case q can be ± 1 or 0. 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. Solutions which approach > > the same minimum for +oo and —oo are the constant solutions <f>c = ±m/g in the $ 4 theory. oo J—oo y The topological quantum number q is defined by 0. . ± l .e.10) we must therefore have that for x — ±oo: 4>c — ±m/g as indicated in Fig. Classical Field Configurations V Fig. (22. 22.552 CHAPTER 22. ± 2 .34) jfc" = e^dv(l since d^k^ = 0. . . .6. n = 0 . (22. » In the <fr4theory defined by Eq. the potentials have degenerate minima (i.c o and another for x — oo. the classical field (bc must approach one minimum for x — .1).
continuous values. Tfl \ (22. one may ask how topology comes into the picture. x—>±oo so that under such deformations the configuration remains in the same topological sector defined by the boundary condition. Then for x — ±oo: > > > V(\4>c) (2 ^ 0 ) ? 4z( 1 .7T (22. A property is. Such a smooth deformation (in the sense that exp(i%((2))<^c depends smoothly on 6. First. and exp(ix(0))<pc reduces to 4>C when x(#) = 0.38) 4 V(\<t>c) >m ~2[l . 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). if it remains unchanged under continuous deformations of the (field) configuration while preserving finiteness of the energy. and we call this "topological stability. Thus we expect the constant solutions (which are characterized by topological number zero) to be nontopological in some sense. The difference Tfl 4>c(oo) — 4>c(—oo) = 27T—An . in general. n = 0 . Thus we consider the family {X(pc(x)} of (field) configurations where A is some parameter which assumes real. termed topological. Thus the condition of finiteness of energy does not permit A</>c deformations in this case other than those with \ = eix(e) for X (0) = O.37) for the onedimensional case here.4 Stability of Classical Configurations 553 Since we call q a topological quantum number. . . We can see this as follows.11) are given by ( so that for x — ±oo: > — J. For such configurations we have the energy oo 2 2 dx ^ A ( ^ ) + y ( A ^ / (22.22. zero is not a meaningful topological number.36) oo Consider the $ theory with 4>c — ±m/g for x — ±oo. In the case of the sineGordon theory the minima of the potential (22. . ± l ± 2 .cos(27rnA)]. . 2g Thus even without varying E with respect to A we can see that E(X(f)c) is finite only for A = ± 1 .A2)2 + 0 except for A2 = 1.
Thus with one space dimension higher there. the factor A = exp[ix(#)] varies over a complete circle). pp. From this we construct the inequality [</>'{x) = Vm^)}2 F Since (cf. The inequality is saturated. W. the energy is minimized. . 22. MiillerKirsten [215]. We observe (a) the topological current k^ of Eq.^ = N. 22.e.554 CHAPTER 22.17)) f 00 > 0. (c) the time component k° is a spatial divergence. H.31) to (22.** (b) the time component k° depends only on <p but not on momenta (again in contrast to the case of Noether currents). Eq. its integral is nonzero only on account of nontrivial boundary conditions. (d) the topological quantum number iV arises in a completely classical context.13b) or ^ ' ( * ) ] 2 = V(</>). Classical Field Configurations involves the difference An = rioo — n.428.34) is conserved independently of the equations of motion (other than Noether currents whose conservation follows from the equations of motion).41) dxy/2V{$)4>'{x). 425 .42) Thus the energy has a lower bound given by the right hand side of this inequality. (22. (22. (For a broader view we consider briefly in Sec. (22.43) **See e.g.5 Bogomol'nyi Equations and Bounds We have seen previously (cf.39) E(cp) = J — 0 0 we can write / dx oo ~1 2(0'(x)) 2 + y(^) (22. (22.9 the foregoing arguments in a higher dimensional context. i. Eqs.33)) that the classical configuration cf)c is the solution of the nonlinear second order differential equation (22. 00 (22.40) (22. J. where n±<XJ are the integers n corresponding to the minima of the potential at x = ±00. 00 (22. if oo / dxy/2V{<l>)<f>'{x).
Eq.n > 1. 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. the set of configurations (fic which interpolate between 0(—oo) = 0 and <f>(oo) = 2nnm/g.44) This first order equation which saturates the inequality (22. We discuss this in the following Example. In the sineGordon theory we can consider.40) is called the "Bogomol'nyi equation" since it was first introduced by Bogomol'nyi. 555 (22.39). (22. m 9 V V / and with x = g4>/m this becomes P IA\ CTnin(0) = = /^ J /^ V ™3 f2W A I A • 2X n—7T I ax\/2ll — cos x) = n—7T / ax* Asm — Z 3Jo 2ir r 9 Jo V 2 2m 2 cos — n — n^[2(l)+2] 2 ~ m3 8m 3 = n—rrin agreement with Eq. 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. n E . for instance.^ Clearly it solves the second order equation or (22. y/V(ft) > 0) oo =2nnm/g r<p=2imm/ g / dxy/2V{<f>)</>'(x) oo /•rj)=2Trm/g = / J<j>=0 d</>y/2V(<f>) = = n Jd>=0 rt=Mmlg n / J<p=o d<f>y/2V(<l>) 2 / ^\ — # W 2 1cos— .19).40)) this occurs when 4>'(x) T V2V(4>) = 0. Here (V(c/)) > 0. Bogomol'nyi [27].5 Bogomol'nyi Equations and Bounds We observe that (cf. B. . (22.22.21). (22.
ij\2 Thus the The potential energy due to gravitation is (maximal for <j> = 7r) mgr(l — COS <f>{Xji. The kinetic energy T of the bob at position xn at time t is (recall " m r 2 r / 2 " ) 1 —mr 2(d<j>{xn. 22. so that oo > u.7 The pendulum analogy.45) We can therefore write the Lagrangian L=TV• r dx fir P 1 2 .t) measured e. Solution: Consider a string along the xaxis as described. Assume that the pendulums can move (i. with mass m and connected with short strings to neighbouring bobs. rotate) only in the (y. Classical Field Configurations Example 22.7. xn+\ — xn = a) 00 a —> dx.g. Chapter XI. z)plane and only such that their elastic strings remain taut the position of any one of the pendulums is determined completely by an angle <p(xn. . t)).e.1: Classical analogy to sine—Gordon theory Consider a string along the xaxis. We can now write down an expression for the energy of such a system of (say) n pendulums. Assuming that the pendulums can move (i.cos</>(x„. Goldstein [114]. V dt „ 1 (d4>_ dx • i^gr(l — cos(j>) (22.t)\2 +mgr{\ . * / dx OO + 2^ V dt 1 K (d<t>\ 2 \7T I + ^9^(1cos 4>) (22.t) <j>(xn. Construct an expression for the energy of such a classical system and its Lagrangian.g. z)plane and only such that their elastic strings remain taut. Fig. Initially the pendulums are suspended freely in the gravitational field.i)) The continuum limit is obtained by t a k i n g ^ (since xn = na.t) 4>{xn. all being identical. rotate) only in the (y.46) **See e. At equidistant points xn = na along this string we attach strings with pendulum bobs. } n= — oo roc —> / L J .t) 2 V dt The energy of interaction with neighbouring bobs is (recall "fcx 2 /2") k[(f>(xn+i. total energy of the system is with the number of pendulums allowed to become infinite *= £ i ''( "' ) .e. H.556 CHAPTER 22. 22. from the yaxis as shown in Fig. ka —> K.2 V dt + k[(f>(xn+i.
(22. ' Equating coefficients of exp(—iw k t). Hence we set ${x. t) is the fluctuation (field).7 linking the pendulum bobs).49) exp(iwkt)ipk(x). We are interested in studying (field) configurations $ in the neighbourhood of the classical solution cpc. wk would be imaginary and so the factor exp(—iwkt) in rj(x.47) into (22.22. is therefore also called "winding numbed. It is instructive to obtain this equation by yet another method.e.t). 22. we see that ipk(x) has to obey the equation 92 dx2 +V"{(t>c{x)) ^ fc (x) = w^k{x). (22.t): {w With the ansatz r)(x.t) = ° (22'48) (22.25) which we obtained as the condition of classical stability for eigenvalues w\ > 0. The equation is called stability equation or equation of small fluctuations. (22.47) where n(x.6 The Small Fluctuation Equation We now return to the Schrodingerlike equation (22. (22. This number.49) would imply an exponential growth in the future t > 0 or in the past t < 0.25) obtained earlier.13a) and (22.t) = <l>c(x) + r}(x.48) becomes ( d2 \ X I [Q^2 + wl)Tpk(x) e*v(iwkt) k ^ X = k ^2v"{4>c{x))iljk(x)exp(iwkt). and so would invalidate the procedure which assumes that rj(x. which represents an expansion in terms of normal modes. 22. Eq. If some w\ were negative. each finite energy configuration beginning and ending with a classical vacuum corresponds to an integral number of rotations about the iaxis (i. t) of Eq. (22. t) = ^ k ~ iL2)71^ + V "^c{x))'q[x.12) we obtain d2 d2 \ Since (j)c{x) obeys Eqs. of the continuous curve in Fig. We can understand what stability means in the present case.t) is a small . (22.50) This equation is identical with Eq.6 The Small Fluctuation Equation 557 Thus if the classical vacuum corresponds to the case when all pendulums are pointing downwards.13b) we have for small fluctuations rj(x. Inserting (22. the topological quantum number.
we obtain V'^c We have m cos ± 4tan" 3±m(:r—xo) (22. we shall see later that the fluctuations n(x. i.20). In fact.50) has the general form of a timeindependent onedimensional Schrodinger equation. and need 9 = tan" 1 j e±m^Xo) 1. (22. is a vanishing eigenvalue w\ = 0 acceptable for stability? Apparently for w\ = 0 the energy is not minimized in the direction of its associated eigenfunction. approaches a constant value which is nonzero in both the <54theory and the sineGordon theory.e.e.18).52) tan 6 := e±m^xo). For x — ±oo the quan> tity acting as "potential". we obtain V"(<t>c) .x0) = —2m + 6m 2 (l — sech 2 m(x — XQ)) 6m Am' cosh m{x — XQ) = (22. Inserting (j)c of (22. the zero mode. The equation of small fluctuations (22.1. i.A \ Inserting <f)c of Eq. This means that the stability is not universal but only local in a certain sense. Thus for stability w\ must not be negative. V"(<b) = 2m2 + 6g2(f>2. . t) have to be orthogonal to the zero mode (which is thereby circumvented) so that the Green's function of the expansion procedure exists. cos46» = 2cos2(2(9) . If that is necessary. V"((/)c(x)).2 m 2 + 6m 2 tanh 2 m(x . In fact.558 CHAPTER 22. Classical Field Configurations fluctuation. 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£^)*.51) (b) SineGordon theory: Here 1 — cos m m 9 V"(<f>) = m 2 c o s f ^ .
Consider the equation fd2 dz2 . ip+.* For such a potential the spectrum can be b o t h discrete and continuous.6 The Small Fluctuation Using the formula cos 20 we obtain cos 40 1 . (22. which satisfy the boundary conditions V>(o) = o. i.2 (22. The eigenfunctions of the discrete case vanish at ± o o and are squareintegrable.22. (22.ijj^ respectively. We can therefore construct solutions of definite parity.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.tan20 1 + tan 2 # Equation 559 1 — e x p { ± 2 m ( x — XQ)} I + exp{±2m(x — XQ)} sinh[=Fm(a. — XQ)] cosh[±m(x — XQ)]: 1 . Those of the other case are complex and periodic at infinity. A  *(* + !) cosh 2 z tp(z) = 0.55) The differential operator is even in z.53) = m(x — (22.e.56) *We shall see later that the PoschlTeller equation is a limiting case of the Lame equation. even and odd solutions. . We can investigate the spectrum as follows.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 . ^+(0) d 0. and as a consequence a limiting form of the small fluctuation equation of all basic potentials. we see t h a t the equation represents a Schrodinger equation with a potential which vanishes exponentially at ±oo.
2n < Z. + l2) = 0.Z ) + ^(A + Z2) = 0.60) terminate after a finite number of terms.58) = (1 + sinh2 z)~l/2x(z). i. i. and in the odd case n+\)U+\l)+\(\ i.61) Discrete eigenvalues are obtained when the series (22.Z ) + (A + Z2) "n+1 (n + l)(n + §) &n (22.e.( 2 n .57) Then the equation for x{z) 1S (22. A * An = .Z)2.( 2 n + l . (22. Since for £ — oo: * the function ^>+ is normalizable provided Z > n. when in the even case n ( n . .e.59) Setting (22. i. 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 + ± . A ^ An = .Z ) 2 . (22.l2 = .e.60) n=0 n=0 oo we find the recursion formulas x+(o = E ^ " > n(nl) = x(o+=l £ £&»£n.560 CHAPTER 22.e.4 n ( n .
11).is normalizable provided 2(n + . i.63) d + K(2ik  2LO)—V ' dijj' + 1(1 + l)V = 0.u2)—^V 2 dio and ip(z) = elkzV{to).e.N<1.54) by setting X = k2 Then 1 . 2 . We can summarize the results in the statement that the discrete eigenvalues are A = (ZAT) 2 .10) and (22. (22. (22.62) we see that the equation possesses discrete eigenvalues w —2 . (22.63) we see that the continuum starts at A= 2 2^ = 0.j < I . (22.2Z = .1: w2 = 0. d w = tanhz. We can now derive the spectra of the stability equation for the potentials (22. m N<2.Ny. (a) ^theory: the equation Inserting (22. From (22.51) into (22. for N = 0.22. m i.. the continuum is w > Am2.50) we obtain with z = m(x — xo) tp(z) = 0. 3m 2 .\=Kl2 From (22..e.65a) _dz2 cosh 2 z l=2.6 The Small Fluctuation Equation 561 and the function ip.2.64) This is the equation of Jacobi polynomials V\a' (u/) with a = —ik of degree I in tanhz.62) where even (odd) iV are associated with even (odd) eigenfunctions. w2 = Am2.( 2 .1. We can see the continuous spectrum of Eq.e.. (22.. °2 i. N = 0.
.. It is now particularly interesting to look at their wave functions. . . for N = 0: to2 = 0 and the continuum starts at w2 = m2. of course. so that its derivative is nowhere zero (except at ±oo).65b) dz2 cosh^J.e. .8 The zero mode as typical ground state.=1A=^_. This property can. In the $ 4 theory we saw that (cf. Classical Field Configurations (b) SineGordon theory: Inserting (22. The zero mode is depicted in Fig.66) We see that this is a nonvanishing function for any finite value of x.2 The discrete eigenvalues are now given by w ^l m^ = (lN)2.8 and we see it has the shape of a typical ground state wave function. d . Thus it has no node and so represents the eigenfunction of the lowest eigenvalue. N<1. (22. m m ipo{x) oc —<fic{x) = ——tanhm(x — XQ) = g cosh m(x — XQ) dx g dx (22.53) into (22.27)) . Eq.562 CHAPTER 22. %(x) Fig. We know that 4>c is a monotonic function. also be deduced from d(j)c/dx. Thus either spectrum contains the expected eigenvalue zero. Thus 4>'c(x) has no node. 22.50) we obtain with z = m(x — XQ) the equation (22. m d . i. and — as expected — this eigenfunction is even.. the zero modes. 22. .
Finally we consider the normalization of the zero modes.oc (22. i.t a n h x . d .z 0 )]oo = 8 . .68) d{mx) m(xx0) 4 m3 3 ^ g Ml J cosh / i.3 .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.o) ±4 m g l __ e ±2m(xx 0 ) 1 cosh m(a. ipo{x) oc —<bc(x) ax oc 4 m d _i Q±rn(xXQ) —tan e g dx m D±m(za. .22. (22.e. (22. without quantum corrections).e. — XQ) OC Hence the conclusion is similar to that in the previous case. Imposing the condition F we have in the <I>4theory: /•oo dx[^o(x)}2 = 1.= E(<f>c). (22. 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 . . .67) J —( Mx)= and hence m w0i^ 1 2 4 m ±—tanhm(x — XQ).70) .6 The Small Fluctuation Equation 563 In the sineGordon theory the zero mode is proportional to . 1 o rrr tanh x — .
They may help. P.g. I.T(4>).g. I. i. (2.41) for the energy of the classical configuration but now with respect to a ddimensional position space.7 Existence of FiniteEnergy Solutions So far we have been considering the case of one space and one time dimensions.564 CHAPTER 22. f G . P(</>) = f ddxV{(f)) > 0. Kastrup [145]. we consider E(<P) which we also write as (22. Under this transformation (the verification is given below) £(</>) ^ E(</>a) = T(cf>a) + P(</>a) = a^T^) since for instance + a" d P(0). One can have different types of scale transformations — see e.e. We now wish to inquire whether static finiteenergy configurations could exist in more than one spatial dimension. *See e. it is important to understand their physical implications.2d. This difference may not always be appreciated later in our very analogous treatment of (topological) instantons and (nontopological) bounces. (22.71) We now consider the scale transformation (with 'a' some number € R) 4>{x) . H. This question has important consequences. Affleck [5] who uses in his Eq.g. to underline the fundamentally different nature of topological and nontopological finite energy configurations. J As a first example we consider the expression (22. ddx {V4>f + v{4>) (22.73) n<(>a) = j ' d^V^)2 = a2~d / dd(ax = J' ddXl{V4>(ax)f d 4>{ax) 5ax . ^For basic aspects of scale transformations see e. Classical Field Configurations 22.72) with T{$) = I' ddx]{V(j))2 > 0.4) <j>(x) —• a<f>(ax). Olive [113]. for if such configurations exist in a realistic theory with three spatial dimensions. H. Derrick [68]. *The considerations of this section extend in part beyond quantum mechanics and into field theory. . Goddard and D.* The general scaling argument used for this purpose was first introduced by Derrick t and is therefore referred to as Derrick's theorem. The arguments of Derrick were later rephrased by several authors whose line of reasoning we shall be using here.> 4>a{x) = 4>(ax). A. however.
a=l (22. we see that the stationarization of E with respect to a is possible only for d = 1 as in the examples we have been considering so far.V(<P) (22.73) we see that this implies (2 . i. to be finite the field <P ~^ 4>oo £ Mo for x — oo. (22. (22.78) For the energy of a static configuration.e.xd). d = 1 minimizes the energy. It is also a virial argument in view of the virial theorem relation T{<j>) = P(<f>).e. In fact. Thus this scaling argument excludes the existence of static finiteenergy configurations of the given theory in more than one spatial dimension. (22.75) Since both T and P are > 0 and d is integral.7 Existence of Finite Energy Solutions 565 We saw that the static configuration with a = 1.22. Next we consider more complicated cases. since d2E(<pa) da2 2(2d)T(<f>). (22..79) In order to emphasize that (f> — (p^ in any direction.e. 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}...71)..\ eMo(22.x2.77) with V(4>) > 0 and x G Rd and the set of minimum configurations Mo := {4>\V{4>) = 0}.74) From (22. i. Consider first the Lagrangian density £[&M = \{d^)(d^) . we write • = lim J r . Can the energy also be minimized for more than one spatial dimension? To see this we stationarize E(<f>a) with respect to a. x = (x1. the second derivative is positive) only for d = 1.d)T(<f>) = dP{4>).76) we see that E(<f>) can be minimized (i.80) . we set 0 dE(<pa) da J a = i (22.
. i. therefore. Classical Field Configurations In the sineGordon theory (d = 1) the equivalent statement is (22. i.81) and (since d = 1) S = ± 1 . is a connected region. 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 . S1.566 CHAPTER 22. x = 00. ( with — = lim <pc(x) 9J *>±oo TYl \ (22. that we have a set of disconnected classical vacua (</>oo><^oo. These observations are summarized in Table 22.e. one set can be mapped into the other. i.1. as is the case for the constant solutions.1 (A) Topologically nontrivial mapping x = +00 : S = +1 x = 00 : 5 = .1 manifold Sd~1\d=i <• manifold Mo <—> 0 O = 27rn^ Q <—> 0_oo = 27rn'^ kink limits (nn'/ 0) const. Thus in this case S^1 is a disconnected set consisting of two points. in fact. of • course. We see. are the same.38).and sineGordon theories we assumed $ to be real. —00. the Lagrangian of these theories therefore does not possess the U(l) symmetry of the complex Higgs model but instead invariance under replacements $ — — $ and $ —• $ + 2TT > > respectively). — 1). 0oo 7^ <>< > in the case of the kink solutions) and /x associated with this a set of disconnected points in space (00. —1. (ptx = <f>oo. If we go to two spatial dimensions (d = 2). —00 or equivalently S = + 1 . the easiest way to visualize points at infinity is to take a circle Sl of radius r and let r — 00.e. Then each point at infinity is characterized by a different value of the polar angle 9.e. (n — n = 0) (B) Topologically trivial mapping x = +00 : 5 = +1 <—> <> o = 27rn^ /o a. Thus in order to obtain a topologically nontrivial case we must have a different classical vacuum in association with each point at infinity. = —00 : S = — 1 <—> ^oo = 27rn^ The topologically trivial case is given when the classical vacua associated with S1 = + 1 . and. Table 22.
map into. (22. we would have only two points at infinity.e. * n reach any other <^(oo) by simple phase shifts.83) . we have w e ca {Mx)}> and 1 3 1. This can be achieved e. Thus. that if the problem is to be topologically nontrivial. to have topologically distinct classical configurations. given any (00) = «Koo)e* = <?[<p(xMx)a2]2 567 say with 9 — 9$. by allowing <fi to have three components in some "internal" (isospin) space. 00. phase transformations 6Q ^> 9Q + 59.e. 9) would depend only on r = x and hence the problem would be effectively onedimensional. Thus.1 . if <p is real. —00 or S = 1 . if we consider a theory in three spatial dimensions and so the unit sphere S2 = {x x{ + xl + xl = 1}. if cj> were real and position space twodimensional. the one point S = + 1 .e. Then the entire set of configurations (f)^ = \4>oo\ exp(i#) would correspond to. i. This will be "nontrivial" in the above sense only if 0oo is independent of 9.3 (^real) v{4>) = / (22. i. assuming invariance of C under the transformations of the group SO(3) in internal space. for instance. the field must map position space into field space at infinity. Similarly. On the other hand.2. i.22. We see.• M0. VMx)) with (0(oo) = linvxx) (p(r. i. .e. If position space were onedimensional.e.g. the manifold of classical vacua would have to possess a similar geometry if the theory is to be topologically nontrivial. therefore.7 Existence of Finite Energy Solutions Suppose.e. i. $(r) : Sl . i.82) Now Mo := \ 4>(oc) ^24>2 = a2\. cj)(r. 9)) Mo = {</>(oo)</>*(oo)0(oo) = a2}.
. $ — > > $ exp{i Xlo=i Pa. L. i. i.2: Derrick's theorem applied to gauged theories Wellknown and important theories involving a gauge field (AM if abelian and A". a = 1. 11 H. Georgi and S. we cover <S^_1 n times. Glashow [108]. circle.„ where G r = 9 M £ .eeabcA*Avc and (V^U = d»<t>a eeabcA^oWrite the energy E(<j>) of such models (the functionals F and V containing no derivatives) £(</>. a theory being described as 'gauged' if a gauge field like the electromagnetic field is involved.Av  dvAf i where 0 is a complex scalar field and the spatial dimensions are 2 or 3.a M J . so that details of these models are irrelevant.e.e. it is effectively a surface (and so a continuous set) S2. .e. invariance under replacements $ — —$ in $ 4 theory.e. the winding number is n. $ — $ + 27 > • T in sineGordon theory. + i ( D " f t W ) a . through each point at infinity). (b) The GeorgiGlashow model I! with nonabelian gauge field in 3 spatial dimensions with Lagrangian density (V as under (a)) £[AM. $ — $exp(ia) in a U(l) symmetric theory. In the following Example the use of Derrick's theorem is demonstrated by application to some gauged theories. going once around S%~1 (i. The theories are: (a) The Nielsen—Olesen (vortex) model (also called Abelian Higgs model) " with Lagrangian density V(4>) = Ug2 2A 9 2 FM„ = du. . A) = T(4>. If.Ta} in an 50(3) invariant theory and so on — determines the transformation from one point at infinity to another (on a line. A) + T{A) + P{<j>).568 CHAPTER 22. if SO (3) nonabelian) are two models defined by the following Lagrangians which we cite here solely for the purpose of being explicit. Nielsen and P. % . B. the subsequent discussion requires only the general form of the Lagrangians. Example 22.3. Summarizing we see that </>(x) is a mapping from S^T1 to <S^1.</>] = ~G^Gai. depending on d) and correspondingly (with n complete windings) the transformation of one classical vacuum into another. sphere. Classical Field Configurations The manifold A4Q is a sphere in a 3dimensional Euclidean space.V{4>a]. The symmetry of the potential (and Lagrangian) — i.2. Olesen [220].. In order to preserve lim y ^ i a\ such rotations must tend to the identity at r = oo. Any 4> £ MQ can be reached from any other cf> e Mo by application of an element of SO(3). the appropriate rotation.
>• A„ x ( x ) = AAM(Ax). ^ A ) = A 2 .e.d)T(4>.22. > Since T>IJ. Solution: Each contribution in (22. i. Mx') = E with a scale transformation x — x' = Ax.87b) We consider explicitly the example of <j> — </>\(x) = <p(\x) for > T ( V(4>):= Jddx(d. P(tf>A) = A " d P ( 0 ) If £ is to allow a static finiteenergy solution. we have {^/»)A^ (22.89) P{4>) are positive semidefinite and not all zero.i T(AX) = X4~dT(A). (22.. (22. T(A) = f ddxV(4>) > 0. i.<^[x) transforms like a vector. A A ) = A2dT(^.d T ( A ) + AdP(</>). A) + (4 . 4>Y We have T«(4>A) = J ddx(d„4>x)2 = J dd . (i)^ 2d T Considering the contributions of (22. (2 . A).e.84) is positive semidefinite and so must be finite by itself.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'>.d T ( ^ .Ax) dX Since T((f>. A.84) in this way one finds that the quantities involved transform T ( ^ . l j ddxGijGai] > 0. and so B ( ^ A . (22. We now consider scale transformations with A £ R. negative. = 0. <0 >• <£A(X) = <KAx). J 4 ) + A 4 . some contribution must be .87a) C?M„(x) — \2G^{\x).d /" d d x ' a</)(x') <9x' d</>(Ax) 3x M A 2 /" ^ ( A z ) d<p(Xx) A / A^ d(Ax M ). (22. (22. Thus we set dE(<fix.A) P{4>) 569 j / i F ( « ( P ^ ) t ( P ^ ) „ > 0.86) (22.84) and investigate the existence of finite energy classical field configurations. it must be stationary with respect to arbitrary field variations and so with respect to the above scale transformations also.A).dP(4>) = 0.d)T(A) .T{A).7 Existence of Finite Energy Solutions where T{<j>.
ft V . a = a(T) = a T —T c . W.90) and we would have to add to Eq. Thus such a contribution could counterbalance T{4>. (e) For d > 5: In this case it is not possible to stabilize Eq. (22.92) The expression ^ ( x ) ! is known as the "order parameter. This case is. We then have a situation of the potential like that described in the context **T.570 CHAPTER 22.t) such that  ^  2 describes the density of Cooper pairs. One can thus define a scalar field or wave function ty(x. . From this development one knows that in the superconducting state of the metal the electrons combine to form pairs called "Cooper pairs'" which then have bosonic properties.89). Ginzburg and L. 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. a > 0.g. Skyrme [253]. is temperature dependent and changes sign at the critical temperature Tc. There are other posssibilities if we permit higher powers of derivatives as in Skyrmion models. (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). H. We first demonstrate — as a matter of interest — that the NielsenOlesen theory for static equilibrium (i. H. Tchrakian and H. (22.89). Equilibrium states of the superconductor are assumed to be described by the timeindependent static wave functions ip(x). if we allow a term T^ 4 ' {<f>) with fourth powers of derivatives. T^{<t>x) = \4dT(</>). 22.8 Ginzburg—Landau Vortices We have already referred to the NielsenOlesen theory. exemplified by soliton and sine—Gordon theories.** e. (c) For d = 3: All three terms must be present as in the GeorgiGlashow model. A) or P{4>) for d = 3.89) the contribution (4 — d)T(4>).e. 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. J. We can therefore use our understanding of the BCS theory in interpreting the GinzburgLandau theory. Landau [111]. timeindependent) field configurations is equivalent to the GinzburgLandau theory of superconductivity. In the following we consider a related theory. D. (b) For d = 2: A theory with all terms would be a candidate — in fact the NielsenOlesen theory is an example. (22. tt The — now established — microscopic BCS theory of superconductivity was formulated some eight or so years after the macroscopic GinzburgLandau theory. (22. L. An explicit and solvable Skyrmion model in 2 + 1 dimensions is considered in D. MiillerKirsten [268]. R. of course.
i.(tp) = 0. Variation as before. Clearly the length (see Eq. Returning to the three dimensional case and switching on a magnetic field (applied from outside of the superconducting material. but for T < Tc it has a double well shape. (22.94) the solution is (in the domain 0 < x < 00) i>{x) = ^/^tanh(^x\ . (22. from x < 0) means — as is familiar from electrodynamics — replacing V by V . is a measure of the distance from the surface of the superconductor to where . Then the static energy density becomes (cf. Sec. Example 22.22.93) J' (22.2): For T > Tc the overall potential has a single well (minimum). 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) = 0.95) This expression can be shown to be equivalent to the integrand in the NielsenOlesen theory.^ 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. 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. assuming there is no superconducting material in this domain).2) £(^.A) = iB 2 + i iXQ + 7 + ^ H 2 + /^4. i. setting S£.e.94) to check the dimension) 1 = I—a.8 GinzburgLandau Vortices 571 of the complex Higgs model (cf. ^(00) = a (22. 22. The parameter 7 serves to put the minima of the double well at V = 0.e.
p. MiillerKirsten [215].98) where M is the magnetization resulting from atomic currents. i. the order parameter \ip\ attains (approximately) its asymptotic value. Classical Field Configurations superconductor (x>0) Fig. The magnetic field on the other hand.g. i. In the Maxwell equation See e.e. 22. The density of the Cooper or super current is given by the typical expression of a current.M. (22. We can see this as follows. V x M = j s . exhibits a completely different behaviour — in fact.e.572 no superconductor CHAPTER 22. 22. i. . as indicated in Fig. h [^DV'V'DV].9 Behaviour of ip and B in the superconductor.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 . 2mi Setting < / > \P\e vp B = ViA.96) ipH. (22. yJ—a/[3. J. inside the superconductor it falls off exponentially away from the surface (this is the MeissnerOchsenfeld effect generally described as the expulsion of lines of force from the superconductor below the critical temperature T c ). (22. W.e.9. H. 450.
(22. Also dD/dt = 0 in a static situation like the one here. i. Approximating p by its equilibrium value. m Here A^ is known as the (London) penetration depth (or length) (in practice this is of the order of 1 0 . There are two types of superconductors.100) we obtain 2 m ( 0\ The ratio XL/XQ is called the GinzburgLandau parameter.div grad" and V • B = 0.99) Inserting (22. This is zero in the present case. K i ™*.M i. 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.5 to 10~6 cm).L B = O. Superconductors of type II are characterized by the fact that when the magnetic field is again increased and beyond a first critical value. and V x (V x B) = / J 0 V x j s . this becomes A B = no V x j 5 . those of the socalled type I and those of type II.97) this becomes (since curl grad = 0) AB = Mo—Vx m A = /x 0 —B. Hence 0 = V x H = V x ( — B . In the onedimensional case we have B{x) = B(0)ex/XL. (22.8 GinzburgLandau Vortices 573 the quantity j is the density of a current applied from outside.e. These vortices therefore carry magnetic flux which is necessarily quantized . 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. Using the relation "curl curl = grad div .. V x B = [i0V x M = /lois.22. m xi A B .e.
x—>oo Thus at spatial infinity </» has the value of a zero of the potential. Classical Field Configurations (see below). 'See the remark in H.9 Introduction to Homotopy Classes In this section we consider topological properties of classical finite energy configurations and introduce the concept of homotopy classes.2 we saw that finiteness of P(V) puts no restriction on the phase of (f> so that r e (22. We can think of (e. . J. 6')r_>00 represents a mapping from the circle at spatial infinity to the circle defined by x(#) (or. *A.102) In the onedimensional soliton case of Eqs. The existence of these vortices has been confirmed experimentally. (22. of two points on a line.g) twodimensional space as being bounded by a circle at r = oo. A. by the element exp[i%(0)] of the internal group A(s) = e * W : SI+SI. i. the space consists of two diametrically opposite points on the circle. G.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. (22.e. Schaposnik [69] after their equation (3. A.36). in fact a periodic function of the polar angle with period 27r. Thus the asymptotic field </>(r. 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). we must have that lim V ( H ) = 0. See also P. In the two models of Example 22. Abrikosov [2]. de Vega and F.101) (p(r. i. 22. In the NielsenOlesen model defined in Example 22. Finiteness of E requires \<t>\ = \<fioo\ on this circle..37).9). (22. in other words.9) ^e^ '>\<poo\.e. De Gennes [67]. 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. where exp(ix(0)) is an arbitrary phase factor.574 CHAPTER 22. In all the models we considered so far we observed that the energy E contains a contribution P(V) = j ' afxVM).
Fig.1). i.X ( 0 = O)]. (22. Misner [96].e.1. W.10 States of winding number 0.2 imply that such a map from a circle to a circle is associated with an integer which we called the winding number n. defines the orientation of this line element.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".22. 22. topological group with the usual multiplication as group operation.10. 0 < 0 < 2TT. We can illustrate the simple lowdimensional example under discussion by reference to a strip with Mobius structure. 2n] remains continuous if the function assumes the same value at the two endpoints. Naturally an integer cannot change continuously. 22.* 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. Sl := {zeC\\z\ = 1}. Then a continuous function of (j> on [0. We may define this as « = ^[X(0 = 2 T T ) . The two theories we considered above in Example 22.103) Later we demonstrate that this winding number n remains unchanged under continuous deformations (those of Eq.102) is called the exponential map of S1 onto S1. The map satisfying X(x + y) = \{x)\{y) is said to be a homomorphism.e. (22.118)). The map A : S1 > S1 defined on S1 C R2 by (22. Thus smooth deformations *See D. Finkelstein and C. X is called the domain of / and y its range. States of the strip corresponding to winding numbers 0 and 1 are illustrated in Fig. . commutative). Suppose <j>. abelian (i. The 1sphere is the unit circle in the space of all complex numbers. In mathematical language S1 is therefore a compact.
. The winding numbers n are such topological invariants. connectivity. (22. We observe that since time evolution is continuous. it retains this twist or hole under the continuous distortion) and is described as a homeomorphism.e.576 CHAPTER 22. i. we required that lim r—>oo w o. no hole. For that reason n is called a "topological invariant?. The different kinds of connectivity of these topological spaces (e. elements in the same class have the same topological characteristic.) remain the same under homeomorphic transformations and are therefore called topological invariants. if it has one twist as above or a hole. finite. one hole. The gauge field required in the asymptotic limit (22.e.g. Classical Field Configurations of the fields which preserve the fmiteness of the energy must preserve the winding number n. i. The continuous deformation or distortion of a set of points does not alter the topological structure of the set (e. i. it cannot be deformed continuously into the socalled "classical vacuum" (the latter is a constant configuration in our soliton example).. r^oo 1 dx(9) n r o0 (22. we can write 1 d h J ee rde w At h r^oo .105) is described as "pure gauge" since it is determined entirely by the phase x(#). the winding number must be a constant of the motion.104) so that d2x hJ Since A has direction e#.105) the corrections falling off faster than 1/r.. Thus two homeomorphic spaces have equivalent topological structures. Homeomorphism subdivides the possible spaces into disjoint equivalence classes (to be explained below). In our consideration of the NielsenOlesen theory we also required the gauge field A^ to behave such that the energy is finite. since otherwise the energy integral would behave like f d x f dr dr 2 J r J r lnr.g.104) to be finite we must demand that 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. In other .
we have B = V x A = curlgradx = 0. if A = V x . (22.e. cannot be pure gauge everywhere. i. However. Suppose two such configurations are given by O) W(9) = ei&)V\ n fixed.<b rd9—4^ = n.106) J g j r d d g Thus the magnetic flux is quantized. into S i . We can calculate the magnetic flux $ through the region with $ = * rd9A0 = . An example is the mapping H(t.9):=t(f>£\9) + (lt)(f)W(9) with te[0. Ffiu. gauge transformation of a static abelian field is A > A' = A + Vx 577 and we know that the The field which is determined entirely by V x is therefore termed "pure gauge". the gauge field. Therefore. i.109) A W(0) = e^ W. the number of flux quanta 2irh/g being the winding number n.108) map the spatial circle S^.9 Introduction to Homotopy Classes words. the group space of (the phase transformations of) U(l) continuously. the vector potential A. It follows that the field Fij. we can go n times around S^. In terms of the phase factor \{x) = eixie\ we can rewrite the relation (22.22. However. (22. for B / 0 . i.110) . we can consider transformations which describe continuous transformations of a configuration with fixed n.107) The fields <f>(r. (22. going once around S^.105) insure that B ^ O . it is determined by a gradient (l/r)d/dd. the higher order contributions in (22. Then we can construct a oneparameter family of maps or transformations which transform </>n (0) continuously into (j>h (6). 9){^ and hence the functions (22. A field configuration with one loop (or twist). Of course. is also zero at r = oo. cannot be deformed continuously into one with two loops (n = 2).l].e. with n = 1.105) as «A(I)'AWV*<I> \(x) = J*V\ ( =• VxW)) (22.e.
the phase factors of (f)(9). n.lQSl^S^ with H(0. (22. (b) symmetry and (c) transitivity.113) . A superposition of classical configurations with different winding numbers does not minimize the energy.110) could be written down for configurations with different winding numbers. we have to show that the map satisfies the three properties of (a) identity. n' ^ n.9) = ^°\0).9) is called a homotopy (emphasis on the second syllable). — S\ > (22. Since the passage of time is continuous.e.e. H. Thus by varying t from 0 to 1 we deform (fin (9) continuously into (fin '(8).111) In principle an expression like (22. (22.112) is an equivalence relation — and as such it is defined by the three properties of (a) identity.e. A brief introduction can be found in the very readable book of W. In fact.0) = $\6). Example 22. Thomas [270]. it is clear that this is also a homotopy. (b) symmetry and (c) transitivity. f T h e standard texts on homotopy theory are N. are said to be homotopic (emphasis on the third syllable) to each other and one writes ^(0)^^(9). (fin (9) in the above example. i.e. to the internal group space of £7(1).3: Homotopy an equivalence relation Verify that the homotopy map H(t.9) = <fiW(9) and H(l.e. We can consider (22. We verify these in the following example. i.1]. Two curves which can be related to each other in this way. The map H(t.T. Reuter [77]. S%. like <pn (0).111) as a map from the space described by 9. 8) = cfinl\9). Solution: As stated. 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.112) are homotopic iff there exists a homotopy connecting (fin (9) and (fin (9).g. the unit interval I = [0.578 CHAPTER 22. Still on the level of a mathematical text is (in spite of its title) the Argonne National Laboratory Report of G. Dittrich and M. e. i. these two maps (fin from S^.8) itself must be a configuration with definite winding number. The relation (22. H. Hu [134]. i. 9) is an equivalence relation. time cannot make a field jump from one homotopy class (see below) to another. i. Steenrod [261] and S. H(1. and the space of t. But this type of combination is excluded sind H(t. Classical Field Configurations which is continuous in the parameter t and is such that H(0.
9) connects <pn (0) with itself continuously. . 0) = ^ (0) and 6(1. (c) We can demonstrate the property of transitivity as follows.0) for for 0<t<. (22 1U) At *=2!: But ^ ( ^ ) = ^ i ( M ) = #2(0. negative or zero). these groups are then called homotopy groups (cf. Hence <f>(0^ ~ <p 2 ). <^(2). a class being a set of equivalent maps.* 22. Assume <p( ' ~ (j> and <p( ' ~ <> ' /( for continuous maps <^(0'.9) = <f>(0\6) and H(l. Hence the set of all such maps breaks up into equivalence classes or homotopy classes. i? is a homotopy connecting </>(°) with <^>(2'. we see that of the set of these maps some are equivalent and others are not.10 The Fundamental Homotopy Group 579 (a) Evidently the homotopy H(t.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. being the continuous map connecting </>(°) with <j>(1'>. Then H(t. ' . Then there exist homotopies Hi and Hi connecting <j>(0> with (j>W and t^ 1 ' with </>(2) respectively. since and tf (1. Returning to the maps \(x) : Si ^ S1^ (i. which therefore demonstrates that the homotopy relationship is transitive. is isomorphic to) the set of integers Z (positive. then </>W ~ <t>{0)• We define an H(t. at t = 1/2 both i?i and H 2 give c/^1'. 9) = <f>W (0). so that H is a homotopy connecting <jyl>{0) with <f>^>(9). \ M)\ H!(2t. On the other hand H(0. Moreover. 0) = tf2(l. 0) = <£{2).9) H(0. Thus 7Ti[C/(l)] = vrifS'1] corresponds to (i. so that H(t. 0). is such that H(0. In some cases the set of homotopy classes possesses group structure (satisfying the group properties).0) = </>(0) is continuous there.ffi(O. 9) := H(l . . 9) = H(l.e.e. as they are called. 0) = 4>W (0). (b) Next we check that if 0<°) ~ <p^. below). U(l)). 9) = H(0. We now define the function H ( f 0 ff .9) = . <f>W .22.t.0). 0). i<t<l.e.e) tf2(2il. 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. Two elements of 7ri [f7(l)] which belong to different classes are homotopically inequivalent. i. Homotopy classes are sometimes also called Chern—Pontryagin classes.
but TTI[S2] = 0: S 1 is a circle. Appropriate maps. Classical Field Configurations Thus there is a denumerable infinity of such homotopic classes or sectors. 52. Fig. which is to be mapped into another circle Si indicated by dashed lines.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. Sketch of idea demonstrating that nilS1] = Z. and then introduce some of the related mathematics. T < 9 < 2TT. .115) Varying t € [0.) why 7i"i[S2] = 0. n=0: t=1: t=1/2: i. to make the result plausible. p.g. This trivial map into a single point is described as a "degenerate map" and is illustrated in Fig. Rajaraman [235]. 22.11 for the special values of t = 1 and t = 1/2 with Xo(0) = to t(2vr 9) for for 0 < 9 < 7T. In the diagrams^ of Fig. T (22.11 the circle drawn with the continuous line is the circle S%. 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. How can we see this? We first sketch the idea.580 CHAPTER 22.e.11 The trivial n = 0 map allowing shrinkage to a point. S 2 the surface of a sphere in a threedimensional Euclidean space. 22. 22. In order to really appreciate how topology comes into the picture here one should understand (e.
. to the case of winding number zero. The maps for these cases are illustrated in Fig. 2 the maps are wound around the dashed circle.e. (22. In the cases n = 1. i.12 The n = 1 and n = 2 maps not allowing shrinkage to a point.10 The Fundamental Homotopy Group 581 n=l: cutting of dashed circle n=2: Fig. 0 < X(P) < 2vr.s. Coleman [55]. (22. the transformations^ (the result may be looked at as 8n of expression (22. 22.I** dOJn*W±e {8) X 2vr J0 — / 2Wo dO(iri) dO \ d6 l. phase factor or element of U(l)): A(x) : Xn(x) = em*W = [Xl(x)]n. i.117) in ±.22.116) In fact from this we obtain (see below) 2  n and this means r.s.117) 2^ d6X(x)\1(x). of (22. is invariant under infinitesimal (continuous) deformations.e.118) is an infinitesimal (continuous) real function on the circle with (5f)o=o = (Sf)0=2n 1 See S.h. For winding number n we have the mapping (i. and hence the winding number.103)) A(x) > \'{x) where 5f(x) = \{x) + i\(x)5f(x) = A(x) + 6\(x). One can prove that the expression (22. 22.e. 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.117).h. d (22.12.
e. there is a theorem which says thatH TTg(Sn) = 0 for q < n. . Hence i /"27r ^ i Thus we have shown explicitly that the winding number remains unchanged under continuous deformations. Considering iri(S2) we recall that this symbol represents the set of disjoint equivalence classes into which the maps A (re) from S1 to S2 (circle to sphere) can be subdivided.e. It is intuitively clear that any circle drawn on the surface of a sphere can be shrunk to a point. Sec. i.582 CHAPTER 22. 1 1 See N.e. the identity class (the group space S2 is simply connected. 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. 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. only one homotopy class. Thus there is only the trivial map into a point. Steenrod [261]. the cyclic group with two elements (since 50(3) is doubly connected).8. Classical Field Configurations Under this transformation n changes by 5n Now. 15. In fact. il**( ^') r2Tr A ( 21 9 2 1) ' de\x2 *(«/W). an integer modulo 2) i. whereas 7ri(50(3)) = Z2 (i.
different from literature. in such a way that the factors appearing in the result are exhibited in a transparent way. and in particular the exponentially small level splittings of the double well potential. We consider such a theory now with action S((p) depending on the real scalar field 583 . Since we have not yet introduced periodic (i. Together with the infinite Euclidean time limit this requires for the calculation of the tunneling contribution to the ground state energy the FaddeevPopov method for the elimination of the zero mode.Chapter 23 P a t h Integrals and Instantons 23. An analogous procedure will be adopted in Chapter 24 in the use of the (nontopological) bounce configuration for the calculation of the imaginary part of the ground state energy of a particle trapped in the inverted double well potential. In each of the cases considered we obtained the result by solving the appropriate Schrodinger equation. our treatment in this chapter is restricted to the consideration of the asymptotically degenerate ground state of the double well potential which is based on the use of the topological instanton configuration. This is an important standard example of the path integral method which serves as a prototype for numerous other applications and hence will be treated in detail and. finite Euclidean time) classical configurations (which are nontopological).1 Introductory Remarks In Chapter 18 we considered anharmonic oscillators and calculated their eigenenergies. 23. 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!).e.2 Instantons and AntiInstantons A onedimensional field theory is a quantum mechanical problem.
m ^ + §^24 ' ^2 > a (23J) The potential V(</>) has — cf. 23. Chapter 6) 1 2 . Gildener and A. i. and ^2(0) : m* .e. "For comparison with literature.584 <)>{t).2 = 2m 2 (here m is the parameter in the potential V{4>) of Eq. 2mo has as the normalized ground state wave function </>o(?) = <<?0) Thus for mo = 1 we have in our notation here a.* /•tf CHAPTER 23. 2 2 4 ™ = ~.2) Fig.v{4>) <?VV m4 __ 1 . n = 0. 23.1 — degenerate minima (positions of classical stability or "perturbation theory vacua") at positions 4>{t) = ±m = ± 0 O . (23. (23.1 The double well'potential. note that here the mass of the equivalent classical particle is taken to be mo = 1. The Hamiltonian of the simple harmonic oscillator (cf.1)) and hence E0 = mh/y/2. Path Integrals and Instantons S(<f>) / dt \i>2 .1. Fig. in particular E.2. Patrascioiu [110] and Chapter 18. ! " ^ j = ^ 2 .
The states of the system here must be even or odd under x — ±x.23. e.x) = x2 V(x). and in the above 4> corresponds to x.2 The wave functions of the lowest states. must have definite parity as in a physical > situation.2 Instantons and AntiInstantons 585 Classically the position <f> = 0 is a point of instability. </.In mechanics of a particle of mass 1 with ~ position coordinate x(t)._(0) = 0.2 — and we can deduce from this the boundary conditions.g. this alone does not yield the level splitting. we have the Lagrangian L(x. but there is tunneling between the two minima if the central hump is not infinitely high. We know from the shape of the potential that the quantum mechanical energy spectrum is entirely discrete (no scattering). 23. as we did in Chapter 18 — but as we saw. i. Fig. 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> ~ * ±</>. These boundary conditions imply nonperturbative contributions to the eigenvalues which yield the splitting of the asymptotically degenerate oscillator . 23. We can develop a perturbation theory around either of the minima. W(0) = 0. This tunneling affects the eigenvalues.e. The corresponding wave functions ip±(x) then must be peaked at the extrema of stability — as in the examples illustrated in Fig.
5) and iS = dLE d dr[d(d(f)/dT)} i. that given by Eq.2 V'(<l>) = . A classical particle starting from rest (0 = 0) at —m/g cannot overcome the central hump and move to +m/g.6) d*l dr 2 . dr m+™ (23. i.7) so that (23.0o. is finite if we go to imaginary.e.7) resembles a Newton equation of motion with reversed sign of the potential as a result of our passage to Euclidean time. Then (0o. In the semiclassical path integral procedure that we want to use here. (23. dLE = 0. we consider the amplitude for transitions between the vacua or minima at 0/^ — ±</>0 = ±rn/g over a large period of time. Path Integrals and Instantons approximations into an even ground state and an odd first excited state. classically forbidden domains then become "classically" accessible. i. For small h we can attempt a stationary phase method. (23.3).2. Euclidean time T = it.tf\ .e.9) .2 m 2 ( l m z (23. The transition amplitude for this. SE.e. i.£'(0) = 0. dr We observe that Eq. We are therefore interested in the amplitude — called Feynman amplitude or kernel as we learned in Chapter 21 — (</>o. But quantum mechanically it can tunnel through the hump from one well to the other.7.e. (23. d(f> d ' dr2 dr It follows that the equation of motion is now given by 55. r .4) J <bi =—(An rTf (ITLE = I JT. (23.<h.586 CHAPTER 23.8) V(4>) + const.7) is to be solved for 0 with the boundary conditions m 0(r) = — for Tf — oo and 0(r) = > m for n oo.3) for tf — oo.U) = J'D{<f>}exp[iS(<l>)/h] (23. n) = f with the Euclidean action rTf SE(<i>) = / J TV f V{<f>}exp[SE(<P)/h} (23.£j — —oo (we assume the normalization factor or metric to > > be handled as explained in Chapter 21).  . Equation (23.
e. The solution (J>C(T) can also be looked at as describing the motion of a Newtonian particle of mass 1 in the potential —V (</>)• This is a very useful description. i.23. i.e. this means the total energy of the particle in Euclidean motion is zero. and having just enough energy to reach the other extremum at 4>f = m/g where it will again be at rest. 23. Its trajectory. Thus we can picture the particle as starting from rest at one minimum of V(</>) or maximum of —V(<p).3 The instanton path.3.e.10) where TO is an integration constant. Fig. . timeindependent) in the 1 + 1 dimensional <J>4theory which we considered previously. the classical path. In the present context of Euclidean time the configuration (f)c is generally described as an "instanton". 23. — (23. If the constant arising in the integration of the Newtonlike equation is chosen to be zero (see below). is depicted in Fig.18). from fa — —rn/g. and hence kinetic energy gained is equal to potential energy lost. we see that the equation is the same as the classical equation for a static soliton (i. in the present case (T — To) = m tanh m(r — TO). Thus we know the solution from there which is the kink solution (22.2 Instantons and AntiInstantons 587 In fact.
Path Integrals and Instantons The Euclidean action of the instanton solution (23.Tj) oc exp 4 m3 m^2 (23. Using Eq. 'Note that this integral may be evaluated for finite limits r = ± T and presents no problem in the limit T oo.]. the constant must be zero.7) admits another configuration which is the negative of (23. Inserting the result (23.12) dimr) cosh m(T — TO) m tanh x tanh x 00 _ 4m*_ (23.5).10) which rises monotonically from left to right is the pseudoparticle configuration called kink in the context of soliton theory and instanton in onedimensional field theory or quantum mechanics with topological charge 1 and Euclidean action given by (23.10) is given by (23.13). This antiinstanton configuration rises monotonically from right to left as shown in Fig. Also note that we can reexpress SE in the form 2 SE f0 c (oo) T Joo \ dr j J0 C ( where ^iw = . Tf = oo.7"/! ). we obtain lim f+ OO.' . We also observe that the expression is singular for g2 —> 0.r 0 ) ' (23.8) we obtain T f SE dr[2V((j)c) + const. The configuration (23.14) But this is not yet enough. The classical equation (23.TJ T fro.10) and is called the antikink or antiinstanton configuration and obviously (following the arguments of Chapter 22) carries topological charge — 1.588 CHAPTER 23. (23.11) For this quantity to be finite for TJ = —oo. 23. (23.^ . thus indicating the nonperturbative nature of the expression.4).13). (23.4. For a nonzero constant we obtain the periodic instantons discussed in Chapters 25 and 26.13) 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.* With this choice and inserting mo = j dr J = m we obtain* SE = m T lm4 2 g2 cosh4 m{r . but with the same Euclidean action (23.13) into Eq. We also need to know how we can calculate the difference AE of the energies of the two lowest levels from this expression.
from the sketch or from the variation of SE. 23. in the case of (23.23. The sum (23.15b) a mending together of an instanton and an antiinstanton at r = T\ with —oo = To < T\ < T2 — 00 for TQ. so correspondingly one defines the instanton or kink with topological charge + 1 and the antiinstanton or antikink with topological charge — 1. 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 (23.15b) a solution of the classical equation? Is it a topological configuration — if so.Writing ••• + L ••• and performing the variation in the usual way with 5<j) = 0 at r = ± 0 0 . Questions which arise at this point are: Is the nonmonotonic configuration (23. 23. and an antiinstanton localized at (say) TQ.g.r0). Clearly we are here interested in configurations of type (23.5. e. / > Looking at the instanton (23.15a) we have a simple superposition.10) we can ask ourselves: Does it make sense to consider the sum of an instanton localized at TQ as indicated in Fig. The terminology is motivated by analogy with the phenomenology of particles: Thus like a proton has electric charge + 1 (in units of the fundamental charge e) and the antiproton has electric charge —1.g.15b) is not exactly a solution. (23.TO far apart and T'Q 3> TQ.15b) In the case of (23. .4 The antiinstanton. (23.TQ are very far apart their sum differs from an exact solution only by an exponentially small quantity. 0c(r) = &(T).15b) since we require configurations from vacuum to vacuum. Both configurations comunicate between the two wells of the potential at < = —m/g and (ft = m/g.15a) or # ( r ) = 0(rro)^c(rro)0(rir)e(TTi)^c(r^)0(T2r). #(r) = </>C(T . as we can see e. b u t if TQ. 23.5.15b) has the shape shown in Fig.2 Instantons and AntiInstantons •<l> 589 ^ ^ ^ 0 ^ ^ ^ _ _ ^ Fig.TO) + ^ c ( r .
vanishes as T\ —> ±oo (of course. In fact. of course.CHAPTER 23. we see that in the domain [—oo.5 Superposition of an instanton and a widely separated antiinstanton. Thus it belongs to the socalled vacuum sector of solutions and its topological charge (determined by the boundary conditions at plus and minus infinity) is zero.15a) can be argued to be zero.2 m ( T o . 5^ In order to answer the second question we observe that (23. T\] with LECIT = 0.16) . Path Integrals and Instantons '. it also does not exactly stationarize the Euclidean action. since (23.e x p { . at oo the variation is <<> = 0). 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 . m/g / o / o ^ 0 1 \ c'\ o\ T2 ' ^^__ m/g Fig.r 0 ) At r = Ti the variation Sep is finite but nonzero so that we are left with an exponentially small contribution which.15a) is a configuration which starts from 4> = —i^/g and ends there. g cosh m(r — TQ) (23.T 0 ) } ] . #(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 . Finally. Ignoring exponentially small contributions the Euclidean action of the configuration (23.15a) is not exactly a classical solution. 23.
23.2 Instantons
and
AntiInstantons
591
and looking at Eq. (23.13) a n d ignoring exponentially small contributions, we obtain (note t h e same values of t h e limits)
<Koo)=(/>(oo) SE dA
Jd> oo)
#A*T
For t h e configuration (23.15b) we have with a similar argument
fTi SE \Joo roo JTx \ / /
rTi
( [ * • • • + [°° •••)LEdr=
( f
\Joo
1
...
f
i— Ti 1
...)LE
J
= 0.
Joo
As a further example we consider a configuration consisting (approximately) of two widely separated instantons (or kinks) and one antiinstanton (or antikink) localized at TQ , TQ and TQ respectively, § # ( r ) = ^ T  T o ^ T  T f M T i  T )
0(r  T I ) ^ ( T  r^VCTa  r)
+^(rr2)^c(rr(S3))0(T3r) with  o o = T 0 < r^ 1} < 2 \ < r^ 2) < T 2 < r ^ < T 3 = oo.
(23.17a)
Fig. 23.6 Two widely separated instantons and one antiinstanton. Again t h e configuration is nonmonotonic and as in t h e previous case one can show t h a t it is only approximately (with exponentially small deviations) a solution of the classical equation. In order t o answer t h e question concerning the associated topological charge of </>(3)(r) we recall t h a t this is determined by t h e boundary conditions of the configuration; thus in t h e present case t h e configuration belongs to t h e instanton or kink sector with topological charge
3
r (3) r
0
( E.g. in the overall range of r = 2T we can choose the locations TQ i ) : 2T/3.
2 T / 3 , T ^ 2 ) = 0 and
592
CHAPTER 23. Path Integrals and Instantons
+1 (like the single instanton or kink). Thus for each of the four charge sectors (corresponding to the two vacua and the kink and antikink sectors) we have a multitude of approximate solutions^ of the classical equation as illustrated in Fig. 23.7 in which (a) and (b) depict the two constant vacuum sectors and (b) and (c) the instanton and antiinstanton sectors respectively.
(a) (b) (c) (d)
Fig. 23.7 Classical solutions in their topological sectors. In evaluating later the Feynman amplitude about such configurations we use the completeness relation of states. Thus in the case of </>c (r) we write
oo poo
/
oo
d<t>2 /
J —oo
d<j>1(<t>f,Tf\(j>2,T2){<l>2,T2\(f>l,Ti)(<l>l,T1\<l)i,Ti).
(23.17b) Evidently the evaluation of this amplitude requires two intermediate integrations.
23.3
T h e Level Difference
The nonrelativistic Schrodinger wave function ip of a discrete eigenvalue problem for a potential V satisfies a homogeneous integral equation which we can write ip(x',t')= dxK(x',t';x,t)tp(x,t), t' > t. (23.18) Here K(x',t';x,t) is a Green's function satisfying the differential equation DK(x', t'; x, t) = iS(x'  x)5(t'  t) with differential operator d Id
2
(23.19)
(23.20)
" B . Felsager [91], p. 143, therefore calls these solutions "quasistationary configurations". Further discussion may be found there.
23.3 The Level Difference
593
(One can convince oneself by differentiation that the integral equation is equivalent to a timedependent Schrodinger equation). In Eq. (23.20) V0(x) = V(x)  SV(x), V = V0 + SV,
where 5V{x) is the deviation of the exact potential V from an approximation VQ. (Since we cannot — in general — solve a problem exactly, we have to resort to some approximation; thus we have to distinguish between the exact problem and the approximate problem which we can solve exactly; the Green's function is constructed from the eigenfunctions of the exactly solvable problem). The Green's function K(x', t'; x, t) can be expanded in terms of the eigenfunctions T/4 (x) of the Schrodinger equation for the potential VQ(X), i.e. (to be verified below)
K(x',t';x,t) =
We have (cf. Chapter 7)
^Y,f 
dE
'•i0)*(x')^0)(x)exp{iE(tlt)} EE&0)
(23.21)
" 1 d2  V0(x>) 2mo dxf2
^<?V) =  i « V )
so that DK(x',t';x,t)
•JL
dt1
V0(x') +
d2 2m 0 dx'2 EE, dE 2^
(0)
^h(x/)^n0)(x)exV[iE(t't)]
n
J
iE(t't)
=
i6(xx')5(t'
t),
(23.22)
as claimed (note that the first delta function results from the completeness relation of the eigenfunctions of the exactly solvable problem). We can go one step further and integrate out the independence of Eq. (23.21). Replacing En in the denominator by (En —ie),e>0,so that the integrand of (23.21) has a simple pole at E = En —ie, and integrating along the contour shown in Fig. 23.8, Eq. (23.21) becomes (with the help of Cauchy's residue theorem) H K(x', t';x,t) = Y, Tp{n}* (x')^n0) (x) exp[i£4°) (*'  t)}9(t'  t).
"This formula is discussed, for instance, in R. J. Crewther, D. Olive and S. Sciuto [60].
(23.23)
594
CHAPTER 23. Path Integrals and Instantons
This is our earlier expression (7.9).
Im E t  t>0
ReE
Fig. 23.8 Integration contour. We now recall from Eqs. (21.10), (21.17), that the Green's function K can be written as a path integral, i.e. that K(xf, tf;xi,U) = J cD{x}eiS/H{tf  ti). (23.24)
Evidently we want to relate Eq. (23.23) to Eq. (23.24) in the case of our onedimensional $ 4 theory. In the derivation of Eq. (23.23) we distinguished between the exact problem with potential V and an approximate problem with potential VQ which can be solved exactly. In an actual application such a subdivision can be a matter of convenience. Thus we could try to approximate V by taking as VQ the harmonic oscillator part of the potential and 5V oc <5 4 then the set {T/4 } would be the set of oscillator eigenfunctions ?>; which is wellknown. Alternatively, we could be interested in a completely different problem, e.g. one with Vo given by Eq. (23.1) and 8V oc <^6. In this case ipn would be the exact eigenfunctions of the Hamiltonian for the (/>4potential (23.1). It is shown for instance in scattering theory that the exact transition amplitude is characterized by poles at the exact eigenvalues (e.g. in the case of the hydrogen atom all discrete eigenvalues are simple poles of the scattering amplitude). Thus if we want to relate (23.24) to the path integral representation of the transition amplitude for the (/)4potential (23.1), we must take in (23.24) the exact (real) eigenfunctions ipn{x) and eigenvalues En for the potential (23.1) which, of course, we do not know. We proceed as follows. From the above equations we obtain K(fi,t';4>,t) = Y,M<i>')Tpn(<t>)eMiEn(tt')/h}
n
= f %>{<!>} exp[iS/h].
•*
(23.25)
23.3 The Level Difference
595
This equivalence of the standard decomposition as a sum on the one hand, and on the other hand with the Feynman integral on the right is the reason for describing the Feynman amplitude also as kernel. We now consider the left hand side of this expression for Euclidean time r = it. Since r' > r, the expression is dominated by the lowest and next to lowest eigenvalues E+,E associated with the ground and first excited state wave functions ip+(<fi) (even) and i/j(<fi) (odd) respectively. Thus the left hand side can then be approximated by (with h = 1)
K(<//,T';0,T)
~
^+(cf>')^+(4>)eME+(T'T)} +^_(0')^(<£) exp[E(r'  r)].
(23.26)
We are interested in the transition with oo and Moreover, V>+W>o) = V+(<fo), so that K(<f>',T';4>,r) ~ {^ + (</>o)} 2 exp[E + (T'T)] {</>_(0o)} 2 exp[£_(T'T)]. We define AE, E0 by** AE == EE+ so that E+ = E0Then
K{<J)Q,T'\<J)Q,T) <t>Q
(23.27)
^(0o) =  ^  (  0 o ) ,
(23.28)
(23.29)
(level splitting),
E 0 =  ( E + + E_),
(23.30)
AE,
E_ = E0 + AE.
(23.31)
~
{V'+(0o)}2exp
 ^  l A ^ ( r '  r ) (r'  r . (23.32)
{V>(<fo)}2exp
 \E0 + AE\
Here the wave functions at the minimum position (fro are comparable to (i.e. approximately given by) the corresponding ground state eigenf unctions of the harmonic oscillator and so (apart from overall normalization) (cf. Chapter 6) tp±{4>) oc < exp 1
(<£ + <
±exp
{<!><t>o?
= ±ip±{<f>) (23.33)
**Here AJE denotes the splitting of the originally degenerate oscillator level. This is therefore twice the deviation of a split level from the originally degenerate oscillator level.
596
CHAPTER 23. Path Integrals and Instantons
with E0 = mh/y/2 ( as we noted at the beginning of this chapter). Hence we can set 1 ± e _2 *° V#o)(23.34) l^±(0o)l = e2*§ ± I Then
K{(J)Q,T'\(J)Q,T)
« =
{V(^o)}2exp[JE;o(r'r)]eA^')/2eAE(^)/2 2{V(^o)}2exp[JB0(r,r)/^]sinh^:AJB(T,r), (23.35a)
where we reinserted h for dimensional completeness (observe that when AE(T' — T) = AEAT ~ h and small, the hyperbolic sine may be approximated by its argument and one returns to (23.25) for Euclidean time). The relation (23.35a) assumes fixed endpoints 4>o, i.e.
d<j>" j #'(V</>X</>" V
foMW.rW
+ 0o)<^'^>
(23.35b)
= <^0o) (</>0,r'  0o, r) (cp0\ip).
v '
Clearly without the delta functions the end points are not fixed. (This situation will be required in Chapter 26 in the case of certain bounce configurations). We now want to deduce the explicit expression obtained with the path integral formula and then extract AE by comparison. In this comparison we require r' — r to be a large Euclidean time interval (—T, T) with T — oo. It is clear from Eq. (23.35a) that we cannot put T = co rightaway; > thus we also require the study of a gentle approach of T to infinity, which necessarily makes the calculation more involved. We also note that since the above expression assumes quantization (discrete eigenvalues), the comparable formula must also involve quantization. Thus we must go beyond the purely classical contribution, i.e. we have to consider field fluctuations about the classical configuration.
23.4
23.4.1
Field Fluctuations
The fluctuation equation
We consider fluctuations n(r) about the instanton (or correspondingly static kink) configuration (pc: </>(T) = MT) + V(T). (23.36)
23.4 Field Fluctuations
597
Of course, the level splitting AE (to be extracted later) must be based on a symmetric treatment of instanton and antiinstanton. Allowing for fluctuations which are not tangential to the path one necessarily introduces more degrees of freedom — in much the same way as in allowing relative motion together with collective motion. Thus the consideration of fluctuations leads to the consideration of a larger (in fact infinite) number of degrees of freedom, i.e. to the continuum. We naturally impose on the fluctuations the boundary conditions 77(T = ± T )  T  O O = 0. (23.37) Clearly we have to expand the action into (23.5) we obtain
SE((J))
about
4>C{T).
Inserting (23.36)
«« />[(K
(ITLE = /
i(g+ £) + m + „ /«/>[K* + £,
dr
oo
1 / d0c \ ~ m"* 2 2 V dr J ' 2g
2
m
2J,2
1 2J,4 re+2*
+
WlA{t) *»**w+tf*
(23.38)
,2J2 J +Sg*<%rjz + 2<?2</>cr?3 +  5 2 T ? 4
Here the first pair of curly brackets {• • • } yields as before the contribution 4 m3
SE(<PC) = V~2
3?
(cf. Eq. (23.13)). In the second pair of curly brackets in Eq. (23.38) we use the identity
s:Am  MW.
dr where we used Eq. (23.37). Hence
—oo •2,
oo
dr2
(23.39)
dr2
•V(j),
+r,2{^924>l ~ m2} + 2</V0C + g2rf
(23.40)
598
CHAPTER 23. Path Integrals and Instantons
The term linear in 7 vanishes as a result of Eq. (23.7). Hence 7 SEW and SE(<j>) = = Here
oo
= SE(<l>c) + SSE(ri)
(23.41a)
JdrLE^c
+ v) + (23.41b)
SE{ci>c)+lJdTr]\T)L'E{4>)W
dr
/
00
\ it) ^^ ~ ^
jvi^eMSE^/h]
T){<p} eMSE(4>c)/h]
+
+2ffW+ .9W (23.42)
\
Hence, since the Jacobian from </>(T) to T?(T) of Eq. (23.36) is unity (with iS = —SE and subscript 1 in the following indicating that the single instanton case is considered),
I
exp[6SE(v)/fi] (23.43)
exp
4 m6 Jv{r)}exp[8SE(ri)/h}, 3g^h,
where we used Eq. (23.13). We consider the approximation in which we restrict ourselves in Eq. (23.42) to terms quadratic in n. This approximation is called the Gaussian or '•'•oneloop''' approximation.* Then
oo r 1 / j„ \ 2
dT
/ oo oo
(23J.0) f 00
\{tr) +'W*S»'>
2\dTj
dT
i/^y
2
+ rf m 2 {3tanh J m(T  r 0 )  1}
(23.44)
1
(7?, Mr?),
where M is an operator.t We set (observe that r acts only as a parameter which parametrizes the classical path; the dynamical — real timedependent — quantity is £n) V(r) = ^2^nrjn(T),
71=0
(23.45)
"The word "loop" refers to internal integrations which this nonclassical contribution requires. ^The factor 1/2 is extracted so that the fluctuation equation assumes the form (23.54b).
23.4 Field Fluctuations
599
where {r/ra(r)} is at every point r of the classical path a complete set of orthonormal eigenfunctions of M with eigenvalues {w^}: Mrtn(T) = wlrin(r), dTrjn{r)r]m{T) = p25nm.
(23.46)
Here we have in mind p 2 = 1. However, for ease of comparison with the literature* we drag the parameter p2 along. Then
(T?, Mr,) (2 = 5) E [°° dr(^Vn, M^Vm) = E l & l W 
(23.47)
Formally, if we insert (23.36) and (23.44) into h of (23.43), we obtain h = I 0 exp with 4 m3' (23.48a)
I0 = Jv{r,} exp
where
2^M??)
(23
^7%
fc=0
exp
n
(23.48b) D{77} = Urn FT dr](rk)
(23.49) (in our earlier discussion (cf. Sec. 21.6): 77(T^) — %, and r?(r) is given by • Eq. (23.45)). With v(Tk) = Y^nVniTk)
n
we have dr](rk) = E d CnVn(r k ),
n
(23.50)
and (the determinant being the Jacobian of this transformation) (23.51)
fc=0 fc=0
Hence I\ = exp 4 m3 3 ^
d
K^)/(n^) e [4? le " 12 ^
•fc=0
(23.52a)
•"•Note that a different normalization of {rj n } as in E. Gildener and A. Patrascioiu [110] (they have in our notation p2 = h/fj,2,fj,2 = 2m2) introduces additional factors on the right hand side of Eq. (23.47) and hence changes some intermediate quantities.
600
CHAPTER 23. Path Integrals and Instantons
Now,
2 2 d^e w P e/2
J
2TT
9 9'
(23.52b)
—c
so that the integral in Eq. (23.52a) can be evaluated formally to give I\ = exp 4 m3 3 ^
•*(%>) fl V C
d
k=Q
2TT
1/2
(23.52c)
provided no w^ is zero. We now show that this condition is not satisfied, i.e. we have one eigenvalue which is zero corresponding to the zero mode associated with the violation of translation invariance by the instanton configuration. This is the case if the zero mode, like all fluctuation modes, is normalized over the infinite time interval, not — however — if a finite interval of T from — T to T is considered, in which case the associated eigenvalue is nonzero (as will be shown below) and Eq. (23.52c) is well defined. Hence first of all we show that the differential operator M is the differential operator of the stability equation or small fluctuation equation. The operator M was defined by (23.44), i.e.
i r<x>
^{V,MV)
2
dr nMrj = \ drr/Mr]
OO oo
\fJTn
'dr]\
K£) + 7 ^ > (23.53) £ + ""<*•> V,
dr
J —(
where in the last step we performed a partial integration and used Eq. (23.37), i.e. n(dtoo) = 0:
dr]
°°
f°°
,
d2
From Eqs. (23.44) and (23.53) we obtain
M v"(4>c
= 6g2(f>c2 — 2m 2 = 6m 2 tanh 2 m(r — TQ) — 2m2 (23.54a) 4m 2 — 6m 2 sech 2 m(r — To).
Thus Eq. (23.46) becomes d2 1 (23.54b)
23.4 Field
Fluctuations
601
We have considered the spectrum {w^} and the eigenfunctions {r]n} of this equation previously; cf. Sec. 22.6. There we found t h a t the spectrum consists of two discrete states (one being the zero mode) and a continuum. In Eq. (23.45) we have to sum over all of these states. We also observed earlier t h a t the existence of the zero mode d4>c/dr follows by differentiation of the classical equation, i.e.
A.
dT2
dr'
:[V
(</>)], implying
dr2
V'itc
dr'
°'
Thus we know t h a t the set of eigenfunctions { ^ ( T ) } of M includes one eigenfunction T)Q{T) with eigenvalue zero. Before we can evaluate the integral (23.52a) we therefore have to find a way to circumvent this difficulty. First, however, we consider in Example 23.1 a gentle approach to t h e eigenvalue zero of the zero mode by demanding the eigenfunction ipo (r) to vanish not at r = oo but at a large Euclidean time T, as remarked earlier. § Some part of the evaluation of the p a t h integral will then be performed at a large but finite Euclidean time T, and it will be seen t h a t this large T dependence permits a clean passage to infinity. Example 23.1: Eigenvalues for finite range normalization
Derive the particular nonvanishing eigenvalue of the fluctuation equation which replaces the zero eigenvalue of the zero mode if the latter is required to vanish not at infinity but at a large but finite Euclidean time T. Solution: The operator of the small fluctuation equation is given by Eq. (23.54b). We change to the variable z = m{r — TO) and consider the following set of two eigenvalue problems: (a) Our ultimate consideration is concerned with the equation ( — j + 6sech 2 z  n 2 J ^0(2) = 0 with However, we consider now the related equation (b) ( —.r+6sech22n2)ipo(z) V dzz J = ^^(z) m2 with ^ n ( ± m T ) = 0, n 2 = 4 and Vo(±°°) = 0.
and our aim is to obtain ui 2 . We proceed as follows. The unnormalized solution of case (a) is (cf. Eq. (23.12)) dt/>c m 2 1 ipo[z) = —— = =—, z = m(T T0). orr g cosh z (Note that this is the zero mode which normalized to p 2 , i.e. / drr/^ = p 2 , is 770 = p(d(pc/dT)/^/SE, where SE is the action (23.13)). We multiply the equation of case (a) by i>o{z) and the equation of case (b) by ipoiz), subtract one equation from the other and integrate from —mT to mT (i.e. over a total T—range of 2T). Then
fmT
JmT
, f . dHo
V
j d 2 Vo\
dz2
t5g fmT
mz
, . r
dz2
j
J mT
This corresponds to a box normalization. trascioiu [110].
See also Appendix A of E. Gildener and A. Pa
602
CHAPTER 23. Path Integrals and Instantons
On the left hand side we perform a partial integration; on the right hand side we can replace (in the dominant approximation) the integral by that integrated from — oo to oo, and since this is the integral over twice the kinetic energy (and hence S B ( ^ I C ) of Eq. (23.13)) we obtain (remembering the change of variable from T to z) , dipo Vo — dz dj)o\ Wo . i mT dz J . diPo
dz
mT
™2 39 2 '
(23.55a)
In order to be able to evaluate the left hand side we consider W K B approximations of the solutions of the equations of cases (a) and (b) above. The W K B solution of case (a) is (c being a normalization constant and the ZQ occurring here a turning point)
^°( z )
=
~TJl e x p ( ~ /
dz 1/2
v J,
v(z)=:
 6sech 2 z
and
V(ZQ) = 0.
Thus for z —• oo we have v1/2{z) — n and tpo(z) —> 0, as required. The corresponding WKB > solution of case (b) which satisfies its required boundary condition, i.e. vanishing at z = mT, is similarly seen to be
*>(*> =
^
exp
Z**1/2)">(* L
/ dzv1/2\ + exp (  1 I
f,V2 dzv
exp
["dzv1/2
J zn
where v = n 2 — 6sech 2 z — (u> 2 /m 2 ). Differentiating the last expression we obtain d ipo(z) = —cv1'4 exp ( dz and so dz But (see ipo(z) above)
•>Po\ z=mT ^ CV 1/4 ' exp rmT
dzv1/2\
exp ( /
,f, dzv1 / 2
1
4'o\z=mT —  2 c 6 1 / 4 exp
rmT
/ Jz0
dzv1!2
dz
/ J z0
dzv1/2).
Hence (with n2 ~ n 2 — (uig/m2)
for T —> oo) and analogously —^o(—mT) ~ 2 — i>o(—mT). dz dz
dz
•4>o(mT) ~ 2 — ipoljnT), dz
It follows that the left hand side of Eq. (23.55a) is
4>o
But
^0(2)
:
mT dj>o mT ~ 2ip0—ip0 dz dz mT mT 4m 2
g cosh z
^
z
for z —> ±00,
so that 8m* e = F 2 i — ^ 0 ( 2 ) — =F — =F2V>o(z) for z —> ±00. dz g Hence ip0(±mT) — i>0(±mT) az = 2 ^ 0 ( ± m T ) — ^ o ( ± m T ) ~ q=4( — dz \ g
2\ 2 4mT
23.4 Field Fluctuations
Equation (23.55a) therefore becomes
603
if,
\
2 c
4mT „
^0 4m4
/
m2 Sg2
We thus obtain the result
wl z, 9 6 m 2 e  4 m T ,
wo ~ voe~2rnT,
vo • 4\/6«
(23.55b)
We observe that this eigenvalue is positive; the configuration associated with the zero mode is approximated by a classically stable configuration. For T — co the eigenvalue vanishes. Note that > here we used a Euclidean time interval of total length IT. This boundary perturbation method may be repeated for higher eigenstates which are then found to possess Tdependent eigenvalues.
23.4.2
Evaluation of t h e functional integral
It is clear from Eqs. (23.52a) and (23.52c) that the infinite product they contain provides a particular difficulty. It is the evaluation of this quantity that we are concerned with in this section. The highly nontrivial method was devised by Dashen, Hasslacher and Neveu* with reference to other sources. ^ We set Jo := det
drj(Tm
din
/(nIW)^',
^
dr
d£k exp
?£i« n\
2
2 2 Wnp
(23.56)
One introduces the following mapping or socalled Volterra transformation for large but finite T at the lower limit
y(T)
=
V
(
T
)

/
Z(~T)=V(T)
T
^ 0,
(23.57)
where N(T) is the zero mode, i.e.
N{T) =
=
HL
1
.
(23.58)
g cosh m(T — r 0 )
One can then show
fvu
*R. F. Dashen, T Integration in I. M. Gel'fand and and W. T. Martin
— see Examples 23.2, 23.3 below — that Vz } Dry exp 1 \1/2
2ith)
lJTAr2(T)]
1/2 T
[N(T)N(T)\
dr
1/2
/
TN*(T)\
(23.59)
B. Hasslacher and A. Neveu [63]. functional spaces and its application in quantum physics is lucidly described in A. M. Yaglom [107]. In particular this paper utilizes results of R. H. Cameron [43], [44] which are required in our context below.
604
CHAPTER 23. Path Integrals and Instantons
Observe that this expression is independent of the normalization of the zero mode N(T). Also observe that the transformation^ (23.57) transforms effectively to a free particle Lagrangian. After evaluation in Example 23.2 of the functional determinant occurring here, the integration requires the introduction of a Lagrange multiplier which enforces the endpoint conditions, i.e. boundary conditions, on the fluctuation rj{r) and hence on 0(T) itself; this latter calculation is done in Example 23.3. The evaluation of IQ with the above formula is then straightforward. We have for T — oo >
m
V K
'
'
1 g2 cosh4 mT
m 24e
4mT
and
T
dr
i
71 m4 m
rmT
/
JmT mT sinh 4z
32
dz cosh z
m°
+
sinh 2z
3z
mT mT
+—
ill
m 32
5
AmT
(
We observe how the largeT dependence cancels out in In and obtain the simple result In = det dr)(rm)
d£n
allfc
n
2n
w
1/2
m \ 1/2
kP2
(23.60a)
Note the variable T at the upper limit of Eq. (23.57) which implies that the transformation is a Volterra integral equation. This is an important point and should be compared with a Fredholm integral equation which has constants at both limits of integration. Consider a differential equation with second order differential operator M and Green's function K(x,x'), e.g. (M  X)u(x) = f(x), MK(x,x') = 5(xx'), a < x < b,
and with boundary conditions B\ [u] = 0, B2M = 0. The solution u(x) of the differential equation can be written u(x) = h(x) + \f
J a,
K(x,x')u(x')dx'
with
h(x) = f
J a
K(x,x')f(x')dx'.
For f(x) 7^ 0 this integral equation is called an inhomogeneous Fredholm integral equation; for f(x) = 0 the equation is called a homogeneous Fredholm integral equation. If one has an integral equation where K(x,x') is a function of either of the following types, K(x,x') = k(x,x')0(xx') or K(x,x') = k(x,x')0(x'  x),
the inhomogeneous Fredholm equation becomes a Volterra equation, i.e.
rx rb rb
u(x) = h(x) + A /
J a
k(x,x')u(x')dx'
or u(x) = h{x) + A /
J x
k(x,x')u(x')dx'.
In the book of B. Felsager [91] this transformation can be found in his Eq. (5.42), p. 190.
23.4 Field Fluctuations
605
Inserting this result into Eq. (23.52c) we obtain for the kernel h the simple expression ' m\ ' 4 m3 (23.60b) irh) 3 hg2 We see that this result is nowhere near to allowing a comparison with our earlier result Eq. (23.35a) for the determination of AE. The reason is that the zero mode has not really been removed, so that all fluctuations, i.e. perturbations, are still present. This has to be changed and is achieved with the FaddeevPopov constraint insertion that we deal with in the next subsection. Example 23.2: Evaluation of the functional determinant
Apply the shift transformation (23.57) to the fluctuation integral (23.43) and evaluate the functional determinant occurring in Eq. (23.59), i.e.
Thj
Solution: We first derive the inverse of the transformation (23.57) by differentiating this which yields
JV(r)
We can rewrite this equation as
Z(T) N(T)
f,{r)
N(T)
N{r)
N2(T)
T)(T)
:
d ( r](r) dr\N(T)
Thus
?j(r) r
N(T)
i= r
J^
T
dr '
Z(T) N(T)
' hand side of this equation yields
When multiplied by N(T),
r) Jj partial integration of the right
r)JV(r) I" K ' ' JT
V(T)
dr'—(—^—)z<T'), dr'\N{T')) y h N(r)£ ,
dr' N(T') 'V \Z{T').
r,{r)
=
z(r) + f(r),
f(r) =
(23.61)
For further manipulations we observe that with Z2(T) (obtained by squaring the expression of the first equation above) we have by differentiating out the following expression:
\ i, M^
dr~
2/ ^NM
JV(r) r,2(r)V"(0c)
,
N
, ,JV(T)
N(T)
, , ,N2M
'N2(T)'
so that since N(r) — V"{tj>c)N{r) = 0 (by definition N(T) being the zero mode)
V2{T)
+ V"(0c)r/2(r) = z2(r) +
^[^(r)
Mr)'
N(T)
606
CHAPTER 23. Path Integrals and Instantons
The derivative term will from now on be ignored since when inserted into the action integral and integrated it yields zero on account of the vanishing of the fluctuation r)(r) at the endpoints T = r 0 =  T , T. Thus 5S,
i
r1
dr
+ v"(<pcW
£
dr
z2(r) 2 v
;
The functional integral to be evaluated is To of Eq. (23.48b) which excludes the classical factor. Thus
h
/^ } exp[/; /T {i(J) 2 + i^(, c )}
jv{z}
D{V} Dr]
f.A\i2{T)]
D{z},
Dz Dr] Dr] Dz
T>z Dr]
Our next step is the evaluation of the functional determinant. Prom Eq. (23.57) we deduce
dr]{r") ) 7
^L=S(TT")+
(T
K(T,T')5(T'
T")CLT'
with
K(T,T')
.
JT
N{r'Y
(23.62)
where we allow for a general r dependence in the Volterra kernel as well. Discretizing the Volterra equation (23.57), i.e. Z(T) = r,(r) + fT K(T,r')r,(T')dT',
or more generally an integral equation of Fredholm type, by setting 5 = Ti+i — T;, i = 0 , 1 , . . . , n with TO = —T,rn — T, we can write the equation
z r
{ n)
= ^2
finiVin)
+
5
5Z
K T
( n>Ti)v(.Ti)
(23.63a)
We observe that K(rn,Ti) = 0 for % > n. The discretization here is a very delicate point. A different and convenient — but not unique — discretization (discussed after Eq. (23.66)) is to rewrite the equation as
n n ..
z T
( n)
=
^2 Snir}(Ti) + 5^2
n r n
K T T
( ^ i)~{v(n)
+
v(ni)}
cn—i
] T SnrtTi)
+  £
K(r,Ti)f,(Ti)
+ ~Y,
K{r,Ti+1)r]{n).
(23.63b)
Equation (23.63a) can be rewritten in the matrix form of a set of simultaneous linear equations: / z(n) \
2(T2)
/
i + i^Cn.Ti)
5K(T2,TI) 5K(T3,T1)
o
1 + <5K(T 2 , SK(T3,T2)
T2)
z( 3)
T
1 + SK(T3,T3)
0 1 +
0 0 0 0
6K(T„,T„)
V{T3)
V 2(r„) /
V
*See also I. M. Gel'fand and A. M. Yaglom [107], after their Eq. (3.4).
. (23.T2) I exp I K(n. 2! . = 1 52 K(n. where a f 1 See R. (23. Martin [43].63b).63a). we obtain exactly the nonvanishing terms on the right of Eq. as in Eq. r ' ) is to be evaluated at r = T ' .64). (23.65) The quantity D is really the Fredholm determinant which is obtained as above by solving the Fredholm integral equation with constant integration limits. the kernel K(r. Thus the factor a inserted above has to be taken as 1/2. the kernels to the left and to the right would be associated with step functions 8(T — r') and 9(T' — T ) . T.Ti) i=l : exp \ £ K(T.TJ) The factorials arise since every determinant of I rows and I columns appears l\ times when i..64) as above and consider the determinant of the expression.lnJV(T)}] N(T) N(T) (23. i.T')dT' (23. Thus D„ = l + 5 ^ K ( r i . whereas the determinant arises only once in the original determinant Dn. 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. r i ) + .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. [44].TI) 0 K(T2.67) i=l v ) = lim exp V In ( 1 + ' L K(T.. Cameron and W.j. X(TI. Felsager [91]. .68) D above. n — oo the expansion becomes > D lim Dn = 1 + n—*oo J K(r1. ^This is the method explained by B.n) o K(TJ.T1)dT1 + J dn J dr2. This determinant can be expanded in powers of K (more precisely in powers of a parameter A which we attach to K). are summed over all values from 1 to n.exp • /_:T JV(T')" dr' = e x p [ . 192.23.n) z=l x (23. H. Thus in our case one of the kernels is zero and we have to decide what we do when.66) •Dr) . . Our final step is to relate the VolterraFredholm determinant to the required functional determinant.a { l n i V ( r ) .e. In the limit S —* 0 . (23. We observe that if we discretize Eq.62) — and with insertion of a discretization dependent factor a still to be explained — we have in the present case D: .N{T) Finally we add the following remarks concerning the discretization (23.n) i=l N lim exp £*(T. n )d? (23.. This case has been considered in the literature^ where it is shown that if the Volterra transformation is considered as a Fredholm transformation whose kernel vanishes on one side of the discontinuity or diagonal.64) D (23. p.n) K(Tj.64) Inserting the explicit expression for the kernel given in Eq.TI) K(T2. 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. the value of the determinant is half (or arithmetic mean value) along the diagonal. so that (with T *T) 1/2 •Dz N(T) (23.7 Y.
70) and this now requires also integration over the endpoint coordinate z(T).0.61) 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).N(T) . (21.— .l + £7t) «i (23. N(T') . I0 = — f dz(T) f 1>{z} f da T'Z x exp T>r\ «/>5^> + ix(*<T> + '"< T >/• T *>m™ 7V2(T') (23. represents a constant translation. T + N(T) / T' = T JT dr' N(T') .^ . Path Integrals and Instantons Example 23. f(T) = N{T) J^ dr'^^z(r') (23.69) Incorporating this condition into the functional integral Jo with a delta function 1 = / dz(T)S[z(T) we have (cf.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. In = dzi dz2 • • • dzn e x p .22)) In = (7T6)"/2 Vn + 1 exp e(n+l) ( Z n + l . Eq. Briefly. (23. (23.J dz(T) J T>{z} J da•Dz hJ dz{T) v{z} I f_A\^+W)i{T)} J H § I"exp [ . n J=.59)) + f(T)] = — f dz(T) f°° dae^W+f™. I2 and using induction one obtains (cf. t h e range of T is subdivided into n + 1 equal elements of length e with (n + l)e = 2T and TO = —T. With partial integration (in the second step) we obtain (since (cf. Thus 0 = 2 (rf) + f(rf) or 0 = z{T) + f(T). (23. rn+i = T. (23. .57)) z(—T) —> 0) ^)/>'£^') = iv(r)/_T/T'A(__l_)2(T0 T JV(T') T . Solution: From Eq.3: Implementation of the endpoint constraints Using a Lagrange multiplier a insert into the functional integral (23.( . In every integration / dzi the contribution €7.2o) 2 rc^O. and hence may be ignored.(T) + /_ dr — . Evaluating Ii.Z ( T ).Z i .608 CHAPTER 23.72) The endpoint integration dz(T) = dz(rn+i) can be combined with this.73) .59) the endpoint conditions »)(±T) = 0 and evaluate the integral. so that (with V^N(T)/N(n)) 7t = J V{z}^[JT_Tdr[\[z(r)+ia^)' — lim n—*oo.e—*0 In. (23. Eq.71) Hence (with h = 1) l = j .
Eq.3 T h e Faddeev—Popov constraint insertion The FaddeevPopov constraint insertion implies the complete removal of the zero mode.. we are left with§ Io 1 1 2TT Vz oo / Vz dee exp oo * r 2 rT 2 dT^n N*(T) L T [N^NiT)]1'2 /2TTh I.£Q) — aTo]dTod£o.12).23. This result verifies Eq.74) where we reinserted h. dxv 0(r) = (f>c(r) + V(r) and set 1 TOO a — const.11).^ 23. 1/2 T W 2 (r)J (23. £n = ~2 dTT](T)7]n(l (23.10) and (21. / — / / We started off with V(T) = ^CnVniT). / d£o <9r0 5[F(TQ. (23." The method consists essentially in replacing the single integration J d£o by a double integration involving a delta function.76) Since the zero mode T]O(T) is proportional to d(f>c{r)/dT with — SE. (23. J co § dr Recall that / f ^ doce^ = yfVfp. Ignoring the remaining metric factor with the reasoning explained > between Eqs. ^E.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. (21. a = const. j / dTT]n(T)r]m(l r . i. Faddeev and V. Popov [89]. 11 L.e.4 Field Fluctuations 609 In the required limit the argument of the exponential is proportional to 1/T and thus vanishes in the limit T = (n + l)e — oo. .4. D. (3. Gildener and A.59) and agrees with that in the literature. N. Patrascioiu [110].
^n{r).J drm(r)Mr + T0) (23. (23. Path Integrals and Instantons _^ ^ w . i. at r = ro. demand validity of) the constraint £o = 0.81) the rdependence is integrated out and hence the integral does not provide the function F(TO. if the set {Vn(T~)} spans the Hilbert space of fluctuation states about 0C (at collective coordinate position r ) .75) requires.e.78) <t>ij) = Y. in the direction T]O(T) at the point TQ of the parameter r which parametrizes the trajectory of the classical configuration </>c(r). in the case of the soliton of the static 1 + 1 dimensional theory this is the kink coordinate.g.80) i r or •i /*oo dr7?o(r)0(r) = 0. / dTm{r)n{T). i /*oo (23.83) **The collective coordinate in the present context is also called "instanton time". this constraint implies that we go to the subspace which is orthogonal to the zero mode. (23. in the direction of the zero mode. The number £o is the component of the fluctuation rj(r). i. In the integral of Eq.£o)/a.76) we set 4>c{r) = ^CnVnir). Thus. Then cn = ~2 / dr(pc(T)rjn(T). In analogy to (23.82) This condition says that (j)(r) is not to possess any component in the direction of T)O(T). (23.e. (23. We consider the following integral / ( r 0 ) := .^O) with dependence on ro which the replacement (23. (23. . We establish the function ^(To>£o) with the following reasoning.£Q) — aro] contained in Eq.e. (23. e.79) (23. The delta function 8[F(TQ. A position ro of this classical configuration is called the collective coordinate** of (the field) 4>{T).610 this means that CHAPTER 23. We now impose (i.75) enforces the collective coordinate ro to be given by F(ro.81) 2 / dr»7o(r)0c(r) = .77) The set {r/ n (r)} constitutes a basis of the fluctuation space about the classical configuration. where L = in + cn.
89) Here the factor App is called the FaddeevPopov determinant.y/S^ro] (23.£O) F(r0^0) we have defined as + ro) + T?(T)] = / ( T 0 ) + p£o.82) we can replace here 7?O(T)0C(T) by 77O(T)T7(T).86) = Jdrrjoir^Mr F(TO.87) l/M) (23.y/SET0 = 0. (23. (23.90) Now tt 0^C(T + TQ) ro=0 <9T0 dr (23.91) " F o r more discussion of this point and related aspects see also B. (23. ./ drri0(T)<pc(T) + PJ (23. We evaluate this quantity as follows in the leading approximation.23.£O) OTQ = = £ " .88) ro0 Then in Eq.Zo)y/SETQ We now define A FP '• =0 • (23. Expanding /(TO) about TQ = 0 we obtain (with (J>'C{T) = T)Q(T)\J13E/'p) /(TO) = = Hence / dTT]Q(T)[(j)c(T + To) . jdrrj0(T)4>c(r) + JdTm(r)(p'c(T) X/SET0.^) .(23. We have AFP = /F(T0..75) 1 = AFP f dT06[{F{rQ. 158. p.4 Field Fluctuations 611 for an infinitesimal translational shift TO of the position of 4>c(r). .<t>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 ) .85) PJ W i t h Eq.84) (23. Felsager [91]. so t h a t with F(TQ.
e. i. 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.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.e.91) into Eq. we obtain (remembering that % = pd(j>c/dT/y/S^) AFP = 4 = f dr ( ^ V \/!5E JT \ dr * = yfe. Inserting (23.94) ' ' T h i s is an important result which will be required later in the explicit evaluation of path integrals. =~ (23.75) now becomes d£0AFP8[f(T0) = = / dr0 / J&T J + p£0 . if the coefficient of 770(70) vanishes. fs d(/)c{ro) = We have also drj(T0) = d£0r]o(To) + ^2d£ir)i{To). in d(j)c(To) + dr/fa). 0 d r (23. .e. Consider. P J AT (23. 770(7"). /(TO) ^/SE~T0} V~SET0 c d£.93). Path Integrals and Instantons Thus a shift TQ in the direction of the zero mode (i.e. ~ There is a quick way to obtain the result (23.90)./ dT0AFP.77) between the zero mode d(^ c (r)/dr and the fluctuation vector in its direction. the "static energy" invariant. 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. (23. iytQ 0C(TO + dr0)  4>C{TQ) = dT0^'c(T0) = dr0 7 0 (r 0 ). tangential to (J)C(T)) leaves this and hence the square of this quantity.612 CHAPTER 23.93) where AT = TT — 00. i. 7 Thus there is no fluctuation along the direction of the zero mode in the sum. if ^ + d£o = 0. i. using the relation (23. P the relative sign being of no significance.0AFP8 4o H P .
We have to take into account also multiinstanton configurations which almost stationarize the action integral and satisfy the same boundary conditions as the instanton. 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. we can set I 2TT ^o = l'o\ wlp2'' which with Eq.43)). (23. We observe t h a t the result (23. (23. (23.4 Fluctuations T h e single instanton contribution 613 We now return to the multiple integral (23. (23. We determine I'Q by observing t h a t this differs from IQ of Eq.96) lim . The source of lack of similarity is attributed to the fact t h a t the contribution of a single instanton to the p a t h integral is not enough.95) P Here £o 2 in the exponential is multiplied by WQ which is very small for large T and approaches zero in the limit of T —>• oo.A T A F P i " o e x p T—>oo p where I'o n^O.4 Field 23. so t h a t h = det r)/(n\ppATexp 4 m3' ATAFPdet(^ \ a£ jexp (23.23.97) is a constant still to be evaluated.55b).93).98) . Since w\ is for finite T given by WQ of Eq.60a) in not involving the zero mode contributions. (23.4.52a) and replace in this J d£o by the expression (23.55b) implies I'o w0p m \ ' LOQP '2K irh) m \ ' irh) 4:\/6mp mAT (23.60a) for I0 and Eq.96) does not yet allow a comparison with (23.35a) since the dependence on Tf — Ti = 2T of both expressions differs significantly. We can now perform the Gaussian integrations and write the result lim T^oo p 4 m3 exp 3^J (23.
the result becomes h lim 2 m r ^ ^ 1 e . Patrascioiu [110].55b)) w0 = v0e VQ 23.r).99) limlATAFF/0f^^eA^exp 4 m3 3 / ^ Inserting the explicit expressions for IQ and App. T0 < 4l) <C Ti « 42) < • • • «C T 2 n + 1 = T — oo (23. ^See E. *For the validity of the calculation the instantons or kinks have to be far apart. 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 . (23.20).100) in such a way that it exhibits the crucial factors of which it is composed.e.TiMiy+'Mr «>.101) 2T > oo (cf. Patrascioiu [110] make the replacements \QP <> 2g 2 and p?GP <» 2m 2 . 0c M.a n t i . (23. i. This means we are concerned with configurations like* 2n+l d(r . Thus — the subscript 1 indicating that it is the Feynman amplitude for the oneinstanton contribution — lim where with A T ATAFPI0 W0 SE{.32).4>C exp h (23.614 CHAPTER 23. (3. Gildener and A.102) *For comparison with the parameters of E. ~ rKo')e{Tx . Gildener and A.i n s t a n t o n contributions As already mentioned.5 I n s t a n t o n . In view of their normalization of the normal modes. in the present case we also have to include multiinstanton configurations in the form of one instanton plus any number of instantonantiinstanton pairs which describe back and forth tunneling between the two wells.4. they have to introduce the inverse of that factor on the far right of their expression (3. .* Finally we rewrite the expression (23. their Eq.2 m T e x p 2 irgh ' T»oo (23. Although we choose p2 = 1. the approximation is then known as the "dilute gas approximation". we drag the factor p2 along for ease of comparison with literature.100) We observe that the dependence on the normalization constant p cancels out.
fo) = Id2 St SE(<Pf.23.<Pi)2 + • • • . (23.<t>ff + h . (23.<j>i) . (23. (n = 0 implies the single instanton solution). The factor ( . .Tl\(j)i.Tf\(p2. 2 .6.4>l)eSE{4>l.102) distinguishes between instanton (rising from left to right) and antiinstanton (rising from right to left).r 0 ( 1 ) = .17b) and there perform first the integrations over the intermediate configurations whose endpoints are not fixed. 7 i = .2 T / 3 .36) (/.4>i.<j>i) + 2 d<& 1 r)2 *? k? + (<f>2 . here (f>2. (3) (23. .4>I) ld2SE SE(c/>i. this means we have to consider the amplitude (23.104) Considering the case n — 1 of a kink with one additional kinkantikink pair.103) For instance for n = 1 one has the configuration <jyc (T) which has the form shown in Fig. Proceeding as before we now consider first in analogy with Eq. ( 2 3 .H)2 + .Ti} OO J — OO /OO OO OC OO d(f>2 J — OO LeSE{<l>f. Consider (with T 0 = T. r 0 ( 2 ) = 0. Taking these configurations into account.r / 3 .(r) = ^ 2 n + 1 ) ( r ) + 77(r) with r ^ ) « 0. are varied.4 Field Fluctuations 615 for n = 1. the total Feynman amplitude is given by (^/.Ti) = X^/' T /I&> r i> ( 2 n+1) n=0 (23.(l>i) (23. T3 = T) d(f>l((pf.<t>2)eSE{<t>2.T2}((t)2.r/l&.105) where we write 'proportional' since only the classical instanton action depends on the endpoint configurations. 23. .l ) i + 1 in the sum of Eq.106b) SE{<t>l.<t>f) + 2 d</>{ SE{4>fi4>i) + ld2SE 2 d(/){ ( h .T2\(f)l.Tl}{4>l. we can expand the Euclidean actions here involved as follows (the first derivative vanishing at a solution): SE{<j>f. Since these endpoint configurations. r0(3) = 2T/3. T2 = T/3. 1 0 6 a ) and SE{(J>I.
105) can in other respects be treated in analogy with the single instanton amplitude. ±m/g (23. i.2d2SE ~c 4>i d(f)i e x p SE^f.4>i) / d(f>2 e x p {<h .108) where the last expression is obtained as follows. On the assumption that the kinks of the instantons are widely separated. (23. (23. i. Thus when in each of the three cases the zero mode elimination has been performed with the help of the FaddeevPopov method.107) with (the evaluation is given below) 7T 1/2 7 = IT 1/2 1/2 li&SE/dfi) {d^SE/dcft) 2m J (23.4>f) d4>\ 4>s72 (23. i.106a).(01 .106b) is to be substituted into the second relation of (23. the dilute gas approximation. rn ^2 J2H 9 1m' ±2gcj)..616 CHAPTER 23. Path Integrals and Instantons where the first expression in Eq.e. the .4>i) { 2 " 2d SE .e. each of the three intermediate kernels contained in Eq. It follows that — always within our approximations — c (^TflfaTjW SE{<t>f. The allowed positions of the individual kinks have to be such that their order is maintained. (23. It has to be remembered.109) The result of these integrations is that the amplitude l2n+i involves 2n intermediate integrations. each of which implies a multiplicative factor 7. of course.2J. in the case of l2n+i therefore ry2n.e.12) we have JM~°°) i.3 9 L 3m 2 and 9SE((/)) ± . dT 9 70(oo) V m J sE(4>) = ± m Hence d2SE{<t>) dej)2 . d2SE{4>) d(j)2 2m. With our remark after Eq.e. that each applies only to a fraction of the overall Euclidean time interval of r = 2T.
Patrascioiu JT/3 ° ir< >2T/3 1 dr.^ .101): (^/.a 3 [110].e.)(3) 1 o A T 3 Ao lim . T ] with ] T 3 = 1 A » T = A T = 2T. (23.r. then sign reversals TQ i ) .V ^0 SE{4>C X exp mAT lim — e ~mAT Sinh T^oo 7 IjATAppIo vo exp h (23.93) has been applied. a 2 0.(*) o ax = 2 T / 3 .Tj)( 3 ) these considerations imply the following result parallel to t h a t of the single instanton calculation (23.105) t h a t each of the three amplitudes contributes a factor e x p [ .112) We now compare this result with expression (23.110) Finally we have to recall from Eq.r/^.4 Field Fluctuations 617 replacement (23. (3) ° JrW: 2T/3 (2T)3 .6 the collective (i) coordinate TQ has to be to the right of TQ . Gildener and A. 23.111) Finally we sum over any number of instantonantiinstanton pair configurations and obtain for the Feynman amplitude oo 1 oo 1 ( {JATAFPIO 2n+l I = lim V n=0 J 2 n + i = lim .(2) (1) (2) (3) _ ( 2 T ) 3 JT JT JT 3! (23. Am3 2m J rW ( r ( 0 and this followed by setting aj.7 3 ^ — .T/0j. we now have to integrate over three collective coordinates.2 T / 3 this product becomes that given by E.35a) which contains AE in the argument of sinh and obtain for the level splitting A J resulting from consideration of the instanton plus any number of instantonantiinstanton pairs A*£ = 2n7AFP/0^=exp 2TT 1/2 4 m3 m V2 / m \ 1/2 4 v / g ^ exp ( 2TT irh _±rrv^ 2h ^With shifts r.m A . In the case of (<^f.A 3 r>oo 7 ' 3! exp SE(<J>C) T6 VQ mAT FP10 3 h (23. and TQ to the right of TQ This implies t h a t in this example we have the product of integrals^ rp (3) (2) . and in the case depicted in Fig. i.23.
To help the compariso