Excitations in a Bose-condensed Liquid - PDF Free Download (2024)

This book gives an up-to-date, systematic account of the microscopic theory of Bose-condensed fluids developed since the late 1950's. In contrast to the usual phenomenological discussions of superfluid 4 He, the present treatment is built on the pivotal role of the Bose broken symmetry and a Bose condensate. The many-body formalism is developed with emphasis on the one- and two-particle Green's functions and their relation to the density-response function. These are all coupled together by the Bose broken symmetry, which provides the basis for understanding the elementary excitations and response functions in the hydrodynamic and collisionless regions. It also explains the difference between excitations in the superfluid and normal phases. The distinction between single-particle excitations and collective zero sound modes, and their hybridization by the Bose condensate, is a central unifying theme. Throughout, an attempt is made to bring out the essential physics behind the formal field-theoretic analysis. A detailed discussion is given of recent high-resolution neutron and Raman scattering data over a wide range of temperatures. Chapter 4 gives the first critical assessment of the experimental evidence for a Bose condensate in liquid 4 He, based on high-momentum neutron-scattering data. This volume will act as a stimulus and guide to work on excitations in superfluid 4 He and 3 He- 4 He mixtures over the next decade. It will also lead to a greater understanding of the dynamical implications of Bose condensation in other condensed matter systems (high-temperature superconductors, exciton gases, spin-polarized hydrogen, etc.).

Excitations in a Bose-condensed liquid

CAMBRIDGE STUDIES IN LOW TEMPERATURE PHYSICS EDITORS

A. M. Goldman Tate Laboratory of Physics, University of Minnesota P.V.E. McClintock Department of Physics, University of Lancaster M. Springford Department of Physics, University of Bristol Cambridge Studies in Low Temperature Physics is an international series which contains books at the graduate text level and above on all aspects of low temperature physics. This includes the study of condensed state helium and hydrogen, condensed matter physics studied at low temperatures, superconductivity and superconducting materials and their applications. 1 2 3 4

D. S. Betts An Introduction to Millikelvin Technology A. B. Pippard Magnetoresistance in Metals R. J. Donnelly Quantized Vortices in Helium II A. Griffin Excitations in a Bose-Condensed Liquid

Excitations in a Bose-condensed liquid ALLAN GRIFFIN University of Toronto

§1 CAMBRIDGE UNIVERSITY PRESS

Published by the Press Syndicate of the University of Cambridge The Pitt Building, Trumpington Street, Cambridge CB2 1RP 40 West 20th Street, New York, NY 10011-4211, USA 10 Stamford Road, Oakleigh, Melbourne 3166, Australia © Cambridge University Press 1993 First published 1993 A catalogue record for this book is available from the British Library Library of Congress cataloguing in publication data Griffin, Allan. Excitations in a Bose-condensed liquid / Allan Griffin. p. cm. - (Cambridge studies in low temperature physics : 4) Includes bibliographical references and index. ISBN 0 521 43271 5 1. Bose-Einstein condensation. 2. Helium. 3. Superfluidity. I. Title. II. Series. QC175.47.B65G75 1993 530.42-dc20 ISBN 0 521 43271 5

hardback

Transferred to digital printing 2004

TAG

92-33320 CIP

Contents

Preface

pcige ix

1 1.1 1.2 1.3

Theory of excitations in superfluid 4 He: an introduction Landau-Feynman picture Role of the Bose condensate and field-theoretic analysis Outline of book

1 3 14 20

2 2.1 2.2 2.3

Dynamic response of Helium atoms to thermal neutrons Response functions: general properties and sum rules Density fluctuation spectrum of superfluid 4 He High momentum transfer and the impulse approximation

25 25 31 39

3 3.1 3.2

Bose broken symmetry and its implications Bose broken symmetry in a liquid Single-particle Green's functions for a Bose-condensed fluid S(Q,co) in the Bogoliubov approximation Mean-field analysis

47 49

67 69

4.3 4.4

High-momentum scattering and the condensate fraction Condensate-induced changes in the momentum distribution Impulse approximation using a Green's function formulation Measurement of the atomic momentum distribution Extraction of the condensate fraction

5 5.1 5.2 5.3

Dielectric formalism for a Bose fluid Dielectric formalism RPA or zero-loop approximation One-loop approximation for regular quantities

3.3 3.4 4 4.1 4.2

53 61 64

75 77 86 92 93 105 112

viii

Contents

5.4 5.5

Gavoret-Nozieres analysis The collective vs. single-particle scenarios

116 123

6 6.1 6.2 6.3

Response functions in the low-frequency, long-wavelength limit Zero-frequency sum rules and the normal fluid density Hydrodynamic (two-fluid) limit The nature of phonons in Bose fluids

127 128 132 141

7 7.1 7.2

Phonons, maxons and rotons S(Q,co): neutron-scattering data S(Q,co): theoretical interpretation

153 154 177

8 8.1 8.2 8.3

Sum-rule analysis of the different contributions to S(Q, co) The Wagner-Wong-Gould /-sum rules Woods-Svensson two-fluid ansatz for S{Q,co) Superfluid and normal fluid /-sum rules

195 195 200 203

9 9.1 9.2 9.3

Variational and parameterized approaches Variational theories in coordinate space Polarization potential theory Memory function formalism

208 209 218 225

10 10.1 10.2

231 232

10.3

Two-particle spectrum in Bose-condensed fluids Two-particle Green's function Evidence for the two-excitation spectrum in neutron scattering Raman scattering from superfluid 4 He

239 249

11 11.1 11.2 11.3

Relation between excitations in liquid and solid 4He Phonons as poles of the displacement correlation function Phonons vs. density fluctuations in solid 4 He Relation between S(Q,to) in superfluid and solid 4 He

257 258 263 266

12 12.1 12.2 12.3

The new picture: some unsolved problems Comments on the development of the new picture Dielectric formalism for superfluid ^ e - ^ e mixtures Suggestions for future research

270 270 277 282

References

285

Author index

295

Subject index

299

Preface

The well known Landau theory of the low-temperature properties of superfluid 4 He starts from a weakly interacting gas of phonons and rotons. This theory is very successful but it is essentially phenomenological since it makes no reference to the Bose condensate. The core of this book is a discussion of the modern microscopic theory of Bose-condensed systems based on finite-temperature Green's function techniques (the dielectric formalism). My emphasis is on developing the language and concepts of this formalism in a way that brings out the essential physics. This book is the first general account of the progress made in the last two decades toward understanding the excitations in superfluid 4 He specifically within the framework of a Bose-condensed liquid. I hope it will be a guide and stimulus to a new generation of experimentalists and theorists studying superfluid 4 He. The book should also be of interest to a much wider audience, since the phenomenon of Bose condensation, with its associated macroscopic quantum effects, plays a central role in modern condensed matter physics (Anderson, 1984). The goal of this book is two-fold: (a) to summarize the field-theoretic analysis of a Bose-condensed fluid and (b) to use this formalism to understand the nature of the excitations in superfluid 4 He. I emphasize how a Bose broken symmetry inevitably leads to certain characteristic features in the structure of various correlation functions, the most spectacular being the phenomenon of superfluidity. A major theme is the way in which a Bose condensate mixes the elementary excitation and density fluctuation spectra. Related to this is an attempt to understand the relation between the excitations in liquid 4 He below and above the superfluid transition. Most previous theoretical accounts have been limited to low temperatures (T ^ 1 K). Apart from Section 9.1,1 do not discuss the results based on variational IX

Preface

many-body wavefunctions. This latter approach has been very successful, but it is effectively limited to zero temperature as far as dynamical properties are concerned and, moreover, the role of the Bose broken symmetry is somewhat implicit. It is expected that this correlated-basisfunction technique will be the subject of a separate monograph in the Cambridge Studies in Low Temperature Physics. Comparing the present book with the recent exhaustive account of superfluid 3 He by Vollhardt and Woffle (1990), one can only be struck by how completely the various excitations in superfluid 3 He are now understood, and what an enormous amount of work is still needed before we reach a similar quantitative stage in dealing with superfluid 4 He as a Bose-condensed phase. At one level, the field-theoretic analysis of Bose fluids is a well developed area of many-body theory (beginning with the work of Beliaev, 1958a,b). But we are not yet at a stage where we can make quantitative numerical predictions about superfluid 4 He at finite temperatures using this approach, mainly because of the difficulties of dealing with a liquid, whether Bose-condensed or not. Consequently, we depend crucially on experimental data for guidance as to which of several theoretically possible scenarios is actually realized in superfluid 4 He. This explains why, in what is essentially a book about theory, I devote considerable space to neutron-scattering experiments, the most powerful probe we have of the excitations in superfluid 4 He. To help experimentalists understand the essential concepts of the microscopic theory, I often try to summarize the theoretical analysis in physical terms. When I need to recall some result of many-body theory or finitetemperature Green's function techniques, I usually give a specific reference to Mahan (Many-Particle Physics, Second Edition, 1990). One deficiency of this otherwise excellent book is that it does not include any discussion of the Beliaev Green's function formulation of Bose-condensed fluids. Classic accounts of the Beliaev-Dyson equations for interacting Bose fluids are given in the well known texts by Abrikosov, Gor'kov and Dzyaloshinskii (1963), Fetter and Walecka (1971) and Lifsh*tz and Pitaevskii (1980). I also recommend the brief account in Inkson (1984). The reader who wishes to follow the theoretical analysis in detail will need to be familiar with one of these accounts and have some background in many-body theory (such as given in the early sections of Chapters 2 and 3 of Mahan, 1990). The first three chapters of the book can be viewed as introductory, and include considerable background material for the non-expert. Chapter 4

Preface

gives a status report of the direct experimental evidence for a Bose condensate in superfluid 4 He. I feel this is necessary in a book devoted to the dynamical consequences of such a condensate. An extended outline of the chapters is given in Section 1.3. Within the general approach based on the role of the Bose condensate, several sections of this book contain material which has not been published before in the literature. I call specific attention to: (a) The detailed comparison of the nature of phonons in the two-fluid and collisionless regions, at both T = 0 and finite temperatures (Section 6.3). (b) The parameterization of the dielectric formalism developed as a concrete illustration of the recent Glyde-Griffin scenario (Section 7.2). (c) The /-sum rules specific to Bose-condensed fluids (Sections 8.1 and 8.3). (d) The unified discussion of how the single-particle, particle-hole (density) and two-particle (pair) excitation branches are hybridized via the Bose order parameter (Chapter 10). (e) The comparison between the microscopic theories of excitations in superfluid and solid 4 He (Chapter 11).

Acknowledgements The present book grew out of a project initiated by Henry Glyde for an introductory book on the theory of neutron scattering from quantum fluids and solids (4He and 3 He). We eventually decided that a modern account of Bose-condensed liquids really deserved a separate monograph. Several chapters have benefited as a result of extensive discussions with Henry. Our collaboration also led to a new interpretation of the phonon-maxon-roton dispersion curve (Glyde and Griffin, 1990) Which is developed in the present account. I would also like to thank Emile Talbot for his useful contributions at the early stages in the writing of this book. Over the last twenty years, I have been strongly influenced by the manuscript preprint of Nozieres and Pines, written in 1964 but only published in 1990. The present book is, in many ways, a development of their approach. I would like to acknowledge several stimulating discussions with Philippe Nozieres. Experimental data taken by the neutron-scattering groups at the Chalk River Laboratories of AECL in Canada and the Institut Laue-Langevin

xii

Preface

in Grenoble, France have been a continual guide and stimulus to my theoretical work. I would like to single out Eric Svensson for countless discussions about the data. I would also like to thank Dave Woods, Varley Sears, Bill Buyers and Peter Martel at Chalk River. At Grenoble, I thank Bill Stirling (now at Keele, U.K.) and Ken Andersen for their assistance. In particular, I am very grateful to them and their colleagues at ILL for making available some of their unpublished high-resolution time-of-flight neutron data for use in this book. I also specifically acknowledge Bjorn Fak (now at CENG, Grenoble) for his careful reading of the manuscript and many useful suggestions for its improvement. My research work on Bose fluids has been generously supported over many years by grants from the Natural Sciences and Engineering Research Council (NSERC) of Canada. I would like to express my appreciation to Karl Schroeder and Jennifer Tarn, who expertly keyboarded the manuscript in TgX and endless revisions of it. Karl had the extra strain of dealing with the many changes which arose as the theoretical interpretations evolved dynamically during the writing of the book. It is with particular pleasure that I thank my wife, Christine McClymont, for her very valuable editorial polishing of the final manuscript as well as for her advice and interest over the three years it took to complete this book. Allan Griffin Toronto, Ontario

Theory of excitations in superfluid 4He: an introduction

The major goal of the present book is to outline the field-theoretic analysis of the dynamical behaviour of a Bose-condensed fluid that has developed since the late 1950's. While we often use the weakly interacting dilute Bose gas (WIDBG) for illustrative purposes, the emphasis is on the dynamical properties of a specific Bose-condensed liquid, superfluid 4 He. We attempt to develop a coherent picture of the excitations in liquid 4 He which is consistent with, and rooted in, an underlying Bose broken symmetry. Recent high-resolution neutron-scattering studies in conjunction with new theoretical studies have led to considerable progress and it seems appropriate to summarize the current situation. The only other systematic account of superfluid 4 He as a Bose-condensed liquid is the classic monograph by Nozieres and Pines (1964, 1990). The phenomenon of Bose condensation plays a central role in many different areas of modern condensed matter physics (Anderson, 1984). Historically it was first studied in an attempt to understand the unusual properties of superfluid 4 He (London, 1938a). In a generalized sense, however, it also underlies much of the physics involved in superconductivity in metals and the superfluidity of liquid 3 He, in which Cooper pairs play the role of the Bosons (see, for example, Leggett, 1975 and Nozieres, 1983). In recent years, there has been increased research on the possibility of creating a Bose-condensed gas, involving such exotic composite Bosons as excitons in optically excited semiconductors, spin-polarized atomic Hydrogen, and positronium atoms. This research, in turn, has re-stimulated theoretical interest in Bose condensation in general and its implications in both liquids and gases. Although 4 He was first liquefied by Kammerlingh Onnes in 1908, it was not until 1928 that Wolfke and Keesom found it could exist in two phases, now called Helium-I and Helium-II, separated by the transition 1

Theory of excitations in superfluid 4He: an introduction Ordinary fluid Solid

40 35

I Liquid Triple 1 point 1 ^ ^

30 25

f 20

Critical

pointy Gas

Liquid He I

10 Critical point 1.0

2.0

Fig. 1.1. The P-T diagram for the condensed phases of 4He contrasted to that for a normal liquid. temperature Tx = 2.17 K at SVP. Kapitza (1938) and, independently, Allen and Misener (1938), found that the fluid flowed without any apparent viscosity below Tx. At higher temperatures (T > Tx), the liquid exists as a "normal" liquid and is called liquid Helium-I. Below Tx, liquid 4 He is a "superfluid" (a term introduced by Kapitza for a fluid having, in a certain sense, zero viscosity) and is called Helium-II. Most of our discussion will deal with this low-temperature superfluid phase, but we will also be interested in how it differs from Helium-I (above Tx). The phase diagram is shown in Fig. 1.1. The study of superfluid 4 He has played a central role in developing our understanding of many-body systems. The modern idea of quasiparticles started with Landau's original theory of superfluid 4 He (1941). Bogoliubov's (1947) derivation of the phonon spectrum in a weakly interacting dilute Bose gas was one of the first treatments of a broken symmetry. In this chapter, we give a brief review of theories about the excitations of superfluid 4 He and how they developed since the pioneering work of London, Tisza, Landau, Bogoliubov and Feynman. In this mini-history, we also introduce the kind of questions with which the rest of this book will be concerned. We assume the reader is already familiar with an elementary account of the properties of superfluid 4 He, such as the first three chapters of Nozieres and Pines (1964, 1990). In Section 1.1, we sketch the well known Landau-Feynman quasiparticle picture (which makes no reference to an underlying Bose condensate). In Section 1.2,

1.1 Landau-Feynman picture

we review the development of the theoretical approach in which the condensate plays a crucial role. It is with this latter approach that this book is concerned. Many experimental probes have given valuable information about the excitations in superfluid 4 He, including Brillouin and Raman light scattering, sound propagation, as well as thermodynamic and transport studies. However, the most powerful technique has been inelastic neutron scattering. Through the dynamic structure factor S(Q,a>), this is a direct probe of the density fluctuations which turn out to be the elementary excitations of superfluid 4 He (but not, as we shall see, of normal liquid 4 He). Moreover, recent high-resolution neutron-scattering data over a wide range of temperatures and pressures have stimulated the theoretical developments which are a major theme of this book. For these reasons, we shall discuss neutron-scattering data at some length. As background, we summarize in Chapter 2 some general properties of the dynamic structure factor which hold for any system and also the main features that S(Q,co) exhibits in superfluid 4 He. References to neutron-scattering results in the overviews given in Sections 1.1 and 1.2 are more fully explained in Chapter 2. For a general account of experimental studies on excitations in superfluid 4 He, the classic review article by Woods and Cowley (1973) is highly recommended. In Section 1.3, we conclude with an extended outline of the contents of the succeeding chapters of this book. 1.1 Landau-Feynman picture The superfluid characteristics of Helium-II are well described by the twofluid model, first suggested by Tisza (1938) and later brought to fruition by Landau (1941). In this model (for authoritative accounts, see Landau and Lifsh*tz, 1959; Khalatnikov, 1965), the liquid consists of two interpenetrating fluids, a normal fluid and a superfluid, each having its own density and velocity fields. The normal fluid has a finite viscosity, contains all the entropy of the liquid, and has a mass density p^ and velocity v#. The superfluid component has no entropy and it has a mass density ps and a velocity v$. If vs is less than some appropriate critical velocity, the superfluid component flows with zero viscosity. The local mass density and the momentum current density are given by the equations p(r)

pN(r) +

ps(r),

J(r)

pN (r)\ N (r) + ps (r)\s (r) .

Theory of excitations in superfluid 4He: an introduction

In He-II, the normal fluid fraction PN/P is a strong function of temperature. As the temperature decreases, the normal fluid fraction goes to zero. Above Tx, the normal fluid fraction is unity, corresponding to the fact that there is no superfluid component. There are many ways of measuring ps and pN, but the most precise experimental results are from measurements of pN from the moment of inertia of a rotating disk (Wilks, 1967). The two-fluid model, consisting of (1.1), the continuity equation, and a set of three other equations describing the flow produced by gradients in the temperature, pressure, and chemical potential, has been extremely successful in describing the macroscopic transport properties and the low-frequency, low-wavelength hydrodynamic modes in Helium-II (Khalatnikov, 1965). One of the great successes of the field-theoretic analysis of Bose-condensed fluids (see Chapter 6) is to show how this two-fluid description can be a natural consequence of a Bose order parameter (Bose condensation). London (1938a,b) proposed that superfluidity in liquid 4 He is a manifestation of Bose-Einstein condensation. Tisza (1938) also conjectured that the superfluid component in his phenomenological two-fluid model could be interpreted as the fraction of the Helium atoms that are Bosecondensed. As we discuss in more detail in Chapters 4 and 6, the condensate fraction and the superfluid fraction are not the same in a Bose-condensed liquid like superfluid 4 He. At zero temperature, the superfluid fraction is 100% (since no quasiparticles are thermally excited) while the condensate fraction is only about 9%. In his classic papers of 1941 and 1949, Landau strongly argued against any connection between a Bose condensate and the two-fluid model. The modern view vindicates both Landau and London in that the microscopic basis of Landau's quasiparticle picture and the two-fluid description lies in the existence of a Bose condensate. The Landau (1941, 1947) theory of superfluidity is based on the lowlying excited states of a Bose liquid. He showed that the low-temperature thermodynamic and transport properties of superfluid 4 He could be understood in terms of a weakly interacting gas of Bose "quasiparticles" (phonons and rotons). This description is in many ways analogous to the phonon picture of an anharmonic crystal (see Chapter 11). This was developed at some length by Peierls, Frenkel and others in the early 1930's. In superfluid 4 He, Landau had no well defined scheme to calculate the dispersion relation of the quasiparticles; i.e., how to relate it to the microscopic forces between the 4 He atoms. Although certain qualitative features could be argued to be consequences of the Bose

1.1 Landau-Feynman picture

statistics of the He atoms, the dispersion relation of the quasiparticles in superfluid 4 He was viewed by Landau (1947) as something to be determined from experiment. We recall his strong resistance (Landau, 1941, 1949) to trying to use a WIDBG as a model for superfluid 4 He, as London (1938a,b) and Tisza (1938) tried to do. A quasiparticle spectrum which was acoustic at low Q but has a roton minimum at large Q has precisely the features needed to explain the temperature dependence of two-fluid thermodynamic functions as well as the velocities of first and second sound. In the two-fluid picture, the normal fluid density describes the thermally excited quasiparticles, while the superfluid density can be thought of (in a rough sort of way) as the probability that the liquid is in its ground state. At low temperatures (T ~ IK), the dominant feature of the dynamic structure factor S(Q, co) in the range 0.1 A" 1 < Q < 2.4 A" 1 is an extraordinarily sharp resonance. This is illustrated by the representative data in Fig. 1.2 as well as in many other figures in this book. (It is to be emphasized that plots of S(Q,co) data almost always still contain instrumental resolution broadening, with a width typically of order 1-2 K.) As first pointed out by Cohen and Feynman (1957), the peaks in S(Q,co) correspond to the creation of the elementary excitations which Landau (1941, 1947) originally postulated to describe the thermodynamic and transport properties of superfluid 4 He. The roton was first observed using neutron scattering by Palevsky et al. (1957). The complete quasiparticle dispersion relation CDQ (see Fig. 1.3) was first determined by Yarnell, Arnold, Bendt and Kerr (1959) and Henshaw and Woods (1961) at T = 1.1 K. For Q values below about 0.6 A" 1 , the dispersion relation is phonon-like (CDQ = cQ), with a slope slightly larger than the thermodynamic speed of sound. It bends slightly upward (anomalous dispersion) before bending over to a maximum energy (referred to as the "maxon") of about 14 K at QM = 1.13 A" 1 . The minimum at QR = 1.93 A" 1 occurs at an energy A = 8.62 K at SVP. The dispersion relation near the minimum is called the "roton" region. For Q within about 0.25 A" 1 of QR, COQ is found to be very well described by the simple Landau expression (Woods, Hilton, Scherm and Stirling, 1977) coQ = A + h(Q-QR)2/2fiR

(1.2)

where the roton "mass" is JUR = 0.13m (m is the bare mass of a 4 He atom). An important early study by Bendt, Cowan and Yarnell (1959) showed that various thermodynamic functions of superfluid 4 He up to about

Theory of excitations in superfluid 4He: an introduction

6 0.7 0.6 0.5 -

o 1.24 K a 1.50 K 2.07 K 2.26 K

+ x

0.4 H

-0.1 -0.2

0.0

0.2 0.4 Energy transfer (THz)

0.6

0.8

Fig. 1.2. A plot of the neutron-scattering intensity vs. the energy at a momentum transfer Q = 1 A" 1 , as a function of the temperature [Source: Stirling, 1991; Andersen, Stirling et al, 1991].

1.8 K could be calculated (with an accuracy of a few per cent) by treating it as a non-interacting gas of Bose quasiparticles with the CDQ dispersion relation determined by the peak position in <S(Q,<x>) measured at 1.1 K. This work is an empirical proof that, at least up to 1.8 K, the resonances in <S(Q, co) do indeed give the energies of the elementary excitations of superfluid 4 He. This is not the case in normal liquid 4 He or in liquid 3 He. In his original paper, Landau (1941) introduced phonons and rotons as two quite distinct kinds of excitations. However, beginning with a later addendum (Landau, 1947), it has been traditional to assume that the quasiparticle spectrum in Fig. 1.3 describes a single excitation branch COQ. From this point of view, while different parts of the dispersion curve are described as phonons and rotons, they are not thought to be qualitatively different types of excitations. In spite of this, while the physics behind the phonon part has been viewed as "obvious" (i.e., a compressional sound wave), the physical nature of the roton has been the subject of much discussion over the years (see, for example, Feynman, 1954; Miller, Pines and Nozieres, 1962; Chester, 1963, 1969, 1975).

1.1 Landau-Feynman picture

12 -

Fig. 1.3. The quasiparticle dispersion curve at T = 1.1 K and SVP. This is a compilation of data from rotating crystal (RCS) and triple axis (TACS) spectrometers [Source: Cowley and Woods, 1971].

The region around Q ~ lA l is often referred to as a "maxon". This region, however, has such a high energy (~ 14 K) that maxons are never thermally excited and thus they play no direct role in the thermodynamic or transport properties of superfluid 4 He. In a rotating vessel of Helium-II, only the gas of excitations rotates with the vessel; thus the angular momentum of the excitations determines the moment of inertia of the system. At temperatures less than about 1 K, when there are few excitations thermally excited, the moment of inertia is greatly reduced from the classical prediction that the entire liquid rotates with the vessel. Landau (1941) was able to express the normal fluid density PM explicitly in terms of the Bose distribution function N(a>) of the thermally excited phonon-roton excitations, (1.3) The normal fluid fraction PN(T) turns out to be a useful quantity since

Theory of excitations in superfluid 4He: an introduction

it gives an effective measure of the number of quasiparticles thermally excited at a given temperature. At T = 0, of course, PN=0 and the entire superfluid component consists of the liquid. Below T < 0.6 K, the low-energy, long-wavelength phonon excitations make the dominant contribution to pN (we note that COQ ~ 1 K at Q ~ 0.1 A" 1 ). On the other hand, PN/P is very small until the temperature reaches about 1.2 K, and then the thermally excited rotons make the overwhelming contribution. The roton's relatively high energy (A ~ 8 K) is compensated by the high density of states at Q ~ QR. Finally, at about 1.7 K or so, the quasiparticle damping becomes appreciable and one expects that the simple non-interacting quasiparticle gas picture, and hence (1.3), will increasingly break down as we approach the lambda point Tx = 2.17 K (SVP). Landau (in collaboration with Khalatnikov) also developed the formalism of "quantum hydrodynamics" to calculate the effect of weak quasiparticle interactions (three-phonon and four-phonon processes, phononroton scattering, etc.) and to use these results to obtain the temperature dependence of the characteristic transport coefficients associated with the normal fluid (the interacting gas of quasiparticles). At temperatures T < 1.7 K, the Landau-Khalatnikov picture gives a very satisfactory account of the transport properties of superfluid 4 He, as described in detail in the well known monographs by Khalatnikov (1965) and Wilks (1967). Landau was the first to make a clear distinction between collective modes (such as first and second sound) and elementary excitations (the phonon-roton quasiparticles). In particular, Landau emphasized that the phonon quasiparticle is conceptually quite different from the first sound hydrodynamic mode, even though the speeds are quite close in magnitude. This distinction between collective modes and quasiparticles is equally important in Fermi liquids like normal liquid 3 He (see Chapter 1 of Lifsh*tz and Pitaevskii, 1980) and was later incorporated into the description of the excitations of an anharmonic crystal (see Chapter 11). The microscopic basis of Landau's picture was developed by Feynman in the period 1953-1957. Feynman dealt directly with the excited states of a Bose liquid, as opposed to Bogoliubov's work on the excitations of a Bose gas. Feynman (1953b, 1954) showed that if atoms obeyed Boltzmann statistics, he could construct wavefunctions describing the motion of a single atom with energy Q2/2m*, where m is some effective mass arising from the interactions with other atoms. On the other hand, he found that even at low Q, such single-particle states develop

1.1 Landau-Feynman picture

an energy gap (relative to the energy of the liquid in its ground state). This gap is a result of the required Bose symmetry of the wavefunction (a pairwise interchange of atoms does not alter the wavefunction) and the fact that in a liquid, motion of an 4 He atom requires that other nearby atoms move out of the way. Feynman estimated that this singleparticle energy gap is of the order of the potential well depth any given 4 He atom moves in (~ 10 K). In contrast with these high-energy singleparticle-like excited states, the excited states corresponding to collective long-wavelength density fluctuations (which involve a large number of atoms moving a small amount coherently) had a sound wave dispersion relation. These brilliant papers by Feynman (1953b, 1954) still deserve careful study. For an assessment of Feynman's work on liquid 4 He from a modern perspective, see Pines (1989). For a lucid summary, we refer to the discussion by Wilks (1967). Feynman (1954) presented a variety of physical arguments to the effect that the excited-state wavefunctions are all described by the expression (1.4) 7=1

where |Oo) is the ground-state many-particle wavefunction and the sum is over the position operators r ; of the 4 He atoms. Carrying out a T = 0 variational calculation with (1.4), it is found that /(r 7 ) = exp(/Qr 7 ) minimizes the total energy. Since the Fourier transform of the density operator p(r) is (see Section 2.1) N

P (Q) = X

**•'' ,

(1.5)

7=1

this means that the Feynman ansatz (1.4) corresponds to the creation of a single density fluctuation of wavevector Q. The energy of this state (relative to the ground state | Oo)) is found to be given by the Feynman-Bijl relation u

S(Q) '

(1.6)

where £Q = Q2/2m and S(Q) is the static structure factor (a quantity whose importance had only recently been emphasized by van Hove (1954) at the time of Feynman's work). Feynman made the following, logically distinct, remarks concerning \Q)F) = /3+(Q)|<X>o): (a) Considering Q as a variational parameter, (1.6) has a local minimum

Theory of excitations in superfluid 4He: an introduction

at 2 — 60 — 27c/ro corresponding to the local maximum in S(Q). Here r0 is the mean distance between the 4 He atoms. (b) In the limit of low Q, (1.6) can be shown to give the correct energy of a state with a single phonon present. (c) |Ojr) = p + (Q)|O o ) is an exact eigenstate of the total momentum operator, which commutes with the total Hamiltonian. It follows that (1.6) gives an upper bound to the energy of any excited state with momentum Q. These three arguments led Feynman to suggest that \0>F) = p+(Q)|) data in solid Helium, as we discuss in Chapter 11. These sorts of questions force us to search for a more fundamental understanding of the elementary excitations of superfluid 4 He and their relation to the structure exhibited by 5(Q,co).

Theory of excitations in superfluid 4He: an introduction 1.2 Role of the Bose condensate and field-theoretic analysis

Beginning with the work of London (1938a,b), a microscopic theory of superfluid 4 He based on a Bose condensate has also been developed. Up to the present time, however, it has not played a large role in experimental research and for this reason it has not been emphasized in standard textbooks on liquid 4 He, which concentrate on the LandauFeynman scenario. We believe that it will be a dominant theme of future work on superfluid 4 He and consequently give it the central role in this book. Since the early 1960's, it has become increasingly accepted that insofar as liquid 4 He exhibits macroscopic quantum effects (as described by the two-fluid model, for example), this was due to a complex order parameter (\p(t)). In addition, the critical exponents near Tx have been found, to a high degree of precision, to be those associated with a two-component order parameter. Finally, many groups have carried out experiments to measure directly the momentum distribution of 4 He atoms by high-energy inelastic neutron scattering and from this to determine the fraction of atoms in the zero-momentum state (the Bose condensate). However, a question hardly ever addressed is how this order parameter is related to the standard picture of superfluid 4 He as a gas of weakly interacting phonon-roton quasiparticles. To ignore the role of the Bose order parameter is equivalent to discussing spin waves in a Heisenberg ferromagnet without mentioning that they are associated with fluctuations of the spontaneous magnetic order (Sz) ^ 0, or discussing phonons in a crystal without reference to the fact that they are vibrations of an underlying lattice. In any modern account of superfluid 4 He, the Bose order parameter must play a central, unifying role. In this book, we show how the field-theoretic analysis of Bose fluids can relate the LandauFeynman quasiparticle picture to the underlying Bose condensate and how one can understand what happens to the "excitations" when liquid 4 He passes through the lambda transition. What is the direct evidence for a Bose condensate in liquid 4 He? Miller, Pines and Nozieres (1962) and Hohenberg and Platzman (1966) first suggested that the condensate density no could be obtained from S(Q,co) data in the high-momentum limit. This stimulated many experimental studies (for a recent review, see Sokol and Snow, 1991) at increasingly high momentum transfers (up to Q ~ 23 A ). To the extent that the impulse approximation is valid, the momentum distribution n(p) of 4 He atoms can be obtained, as discussed in Section 2.3 and Chapter 4.

1.2 Role of the Bose condensate and field-theoretic analysis

The magnitude and temperature dependence of the condensate fraction no(T)/n can then be extracted from n(p). In Chapter 4, we discuss the problems involved in this analysis in some detail. While there is still some uncertainty about the precise magnitudes, recent experimental results convincingly show that no is zero above Tx and builds up to a value (as T is lowered) which is consistent (within about 1%) with the best Monte Carlo computer simulation values at T = 0. A major recent development concerning the Bose condensate is the Monte Carlo study at finite temperatures carried out by Ceperley and Pollock (1986) and Pollock and Ceperley (1987). This uses an imaginarytime path-integral formulation as initiated by Feynman (1953a). Their work is the culmination of a long history of research using Monte Carlo techniques to study liquid 4 He, starting with the estimate of no at T = 0 by McMillan (1965). Ceperley and Pollock found: (a) The specific heat shows the beginnings of the characteristic 2-singularity close to Tx = 2.17 K (see Fig. 1.7). (b) The calculated condensate fraction no(T)/n is essentially zero above Tx and then rapidly builds up below Tx to a value consistent with 9% at T = 0 (see Fig. 1.8). (c) The calculated normal fluid density PN(T) is in excellent agreement with experimental results (see Fig. 1.9). Although the small sample size (64 atoms) leads to finite-size rounding near Tx, these results are a major watershed in our understanding of superfluid 4 He. The successful ab initio calculation of the normal and superfluid densities is especially important. Pollock and Ceperley (1987) started with an exact relation which expresses ps(T) as the difference between the long-wavelength, zero-frequency momentum current-response functions to transverse and longitudinal perturbations. As we discuss in Section 6.1, this fundamental definition of ps(T) = p — PN(T) has been understood since the early sixties, as well as the fact that ps ^= 0 is a direct consequence of a Bose broken symmetry (ip(r)) ^ 0. This generalized definition of ps(T) and PN(T) is not dependent on a weakly interacting quasiparticle picture being valid (as it isn't near Tx) and thus is much more general than Landau's formula in (1.3). The results of Ceperley and Pollock give perhaps the most direct "evidence" that the superfluid transition in liquid 4 He is associated with a Bose broken symmetry. Their work, in conjunction with the experimental determination of the condensate from high-momentum neutron scattering (see Chapter 4), forces us to examine an aspect of superfluid 4 He

Theory of excitations in superfluid 4He: an introduction

16 6 5

4 -

5.3-

Fig. 1.7. PIMC values of the specific heat at SVP in the region near T;. The full line is the experimental data as given in Wilks (1967). The unlabelled arrow shows the superfluid transition temperature TA [Source: Ceperley and Pollock, 1986].

which has been largely untouched - namely, the changing properties as the temperature increases through Tx as the direct consequence of the condensate decreasing in magnitude and vanishing at Tx. In concluding this section, we summarize some highlights of the development of the field-theoretic analysis of a Bose-condensed liquid. Fritz London first understood that the superfluidity of liquid 4 He must have its explanation as a macroscopic quantum effect (see the introduction of the monograph by London, 1950). Unfortunately, the ideal Bose gas (the remarkable properties of which were first delineated by London) does not exhibit the needed long-range phase coherence. London never appreciated that the next step in his programme was in fact given by the work of Bogoliubov (1947) and Penrose (1951), in which the order parameter (tp(r)) formed the microscopic foundation for understanding superfluid 4 He (see Section 3.1). More specifically, the T = 0 study by Bogoliubov (1947) of a weakly interacting dilute Bose gas (WIDBG) was pivotal in that it showed how a Bose condensate in an interacting gas could change the dispersion relation from particle-like to phonon-like at low momentum. As we discussed earlier, this is sufficient to lead to superfluidity (i.e., a finite critical velocity, by Landau's argument). Bogoliubov's work was immediately appreciated by Landau (1949) as an example of a microscopic calculation which exhibited the required

1.2 Role of the Bose condensate and field-theoretic analysis

10 -

8 -

\ o \ \

6 -

\ \

1 \

4 --

2 -

0 0

2 1 Tx

T(K)

Fig. 1.8. Percentage of 4 He atoms with zero momentum at SVP, as given by PIMC calculations. The dashed line is a guide to the eye. The cross is the GFMC ground-state value at T = 0 as calculated by Kalos, Lee, Whitlock and Chester (1981) [Source: Ceperley and Pollock, 1986].

phonon spectrum at low Q. It was far ahead of its time, being one of the first studies of broken symmetry with a symmetry-restoring Goldstone mode (Anderson, 1984). As an historical aside, however, it seems that Bogoliubov's work was largely unappreciated in the West until about 1956, when Lee, Yang and Huang rederived many of the results independently by a different approach (the latter work is nicely reviewed by Huang, 1964). There is not a single reference to Bogoliubov's work in London's well known monograph (London, 1954) or in any of Feynman's papers, and only a single passing reference in Lee, Huang and Yang (1957) in the well known series of papers associated with these authors. Bogoliubov's discussion of a dilute Bose gas was generalized by Beliaev (1958a) using Green's function techniques so as to be applicable to Bose liquids. The next five years or so was a period of intense research developing this approach, the most important papers being those of

Theory of excitations in superfluid AHe: an introduction

•

l.o -

1 5 /

•

;

^0.5

- ZA 1

T(K) Fig. 1.9. PIMC results for the normal-fluid fraction vs. temperature. The full line is the experimentally measured value. The theoretical values were obtained from evaluating the "winding number" (open circles) as well as the current-response functions (solid dots) [Source: Pollock and Ceperley, 1987]. Hugenholtz and Pines (1959), Bogoliubov (1963, 1970), Gavoret and Nozieres (1964) and Hohenberg and Martin (1965). Still useful are the excellent summaries of the implications and accomplishments of the fieldtheoretic approach up to 1965 given by Hohenberg and Martin (1965), Martin (1965), Pines (1965) and Nozieres and Pines (1964, 1990). In particular, these studies gave us an understanding of: (a) How the two-fluid equations are a natural consequence of a Bose broken symmetry, without the need for a well defined quasiparticle spectrum as in Landau's original theory. (b) How the Bose broken symmetry leads to the poles of the singleparticle Green's function G(Q, a>) being the same as the poles of the density correlation function S(Q,a>). In particular, this was exhibited by the tour de force calculation of Gavoret and Nozieres (1964) who explicitly proved by diagrammatic methods that the phonon is the dominant excitation in all correlation functions in the longwavelength, T = 0 limit. This explained, at least in one important limit, why one could measure the elementary excitation spectrum by inelastic neutron scattering, but only in the Bose-condensed phase (this point seems to have been first explicitly noted by Hohenberg

1.2 Role of the Bose condensate and field-theoretic analysis

and Martin, 1965). In normal liquids (including liquid He and liquid 3 He), the density fluctuation modes are not the elementary excitations. An important development was initiated by Ruvalds and Zawadowski (1970), who pointed out that any structure in the two-quasiparticle spectrum would also be coupled into the single-particle Green's function through the effect of the Bose broken symmetry. As a result, the singlequasiparticle spectrum is hybridized with the two-quasiparticle spectrum, with a strength related to the condensate density no. Their work was a natural continuation of the pioneering field-theoretic analysis of Pitaevskii (1959) on the stability of excitations at large Q. It led to a new interpretation of the results shown in Fig. 1.6, in terms of a single-particle excitation (such as given by (1.6)) crossing a relatively dispersionless pair-excitation branch describing two rotons. In this interpretation, the natural continuation of the unrenormalized roton dispersion curve is the free-particle excitation at large wavevectors as in (1.6), rather than as the broad resonance at 2A for Q > 2.5 A" 1 assumed in Fig. 1.3. Ruvalds and Zawadowski emphasized the crucial role that a Bose broken symmetry has on the coupling of the one- and two-excitation spectrum at large co (see Chapter 10), complementing the work of Gavoret and Nozieres (1964) and others at low co. The next major period of research progress after the "golden era" of the early 1960's was the early 1970's when the dielectric formalism (this name arose because the theory was first developed by Ma and Woo (1967) for a charged Bose gas) was used to analyse the structure of the one- and two-particle Green's functions and shows more clearly the role of the Bose order parameter. Key papers were by Ma, Gould and Wong (1971), Wong and Gould (1974, 1976), Griffin and Cheung (1973), and Szepfalusy and Kondor (1974). This work led to: (a) Consistent finite-temperature approximations for both G(Q, co) and 5(Q, co) in terms of irreducible, proper diagrammatic contributions, which allowed one to see how a Bose order parameter leads to mixing of the density fluctuations with the more fundamental single-particle spectrum. (b) A realization of the qualitative differences between the spectrum exhibited by a Bose-condensed gas at T = 0, in which almost all the atoms are in the condensate, and at finite temperatures where the dynamics of the non-condensate atoms play a crucial role. This led to simple models which were more appropriate to compare with

Theory of excitations in superfluid 4He: an introduction

superfluid 4 He than the traditional weakly interacting dilute Bose gas at T = 0 with n0 ^ n. (c) Understanding how the excitations of the superfluid merge with those of the normal phase as the liquid passes through the transition temperature. In the traditional Landau-Feynman quasiparticle picture (Section 1.1), the quasiparticle dispersion relation in superfluid 4 He continues to exist right up to about Q ~ 2.5 A" 1 (see Fig. 1.3). That is, it is assumed that there is a smooth transition from the low-Q phonon to the high-Q maxon-roton part of the spectrum. In a scenario developed by Glyde and Griffin (1990) based on the dielectric formalism, the (zero sound) phonon branch at low Q broadens substantially for Q ^ 1 A" 1 and effectively disappears. In contrast, the sharp component which is observed in S(Q,co) at Q ^ lA" 1 is viewed as a single-particle excitation branch which has been mixed into the density fluctuation spectrum via the Bose condensate. In this new interpretation of the excitations in superfluid 4 He, the existence of the Bose broken symmetry leads to a hybridization of the single-particle and collective density excitations, with the observed phonon-maxon-roton spectrum being the final result. This scenario is developed at length in Chapter 7. The condensate-induced hybridization of single-particle and density excitations makes S(Q,co) for superfluid 4 He especially interesting from the point of view of neutron scattering, since it explains why one can probe the elementary excitations through a study of S(Q,OJ). The microscopic theory needed to deal with this condensate-induced hybridization is reviewed in Chapters 5 and 6. At a more phenomenological level, of course, the excited states in superfluid 4 He at T = 0 are well described by variational many-body wavefunctions first introduced by Feynman (1954) and developed by many others since. However, as we discuss in Section 9.1, the underlying physics involved in such variational wavefunctions is not easily extracted, especially as concerns the role of the condensate. Moreover, it is very difficult to use this wavefunction approach at finite temperatures, when damping of excitations is important.

1.3 Outline of book Few substances have been investigated as thoroughly as superfluid 4 He. It is a vast subject and any monograph must have focus. Our major theme will be the difference between single-particle modes (elementary

1.3 Outline of book

excitations) and density fluctuations (collective modes) and how these are intertwined in superfluid 4 He due to the Bose broken symmetry. As should be clear from Section 1.2, we put emphasis on the fieldtheoretic approach which has led to what we call the dielectric formalism (Ma and Woo, 1967). The recent scenario of Glyde and Griffin (1990) concerning the nature of the phonon-maxon-roton excitations is a natural development within this dielectric formulation, elements of which have been developing over the last 30 years. It starts with the premise that superfluid 4 He has a Bose condensate and that this must underlie the unique nature and role of the excitations in this liquid. In this book, we will always work with a spatially uniform Bose order parameter. In particular, we ignore all effects related to the appearance of quantized vortex filaments or the associated vortex oscillations. For further discussion of this topic, see Nozieres and Pines (1964, 1990), Lifsh*tz and Pitaevskii (1980), and the recent monograph by Donnelly (1991). The dielectric formalism can be applied to situations involving vorticity but this must be left to future studies. The dielectric formalism is a well defined many-body formalism which allows detailed calculations in a consistent manner. It can be used at any temperature (above and below Tx) and is valid in the hydrodynamic two-fluid region as well as the high-energy collisionless region probed by neutrons. Needless to say, controlled microscopic calculations are difficult once we go away from a weakly interacting dilute Bose gas. However such model calculations (at finite T) do indicate the general structure imposed by the Bose broken symmetry on the various response functions in any Bose-condensed fluid. This structure must be allowed for in variational and phenomenological approaches if one is to come to grips with the underlying Bose broken symmetry of superfluid 4 He. Because it is this aspect which has been crucial in recent research in understanding the physical basis of the excitations in superfluid 4 He, we also emphasize the difference between the dynamic structure factor above and below the superfluid transition temperature. Related to this, we give a detailed analysis of the neutron-scattering data as a function of the temperature. This is in contrast to most of the theoretical literature on superfluid 4 He, which has concentrated on either the low-temperature region (< 1 K) or the extreme critical region (within a few mK of the superfluid transition). Using the renormalization group formalism and the concept of universality classes, the static and dynamic behaviour of superfluid 4 He near Tx appears to be very well understood in terms of a fluid with a two-

Theory of excitations in superfluid 4He: an introduction

component order parameter. We will not be concerned with this subject in the present book, but note that it gives additional powerful evidence that the superfluid phase is indeed associated with a Bose broken symmetry. Neither will we discuss the extensive work on the phonon dispersion relation and damping at very low T ( 0.6 K), where one can work with phenomenological model Hamiltonians describing well defined weakly interacting quasiparticles. As discussed by Khalatnikov (1965), in this region one can calculate the thermodynamic and transport properties of superfluid 4 He with considerable success, as well as the stability of excitations due to spontaneous decay processes allowed by kinematics. We now briefly outline the contents of each chapter. As we have mentioned, Chapter 2 gives a brief summary of the neutron-scattering data for <S(Q, CD) in superfluid and normal liquid 4 He. We also introduce the impulse approximation for S(Q, co) valid at large momentum transfers, which will be needed in Chapter 4. In Chapter 3, we review the concept of Bose broken symmetry and how it modifies the nature of excitations. We also introduce a major theme of this book: namely, the difference between a Bose-condensed gas at low T and a Bose-condensed liquid like superfluid 4 He. In the latter, the condensate fraction is never more than about 10%, and thus it is crucial to include density fluctuations arising from the non-condensate atoms. The latter are neglected in the simple Bogoliubov approximation discussed in most textbooks. We discuss the atomic momentum distribution n(p) in liquid 4 He in Chapter 4, both theoretically and experimentally. High-momentum neutron-scattering studies in recent years (especially by Sokol and coworkers) have given very strong confirmation of the Monte Carlo estimates of the condensate fraction. We also give a detailed review of the condensateinduced changes in n(p) at low but finite momentum. Chapter 5 is the heart of this book. We introduce the diagrammatic analysis in terms of irreducible, proper contributions and show how the single-particle Beliaev Green's functions Ga^(Q,co) and the densityresponse function Xnn(Q,o)) are coupled via the Bose order parameter. This formalism is illustrated using simple Bose-gas model calculations and we show it can lead to a coupling of the single-particle (SP) pole of G(Q,co) with the collective zero sound (ZS) pole of %nn(Q,co) when nQ =£ 0. The related analysis (at T = 0) of Gavoret and Nozieres (1964) is also summarized. We argue that the dielectric formalism is the method of choice for future microscopic studies of the dynamics of Bose-condensed fluids.

1.3 Outline of book

The difference between low-frequency hydrodynamic and high-frequency collisionless phonon excitations is the main subject of Chapter 6. It is important that the field-theoretic analysis we use to discuss the collisionless region (as probed by neutrons) also leads to hydrodynamic correlation functions which exhibit the characteristic behaviour associated with superfluidity (as probed by Brillouin light scattering and ultrasonic studies). We compare first and second sound in the low-co region with phonon excitations in the large-co region, both in a Bose gas and in Bose liquid. In Section 6.3, we also give a critical analysis of the Gavoret-Nozieres results in the limit of g,co —> 0. Somewhat surprisingly, the physical interpretation of the low-energy phonon region in a Bose-condensed liquid is more complicated than the high-energy roton region. In Chapter 7, we review high-resolution neutron-scattering data for S(Q,co), with emphasis on how the line shape varies with temperature from 1 K to well above Tx. Recent experiments at ILL (Talbot et al, 1988; Stirling and Glyde, 1990; Andersen, Stirling et al, 1991) have confirmed earlier work at Chalk River (Woods, 1965; Woods and Svensson, 1978) that the effect of temperature is quite different on the phonon and maxon-roton excitations as one goes from the superfluid to normal phase. The dielectric formalism naturally leads to a picture in which the elementary excitations below Tx involve the hybridization of a low-Q zero sound (ZS) phonon with a high-Q high-energy single-particle (SP) maxon-roton. This scenario of Glyde and Griffin (1990) gives a plausible connection between the underlying Bose condensate and the phonon-maxon-roton dispersion curve. The well known /-sum rule for S(Q, co) plays a very important role in analysing neutron-scattering data as well as constraining phenomenological theories. In Chapter 8, we discuss similar sum rules which are unique to Bose-condensed fluids (Wagner, 1966; Wong and Gould, 1974; Stringari, 1992). We also critically review the Woods-Svensson (1978) two-component formula for S(Q, co). Their ansatz is important historically as the first attempt to describe the fact that the maxon-roton has a weight in S(Q,co) which vanishes at Tx. In Chapter 9, we review three alternative theoretical approaches to describing the excitations in superfluid 4 He, mainly from the point of view of making some contact with the dielectric formalism. These are variational many-body wavefunctions (the correlated-basis-function method), the polarization potential approach of Pines and coworkers, and the memory function formalism. The role of the Bose order parameter is

Theory of excitations in superfluid 4He: an introduction

somewhat implicit in all these techniques (which so far have been mainly developed a t T - 0 ) . In Chapter 10, we discuss the high-energy multiparticle spectrum (including the two-roton bound state) which shows up in the neutronscattering intensity but which can also be studied more directly by Raman light scattering. We review the condensate-induced hybridization between the two-particle and the single-particle spectrum, as first studied by Pitaevskii (1959) and later by Ruvalds and Zawadowski (1970). In Chapter 11, we review the distinction between the poles of the displacement correlation function (phonons) and the density fluctuations described by S(Q,co) in highly anharmonic crystals (Ambegaokar, Conway and Baym, 1965). We call attention to certain similarities between the modern theory of excitations in quantum crystals like solid 4He and the dielectric formalism used to describe superfluid 4He, and suggest that this may be the basis for understanding the striking similarity of S(Q,co) data in these two different phases. Finally, in Chapter 12, we summarize and assess the new interpretation of the phonon-maxon-roton dispersion curve based on a condensateinduced hybridization of two different excitation branches. We discuss the history of this scenario and some of its implications, for both normal and superfluid liquid 4He. We briefly sketch the dielectric formalism results for superfluid 3 He- 4 He mixtures and conclude with several suggestions for future work.

Units used in this book In neutron-scattering data for S(Q,co) which we quote, energies are given interchangeably in milli-electron-volts, kelvins and terahertz. The conversion factors are 1 THz = 48.0 K = 4.14 meV, 10 K = 0.208 THz = 0.863 meV, 1 meV= 0.24 THz = 11.6 K. In general, we set h = 1 but occasionally we insert it in final formulas or when it will make the discussion more physical.

Dynamic response of Helium atoms to thermal neutrons

Much of this book is concerned with the density fluctuation spectrum of superfluid 4 He. As background, in this chapter, we review some standard material concerning the dynamic structure factor S(Q,a>) and give an introductory summary of the extensive inelastic neutron-scattering data on liquid 4 He (see also Section 7.1). A systematic formulation of neutron scattering as a microscopic probe of atomic motions in condensed phases was first developed in the early 1950's by Placzek and van Hove. This led to the epochal work of van Hove (1954), who showed that the inelastic scattering cross-section (in the first Born approximation) of thermal neutrons is directly related to the Fourier transform of correlation functions involving local operators at two different space-time points. In liquid 4 He, the most important of these are the coherent S(Q,co) and incoherent S[nc(Q9co) dynamic structure factors. In Section 2.1, we define these correlation functions and discuss the symmetry relations and rigorous "sum rules" which these functions satisfy. In Section 2.2, we discuss the characteristic features exhibited by S(Q,a>) in superfluid 4 He and their traditional interpretations. In Section 2.3, we consider the large Q and a> limit of 5(Q,co), and introduce the basic physics of the so-called "impulse approximation" (IA). 2.1 Response functions: general properties and sum rules Neutrons interact only with the nucleus of an 4 He atom, which has zero spin. As discussed in standard texts (see Marshall and Lovesey, 1971), a low-energy (thermal) neutron impinging on a sample of liquid Helium sees a total scattering potential ^ r ) , 25

(2.1)

Dynamic response of Helium atoms to thermal neutrons

where r ; is the position operator of the j t h 4 He nucleus (mass m) and b is the "bound" scattering length describing the neutron-Helium-nucleus interaction at the thermal energies of interest. Treating this potential in the first Born approximation, one finds that the double differential scattering cross-section per target atom is given by (van Hove, 1954) dndE

(2 .2)

,),

h i.

where the dynamic structure factor is defined as

S(Q,co) = -±- f°dt eiM

(2.3)

2nN J_ao Here ko(ki) is the incoming (outgoing) neutron wavevector and HQ = frk0 -

„

i ,

^k\ >

are, respectively, the momentum and energy transfer to the fluid which take place in the scattering process; also

p(Q) = J V " ^ = p+(-Q)

(2.5)

is the Fourier component of the 4He number-density operator (r-r,) .

(2.6)

It is sometimes convenient to introduce the time-dependent "intermediate" scattering function Scoh(Q, t) = i (p(Q, t)p(-Q, 0)) .

(2.7)

At high momentum transfers, S(Q,o) in (2.3) may be approximated by the "incoherent" structure factor Sinc(Q,co) = 1 - r 27C J-oo

dt j» B

(2>8)

This describes the contributions to S(Q,a>) from the terms k = j . Sinc(Q,&>) m (2.8) is the same for all atoms (i.e., it is independent of the label j). This correlation function describes the motion of a single atom and is thus of independent interest in the theory of liquids (Hansen and McDonald, 1986).

2.1 Response functions: general properties and sum rules

The static or instantaneous structure factor is defined as S(Q) = S(Q, t = 0) = — 2^(^

Tie

'Tj) -

(2-9)

This is related to the static pair correlation function g(r) through the expression (p. 371, Marshall and Lovesey, 1971) S(Q) = 1l ++ nny/ ^ Q r [ g ( r ) - 1] ,

(2.10)

where g(r) describes the instantaneous correlations between He atoms. It is proportional to the probability that, if there is an atom at the origin, then there is simultaneously another atom at position r. Measurements of S(Q) can be deconvoluted using (2.10) to give g(r). An equivalent definition of S(Q) is S(Q)= = H dcoS(Q9a>) ,

(2.11)

J—

the dynamic structure factor integrated over all energy transfers. We also note explicitly that, with the definitions (2.8) and (2.9), (2.12) dcoSinc(Q,co) = l . ) This is often used to determine how high the momentum transfer must be before S(Q, co) can be approximated by Sinc(Q, co) in (2.8). The oscillations observed in S(Q) in (2.9) are associated with the terms i =£ j . In liquid 4 He, these oscillations have disappeared for Q ^ 8 A" 1 (Svensson et al, 1980) and this is interpreted to mean that the i ^ j terms are also not important in S(Q,co) for large Q. We can express S(Q,co) explicitly in terms of the exact many-body eigenstates \n) of the effective Hamiltonian Hr = H — fiN, where \i is the chemical potential. Defining

we obtain (see pp. 148ff, Mahan, 1990) S(Q,co) = — 2^ —^-\(™\P

(Q)|n)l S[co - (Em-En)]

(2.14)

n,m

Here Z is the grand canonical partition function and /? = l/Zc^T. In (2.14), the states \n) and \m) have the same number of atoms since the effect of the operator p+(Q) does not involve changing the number of atoms (Nn = Nm). At T = 0, the sum over |n) reduces to the ground

Dynamic response of Helium atoms to thermal neutrons

state corresponding to the correct value of the chemical potential \i. The exact spectral representation (2.14) is useful in deriving various general properties of S(Q,co). In addition, it emphasizes that S(Q,co) involves the contribution of all energetically possible transitions |n) —> \m), each weighted by the absolute square of the matrix element (ra|p + (Q)|n). Physically, this says that the density operator p + (Q) acting on a state \n) must produce a state with a. finite overlap with the many-body state \m) if these states are to be important in the summation involved in (2.14). In the isotropic fluid case, one can introduce operators which create or destroy atoms of specified momentum. In this second quantized language, (2.5) is equivalent to +

J X .

(2.15)

This form naturally leads to an interpretation in terms of destroying an atom of momentum h\ and creating one with momentum hk + ftQ, that is, a particle (k + Q)— hole (k) excitation. This language is familiar in describing interacting Fermi systems (Pines and Nozieres, 1966). It is equally useful in discussing density fluctuations in Bose fluids and we shall often use it in this book. Thus <S(Q,co) in (2.3) may be viewed as related to the propagator of particle-hole (p-h) pairs created at t = 0 and destroyed at time t. The elementary excitations are described by single-particle Green's functions. It is important to understand the distinction between these functions and the density correlation function in (2.3) which is measured by inelastic neutron scattering. A generic example of a single-particle correlation function is given by (see Section 3.2) A(Q,t) = (aQ(t)a+(0)) .

(2.16)

In contrast to S(Q,t) defined in (2.7), this correlation function describes how a single atom of momentum hQ propagates over the time interval t. In terms of an exact eigenstate representation used in deriving (2.14), one obtains (see p. 142 of Lifsh*tz and Pitaevskii, 1980) ^

e~PE'»

A(Q,co) = YJ-^-\(Mtyn)\2S[cD -n-(E'm-Efn)\

(2.17)

m,n

Clearly only states \m) which contain one more atom than \n) can contribute to the summation in (2.17): Nm = Nn + 1. Thus the excited states \m) which contribute to A(Q,co) will, in general, be different from those which contribute significantly to S(Q,a>) in (2.14). In more technical

2.1 Response functions: general properties and sum rules

language, the single-particle spectrum does not usually overlap strongly with the p-h or density fluctuation spectrum. However, we shall see (Chapters 3 and 5) that in the presence of a Bose broken symmetry, these two functions are closely related to each other. Thus in superfluid 4He, we will be interested in v4(Q, co) and S(Q, co) as inter-related correlation functions. At high energies, these functions are in turn coupled into the two-particle spectrum discussed in Chapter 10. In explicit calculations of the correlation function <S(Q,co), it is convenient to work with the associated density-response function ^

(2.18)

where 6(t) is the step function and Q is the sample volume. The square bracket is the commutator [A,B] = AB — BA. xm(Q,t) is interchangeably called a retarded Green's function, a dynamic susceptibility or a response function. As discussed in standard texts, many-body techniques most naturally give the Fourier components ^nn(Q,co). At finite temperatures, one works with imaginary times in the interval 0 < T < /? and discrete Matsubara imaginary frequencies icon = ilnksT\ n = 0,±1,±2,... (see Chapter 3 of Mahan, 1990). Evaluating (2.18) in the exact eigenstate representation used in (2.14), one easily finds (n is a positive infinitesimal) 1 ^ p

Xnn(Q,« + in) = s Z -^\^P mn)\\\.n_{Elm_ln)

(2.19)

and hence Im Xnn(Q, co + irj) = -nn(l - + it,) , nn where N(co) is the Bose distribution

(2.20b)

An important relation between negative and positive energy transfers immediately follows from (2.14), namely

J^l

(2.22a)

Dynamic response of Helium atoms to thermal neutrons

This is called the principle of detailed balance. In an isotropic liquid, S(Q,co) depends only on the magnitude of Q. In this case, it follows from (2.22a) and (2.20a) that Im Xnn(Q, co + in) = - I m Xm(Q, -co + in) .

(2.22b)

In approximate calculations, it is important that (2.22) be satisfied. According to (2.22a), the ratio of the intensity for an energy loss of hco by the system to the scattered neutron (anti-Stokes) to the corresponding energy gain of hco by the system (Stokes) is given by exp(—jSfao)- At T = 0, the anti-Stokes component vanishes, reflecting the fact that the system is in its ground state and thus can only gain energy. To avoid later confusion, note that the single-particle spectral density A(Q,co) in (2.17) does not satisfy (2.22a). There is no general relation between A(Q,co) and A(Q, —co). In the low frequency classical region (hco kBT. The dynamic structure factor satisfies various sum rules or constraints which involve frequency moments defined by /•OO

(co11) = /

dcoconS(Q,co) .

(2.23)

J — OO

For n > 1, the main contribution to these moments is from large co, which means that they are related to the sma//-time behaviour of <S(Q, 0The well known longitudinal /-sum rule (Placzek, 1952) (co) = /

dcocoS(Q,co) = y - .

J—oo

(2.24)

is derived, for example, on p. 365 of Lifsh*tz and Pitaevskii (1980). Higherorder moment sum rules can also be derived in a similar manner (Rahman et al.9 1962; Puff, 1965). Such sum rules are very useful constraints in experimental studies since the data can be used to compute (con) to see if the results satisfy the sum rules. Sum rules unique to Bose-condensed fluids are discussed in Sections 6.1 and 8.1. The frequency moments in (2.23) are given directly by a high-frequency expansion of #nn(Q,co). We recall the spectral relation (n

Xnn(Q,C0) = -

f00 dCQ'ImXnn(Q,COf) —

(D_CD!

(2'25)

a result which may be explicitly verified using (2.19). Inserting (2.20a)

2.2 Density fluctuation spectrum of superfluid 4He

into (2.25) and using (2.22), one obtains poo

XnniQ,(0)

'S!(f*\

2nJ_Ja>>-r-^

= ^ ( w ) + ^ ( a ; 3 ) + ... for co -+oo . CO1

(2.26)

CO*

This high-frequency expansion is a useful constraint in model calculations of x(Q, co), as we shall see in later chapters. The zero-frequency limit of (2.24) gives

Xnn(Q,co = 0) = -Inp

dco'^^

(2.27)

This inverse frequency moment clearly will be dominated by the lowfrequency behaviour of S(Q,co), i.e., the long-time dynamics of the 4 He atoms. The low co and Q region is usually referred to as the hydrodynamic domain (Section 6.2). One can relate the long-wavelength limit of the zero-frequency response functions to thermodynamic derivatives. In particular, one finds (see p. 136 of Pines and Nozieres, 1966)

Um Znn (Q,c = 0) = - £

(2.28)

where c is the isothermal sound velocity. This result is known as the compressibility sum rule. In connection with the incoherent dynamic structure factor (2.8), it is also useful to define central frequency moments do)(o)-o)R)nSinc(Q,co)

Mn(Q)=

J—oo

These are centred around the free-atom recoil frequency COR = HQ2/2m and hence are useful at high Q when Sinc(Q,co) is peaked near COR. These central moments satisfy various sum rules (Rahman et al, 1962). The first central moment sum rule M\ (Q) = 0 is easily seen to be equivalent to the /-sum rule for S[nc(Q,co) when (2.12) is taken into account.

2.2 Density fluctuation spectrum of superfluid 4 He In this section, we expand on our introductory remarks in Section 1.1 concerning the density fluctuation spectrum observed in superfluid 4 He. A more detailed, critical analysis of recent neutron-scattering data is given in Chapter 7.

Dynamic response of Helium atoms to thermal neutrons

As orientation, we first discuss the density-response function of a non-interacting Bose gas. In this case, (2.19) reduces to

where ek = k2 /2m is the free-particle energy and Q is the sample volume. This expression is derived in most many-body texts (see, for example, Mahan, 1990; Fetter and Walecka, 1971; Nozieres and Pines, 1964, 1990) as well as in Section 5.2 as a limiting case. Inserting (2.29) into (2.20b), we obtain that the dynamic structure factor S°(Q,a>) for an ideal Bose gas can be written in several equivalent forms:

= l- [N(CO) + l] i £ [jV(k) - JV(k + Q)] S (co - [sk+Q - ek])

(2.30a)

= -n [N(CO) + l] ^ ] [ > ( k ) {8(a> - [ek+Q-ek])-S(co k

=^Z

k k

[ek+Q-ek])} (2.30b)

i V ( k ) l N{k + Q ) 5

] (^ -

where N(k) = N(8k — /i). In the last step, we have used the identity satisfied by the Bose distribution, l+N(co) = -N(-co)

(2.31)

We call attention to the appearance of the statistical factors in (2.30c) associated with creating and destroying Bosons. Clearly the first term in (2.30fo) is the Stokes and the second term is the anti-Stokes term. One can also easily verify that the expressions in (2.30) satisfy the /-sum rule (2.24) as well as the detailed-balance condition (2.22). In the special case of a Bose-condensed gas, (2.30c) reduces to

S°(Q,o>) =

N{k+Q) s

^ ~[8k+Q ~Sk]^'

where no is the density of atoms in the k = 0 state (the Bose condensate). The second line in (2.32) describes scattering of atoms from state k to k + Q. As with an ideal Fermi gas, the latter is a broad continuum arising from scattering of incoherent particle-hole (p-h) excitations. In contrast, the first line of (2.32) describes density fluctuations associated with atoms

(Z32)

2.2 Density fluctuation spectrum of superfluid 4He

coming out of (or going into) the Bose condensate "reservoir". This is a highly coherent process and gives rise to sharp single-particle peaks at ±SQ, with an overall weight proportional to the condensate fraction no/n. With these results for an ideal Bose gas as background, we now turn to liquid 4 He. S(Q,co) has been measured by neutron scattering over a wide range of momentum transfers (0.1 < Q 70 K). The general features of this additional scattering are shown in Fig. 1.6. It is customary to take S(Q,co) as the sum of a quasiparticle part Si and a multiparticle part Sn, S(Q,G>)

SI(Q,G>)

+ Sn(Q,G>) .

(2.33)

The existence of these two distinct contributions was first discussed (at T = 0) by Miller, Pines and Nozieres (1962) in a landmark paper. It is important to emphasize that writing down the two-component expression (2.33) for S(Q,co) must be justified theoretically and, as we shall see in Chapter 7, its use prejudices what we mean by a "quasiparticle." At low temperatures, an empirical decomposition into these two parts is easy to make by "eye", if the single-quasiparticle and broad multiparticle background are well separated in energy. As the temperature increases, the quasiparticle resonance broadens and a precise separation of Si and Sn becomes more difficult. At temperatures T ^ 1.7 K, the quasiparticle resonance at CDQ is sharp

Dynamic response of Helium atoms to thermal neutrons 1

1.0 - + 0.8 _

TACS

0.6 -

9 0.4-

0.2

0 /

A /\ -

RCS

j RCS f

;

1.0

2.0

3.0

4.0

Wavevector Q (A"1)

Fig. 2.1. Intensity of the quasiparticle peak in S(Q,a>) as a function of g, at 1.1 K and SVR This is a compilation of data using different spectrometers [Source: Cowley and Woods, 1971].

and often fitted by (see Appendix of Talbot et a/., 1988)

Q (2.34) where the quasiparticle half-width at half maximum (HWHM) is denoted by TQ. If the damping is negligible, as at low temperature, (2.34) simplifies to Si(Q9co) = [N(co) + 1] Z ( 0 [3(co -coQ)-d(co

+ coQ)]

(2.35)

Here Z(Q) gives the weight of this excitation in S(Q,co), with the associated static structure factor being Si(Q)=Z(e)[2JV(a) Q )

(2.36)

The double Lorentzian in (2.34) is consistent with detailed balance (2.22); in much of the older literature, only the first (or Stokes) term in (2.34) is used in fitting data. In Fig. 2.1, we plot the relative weight Z(Q) of the so-called "one-phonon" part of S(Q,co) at low temperatures. We note

2.2 Density fluctuation spectrum of superfluid 4He 1

1 '

24 —

—

20 —

—

16 —

•

y^—"^NL

—

12 —

1/ f

8 _ 4

25.3 atm

SVPX^ **

—

f ,

•

1 . 1 , 1 , 1 , 0.4

0.8

1.2

1.6

2.0

1 , 2.4

| 2.8

Momentum change Q (A" )

Fig. 2.2. The effect of pressure on the quasiparticle dispersion curve at T = 1.1 K. The dashed lines give the measured sound velocity [Source: Woods and Cowley, 1973].

that the full temperature dependence of the static structure factor S(Q) has been measured with high precision (Svensson et al, 1980). To the extent that Si(Q,a>) can be unambiguously extracted from the full S(Q,co), the quasiparticle peak position and width can be obtained from neutron-scattering data. At T < 1.2 K, the quasiparticle peak width is extremely small and is mainly determined by kinematically restricted decay processes since there are few thermally excited quasiparticles. At T < 1 K, the phonon width is found to decrease abruptly for Q larger than Qc < 0.55 A" 1 . This is consistent with the fact that phonon decay into two phonons is possible only because of (pressure-dependent) anomalous dispersion in the region Q < Qc (for further discussion see Maris, 1977, and Stirling, 1983). At higher temperatures, the quasiparticle width 2TQ is apparently mainly due to scattering from thermally excited rotons since it scales roughly with NR(T), the number of rotons. A width having this temperature dependence was first predicted by Landau and Khalatnikov (1949) for rotons (Q ~ QR) but it has been found

Dynamic response of Helium atoms to thermal neutrons

0.03

0.02 -

3 O 0.01 -

Fig. 2.3. Smoothed scattering data vs. frequency, for Q = 0.8 A"1 and SVP, at several temperatures [Source: Woods and Svensson, unpublished; Griffin, 1987]. experimentally to be approximately true for all wavevectors (Cowley and Woods, 1971; Woods and Svensson, 1978; Mezei and Stirling, 1983). There has been a continuing effort studying the dynamics of superfluid 4 He under pressure. In Fig. 2.2, we show the pressure-induced changes to the phonon-maxon-roton curve at low temperatures. As we shall see in Chapters 7 and 10, studies of S(Q,co) as a function of pressure have played an important role in disentangling various contributions and also in understanding the role of the Bose condensate. There has been increasing interest in how the quasiparticle line shape changes in S(Q,co) as a function of the temperature. For Q = 0.4 A" 1 (the phonon region), recent high-resolution studies (Stirling and Glyde, 1990) have shown that while the width of the phonon peak steadily increases, there is no qualitative change as we go from below to above Tx. In particular, there is little change in the peak position (see Fig. 1.4). This was first noted in the pioneering study by Woods (1965b). In the rotonmaxon wave vector region (0.8 ^ 2 ^ 2.4 A" 1 ), the situation is quite

2.2 Density fluctuation spectrum of superfluid 4He

600

0.4 0.6 v (THz)

-0.2

200

1 15 °

m*o

• 7=1.90K . T=2.05K *T=2.96K oT=3.94K .

ensit;

^100 5 50 1 •

0 50 -0.2

O _ • — — «_

0.2

-. — — _ _

0.4 0.6 v (THz)

_ - - .

0.8

—

T ^ - -

1.0

1.2

Fig. 2.4. Neutron-scattering intensity vs. frequency, for Q = 1.13 A" 1 and 20 atm pressure. The top panel shows superfluid-phase data below TA while the bottom panel shows data from Tk (=1.928 K) up to 3.94 K [Source: Talbot, Glyde, Stirling and Svensson, 1988].

Dynamic response of Helium atoms to thermal neutrons

Fig. 2.5. Smoothed scattering intensity given as a contour map of the energy transfer and temperature, for Q = 1.3 A" 1 and SVP. The maxon peak is on a reduced scale [Source: Andersen, Stirling et a/., 1991].

different, as first observed by Woods and Svensson (1978). Above Tx, S(Q,co) exhibits a broad distribution with a width which increases with wavevector Q but whose general shape is fairly temperature-independent for Tx < T < 3 K. As we go below Tx, however, a sharper component seems to appear, sitting on this broad background (Fig. 2.3). As the temperature decreases, the width of this peak rapidly decreases while its weight increases. The data of Woods and Svensson (1978) gave the first evidence that the roton-maxon quasiparticle has weight in S(Q, co) only below Tx, in contrast to the phonon branch. Recent high-resolution data confirm these observations (Talbot, Glyde, Stirling and Svensson, 1988; Stirling and Glyde, 1990). The high-pressure data in Fig. 2.4 probably gives the most striking evidence. In Figs. 2.5 and 2.6, we show some recent time-of-flight high-resolution ILL data giving the scattering intensity as a contour map of the frequency and temperature at two values of Q. Such plots summarize the huge amount of detailed information which is now available. They provide a real challenge to our understanding of superfluid 4 He and have been a major stimulus for the present monograph.

2.3 High momentum transfer and the impulse approximation

Fig. 2.6. Smoothed scattering intensity plotted as in Fig. 2.5, for Q = 2.0 A and SVP [Source: Andersen, Stirling et al, 1991].

2.3 High momentum transfer and the impulse approximation

For wavevector Q ^ 2.5 A" 1 , the quasiparticle dispersion relation seems to flatten out abruptly as shown in Fig. 1.3, saturating at about twice the roton energy 2A for large Q. There is also a rapid loss in intensity (Fig. 2.1). As can be seen from Figs. 2.7 and 2.8, for Q larger than 2.5 A" 1 , the main contribution to the scattering intensity is identified with the multiparticle background Su. In the high-momentum region Q ^ 3.5 A" 1 , S(Q,a>) can be increasingly well described by an expression similar to that of a non-interacting gas of Bosons given by (2.32), with the Doppler-broadened peak centred at free-atom energy Q2/2m but with a width determined by the momentum distribution n(p) of the 4 He atoms. To the extent that this "impulse approximation" for S(Q,co) holds, the momentum distribution n(p) of the 4 He atoms can be extracted (see Fig. 2.9) at various temperatures (Sears, Svensson, Martel and Woods, 1982). In turn, this momentum distribution can be used to obtain information about the fraction of atoms no in the zero-momentum state (the Bose condensate).

Dynamic response of Helium atoms to thermal neutrons 800 600 400 1.88 A"1

200 0 300

2.44 A-1

200 100 0

120

2.89 A"1 80 40 0 80 40 -10

Neutron energy loss (K)

Fig. 2.7. Scattering intensity vs. energy for intermediate momentum transfers, at 1.6 K and SVR For more recent high-resolution data, see Figs. 7.19 and 7.20 [Source: Woods, 1965a].

It is convenient to introduce here the basic physics which leads to the impulse approximation (IA). The derivation is done in terms of how a specific 4 He atom moves (recoils) over small times (see pp. 809ff of Mahan, 1990). An alternative (but physically equivalent) derivation in frequency space is given in Section 4.2. Scattering at high momentum transfers requires a high-momentum neutron. Thus the incident neutron wavelength must be short, even relative to the interatomic spacing. As

2.3 High momentum transfer and the impulse approximation

Q (A-1) Fig. 2.8. The wavevector dependence of the quasiparticle weight Z(Q) and the full static structure S(Q), at 1.1 K and SVP [Source: Cowley and Woods, 1971].

a result, the neutron can interact with only a single Helium nucleus and consequently we expect S(Q,co) in the scattering intensity (2.2) to be well approximated by Sinc(Q,co) given in (2.8). As we noted after (2.12), this high-g region is reached for Q ^ 8 A" 1 in liquid 4 He. At large g, the scattering time TS is very short and the scattered atom can travel only a short distance within this time. This suggests an approximation in which the potential energy of a struck atom does not change appreciably during this scattering time. It is convenient to recast the incoherent intermediate scattering function in terms of a time-ordered correlation function (Rahman, Singwi and Sjolander, 1962)

(2.37)

where COR = Q2/2m is the kinetic energy of the recoiling 4 He atom. If the important values of t are small, we can approximate p ; (t') in the exponent by its initial value py(0) = p, in which case (2.37) immediately

Dynamic response of Helium atoms to thermal neutrons 0.20

0.15

•

LOOK

2.12 K

2.27 K

° o

4.27 K

0.10

3. 8

0.05

0.00

Fig. 2.9. The atomic momentum distribution in the normal and superfluid phases of liquid 4 He. These results are obtained from an analysis of high-momentum neutron scattering using a method discussed in Section 4.3 [Source: Sears, Svensson, Martel and Woods, 1982].

reduces to (2.38) Here the average is over the thermal equilibrium momentum distribution rc(p) of the 4 He atoms and thus the Fourier transform of (2.38) leads directly to

(2.39) where rc(p) is normalized to unity. This result is the impulse or independent-particle approximation (IA). In this limit, 5(Q,co) is seen to depend only on the equilibrium atomic momentum distribution. Further insight into the physics of the IA result (2.39) as a small-time

2.3 High momentum transfer and the impulse approximation

approximation can be obtained by expanding (2.38) and keeping only terms of order t2. This gives

= *>-r2/T' .

(2.40)

In the last line, we have introduced an explicit definition of the scattering time TS as the relaxation time of SIA(Q, t). We see that TS OC 1/g and thus the IA should become a better and better approximation as Q increases. Morever we note that if (2.40) is valid, 5IA(Q, CO) in (2.39) is a Gaussian centred at COR, SIA(Q,O>)

- -L=e-«°-2

(2.41)

where a2 = 2/T2S. Parenthetically, we note that if the atomic momentum distribution rc(p) were Gaussian, (2.40) and (2.41) would be exact. This can be proven by expanding \n(ex) and noting that all cumulants higher than the second vanish for a Gaussian average. This is the reason why S(Q,co) is often taken to be the Gaussian form (2.41) when analysing high-momentum neutron-scattering data (Sokol, 1987). One can show that SIA(Q,CD) in (2.39) does not depend on Q and co separately but only on the combination (West, 1975; Gersch and Rodriguez, 1973; Sears, 1985) (^\

(2.42)

This scaling variable Y is very useful in deep-inelastic scattering studies. One can easily carry out the angular integration in (2.39) to obtain ^

(2.43)

where the "Compton profile" J(Y) for the IA is given by = -6(Y) + r dp pn(p) . (2.44) n J\Y\ (We have explicitly allowed for a condensate at p = 0 in n(p).) Such Compton profiles are commonly used in the study of electronic systems as well as in nuclear and particle physics (for further references, see Silver and Sokol, 1989). The fact that the IA Compton profile depends only on Y ("Y -scaling") can be traced back to the fact that the peak position in 2 SIA(Q?CO) is proportional to Q while its width goes as Q. The Compton JIA(Y)

Dynamic response of Helium atoms to thermal neutrons 0.6

Fig. 2.10. The measured Compton profile J(Y,Q) vs. the scaling variable Y defined in (2.42), for momentum transfers Q = 7 and 12 A"1 (at T = 1.0 K) and Q = 24 A"1 (at T = 0.32 K). The results illustrate 7-scaling behaviour [Source: Sosnick, Snow, Silver and Sokol, 1991]. profile of liquid 4 He does exhibit 7-scaling fairly well (see Fig. 2.10), which is usually viewed as evidence that the IA is valid. However, we remember that Y -scaling will also occur as long as corrections to the IA are only functions of Y. It is clear that one needs to understand how large Q must be for the IA to be a sufficiently valid approximation if we use (2.39) to "extract" accurate results for n(p). As a qualitative way of discussing corrections to the IA, one might expect that more generally, (2.39) will be replaced by an expression like dp n(p)

S{Q,co) •

[co-coR-t£-

(2.45) This includes the collisional broadening described by T as well as the change in the potential energy A of the struck atom between thefinaland

2.3 High momentum transfer and the impulse approximation

initial states. The presence of A in (2.45) emphasizes that the distinction between "initial" and "final" state corrections is not fundamental and we refer to corrections to the IA generically as "Final-State" (FS) effects. In the pioneering study by Hohenberg and Platzman (1966) of the IA in superfluid 4 He, A was set to zero and the broadening approximated by T(0 = n (^\

G[Q) ,

(2.46)

where o{Q) is the atomic cross-section for two 4 He atoms. Martel, Svensson, Woods, Sears and Cowley (1976) later used this approximation to analyse neutron-scattering data at intermediate momentum transfers (Q ~ 5 A" 1 ). However, in general, T (and A) can be expected to be a function of co also and thus the integrand of (2.45) may deviate significantly from a simple Lorentzian. The difference of S inc (Q,0 in (2.37) from the IA in (2.38) due to FS contributions may be conveniently isolated by introducing a new function R as follows: Sinc(Q, t) = SiA(Q, 0 # F S ( Q , t) .

(2.47a)

In this form, Sjnc(Q,ct>) is given as a convolution over the final-state resolution function Sinc(Q, co) = r J—oo

dco'SlA(Q,

O/)*FS(Q,

co-co')

(2Alb)

As an illustration, the Lorentzian form

gives rise to an integrand of S[nc(Q, co) of the kind assumed in (2.45). The fact that the impulse approximation SiA(Q,<x>) already satisfies the first three central moment sum rules (n = 0,1,2) for Mn(Q) as defined at the end of Section 2.1 means that RFs(Q,co) must be negative in its high-frequency wings. In particular, this means a simple Lorentzian expression like (2.48) is inconsistent with the n = 2 central moment sum rule. A detailed theory of RFS(Q,CO) was given first by Gersch and Rodriguez (1973), and more recently by Silver (1988, 1989), Rinat (1989), and Carraro and Koonin (1990). Simple approximations to (2.45) based on (2.46) leave out the important short-range spatial correlations present in a liquid (as described by the pair distribution function g(r) defined in

Dynamic response of Helium atoms to thermal neutrons

(2.10)). These strongly modify how a recoiling atom moves over atomic distances after it is hit by a high-energy neutron. FS contributions are especially important if #FS(Q>&>) is broad compared to the width of any low-frequency peak in <SIA(Q,CO). Specifically, if SIA(Q,CO) contains a sharp component noS(co — coR) due to a Bose condensate, this component will be broadened to a width given by that of KFs(Q>&>). This broadening spreads the intensity due to the condensate component into the regions of co which overlap with the Doppler-broadened contribution from non-condensate atoms. This is a major source of difficulty in extracting information about the condensate fraction. It is clear that determination of n(p) in superfluid 4 He using high-momentum neutron-scattering data requires a careful removal of FS effects (see Chapter 4). By way of contrast, the momentum distribution is found to be broad and nearly Gaussian in normal liquid 4 He, solid Helium, and most classical liquids. The influence of FS corrections to the IA at high Q is relatively less important in these cases.

3 Bose broken symmetry and its implications

In this chapter, we begin our analysis of the dynamical correlation functions in a Bose fluid. Field-theoretic techniques and Green's functions are the most powerful ways of understanding the effect of a Bose broken symmetry. In Section 3.1, we introduce the order parameter associated with this broken symmetry and show how it couples the single-particle excitations and the density fluctuations. In Section 3.2, we review the formal structure of the single-particle Green's functions Ga^(Q,co) and then illustrate this with the well known Bogoliubov approximation. This model approximation really only describes a weakly interacting dilute Bose gas (WIDBG) at low temperatures but it already exhibits characteristic features of superfluid 4 He. In Section 3.3, we evaluate the densityresponse function Xnn(Q,w) in a Bose-condensed fluid within the simple Bogoliubov approximation in order to illustrate these features. Finally, in Section 3.4, we use a simple mean-field approach to illustrate how Ga£ and Xnn share the same poles when there is a Bose condensate. This sets the stage for the more systematic field-theoretic analysis given in Chapter 5. For orientation, we first summarize the properties of a non-interacting Bose gas (see, for example, pp. 38fT of Fetter and Walecka, 1971). The number of atoms in a free Bose gas with energy Sk is given by

where (.. .)o is an average in the grand canonical ensemble. The chemical potential \i is defined by the condition

n). 47

(3.2)

Bose broken symmetry and its implications

One can show that \i must be < 0. If \i is fixed and the temperature decreases, {hk)o decreases and hence the total number of atoms decreases. Indeed, for any finite \i, the value of N must decrease to zero as T —• 0. Since N is fixed, as the temperature drops, fi must become zero at some finite temperature TBE given by hBTBE

= 3.31

h2n2^ m

(3.3)

where n is the density of atoms. The scenario is therefore as follows: to keep N fixed as the temperature decreases, the chemical potential approaches zero until, at TBE, it reaches zero. For temperatures below TBE, Einstein (1925) first pointed out that extra atoms can go into the k = 0 state, which then becomes macroscopically occupied. We have (a+a o ) o = No ,

(3.4)

such that no = NQ/Q remains finite in the thermodynamic limit (the sample volume Q —• oo). Separating the atoms with k=0 and k ^ 0 explicitly, we find that (3.2) takes the form (note that \x = 0 below Tx) N = No + N ,

One refers to this macroscopic occupation of the k = 0 state as BoseEinstein condensation or more simply, Bose condensation. For T < TBE, the number of "excited" atoms N decreases as T 3 / 2 until at T = 0, Bose condensation is complete and there are no excited atoms (N = No, N = 0). The prediction of Bose condensation by Einstein (1925) was ignored until London (1938a,b) suggested that this kind of phenomenon might be involved in the then recently discovered superfluid phase of liquid 4 He. London noted that using the density of liquid 4 He, TBE as given by (3.3) is 3.1 K, very close to the observed lambda transition at 2.17 K (SVP). Since liquid 4 He is a strongly interacting system with all the ensuing complications, London's suggestion that a Bose condensate is involved in the superfluid transition was hard to prove or disprove theoretically (see, however, Feynman, 1953a). It was controversial for many years and, in our view, was only finally settled by the finite-temperature Feynman path-integral Monte Carlo calculations of Ceperley and Pollock (1986), summarized in Section 1.2.

3.1 Bose broken symmetry in a liquid

3.1 Bose broken symmetry in a liquid A formal definition of Bose condensation in an interacting Bose fluid was first provided by Penrose and Onsager (1956), developing earlier work by Penrose (1951). They generalized the criterion for Bose condensation in a gas used earlier by Bogoliubov (1947) to the condition that the oneparticle reduced density matrix in a liquid (or equivalently, the equal-time single-particle Green's function) pi(r,r') = p\(r — r') = (\p+(r)ip(rf)) does not vanish at large separation |r — r'|. Here \p(r) and ip+(r) are the field operators which destroy and create, respectively, 4 He atoms at position r. In modern terminology, the average ( ) involves a broken symmetry (or restricted ensemble) such that = 0>(r) ^ 0 , , }

(3.6)

below Tx. Thus this generalized criterion for Bose condensation can be stated in the form lim |r-r'|-KX)

Pl(r,O

= **(r)*(r / ) ^ 0

(3.7)

and /No- w i t h n o loss of generality, we can set the phase (j) of the uniform condensate to zero. The Bose broken symmetry is physically equivalent to "off-diagonal long-range order" (ODLRO) as formulated by Yang (1962). However the usual discussions of ODLRO do not give much insight into the dynamical implications of a Bose broken symmetry, which is our major interest in this book.

Bose broken symmetry and its implications

Penrose and Onsager (1956) had a tremendous influence on further work because they also gave the first numerical estimate of the condensate density no = |O(r)| 2 , using a crude ground-state variational wavefunction for hard-sphere Bosons originally introduced by Feynman (1953a). For further details, we refer to pp. 313ff of Huang (1987). They concluded that approximately 8% of the 4 He atoms are in the zero-momentum state at T = 0. However, Penrose and Onsager did not spell out the precise relation between the existence of the Bose condensate and superfluidity. It was later shown by Bogoliubov (1963, 1970) as well as Hohenberg and Martin (1965) that the Bose broken symmetry described by (3.7) does indeed lead to the two-fluid equations and superfluidity (see Chapter 6 for further discussion). To what extent superfluidity implies the existence of a Bose condensate is, logically, a separate question; but this is clearly of much less interest once a Bose condensate is known to exist in a given Bose fluid. The condition (y>(r)) ^ 0 describes a breaking of the gauge symmetry associated with the conservation of particles and is analogous to the broken-symmetry condition (Sz) = m ^ 0 in a ferromagnet. The main difference between (3.6) and the broken-symmetry condition in a ferromagnet is that a state of net magnetization is more easily visualized than a macroscopic wavefunction describing a state having a specific phase but not a fixed number of atoms. From a physical point of view, one can understand (3.6) by noting that the physical average of the phase of the field operator xp(r) is still undefined: Bose condensation corresponds more precisely to enforcing a well defined relation between the phase of \p at r and that at r', as given in (3.7). Without the "gauge fixing" or clamping implied by (3.6), the average over all possible phases would result in (N). For further insight into Bose broken symmetry, we refer the reader to the classic article by Anderson (1966), as well as Chapter 10 of the monograph by Forster (1975). The introduction of an explicit symmetry-breaking term in the Hamiltonian gives one a "hunting license" to look for a new thermodynamic phase (Bogoliubov, 1963, 1970) within a scheme where we can use the usual many-body techniques of finite-temperature perturbation theory. Landmark papers on the quantum field theory of Bose-condensed fluids include those by Beliaev (1958a,b), Hugenholtz and Pines (1959), Bogoliubov (1963, 1970), Gavoret and Nozieres (1964), Hohenberg and Martin (1965) and Ma and Woo (1967). All of these papers treat T = 0, but the formal extension to finite temperature is straightforward using the technique of imaginary-time Green's functions (Mahan, 1990). A fundamental implication of the Bose broken symmetry (3.6) is that the single-particle spectrum appears directly in the density fluctuation spectrum, as we shall now show. In second-quantized form, the number density operator defined in (2.6) is given by p(r)=v+(r)v(r) .

(3.9)

In a Bose-condensed system, it is useful to decompose the field operators as follows (Beliaev, 1958a):

where the ip, xp+ operators only involve atoms outside the condensate. Using (3.10) in (3.9), we obtain p(r) = |O(r)|2 + * » y ( r ) + *(r)y + (r) + i^+(r)^(r)

= HO + V ^ Gift) + V+M] + V+(r)v(r) .

(3.11)

Clearly the non-zero value of the condensate couples the single-particle operators directly to the density operator. In momentum space, (3.11) is equivalent to (Q ^ 0)

Bose broken symmetry and its implications

where the prime means that the second term (the "normal" density fluctuation operator) involves only atoms outside the condensate. The first term in (3.12) describes density fluctuations involving atoms scattering into and out of the condensate. This is summarized by rewriting (3.12) in the form (Q ^ 0) (3.13) where A Q = aQ + a+_Q

(3.14)

describes single-particle excitations. One may think of the separation in (3.13) in analogy with (3.5). Consider a system of interacting Bosons (in a volume Q satisfying the boundary conditions) with a uniform Bose condensate described by (3.8), with (So) = \/No. The second-quantized Hamiltonian is H = £ ( 8 , - fi)atak + ^

(Q)P(Q)P(~Q)

(3-15)

where p(Q) is defined in (3.12). As usual, it is convenient to include the chemical potential \i in our effective Hamiltonian as in (3.15) since we work in the grand canonical ensemble. We have not explicitly included the symmetry-breaking perturbation in (3.15). Inserting (3.13) into (3.15), we obtain H =

.Qi

(3.16)

where the prime on the summations again means that atoms with zero momentum are excluded. The first term in the second line of (3.16) shows how the presence of a condensate leads to coupling of the single-particle (AQ) and the "normal" density (p@) fluctuations. This coupling will play a crucial role in the subsequent analysis in this and succeeding chapters. Turning to the dynamic structure factor (see Section 2.1) S(Q,co) = (271ATT1 r

dteiait(p(Q,t)p(-Q))

J — oo

(3.17)

3.2 Single-particle Green's functions for a Bose-condensed fluid 53 we find (3.13) immediately leads to (Hugenholtz and Pines, 1959) S(Q,a>) =

;) .

(3.18)

S\ describes the density fluctuations associated with scattering atoms into or out of the Bose condensate and shows the direct role the single-particle excitations play in S(Q, co). However, in the presence of a condensate, all terms in (3.18) are coupled and thus exhibit the same poles (as we illustrate in Section 3.4). This motivates the next section, where we introduce the matrix single-particle Green's function Gap(Q,co). We note in passing that in the Bardeen-Cooper-Schrieffer theory, superconductivity is associated with the finite value of anomalous correlation functions of the kind (ip^(r)y)^(r)). This order parameter describes Cooper pairs which are Bose-condensed in the sense that all electron pairs have the same two-particle bound-state wavefunction (see, for example, Schrieffer, 1964). Because both theories invoke a broken gauge symmetry involving non-conservation of particles, there are many similarities between the descriptions of BCS superconductors and Bose-condensed fluids. Indeed, it was probably the BCS theory in 1957 that stimulated a wider appreciation of the original work of Bogoliubov (1947). Among many articles stressing the useful analogies between BCS superconductors and Bose-condensed fluids, we call attention to those of Anderson (1966), Nozieres (1966) and Vinen (1969). However, in contrast to superfluid 4 He, the single-particle Fermi excitations (quasiparticles) are quite distinct from the collective modes in superfluid 3 He. This distinction between Bose and Fermi superfluids is an important one to remember. A full account of the modern theory of superfluid 3 He, with emphasis on the broken symmetries involved, is given by Vollhardt and Wolfle (1990).

3.2 Single-particle Green's functions for a Bose-condensed fluid In this section we discuss the general structure of the single-particle Green's functions taking into account the Bose broken symmetry (3.6), as first worked out by Beliaev (1958a). In addition, we analyse the simplest non-trivial model calculation, for a WIDBG at T = 0, due to Bogoliubov (1947). This is discussed in all the standard texts. Finally, we briefly review theories which go past the Bogoliubov approximation.

Bose broken symmetry and its implications

Fig. 3.1. The four different vertices involved in a Bose-condensed fluid. The jagged line represents a condensate atom and the dashed line is the two-particle interaction. The single-particle spectrum of the non-condensed atoms is given by the poles of the single-particle Green's function, which for real times is given by (for 2 ^ 0 )

where T [...] denotes Wick's time-ordering operator. Since the number of particles is not conserved by the symmetry-breaking term, we also need to allow for the anomalous Green's functions (Ta+ e (t)flg(O) and (TaQ(t)a-Q(tf)) in the Bose-condensed phase. A generalized 2 x 2 matrix single-particle Green's function can be defined as (see pp. 249ff of Rickayzen, 1980) /_;Gll =l \-iGn

-iGaP(Q9t)

_iG21\

/ (ThQ(t)&+) (TatQ(t)a+)

-1G22J

\{TaQ(t)a_Q)

{TaZQ(t)a_Q)J

(3.20) It is common to introduce the notation (Gavoret and Nozieres, 1964) tq

if a = 2 or — .

V • ;

We can then write the matrix Gap in (3.20) in the compact form —iGap(Q9t) = (TaQa(t)a^p). In Fig. 3.1, we show the four interaction vertices which are involved in (3.16). Switching to imaginary time T = it (0 < T < /? = l/kBT), we define the imaginary-time matrix Green's function by (see Chapter 3 of Mahan, 1990) G

(f\

— _\/rrPj

(r\r& \ — la )la+ \/S

H ??)

where a

Qa\Z)

= e

^Qae

•

(j.23)

These finite-temperature Green's functions are periodic in r with period

3.2 Single-particle Green's functions for a Bose-condensed fluid

and we can expand Ga^(Q,r) as a Fourier series ajn) ,

(3.24)

where &>„ is the Bose discrete Matsubara frequency 2nnkBT integer) and the Matsubara Fourier coefficients are given by rl/kBT

G./KQ,icon) = / Jo

dx e^xGap(Q9T)

(n is any

(3.25)

The Fourier transform of the more physical real-time Green's functions is obtained by the standard technique of analytic continuation of the imaginary Matsubara frequencies to the real frequency axis (icon —• CD + it]). The terms subtracted in (3.22) are only important when dealing with the dynamics of the condensate atoms at Q = 0 (see the end of Section 5.1 and Section 6.3). Finite-temperature many-body perturbation theory can be used to evaluate Gap. In particular, the equations of motion of Gap can be conveniently written in the form of a Dyson equation involving a 2 x 2 matrix self-energy Ea0, = 0) + n0V(Q)

We emphasize that (3.31) describes only a subset of the complete HartreeFock self-energies, since it ignores the terms arising from interactions between excited atoms. The determination of the chemical potential ft within any given approximation is non-trivial but it can be proven that it must satisfy the relation (Hugenholtz and Pines, 1959; also Section 6.1) fi = Zn(Q - • 0,o) = 0) - Zi 2 (Q -> 0,o) = 0) .

(3.32)

As can be shown from (3.30), to satisfy the Hugenholtz-Pines "sum rule" (3.32) means that the spectrum of Gap in (3.29) must be gapless in the long-wavelength limit Q -» 0. Using (3.31) in (3.32) gives ^ = no7(p = O)

(3.33)

for the Bogoliubov model approximation. Using (3.31) and (3.33), the denominator (3.30a) reduces to B B ) = (iwny - cozQ ,

(3.34)

where the Bogoliubov quasiparticle dispersion relation is given by Q. This can also be seen from the general structure

Bose broken symmetry and its implications

v 0 »> • »

Fig. 3.3. The Bogoliubov approximation (3.31) for the self-energies The external propagator lines are not part of the self-energy. of (3.29) and (3.30). In a free Bose gas, UQ=1 and VQ=0 and hence only the positive-energy pole at co = SQ has any weight. Eq. (3.35) shows how the condensate changes the single-particle spectrum from particle-like at high momentum to phonon-like at low momentum. The Bogoliubov approximation corresponds to keeping only the terms in the first line of (3.16). This simplified Hamiltonian can be diagonalized by the well known Bogoliubov transformation (see p. 314 of Fetter and Walecka, 1971), with results equivalent to (3.36). One can use G^ in (3.36) as the new unperturbed Green's function in a new renormalized matrix Dyson-Beliaev equation in which the self-energies entirely arise from the terms in the last line of (3.16). The resulting diagrammatic expansion is discussed on pp. 249ff of Rickayzen (1980). As discussed in the standard texts on many-body problems, one can generalize (3.32) to include multiple scattering of two free Bose atoms within the "ladder approximation." In this approximation, the HartreeFock self-energies take the same form with V(Q) replaced by an expression involving /(p,p') ? the exact scattering amplitude for two free atoms (see Eq. (11.14) of Fetter and Walecka, 1971). In the limit of low momentum, V(Q = 0) is replaced by /(p = 0,p' = 0)/m = 4na/m, where a is the s-wave phase shift. Within this approximation (for hard spheres, a is the diameter), the quasiparticle spectrum in (3.35) has the following limiting behaviour: COQ = c0Q,

is^r

for Q < 2mc0 ,

(3.38)

for Q

(3 39)

* 2mco'

where the Bogoliubov phonon speed is given by CQ = 4nnoa/m . In a dilute gas at T = 0, the small expansion parameter is (na3)1^2, that is the

3.2 Single-particle Green's functions for a Bose-condensed fluid 59 spacing between atoms must be much larger than the interaction range a. The Bogoliubov approximation results (3.32)-(3.37) are studied at considerable length in standard many-body texts (see especially Chapter 6 of Fetter and Walecka, 1971). This approximation deserves attention since it exhibits the characteristic structure imposed by a Bose broken symmetry on the single-particle fluctuation spectrum. Unfortunately many studies have taken the Bogoliubov approximation (and minor variations of it) as a realistic model for a strongly interacting Bose liquid like superfluid 4 He, which is quite unjustified. This also leads to incorrectly assessing the value of the Bogoliubov approximation (3.36) in terms of how well it reproduces the phonon-roton spectrum of liquid 4 He (see, for example, Section 10.1 of Mahan, 1990), rather than for the qualitative insight it gives into the role of the Bose condensate on the excitation spectrum. The structure which the single-particle Green's functions (3.36) exhibit in the Bogoliubov approximation already captures some of the essential features of the exact expression. This is also nicely illustrated, for example, by the renormalization group analysis of the scaling properties of a WIDBG within the Bogoliubov approximation (Weichman, 1988). Beginning in the late 1950's, there have been many attempts to improve the Bogoliubov approximation by treating the multiple scattering in a self-consistent t-matrix approximation (see, for example, Brueckner and Sawada, 1957; Parry and ter Haar, 1962; Brown and Coopersmith, 1969). That is, the ladder diagrams describing the interaction between excited atoms are calculated using renormalized propagators which include (in an approximate way) the effect of a condensate. In this way, it was hoped that one could include the strong renormalization effects on the excitation spectrum expected in a liquid, as well as obtain an estimate of the depletion of the condensate fraction. The limitations of this approach were noted by Hugenholtz and Pines (1959); the dielectric formalism of Chapter 5 pinpoints the key inconsistency. Essentially, the usual ^-matrix studies attempt to treat the diagonal self-energies I n in a way more appropriate to a liquid but only keep the simplest off-diagonal self-energy term En appropriate to a gas, namely that considered in the Bogoliubov approximation (3.31). Developing better approximations for the self-energies Hap is not an easy task. Even for a WIDBG, (3.31) is inadequate at higher temperatures where the condensate is thermally depleted (no 0 limit are phonon-like, even though the excitations of the single-particle Green's functions used to generate it may have an energy gap at Q = 0. In this kind of conserving approximation, the twoparticle Green's functions G2 (such as inn) are generated by functional differentiation of the one-particle Green's function G\. Schematically, we have Gi = dG\[W]/8W9 where W is some appropriate external field set to zero at the end. More precisely, the equation of motion for Gi is obtained from functional differentiation of the equation of motion for G\. Any approximation for the single-particle self-energy Z defines a G\ and hence £ may be viewed as a functional of G\. One finds that G2 = G\G\ + G\G\TG2, where the interaction vertex is given by T = (SX[Gi]/(5Gi. Thus the self-energy S determines Gi as well as T. The two-particle Green's function G2 given by this procedure is guaranteed to be consistent with various conservation laws but clearly the poles may be quite different from the G\ used to generate it. Examples of such conserving approximations have been worked out for Bose fluids by Hohenberg and Martin (1965) at T = 0 as well as by Cheung and Griffin (1971b) at T ± 0. The coupling of the single-particle fluctuations with density fluctuations (as exhibited at the end of Section 3.1) means that Gajg(Q,co) and the density response function xnn(Q9(o) share the same singularities, although with different weights. This will be shown using a simple approach in Section 3.4 and in more general terms in Chapter 5. However, while it is clear that an approximate calculation should be consistent with this requirement, it is not easy to satisfy if we simply include more self-energy diagrams in an ad hoc manner.

3.3 S(Q,co) in the Bogoliubov approximation

As we mentioned in connection with (3.18), the intimate connection between Gap and Xnn was first pointed out by Hugenholtz and Pines (1959). In particular, they showed that in the Beliaev (1958b) approximation, the phonon pole of Gap had a velocity corresponding precisely to the compressional sound velocity (as determined by the thermodynamic derivative of the ground-state energy computed in the same approximation). This result is of great importance since it justifies the key assumption of Landau and Feynman that "sound waves" play the role of elementary excitations. In this regard, we also call attention to the important field-theoretic results of Gavoret and Nozieres (1964). Working to all orders of diagrammatic perturbation theory, they evaluated the single-particle Green's function as well as the density and current response functions. At T = 0 and in the long-wavelength limit (g, co —• 0), they showed explicitly that Gajg and Xnn exhibit the same phonon resonance in a Bose-condensed fluid. This generalized the relation noted by Hugenholtz and Pines (1959). The work of Gavoret and Nozieres still provides one of the few rigorous calculations we have for the excitations of a Bose fluid based on an explicit many-body calculation. It is discussed in detail in Sections 5.4 and 6.3. In Chapter 5, we develop an approach for calculating both G^ and %nn which guarantees from the beginning that they share the same spectrum, as they must in the presense of a Bose condensate.

3.3 S(Q,(o) in the Bogoliubov approximation It is instructive to evaluate also the dynamic structure factor S(Q,a>) in the simple Bogoliubov approximation. We recall that S(Q,co) is related to the density response function Xnn by the relation (2.20b), S(Q,o>) = - — [N(co) + l]Im Xnn(Q,a> + in) . (3.40) nn Here Xnn is the usual analytic continuation of the imaginary-frequency Fourier component (icon -> co + in) - XnniQ, icon) = ^ / " Jo

dre^

[) = — (UQ - VQf{ [N((OQ) + l]S(C0 - CDQ) + N(CDQ)3(CO + CDQ)] } 1 f dp + " / 7 ^ T 3 ^ ( 0 ) p ) [ l + N(C0p+Q)](upUp+Q + VpVp+Q)Zd(C0 - [cOp+Q - COp])

- / 7^)3 2 N(cop+Q)][l

+ N(cop)]S(co - [cop+Q + cop])} .

(3.45)

3.3 S(Q,co) in the Bogoliubov approximation

We have carried out the standard Bose Matsubara frequency sums using (see pp. 167ff of Mahan, 1990) 1V

_ — G)p+Q

0n — top iOJn — iCOl + Wp+Q

(°)P+Q)

~ N(°)P)

u <w, \

iCOl — [(Op+Q — O)p]

i(Di — [(Op+Q

•£&•

-*««>•

(3.46) In addition, we have used the Bose distribution identities given by (2.31) as well as [N((o2 - coi) + 1][N(coi) - N(co2)]

= N(coi)[l + N(co2)]

in simplifying the final expression given in (3.45). The frequency sums Ri and R2 in (3.46) will also be needed in later chapters. Factors such as (UpUp+Q + VpVp+o) and (upvp+q + UPVP+Q) in (3.45) are referred to as Bose coherence factors. They describe the complicated interference effects associated with the mixing of positive and negative energy poles of G^ and are the signature of Bose-condensed fluids. The expression in (3.45) is only an illustration of the kind of structure S(Q, (o) which exhibits when the Bogoliubov approximation is a good starting point. The first line of (3.45) corresponds to Si in (3.18) and describes the creation (iV(co) + 1) or destruction (N(co)) of a single Bogoliubov excitation. The remaining terms correspond to S in (3.18). The second line of (3.45) describes the thermal scattering processes destroying an excitation with energy cop and creating one with energy COP+Q. These first two lines in (3.45) have their analogue in a free Bose gas given by (2.32) of Section 2.2. The third line of (3.45) describes the creation (or destruction) of two quasiparticles, with total energy cop + cop+g. This term is associated with the existence of negative and positive energy poles of G^ in (3.36). As with the first line of (3.45), this multiparticle term disappears when no=O (since VQ vanishes) and thus it is characteristic of a Bose-condensed fluid. The "two-phonon" or multiphonon contribution in (3.45) will give rise to a broad frequency spectrum, in contrast to the "one-phonon" terms which are sharp resonances at +COQ. At T = 0, all the Bose occupation factors N(co) vanish, in which case we can only create one or two quasiparticles. There are no thermal scattering terms at T = 0. The two-excitation or pair spectrum in liquid 4 He is discussed in detail in Chapter 10. Eq. (3.45) is, at best, only appropriate in the "weak-coupling" limit.

Bose broken symmetry and its implications

In using (3.43), one has completely ignored any collective (zero sound) density fluctuations, as well as the interference terms coupling such fluctuations into the single-particle terms described by the first line in (3.45). To include such effects, we need a more systematic procedure for calculating Xnn past the Bogoliubov approximation. In Chapters 5 and 6, we show quite generally that Xnn can always be separated into two parts, one of which is directly proportional to the single-particle Green's function: Xnn(Q,co) = XA*(Q,a)Gap(Q9(D)Ap(Q9) •

(3-47)

The result in (3.43) may be viewed as an illustration of this. The Bose vertex functions Aa(Q,co) in (3.47) determine the strength with which the single-particle excitations appear in the density-response function. These symmetry-breaking vertex functions vanish with no, while Xnn S o e s o v e r into the full response function of the normal Bose fluid. Such rigorous decompositions of Xnn into what one might interpret as condensate and normal contributions were first derived (at T = 0) by Gavoret and Nozieres (1964) and Hohenberg and Martin (1965). Formulas such as (3.47) will play a central role in future chapters.

3.4 Mean-field analysis All three contributions in (3.18) are strongly modified by the effect of the p-A coupling terms in the second line of (3.16). They are all related, with the result that the single-particle spectrum of Ga^(Q,co) contains the density fluctuation spectrum of S(Q,co) and vice versa. This is the key feature of a Bose-condensed fluid, as is already apparent from (3.13). To understand the details of the resulting hybridization of the spectra of iS(Q,co) and Gajs(Q,<x>) requires a fairly sophisticated diagrammatic analysis in terms of proper, irreducible diagrams. This is called the "dielectric formalism" and is the subject of Chapter 5. As an introduction to the basic physics involved, in this section we calculate both XnniQ, co) and Gajs(Q,co) using a simple mean-field analysis (Griffin, 1991). We consider the linear response of a Bose fluid to an external scalar potential (50° which couples into the variable PQ and an external brokensymmetry potential Sn° which couples into the variable AQ (see (3.13) and (3.14)). This perturbing Hamiltonian is given by

= J ^3^°(Q,a>)p e + J

(3.48)

3.4 Mean-field analysis

Thus we have Sn(Q9(o) = Xnn(Q,o)d) + XnA(Q,co)dn°(Q,(o) ^ X%i(Q,a>) [S^°(Q9co) + V(Q)5n(Q9co) + y/f*V(Q)SA(Q9a>)] , (3.49a) SA(Q,CD)

= XAA(Q,co)dn°(Q,co) +XAn(Q,aj)dkBT , Cp ksT). By way of contrast, the observed momentum distribution np in the region p ^ 0.8 A" 1 is found to be a remarkably temperature-independent Gaussian (see Fig. 2.9). As with a classical Maxwell-Boltzmann velocity distribution, one can thus write np in the form np^e~^t

(4.18)

where {K)

k is the energy of the excited state |k) (relative to the ground-state energy). By explicit calculations based on correlated basis functions using Feynman-Cohen eigenstates (see Section 9.1), MPU

4.2 Impulse approximation using a Green's function formulation

0.25

0.20

0.15 —

0.10

0.05

0.00

Fig. 4.2. The atomic momentum distribution n(p) as given by three methods: variational; Green's function Monte Carlo GFMC (Whitlock and Panoff, 1987); and high-momentum S(Q,co) data as analysed by Sears et al, (1982) [Source: Manousakis, 1989]. show that the expression in (4.21a) leads to np(T) = np(T = 0) + Snp(T) 1 n0 me n0 me = + N(cp) , 2 n p n p

(4.21b)

in precise agreement with the result in (4.8).

4.2 Impulse approximation using a Green's function formulation We now derive the impulse approximation formula (4.2) directly from a Green's function approach, working in momentum and frequency space. This allows us to see explicitly how the unique coherence factors associated with Bose condensation enter into the analysis which ultimately leads to the IA. This approach complements the usual derivation given in Section 2.3. In the IA, the starting assumption is that the scattering cross-section can be approximated by the product of the single-particle propagators of

High-momentum scattering and the condensate fraction

the initial atom (with momentum p) and scattered atom (with momentum p + Q). Working at finite temperatures, the corresponding densityresponse function is given by

~~7j X / Q,icon + w)l)] .

(4.22)

This form is formally identical to the Bogoliubov approximation in (3.43) and (3.44), except that now the Gap in (4.22) are the full Beliaev singleparticle Green's functions of the interacting system, as given by (3.29). Basically, in starting from (4.22), we allow the incident and scattered atoms to interact with the rest of the liquid but ignore any interactions with each other. Using spectral representations (as defined in (6.25)), the Matsubara frequency sums in (4.22) can be carried out using (3.46), and we obtain

7 (2^)3/_«, ^T/.oo liT |>|-(a>"-a)' )An{* + Q,a>")] •

(4.23)

Here A\\ = A is the spectral density of G\\ and An is the spectral density of G12. Making the change of variable p + Q —• —pr and cof «-> co", one can rewrite the first factor in the integrand of (4.23) as

The characteristic simplifying feature of high momentum transfers is that for the scattered atom, we can use the approximations ~ 2nd(co"-sp+Q)

That is to say, the Bose coherence factors disappear at high values of momentum and so does the anomalous Beliaev Green's function GnInserting (4.25) into (4.23), we obtain Xnn(Q, m)

= no [G°n (Q, icot) + G°22(Q, i,T)

(4.36)

where (5n*(p, T) is not dependent on no. Combining (4.36) with (4.34), and integrating up to some wavevector pc, one obtains (Sears et al, 1982; Sokol et al., 1989) _ (T n x no(T,pc) = where we have defined

cc(T,pc)-a(Tx,pc) ——————- , -0L(T,pc)+y(T,pc)

a(T9pc) = f0Pcdp4np2n(p,T) Pc

y(T,pc) = f0 dp4np Sn(p,T)

(4.37)

, .

The value of the condensate fraction HQ(T) given by (4.37) should be independent of the value of the cutoff pc chosen; this is not the case, as shown in Fig. 4.11. However, the decrease in the value of no(/?c) exhibited in Fig. 4.11 at large and small values of pc is expected. The low-/? behaviour of n(p) in the superfluid phase is not correctly approximated by the two-Gaussian functional form (4.33) used for n(p, T). At high /?, the difference between the superfluid and normal phase momentum distributions is very small. We agree with the argument of Sokol et al. (1989) that the peak value of no as a function of pc gives the most accurate estimate. This always occurs at pc in the region 0.6-0.8 A" 1 , where (4.9) and (4.36) should be valid (Griffin, 1985). Using these, one finds y(T,pc) = 0.85/?;:, where pc is measured in A" 1 . Fig. 4.12 shows the values for no(T) obtained by Sokol, Sosnick and Snow (1989) by the procedure described above. For comparison, the Monte Carlo path-integral results of Ceperley and Pollock (1986) are also plotted. Earlier estimates by Sears et al. (1982) and Mook (1983)

4.4 Extraction of the condensate fraction

0.10

0.08

0.06 —

0.04 —

0.02 —,

0.00

MA" 1 ) Fig. 4.11. Inferred values of n0 at several temperatures as a function of the cutoff momentum pc. See discussion after (4.38) [Source: Sokol, Sosnick and Snow, 1989].

used an inadequate treatment of the key y(T,pc) contribution in (4.37), which led to an overstimate of no (Griffin, 1985). We have given a careful analysis of the expression in (4.37) because it was the basis of the first adequate analysis of S(Q,co) data to extract the condensate density. However, (4.37) is based on the form (4.34), which assumes that the non-condensate atom momentum distribution in superfluid 4 He has a component which scales with that in the normal phase. An improved version which does not rely on this assumption has been developed by Sokol and Snow (1991). They assume that n(p) is given by

(4.39)

A2e~p2

where nof(p) is the condensate-induced low-/? anomaly in rc(p). This form for n(p) incorporates what is known rigorously from microscopic theory (see Section 4.1): it is positive, normalized to unity, continuous at pc and symmetric about p = 0. Fortunately, the parameters in (4.39) are mainly

High-momentum scattering and the condensate fraction

12 —

x Expt.. O PIMC

—

D GFMC

-• >
co) + p] .

(5.15a)

(5.15c)

It is crucial that approximations are consistent with these exact relations. The irreducible functions also satisfy the same identities as in (5.15), with all quantities replaced by the corresponding barred functions. We note that (5.15fr) and (5.15c) can be combined to give the "continuity equation" Xm =

l / j j + p ]

{5Ma)

The irreducible version of these equations gives the same relation between Inn and Xn,

This last relation enables one to compute Xnn (and hence Xnn) from yfjj in such a way that the /-sum rule (2.24) is obeyed. Using these relations in conjunction with (5.8)—(5.13), we see that the regular (proper and irreducible) functions satisfy the following important identities (Wong and Gould, 1974): C0A a (g, CO) = — A^(g, (O) + V ^ [(O - (X(SQ - fi)] m

->/no Pi,a(6,<w) - £^(G,fi>)] > (otniQM

= -TfjniQM m «Zj*(e,») = -GjjiQM m

(5-17)

~ V ^ [A,(Q,a>) - A2(Q,©)] ,

(5.18)

+ P)-V*o

(5.19)

[A((e,CB) -

Ai(Q,a>)] .

In deriving these results, we have used Dyson's equation for the inverse irreducible matrix Green's function (see (3.26)) G^p = <W M

- (£)

~ where the denominator is defined as

C(Q,a>) SEE De = D \eR

= DeR - VUDAtNmAp

(5.22)

Turning to the denominator of Gap = Nap/D, one can separate out the irreducible (proper) self-energy £aj? from the reducible proper self-energy as in (5.12). After some calculation, one finds D = D - NaPI,cpgL

M^l

(5.23)

Comparing (5.22) and (5.23), we arrive at an important identity relating e and D, namely eRD=eD; this shows that the poles of both Gap and Xnn are given by the zeros of C(Q,co) in (5.22). The crucial role of the condensate is manifest through the Bose vertex functions Aa in (5.22) and (5.23). In the absence of a Bose broken symmetry (Aa = 0), the pole a>2 of Gap (given by D(Q, a>2) = 0) and the pole cb\ of %nn (given by eR(Q,(bi) = 0) are uncoupled. "Turning on" the condensate (Aa ^ 0) hybridizes these modes, as shown by (5.22). The renormalized zeros of C(Q,a>) will be denoted by co\ and coi- (In Griffin and Cheung, 1973, the definitions of co, and a), are interchanged.) From (3.12), we see that in the presence of a condensate, the densityresponse function Xnn will contain contributions which are proportional

5.1 Dielectric formalism

101

to a single-particle propagator Gap as a separate factor (i.e., as an intermediate state). Gavoret and Nozieres (1964) refer to these contributions as the "singular" part of /„„. Within the dielectric formalism, the GavoretNozieres decomposition of Xnn (and other correlation functions) into "singular" and "regular" parts corresponds to splitting diagrammatic contributions into improper and proper parts (see Fig. 5.6)

where the first term is equivalent to the Gavoret-Nozieres singular term. This key formula will now be derived. First of all, (5.2) can be split as follows: _ Xnn —

Xnn _

\Xnn ~^~ Xnn)\^ ~

*Xnn)

€ yC I yR (I _ Ann T Ann\ Ann

yyC _ Ann

WyR Ann

i Ann

where xnn has been defined in (5.11). To prove that the first terms in (5.24) and (5.25) are identical, we note that the irreducible parts of the numerator Nap of G^ given in (3.29) can be separated out, giving

tf^-p (5.26) (recall that summation is only over repeated Greek subscripts, which do not occur in (5.26)). Using the trivial identity

£aA a A_ a = 0 ,

(5.27)

one easily obtains from (5.26) the surprising result AaNapAp = Aa Note that there is no contribution from the second term in (5.26). Using this fact and the key identity eRD = eD derived above, the first term in (5.25) can be transformed as follows: AaGapAp €R€

Aa Nap Kp

€R

which completes the proof of (5.24).

€R

A^ H

€R

102

Dielectric formalism for a Bose fluid

Fig. 5.6. Diagrammatic structure of the proper and improper contributions to Xnn as given by (5.24). See Fig. 5.4 for the Bose vertex function Aa. A result analogous to (5.24) can also be derived for the density singleparticle correlation function Xan- Starting from (5.6), we find D

(5.29)

using (5.26) and (5.27), as well as the identity given after (5.23). Recalling (5.9), the last result is seen to be equivalent to (5.8), i.e., xan = ApGp^. We have shown how the density-response function XnniQ,co) naturally splits into two parts as in (5.24), with a "singular" or "condensate" part having Gap as a separate factor. The analogous decomposition of the longitudinal momentum current-response function Xyj(Q,a>) in (5.6) into improper and proper parts is given by Eqs. (2.5) and (2.6) of Talbot and Griffin (1984b). Such decompositions are useful because the proper term which is separated off may be loosely viewed as the "normal" contribution. This terminology is justified because, above 7^, only the second term in (5.24) remains. However, below T^, G^ and Xnn have singularities given only by the zeros of C(Q,co) in (5.22) and not by the zeros of eR or D. This fact emphasizes that one should not automatically think of the two contributions in (5.24) as distinct. This inter-relation is already clear when one realizes that they can always be summed to give the expression in (5.2). The key point is that eR enters both terms in (5.24) explicitly, as well as implicitly in Gap via the reducible self-energy in (5.12). Talbot and Griffin (1984a) have shown there are cancellations between the two contributions in (5.24) when there is significant structure arising from zeros of eR(Q,co). Whether we start from (5.2) or (5.24) is central to the interpretation of the structure exhibited by S(Q,a>) in superfluid 4 He, as we discuss at the end of Section 5.5 and in Section 7.2. The Dyson-Beliaev matrix equations in (3.26) can also be rewritten

5.1 Dielectric formalism

103

with Gajg playing the role of G^ (Szepfalusy and Kondor, 1974). One finds Gap = G^p + GasZc$yGyp ,

(5.30)

where the reducible matrix self-energy l Xnn- Starting from some choice for Aa and Xnn is less preferable because (5.17) and (5.18) do not give any information about Aa or xRn a * z e r o frequency. In contrast, (5.17)—(5.19) give exact constraints on the zero-frequency limit of A^ and y}jj ( se e Section 6.1). A convenient way of summarizing this section is to give the explicit procedure for calculating Gap and #„„, starting from a specific approximation to the "building blocks" A£, Xn an SQ. This result illustrates how we can end up with a temperatureindependent zero sound phonon frequency using the SPA, even though its origin lies equally in the condensate and normal parts of xnn which are individually very temperature-dependent. We also note that condensate self-energy in (5.42) reduces to ZC(Q, co = cQ) = nV(Q) using (5.47) and (5.51). This has the effect of ensuring that the strong-coupling singleparticle spectrum is renormalized to cQ, with the velocity going as y/n 9 instead of y/Tio as in the weak-coupling limit (5.46). In later chapters of this book, (5.50) in a suitably generalized version will be used to suggest how the phonon mode frequency is essentially temperature-independent in superfluid 4 He. As far as Xnn is concerned, the derivation of (5.51) based on (5.50) goes back to a suggestion of Pines (1966). The analogous discussion of a temperature-independent phonon pole of G^ is due to Tserkovnikov (1965). To summarize, we emphasize that while the phonon spectrum exhibited by both Gap and Xnn is quite different in the WC and SC limits, in both limits it is given by the zeros of e(Q,co) in (5.2). In the limit of low ). We can summarize the SC and WC limits as follows: SC: The collisionless phonon orginates as a pole of %„„, with a temperature-independent speed. It is the natural extension of first sound and is best viewed as a zero sound mode involving all the atoms. This phonon has a weight in S(Q,co) proportional to n but a weight in Gap(Q, co) proportional to no(T). This description is most appropriate in a liquid but can also be formally considered in a gas (Szepfalusy and Kondor, 1974; Griffin, 1988). One expects that as the temperature increases, additional single-particle states must appear to take over the spectral weight in A(Q,co). This is nicely illustrated by the Bose gas model calculations of Szepfalusy and Kondor (1974), where overdamped high-energy excitations leave the imaginary frequency axis at some intermediate temperature and, above 7^, ultimately become the free-particle states of energy Q2/2m. The analogue of this in a Bose liquid has not been investigated. WC: This limit naturally arises in a WIDBG at intermediate temperatures (SK, 1974; Payne and Griffin, 1985). In this domain, the phonon mode is completely associated with the condensate mean field even at finite temperatures. It thus has a speed which is very temperature-dependent, decreasing as ^no(T), as predicted by (5.46). It is a soft mode and is seen to be the natural extension of second sound in a Bose gas (Gay and Griffin, 1985) into the collisionless domain. It has a weight in S(Q,co) proportional to no(T). In a WIDBG described by this WC limit, there is no well defined zero sound mode involving a non-condensate mean field. An expanded discussion of the nature of the low-g phonons in a Bosecondensate fluid is given in Section 6.3. The discussion there is quite general and not based on an analysis of dilute Bose gas models used in this section.

112

Dielectric formalism for a Bose fluid 5.3 One-loop approximation for regular quantities

In Section 5.2, we discussed the simplest diagrammatic approximation for the regular functions, namely those of a (Hartree) Bose gas. We now briefly comment on the next level of approximation for the regular functions A£, yf/j a n d £«£> given by the "one-loop" diagrams. As with the SPA discussed in the preceding section, our main interest in the one-loop approximation is to develop further insight into superfluid 4 He, rather than using it to find perturbative corrections to the Bogoliubov approximation. In Section 6.1, we show that several non-trivial zerofrequency sum rules are satisfied by this one-loop approximation, which emphasizes that it captures the right physics. The one-loop regular diagrams are shown in Fig. 5.8. Referring to Wong and Gould (1974) as well as Talbot and Griffin (1983) for details, the rules of diagrammatic perturbation give £n(Q,o>) =

- R2(co)] + v2p+Q[R{(-co) -

R2(-co)]

Q)(up - vp)(up+Q - vp+Q) X{VPUP+QRI(CO)+UPUP+QR2{CO)-\-UPVP+QR1(-CO)-\-VPVP+QR2(-OJ)}

(5.52)

{R{(CD)

R2(CD)

+ Ri(-co) -

R2(-co)}

(2n

(5.53)

--{upvp+Q

- vpup+Q)2[R2(co) + R2(-co)]}

x {(up - vp)(upUp+Q - vpvp+Q)[up+QRi(o))

(5.54)

vp+QRi(-co)]

5.3 One-loop approximation for regular quantities -(up

- vp)(upvp+Q

- VPUP+Q)[UP+QR2(CO)

+ vp+QR2(-co)]}

113 .

(5.55)

The single-particle distributions are (in the Bogoliubov approximation) GBn(p, kon) = v2p + (u] + v2p)N(cop) ,

(5.56)

ihp = « a i p ) B = - - ^ Gf2(p, iG)B) = - Mp i, p [l + 2N(cop)] ,

(5.57)

np = (a+ap )

= - -- £

and the functions R\ and #2 are as defined in (3.46). While looking somewhat daunting, these one-loop expressions show several characteristic features of regular functions. Strictly speaking, we should calculate the one-loop diagrams in Fig. 5.8 using the full zero-loop or SPA single-particle Green's functions for the internal propagator lines. As we discussed in Section 5.2, however, the SPA G^p is only simple in the single-particle (SP) limit, where it can be approximated by the Bogoliubov form given by (3.35)—(3.37). The one-loop results written down above have, in fact, been derived using this SP approximation to Gap. Thus the quasiparticle dispersion relation cop which enters these expressions (including the up and vp amplitudes) is given by (5.46). We refer to this as the SPA-Bogoliubov approximation for Gap. In the zero sound (ZS) limit for the SPA Gap, the structure is much more complicated than the single-particle SP limit. There is not much pedagogical value in working out the one-loop diagrams with such improved propagators. The basic physics is already clear using the zero-loop or SPA propagators. The one-loop regular functions based on the SPABogoliubov propagators provide a well defined microscopic model which is still simple enough to understand the structure of a non-trivial approximation past the Hartree Bose gas expressions in (5.34). Terms to first order in the interaction V in £a£ ((5.52) and (5.53)) correspond to the complete Hartree-Fock proper self-energies (in contrast with the Bogoliubov approximation discussed at the end of Section 3.2 and shown in Fig. 3.3). In the normal phase (T > 7^), the one-loop approximation consists of the Hartree-Fock self-energies. The new physics involved in the one-loop approximation comes from the terms which are second-order in the two-body interaction. Referring to (3.46), R\ clearly describes scattering processes involving thermally excited quasiparticles (p —• p + Q), which are only present at finite temperatures. In contrast, R2 describes the creation (or destruction) of two quasiparticles, processes which occur even at T = 0. The regular functions given by (5.52)—(5.55) lead to expressions for both Gaj?(Q,co)

114

Dielectric formalism for a Bose fluid

fc--^-

+ GB

Fig. 5.8. The proper, irreducible diagrams for £a£, A£ and XJJ used in the one-loop approximation. and Xnn(Q>tt>) (using the procedure given at the end of Section 5.1) which will involve two-quasiparticle processes. In contrast, the Bogoliubov approximation in Section 3.2 includes the creation or destruction of only a single quasiparticle. In Fig. 5.9, we show an example of a second-order reducible self-energy diagram which is included in X ^ in (5.12) when we work within the oneloop approximation. It may be viewed as a broken-symmetry vertex correction to the simplest reducible self-energy exchange diagram shown in Fig. 3.3. It arises from using a Aa in (5.11) which is associated with the A^ given in Fig. 5.8. We note, however, that the one-loop approximation does not include the normal second-order contributions to S a ^ shown in Fig. 5.10. Since the one-loop approximation does not include the effect of such real two-body collisions, it cannot be used to discuss hydrodynamic modes at finite temperatures. While (3.44) and (3.45) of Section 3.3 illustrate the kind of structure expected in x%n, we note that calculating xm from the related currentcurrent response function using (5.16b) is a superior approach. This automatically includes "backflow" effects and thus gives a density-response

5.3 One-loop approximation for regular quantities

115

Fig. 5.9. A second-order reducible self-energy diagram included in the one-loop approximation for S£».

Fig. 5.10. Second-order irreducible self-energies not included in the one-loop approximation for £a£. function Xnn which is consistent with the /-sum rule (2.24), in contrast with (3.45). This approach was first used by Miller, Pines and Nozieres (1962) as well as in Section 7.3 of Nozieres and Pines (1964, 1990), and was later emphasized by Wong and Gould (1974) in the context of the dielectric formalism. Using the one-loop expression for xf/j given in (5.54), the second and third lines in (3.45) are then replaced by Eqs. (A7) and (A8), respectively, of Griffin and Talbot (1981). Wong and Gould (1974, 1976) have used the one-loop approximation in a WIDBG at low temperatures as a systematic way of calculating the temperature-dependent corrections to the lowest-order Bogoliubov results discussed in Section 3.3. Since they assumed that there is no well defined zero sound mode in the non-condensate atoms, they were justified in using the expansion — ~ 1 + V(Q)Xnn+--

(5-58)

in both (5.24) and (5.12). That is, Xnn in (5.24) was expanded around the Bogoliubov SPA approximation, whose spectrum is given by (5.46) in the SP limit. Such an expansion gives a result for S(Q,co) which has a "one-phonon" resonance (pole of G^) as well as a "two-phonon" continuum from the second term in (5.24) and the Bose vertex functions Aa/eR. The latter are often called "interference" contributions and

116

Dielectric formalism for a Bose fluid

correspond to the pA contributions in (3.18). Wong and Gould (1974) have shown how these provide a microscopic basis for understanding how such "backflow" effects (first discussed by Feynman and Cohen, 1956, in a different context) modify the elementary excitations in a Bose-condensed fluid. In connection with the approximation (5.58), the work of Cheung and Griffin (1971b) is of interest. They compute Xnn in a conserving approximation using functional differentiation of the Hartree-Fock selfenergies (as given by Girardeau and Arnowitt, 1959). Cheung and Griffin then prove that the pole of this Xnn is identical to that of G^ computed in the second-order Beliaev (1958b) approximation, for all T and Q. To show this equivalence, however, it was necessary to assume that there were no zeros of eR(Q,a>) so that one could use an expansion analogous to (5.58). It is no accident that such a "regularity assumption" leads to a single identical mode in both Gap and Xnn- A similar assumption was made by Beliaev (1958b) in working out the explicit form of the poles of An expression like (3.45) is often implicitly assumed in discussing the observed spectrum of S(Q,a>) in superfluid 4 He. In particular, it is the basis of the classic paper by Miller, Pines and Nozieres (1962). However, this sort of approximation to (5.24) breaks down when the dynamics of the "normal fluid" excitations play a crucial role, in which case one cannot ignore the structure arising from the zeros of eR. If the ZS limit is appropriate, the decomposition in (5.24) (and simple approximations to it such as (3.45)) is not as useful as (5.2) as a basis for interpreting S(Q,a>). The low-g phonon region in liquid 4 He is a case in point (see Section 6.3).

5.4 Gavoret-Nozieres analysis In this section, we summarize the seminal work by Gavoret and Nozieres (GN, 1964). Their analysis gave the first rigorous treatment of the role of the Bose condensate in coupling the single-particle spectrum with the density fluctuations. They obtained expressions of the type given in (3.47) within a well defined diagrammatic scheme. In fact, GN analyse the structure of an arbitrary two-particle Green's function G2. While this section concerns itself mainly with the particle-hole correlation function (involved in the density and current response functions), GN's general expressions for G2 will be useful in Chapter 10, where we discuss the two-particle (or pair) spectrum in a Bose-condensed fluid. The formal

5.4 Gavoret-Nozieres analysis

ill

results of GN are valid at all Q and co and are easily rewritten at finite temperatures (using imaginary Bose frequencies in the usual way). In our summary, we work at T = 0 and follow GN's notation closely. A second, equally famous, aspect of the GN paper is their determination of the explicit forms of Gap and %nn in the Q, co —> 0 limit. Using regularity arguments specifically restricted to T = 0, they argue that Xnn in (3.47) and the single-particle self-energies S a^ are non-singular in the limit g, oo —> 0. This is shown to be consistent, in that a direct evaluation of the pole of Gap(Q,co) gives a phonon with the expected (compressional) sound velocity. We defer further discussion of these results (and their extension to finite T) to Section 6.3, after we have derived various rigorous results in the zero-frequency limit in Section 6.1. Using the matrix notation for field operators given in (3.21), GN define a general two-particle Green's function (a, fi,y,d = +, —) at*) = (fapdS(td)aM(ty)a^,^/O^aCa))

(5.59)

In the Bose-condensed phase, there are 2 4 or 16 different two-particle Green's functions. We recall that a^a = a_q_0L. The Fourier transform of (5.59) is proportional to Kl^(ps,py;pp,pa), where it is convenient to use a 4-momentum notation pa = (pa, coa), etc. Using the fact that K^ is invariant under translations in space and time results in the requirement that ps + py = pp + p a , i.e., only three of the "4-momentum" vectors are independent variables. Following GN, we use this fact and work with (see Fig. 5.11)

P+ f >P+ f ;/>' + f ,-* + f ) - K$tf,P',Q) ,

(5.60)

where Q = (Q, co) is the total 4-momentum of the pair of excitations. As an illustration of this notation in a concrete case, we note that the density-response function is given by (at T = 0) Xnn(Q,co) = -i f"

^ei(at(tp(Q,t)p(-Q,0))

where

For further details, we refer to §15 of Lifsh*tz and Pitaevskii (1980) and Chapter 6 of Nozieres (1964).

118

Dielectric formalism for a Bose fluid y

5 Q

Fig. 5.11. Two-particle Green's functions Kl^(pr,p;Q) defined in (5.60).

As discussed in Section 5.1, the diagrammatic analysis of Klp(p',p;Q) in (5.60) leads naturally to improper contributions which carry a single (renormalized) propagator. In terms of three-point kernels Pj!e(p\Q) and PJ- (p,Q)9 these improper contributions are shown in Fig. 5.12 and form what GN call the singular or condensate contribution to K^: c

Kptf,p;Q)

= Q%{p',Q)GPo{Q)Qld{p,Q) .

(5.63)

As usual, the repeated Greek indices (p, o) are summed over. The Bose vertex functions in (5.63) are clearly related to the Aa(Q) vertex functions introduced in Section 5.1, with, for example,

+<M2*)4 0 limit parallels the discussion of zero sound in a Fermi gas. Working with JJJ in (5.81) has a technical advantage in that the integrand has two extra factors of momentum, which makes it better behaved than %nn in the low-Q limit. After the work of Gavoret and Nozieres (1964) and Hohenberg and Martin (1965), it was implicitly accepted in much of the theoretical literature that two-component expressions such as in (3.47) were the key to understanding superfluid 4 He. The scenario was that in superfluid 4 He, Gaj? had a sharp single-particle excitation and that this appeared as a resonance in Xnn with a strength related to the Bose vertex function Aa. This scenario seems to ignore the condensate-induced hybridization of the single-particle and density fluctuations, whose possibility the dielectric formalism of Section 5.1 exposes so clearly. It ignores the fact that both

126

Dielectric formalism for a Bose fluid

terms in (3.47) or (5.24) are intimately connected and can not always be viewed as two separate terms. This is shown most dramatically by the fact that (5.24) is formally equivalent to (5.76). We shall argue that the answer to the question of which expression is most appropriate as a starting point in discussing xwn(Q,ew) depends very much on the wavevector region one is dealing with. Broadly speaking, one can distinguish the resulting spectra of Xnn and Ga(3 by whether or not there is any significant structure above Tx arising from Jm m (5.10) or (3.47). A zero of eR corresponds to a zero sound pole. If there is no such structure, the two terms in (5.24) have a distinct physical significance and (5.24) then gives a better representation of the physics than (5.2). The Xnn spectrum is modified below Tx by the appearance of the single-particle excitation peaks (appropriately renormalized by the Bose condensate) which are associated with the first term in (5.24) or (3.47). Besides the scattering from thermally excited particle-hole excitations, which have their analogue in the normal phase, contributions from pair excitations appear with an intensity dependent on the condensate. We call this the single-particle or SP limit. The simplest illustration of this SP scenario is given by (3.45). In the opposite limit, if there is a zero sound density fluctuation in Xnn for temperatures above Tx, there are important cancellations between the two terms in (5.24). In this case, the expressions in (5.76) and (5.81) are then more useful than (3.47) or (5.24). As illustrated by (5.50) and (5.51), it is possible that the dominant pole of both Xnn and Gajg below Tx will be a renormalized zero sound mode involving both the condensate and non-condensate mean fields. For this reason, we call this the collective zero sound or ZS limit. Which scenario is more relevant in superfluid 4 He is not at all obvious, and here experimental data on S(Q,co) play a crucial role. In the context of the Bose gas models discussed in Section 5.2, the strong-coupling limit corresponds to the ZS scenario defined above and the weak-coupling limit corresponds to the SP scenario. However, the SP and ZS scenarios are not limited to small Q. Moreover, we assumed in the dilute Bose gas models that SP structure arising from Gap was at low energy compared to the phonon frequency. In liquid 4 He, in contrast, we shall argue, that the maxon-roton excitations are SP modes of the normal phase and, due to their high energy, continue to exist in the superfluid phase largely unchanged. At high Q ^ 1 A" 1 , these SP modes will also be the dominant poles of Xnn as a result of condensate-induced hybridization. We refer to Section 7.2 for further discussion of this scenario.

Response functions in the low-frequency, long-wavelength limit

In this chapter, we use the formalism developed in Sections 3.2, 5.1 and 5.4 to discuss various correlation functions in the long-wavelength, lowfrequency limit. It is important that the microscopic theory based on a Bose broken symmetry used to describe the high-frequency excitations probed by neutrons also explains the low-frequency behaviour which characterizes superfluidity. In Section 6.1, we show how the generalized Ward identities given in Section 5.1 lead in a simple way to several rigorous zero-frequency sum rules. We discuss the structure of the lowfrequency, long-wavelength response functions and make contact with the two-fluid description of Landau. In Section 6.2, we discuss the structure of the correlation functions in the hydrodynamic region, as given by the two-fluid equations of Landau (see Khalatnikov, 1965). While the hydrodynamic region of S(Q,co) is difficult to probe by thermal neutron scattering, it can be studied by inelastic Brillouin light scattering (for excellent reviews, see Stephen, 1976; Greytak, 1978). In Section 6.3, starting from the Gavoret-Nozieres formalism summarized in Section 5.4, we review GN's explicit calculation of the phonon spectrum of G^ and Xnn (at T=0). We comment on the significance of the infrared divergences in the g,co —• 0 limit first noted by Gavoret and Nozieres (1964) and clarified in later work, by Nepomnyashchii and Nepomnyashchii (1978), Popov and Serendniakov (1979), and Nepomnyashchii (1983). We also discuss the relation between first and second sound which occurs in the hydrodynamic region and the phonons which arise in the collisionless region.

127

128 Response functions in the low-frequency, long-wavelength limit 6.1 Zero-frequency sum rules and the normal fluid density A microscopic derivation of the two-fluid model was accomplished in the early 1960's, especially by the work of Martin and Hohenberg (1965), Gavoret and Nozieres (1964) and Pines (1965). The crucial step is to formulate a general definition of the normal fluid density PM which is valid for all temperatures 0 < T < Tx and which is not dependent on the existence of long-lived quasiparticles. The appropriate definition of pM originates in Landau's argument (see Section 1.1) that in a slowly rotating bucket of superfluid 4 He, only the normal fluid component rotates. Thus pN is simply related to the moment of inertia of the rotating fluid. By considering the momentum current produced by a constant (transverse) Coriolis force, standard linear response theory gives

tiAQ, = 0 ) .

(6.1)

The normal fluid density is then formally determined by the longwavelength limit of the zero-frequency transverse current-current correlation function, as defined in (5.4) and (5.5). Taking the co = 0 limit, the continuity equations in (5.16a) and (5.166) reduce to the requirement P = -XJAQ,

to = 0) = -XJJ(Q,

m = 0) .

(6.2)

This is equivalent to the /-sum rule for S(Q, co) given by (2.24). Combining (6.1) and (6.2) gives a general definition of the superfluid density ps = p- pN9 namely ps(T)

= - U r n [/JJ(Q>

(o = 0)-

X J A Q , CO = 0 ) ] .

(6.3)

Thus ps is given by the difference between the zero-frequency longitudinal and transverse momentum current-response functions in the longwavelength limit. In a normal Bose fluid, these are identical and thus Ps=0. For a detailed account of the derivation of (6.1) and the long-range spatial correlations which are implied by the fact that XJJ ^ XJJ, we refer to the excellent review article by Baym (1969). Chapters 4 and 6 of Nozieres and Pines (1964, 1990) also give a lucid description of the physics implied by the relations (6.1)—(6.3). As we reviewed in Section 1.2, Pollock and Ceperley (1987) have successfully evaluated ps(T) using a Monte Carlo path-integral technique (see Fig. 1.9). Their work starts with the equivalent of (6.3) written as an integral over real-space current response functions. This paper by Pollock and Ceperley is highly recommended for

6.1 Zero-frequency sum rules and the normal fluid density

129

the physical picture it gives concerning the long-range spatial correlations characteristic of a Bose superfluid. Making use of the dielectric formalism of Section 5.1, we note that for arbitrary (Q,co), all diagrams contributing to XJJ(Q,CO) are regular (proper and irreducible), and hence xljj — X!JJ = Xn ( s e e (5.13b)). Moreover, the diagrams contributing to XJ*J and Xj^ are similar except for their external points and, in the long-wavelength limit, these two functions are identical. Thus we can rewrite (6.1) in the equivalent form pN(T) = -Urn xn(Q,© = 0) .

(6.4)

Combining this result with the first equation in (5.13b) and (6.2), we obtain ps(T) = -Urn ]T A£(Q,co = 0) Ga/KQ,co = 0) A£(Q,co = 0) .

(6.5)

This shows very explicitly how the superfluid density is related to the anomalous Bose current-field correlation functions A^ which describe the Bose broken symmetry. More technically, ps is given by the sum of the improper diagrams contributing to the long-wavelength limit of fjj(Q9co

= 0), i.e., ps = -(XJJ

~ XJRJ).

We now turn to the implications of the generalized Ward identities given by (5.17)—(5.19) in the co = 0 limit. One can show that (5.17) reduces to (Talbot and Griffin, 1983) A^(Q,co = 0) = ^ p a [ s

+ B(Q)] ,

(6.6)

where we have defined B(Q)=

£n(Q,co = 0 ) - £ 1 2 ( Q , c o = 0 ) - i u

(6.7)

and have made use of the equivalence Z 2 i(Q, co = 0) = 2i2(Q,co = 0) (see (3.28)). At co = 0, (5.19) reduces to X

(Q, co = 0) = -p + ^ p [Af (Q, co = 0) - A£(Q, co = 0)] . (6.8)

Combining (6.7) and (6.8) with (6.6), we obtain two very important long-wavelength results (valid at arbitrary temperatures): (6.9) lim B(Q) = ^- — - 1 . e-o y^ 2m[mn0 \

(6.10) v

}

130 Response functions in the low-frequency, long-wavelength limit We note that (5.17) does not allow us to derive any exact result for Using (5.12) and the relation Ai(Q,co = 0) = A2(Q,co = 0), one obtains in the Q -> 0 limit Sn(Q,a> = 0) -Si 2 (Q,a> = 0) = £n(Q,co = 0) - £i2(Q,a> = 0) (6.11) where the second line follows from (6.10). The results (6.9)—(6.11) were first derived at T = 0 (where ps = nm) by Gavoret and Nozieres (1964) using a direct analysis of Feynman diagrams. It may be viewed as a generalized version of the Hugenholtz-Pines theorem quoted in (3.32). The relations (6.6)-(6.8) emphasize the role which £ ^ , Xn a n d A£ play as the basic building blocks of the theory (see discussion at the end of Section 5.1). A direct consequence of (6.11) is that the regular self-energies £ a ^ satisfy the Hugenholtz-Pines theorem (3.32). This means that the poles of Gap have no energy gap in the Q —> 0 limit. Moreover, if one can expand £ a ^(Q, co) around Q = 0, co = 0, the modes will be phonon-like (see, however, Payne and Griffin, 1985). The preceding derivation emphasizes that (6.11) is a direct consequence of the generalized zero-frequency Ward identities and hence of the continuity equations (5.16a) and (5.16b). This close relationship between the Hugenholtz-Pines relation (3.32) and the continuity equation was first demonstrated by Hohenberg and Martin (1965) and Huang and Klein (1964). Since Gap is given by the same equations (3.29) as Gap (with £a£ replaced by £ajg), one finds (in the limit Q —• 0) Gn(Q, co = 0) - Gi2(Q,co = 0) = Gn(Q,a> = 0) - Gi2(Q,co = 0) 1 eQ+B(Q) 2m mn0 {6A2)

where we have used (3.29), (3.30) and (6.11). This last result was first obtained by Bogoliubov (1963, 1970) and Hohenberg and Martin (1965). The 4 He atom momentum distribution nQ for small Q can be directly related to (6.12), as discussed in Section 4.1. We call attention to the fact that pN and ps arise naturally when we work with correlation functions involving the momentum current density (rather than the number density). It will be especially important when

6.1 Zero-frequency sum rules and the normal fluid density

131

working in the two-fluid frequency region to base the discussion on Xjj and A£, rather than Xnn a n d Aa. We also note that the high-frequency limit, A£(Q, co —• oo) is identical to the zero-frequency result (6.9) but with the superfluid density ps replaced by the condensate density ranoAs Talbot and Griffin (1983) have emphasized, the above systematic derivation shows that the exact zero-frequency results (6.12), (6.11), and (6.9) are all really equivalent to (6.4). This follows when one takes into account that the generalized Ward identities (see Section 5.1) are direct consequences of the equation of continuity in a Bose-condensed system. This shows the power the dielectric formalism has in exposing how the Bose order parameter determines the structure of various correlation functions. The relation in (6.12) is of especial interest since it relates the zero-frequency, long-wavelength limit of the single-particle Green's functions to the ratio of the condensate density no(T) and the superfluid density ps(T). In connection with the rigorous result (6.9), Talbot and Griffin (1984b) have used (6.12) to prove that the total mass current density associated with a moving condensate (velocity \s = Q/wi) is indeed equal to the two-fluid result, <J)e = Psv s .

(6.13)

Talbot and Griffin (1983) have calculated the zero-frequency regular functions using the one-loop diagrams given in Section 5.3. Starting from the expression for p^ in (6.4), one can verify that only the thermal scattering part of (5.54) remains in the Q,co —• 0 limit (Fetter, 1970)

Recalling that uj — Vp = 1, this reduces to Landau's well known quasiparticle formula for PN(T). In addition, one can show that Af in (5.55) gives (6.9) while the self-energies in (5.52) and (5.53) reproduce the Gavoret-Nozieres sum rule (6.11) with a superfluid density ps = p — piv,with pN again being given by (1.3). The (somewhat lengthy) calculations leading to these results (see Talbot and Griffin, 1983) show explicitly how the one-loop corrections in (5.52)—(5.55) provide additional terms which result in the natural appearance of a superfluid density ps(T) which is different from mno(T). Taken together, these results give further weight to the argument made in Section 5.3 that the one-loop diagrams for the regular functions (in the dielectric formalism) give a consistent, non-trivial description of an interacting Bose-condensed fluid at finite temperatures.

132 Response functions in the low-frequency, long-wavelength limit In this section, we have emphasized the close relation between a Bose broken symmetry and superfluidity. As is well known (for a review of Kosterlitz-Thouless ideas, see Nelson, 1983), there is no long-range order (LRO) in a two-dimensional (2D) Bose fluid (at T =£ 0), even though it exhibits superfluidity as described by ps and vs. However, quasiLRO and superfluidity in 2D Bose fluids can still be understood as the consequence of a macroscopic occupation of some state. The difference is that because of strong phase fluctuations, the "condensate" in 2D is no longer simply associated with the fraction of atoms in the zeromomentum state but rather with long-range spatial correlations which decay with a characteristic power law. For further discussion, see p. 1372 of Griffin (1987) and Popov (1983, 1987).

6.2 Hydrodynamic (two-fluid) limit In discussing the excitations of quantum liquids, a very basic distinction arises between the low-frequency "hydrodynamic" region and the highfrequency "collisionless" region. These two domains are most easily distinguished by introducing the average lifetime f of the elementary excitations which make the dominant contribution to the thermodynamic and transport properties of the fluid in question. Roughly speaking, T is the lifetime of the elementary excitations with thermal energies of order /c#T (measured with respect to the chemical potential /J). The two frequency domains can be defined as a) < 1/T : hydrodynamic,

(6.15a)

co > 1/T : collisionless.

(6.15b)

Clearly the cross-over frequency cb = 1/T is very dependent on the temperature, since 1/T usually increases as some power of the temperature. In particular, the distinction between the two regions is lost at T = 0. In superfluid 4 He in the temperature region ~ 1 K, a typical value of cb might be of order 109 Hz. (For a more detailed discussion of quasiparticle lifetimes, see Khalatnikov, 1965.) We conclude that inelastic (Brillouin) light scattering probes the hydrodynamic modes (described by the two-fluid equations), while inelastic neutron scattering probes the collisionless modes. The distinction between the collisionless and hydrodynamic domains has been extensively discussed in the case of liquid 3 He using the well

6.2 Hydrodynamic (two-fluid) limit

133

known Landau kinetic equations that describe the quasiparticle dynamics of a Fermi liquid (see, for example, Chapter 1 of Pines and Nozieres, 1966). In a Bose-condensed fluid, the equivalent kinetic equations governing the elementary excitations are available for a WIDBG only in the weak-coupling limit (Kirkpatrick and Dorfman, 1985). However, the basic concepts involved in distinguishing the high- and low-frequency regions are the same in all quantum liquids and solids (solid 4 He is discussed in Chapter 11). The various kinds of collisionless and hydrodynamic modes have been worked out in great detail in superfluid 3 He. The mode structure in this case is very rich due to the p-wave Cooper, pair condensate (for further discussion, see Vollhardt and Wolfle, 1990). In this section, we examine the structure of G^ and Xnn for a Bosecondensed fluid in the hydrodynamic region. The discussion uses the two-fluid equations of motion (Khalatnikov, 1965) to obtain explicit expressions for the correlation functions in the region of low Q and co, where this description is correct. Only in Section 6.3 do we turn to the question of deriving the low Q, co correlation functions directly from a diagrammatic analysis of the structure of a Bose-condensed fluid, which shows the role of the Bose condensate more clearly. The hydrodynamic domain is also referred to as the "collision-dominated" region. Assuming that we are dealing with an acoustic dispersion relation, one sees from (6.15a) that for hydrodynamic modes there are many collisions between the elementary excitations in one wavelength A = 2n/Q of the collective mode. For this reason, the hydrodynamic limit is often difficult to derive in a liquid, when one starts from the microscopic or atomic level. From another point of view, it is simple, in that this limit can be described by equations of motion for "coarse-grained" local variables like pressure, temperature and velocity. These equations are basically conservation laws plus constitutive equations involving transport coefficients (such as thermal conductivity and the various viscosities). This description can be written down from general considerations, a detailed microscopic theory being needed only to evaluate the various thermodynamic derivatives and transport coefficients which occur in the hydrodynamic equations. In the case of a Bose-condensed fluid, these are the well known two-fluid equations as developed by Landau and Khalatnikov (Khalatnikov, 1965). They predict two kinds of wave-like phenomena to occur (first and second sound). Second sound has been studied in great detail in the context of superfluid 4 He, but it also exists in a WIDBG, as discussed by Ma (1971), Popov (1983), Kirkpatrick and Dorfman (1985) and Gay and Griffin (1985).

134 Response functions in the low-frequency, long-wavelength limit Starting with the linearized equations of motion describing the hydrodynamic domain of any system, one can evaluate the various correlation functions in the low Q and a> region (Kadanoff and Martin, 1963). The calculation is straightforward, but algebraically rather lengthy. The correlation functions predicted by the two-fluid equations were first worked out by Hohenberg and Martin (1964, 1965) as well as by Bogoliubov (1963, 1970). We refer the reader to Chapter 10 of Forster (1975) for more details, as well as the classic review article by Martin (1968). In our subsequent analysis, we omit the transport coefficients (dissipation) in order to expose the structure of the correlation functions in as simple a manner as possible and, moreover, we neglect vorticity in the superfluid velocity field (Vxvs = 0 ) . The dynamic structure factor S(Q,co) for Q < 10~3 A" 1 can be studied by using inelastic (Brillouin) light scattering as well as ultrasonic measurements, as discussed in Section 4 of the review article by Woods and Cowley (1973). The theory of Brillouin light scattering in liquid 4 He is reviewed by Stephen (1976). One is usually probing the hydrodynamic region (6.15a) when one uses these experimental techniques. For T > 7^, S(Q,co) as measured by light scattering will exhibit a Rayleigh central peak at co ^ 0 due to scattering from (diffusive) temperature fluctuations, in addition to two Brillouin sound-wave peaks at co = +cQ, where c is the first sound velocity (see Pike, Vaughan and Vinen, 1970). Lightscattering experiments on liquid 4 He are intrinsically difficult since the electronic polarizability of the closed-shell He atom is very small and thus the scattering is weak. Moreover, the intensity of the diffusive central component relative to the sound-wave peaks is proportional to the Landau-Placzek ratio y — 1 (where y = CP/Cv), which rapidly decreases as the temperature is lowered below the gas-liquid critical point (see Fig. 6.1). In spite of the kinematical difficulties, neutron scattering has also been successfully used (Woods, Svensson and Martel, 1975, 1978) to measure S(Q,co) at T = 4.2 K and SVP, for momentum transfers as small as 2 — 0.1 A" 1 . The results are consistent with the expected hydrodynamic structure discussed above. Neutron scattering can be used to show the cross-over from the hydrodynamic to the slightly higher collisionless sound velocity (Woods, Svensson and Martel, 1976). At T = 2.3 K, the velocity increases from 220 ms" 1 to 255 ms" 1 at around Q ~ 0.25 A" 1 (see Fig. 6.2). In the superfluid phase below 7^, the main difference in the hydrodynamic region is the appearance of propagating second sound modes in

6.2 Hydrodynamic (two-fluid) limit

3.0 Temperature (K)

135

4.0

Fig. 6.1. The temperature dependence of the Landau-Placzek ratio in normal liquid 4 He as given by thermodynamic data up to the boiling point at 4.2 K (1 bar) [Source: Pike, 1972].

place of the diffusive central peak. The dynamic structure factor implied by the two-fluid equations is given by (Ginzburg, 1943; Hohenberg and Martin, 1964, 1965) O2 S(Q,co) = ^ m where the weight of the first sound mode is

- 4 ^ 2 ) ] , (6.16)

u\-v2 u\ - u2n

(6.17a)

with (s is the entropy and Cy the specific heat) 2

v = T— — . Cy

(

The first and second sound velocities in (6.16) satisfy the exact relations 2 2 «

(6.18)

136 Response functions in the low-frequency, long-wavelength limit

0.30

0.20

0.10

0.00

Fig. 6.2. The phonon dispersion relation at T = 2.3 K showing the transition from first sound to zero sound. The dashed line is drawn as a guide to the eye. The full line has a slope equal to thermodynamic sound velocity [Source: Woods, Svensson and Martel, 1976].

In the case of a dilute Bose gas, where (Cp/Cy) — 1 is appreciable, one finds that both first and second sound have appreciable weight in the density fluctuation spectrum in (6.16) (see, for example, Gay and Griffin, 1985). In superfluid 4 He at SVP, in contrast, we have Cp ~ Cy to a very good approximation, whence it follows that u\\ ~ v and u\ ~ (dp/dp) 2. Consequently we have Z\ ~ 1 in (6.16) and thus the second sound mode has negligible weight in S(Q,co) in superfluid 4 He. In physical terms, this means that second sound at SVP is mainly a temperature wave, which is only weakly coupled into the density fluctuation spectrum (Khalatnikov, 1965). However, the thermal expansion coefficient and hence the LandauPlaczek ratio y — 1 diverges at Tx; moreover, this expansion coefficient is considerably increased as we approach the superfluid transition by working under high pressure (Ferrell et a/., 1968). This feature has been used by several groups to study the behaviour of the second sound component in S(Q,co) near Tx in great detail (for reviews, see Greytak, 1978;

6.2 Hydrodynamic (two-fluid) limit

137

0.15

0.10

5 atm -

0.05

J2.7 atm

1.8 2.0 Temperature (K)

Fig. 6.3. The Landau-Placzek ratio (CP/CV) — 1 vs. temperature for various pressures in superfluid 4 He. The vertical dashed lines indicate the value of the transition temperature TA(p) at various pressures [Source: Vinen, 1971].

Stephen, 1976). The ratio of the second sound to first sound scattering intensity can be approximated by the Landau-Placzek ratio y — 1 within a few per cent if we are outside the critical region (AT ^ lmK), as discussed by Hohenberg (1973) and O'Connor, Palin and Vinen (1975). In Fig. 6.3, we plot y — 1 as a function of the temperature for different pressures, as given by thermodynamic data (Vinen, 1971). These results may also be relevant to neutron-scattering studies of S(Q,co) in superfluid 4 He at very small momentum transfers, which are usually analysed (see Section 7.1) without considering any residual contribution from second sound. We recall that the second sound component is the critical mode associated with the Bose broken symmetry near 7^, where its behaviour is somewhat complicated (see Fig. 6.4). For further details and references, we refer to Tarvin, Vidal and Greytak (1977).

Green's functions in the two-fluid region

A unique aspect of a Bose-condensed fluid is that the single-particle Green's function Gap can be directly related to correlation functions such as Xnn and yf33 in the hydrodynamic domain. This is a consequence of

138 Response functions in the low-frequency, long-wavelength limit s «2, w)

= u2(T)Q

Hydrodynamic region

Critical region

Limiting structure

factor (7 = Trf

Fig. 6.4. Schematic plot of the observed second sound contribution to 5(Q, co) in the superfluid phase just below TA. The fluctuations in the Bose order parameter are characterized by a correlation length £(T), which diverges as TA is approached. For wavevectors probed in light scattering, Q£(T) becomes of order unity for temperatures within about 1 mK of TA [Source: Tarvin, Vidal and Greytak, 1977].

the fact that the local superfluid velocity (a hydrodynamic variable) is related to the gradient of the phase fluctuations. The following heuristic discussion captures the essential physics (see Appendix B of Griffin, 1981; p. 107 of Lifsh*tz and Pitaevskii, 1980; Chapter 10 of Forster, 1975). Quantum field operators can be expressed in terms of amplitude and phase operators. A systematic study shows that the slow, long-wavelength phase fluctuations dominate over the amplitude fluctuations of an interacting Bose-condensed system; hence one can use

in discussing the dynamics in the limit of small Q and co. Using (6.19) in the usual quantum mechanical expression for the current operator, one can identify the part of the current directly associated with the motion of the condensate as J s (r) =

(6.20)

6.2 Hydrodynamic (two-fluid) limit

139

where the superfluid velocity operator is given by the gradient of the phase

vs(r) = i v & r ) .

(6.21)

(As Talbot and Griffin (1984b) have discussed, the additional current associated with the non-condensate atoms which are dragged along by the condensate atoms combines with the current in (6.20) to give (6.13).) With (6.21), the longitudinal part of the superfluid velocity correlation function (as defined in (5.4) and (5.5)) is directly related to the phase fluctuation spectrum /VsVs(Q,co) = ^X^(Q,oj)

(6.22)

Moreover, using (6.19) in (3.10) and expanding the exponentials for small phase fluctuations, we find (6.23) Thus the single-particle Green's functions reduce to G n (Q,co) = -Gi 2 (Q,(o) (6.24) where we have used (6.22) in the second line. The expression (6.24) gives a direct relation in the small Q, a> limit between xisVs ( a s determined by the two-fluid equations) and Gap. This important relation arises only because of the Bose broken symmetry. A systematic way of separating out the slow, long-wavelength fluctuations from the fast, short-wavelength motions in field-theoretical calculations is discussed in Sections 18 and 19 of Popov (1983). His procedure gives a more precise basis to the preceding heuristic proof of (6.24). Gn(Q, icon) can be expressed in terms of the single-particle spectral density (see pp. 150ff of Mahan, 1990)

=f ^ J_oo 27C

(6.25) -

This spectral density has been discussed at length in Section 4.1. The single-particle spectral density A(Q,co) can be obtained from the twofluid hydrodynamic equations once we have the key relation given in (6.24). From our general analysis in Sections 5.1 and 5.5, G^ can be

140 Response functions in the low-frequency, long-wavelength limit expected to exhibit the same resonances as /„„, corresponding to first and second sound (albeit with different weights). The expression for A(Q,a>) valid in the two-fluid domain has been discussed with some generality by Hohenberg and Martin (1964). In superfluid 4 He away from Tx, one can ignore terms of order y — 1 (i.e., CP ~ Cv), in which case the general expression simplifies to (Cheung and Griffin, 1971a) A(Q, co) = 2n^sgn ri

co [^u\6(co2

- u2Q2) + ^u2uS(co2

- u2uQ2)]

(6.26) where u\ and u\\ are defined in (6.18). Inserting (6.26) into (6.25), one may verify that at zero frequency ^

(6.27)

This is, as expected, in perfect agreement with the Bogoliubov-Hohenberg-Martin sum rule (6.12). In contrast to the expression for <S(Q, co) given in (6.16), second sound has appreciable weight in the single-particle spectral density A(Q,co) given in (6.26) within the hydrodynamic domain of superfluid 4 He, as long as T ^ 0. (In the limit where (6.26) is valid, u\\ reduces to v in (6.17b)). The strong amplitude of second sound in (6.26) emphasizes that this acoustic mode is associated with the Bose order-parameter fluctuations, which are in turn closely related to the field fluctuations described by the singleparticle Green's functions. The second sound pole dominates ,4(Q, co) in the region near Tx (see, for example, Tarvin, Vidal and Greytak, 1977). In comparing the spectrum exhibited by Gajg and /„„ in superfluid4 He in the two-fluid region, we can do no better than quote the classic paper by Hohenberg and Martin (1964): (1) "When T = 0, the second sound velocity u\\ approaches a finite value but its contribution to all correlation functions (i.e. the residue at co = u\\Q) vanishes." (2) "At T = 0, both Xnn and Ga^ are both sharply peaked about co = u\Q, where u] = dp/dp. This pole exhausts the /-sum rule for /„„ but not the corresponding sum rule for Ga£." (3) "With increasing temperature, Xnn is not greatly altered; the oscillation at co = u\Q continues to dominate it, merging smoothly at T = Tx with the dominant contribution from ordinary sound in the normal phase. There is also a contribution to Xnn (with relative weight (Cp — Cv)/Cv < 1) from second sound (the mode of tempera-

6.3 The nature of phonons in Bose

fluids

141

ture transport for T < Tx). This contribution to Xnn merges with the corresponding one for T < Tx (the thermal conduction mode co = iDTQ2V (4) "As the temperature increases, the correlation function Gap is substantially altered. In fact, at T = Tx, the oscillation at co = u\Q has vanishing weight in Gap. Likewise, the contribution at co = u\\Q vanishes as it 'joins' the oscillation O(Q2) describing a single-particle energy of a non-superfluid system." 6.3 The nature of phonons in Bose fluids As we mentioned in our overview in Section 1.1, the fact that the quasiparticle spectrum at low Q is phonon-like in superfluid 4 He was usually viewed as obvious in the older literature. This phonon mode was identified as a compressional sound wave familiar in all liquids and gases. In the microscopic field-theoretic literature, however, this phonon spectrum has always presented a challenge. The key requirement is to prove that because of the Bose condensate, the phonon pole exhibited by the density-response function XnniQ, &>) is indeed the only low-energy pole of the single-particle Green's function Ga^ (Q, co); that is to say, the elementary excitations, which completely determine the thermodynamic and transport properties at low temperatures, are exhausted by the compressional density fluctuations. Moreover, one must also show that this phonon frequency is essentially temperature-independent in superfluid 4 He (see Section 7.1 for discussion of neutron data concerning this), and smoothly transforms into the zero sound pole of Xnn above Tx. We have seen in Chapter 5 that a Bose broken symmetry naturally leads to a mixing of the spectra exhibited by Gap and Xnm as shown by (3.47) or, more explicitly, (5.24). This sharing of poles is also present in the hydrodynamic region described by the two-fluid equations, as shown by the results in Section 6.2. As we mentioned in Section 3.2, Gavoret and Nozieres (1964) gave the first convincing proof that both Gap and Xnn exhibited the same phonon mode in the Bose-condensed phase. Their argument proceeds as follows. As indicated in our review of GN in Section 5.4, a key role is played by the singularities arising from the product of two fully renormalized Beliaev propagators. These occur in the intermediate state of Feynman diagrams contributing to F,P and Ea0 (see (5.68)-(5.70)). In Fermi liquid theory, similar products of two propagators give rise to singularities related to the low-energy particle-hole excitations. The

142 Response functions in the low-frequency, long-wavelength limit well known Lindhard function is given by (the variables represent a 4-momentum, as in Section 5.4)

- i £ Go (p + f) Go (-p + f) = -i X G0(p + G)G0(p) . (6.28) The analogue of this expression in a Bose-condensed system is given by (5.67). At T = 0 (to which GN restrict themselves), there is no contribution to (5.67) from thermally excited quasiparticles (particlehole excitations) but only from the creation of two quasiparticles. This is illustrated explicitly by the model expression in (3.45). Carrying out the 4-momentum integration in expressions like (5.67), GN find a logarithmic divergence G

n* n* (P (P ++ ff )) % % (-P (-P ++ §§ )) ~~ In In II OJ OJ22 -- c2Q2 \ .

(6.29)

This infrared divergence is a direct consequence of assuming that Gap (p, co) has a phonon pole G g/? (p,ca)~

CD2 —

Clpl

(6.30)

It comes from the fact that the pair energy cop+ iQ + cop_i^, which occurs in the denominator of the integrand of (6.29), becomes vanishingly small in the limit of small Q and p. However GN note that such divergences cancel out in the final results for Gap and Xnn in the Q, co —• 0 limit and hence conclude it is safe to eliminate these divergences by introducing a fictitious energy gap A into the excitation spectrum of Gap. This "regularity assumption" allows perturbative calculations about Q = 0 and co = 0, a procedure which makes sense since the final results are well defined when the gap A —• 0 is set to zero at the end. Working within the above scheme, GN argue that it is sufficient to evaluate Xnn at Q = 0, co = 0. Moreover they compute the threepoint kernel P^ given by (5.69) and the Beliaev self-energies Ea^ using Taylor expansions about Q = 0, co = 0. Making a careful diagrammatic analysis, the terms at Q = 0, co = 0 as well as the Taylor expansion coefficients all reduce to various thermodynamic quantities and associated derivatives. This whole procedure makes use of exact Ward identities and is, technically, very similar to the Green's function analysis of the response functions of a Fermi liquid in the Q, co —• 0 limit (see Chapter 6 of Nozieres, 1964). The final results of this very lengthy procedure are deceptively simple

6.3 The nature of phonons in Bose

fluids

143

(at least at T = 0): GH(Q,Q>)

= G22(Q,co) = -G 12 (Q,co) = ^ , 2n2 n co2 — c2Q2

(6.31)

where c2 is the sound velocity given by compressibility zero-frequency sum rule (2.28) and n = n — nQ is the depletion. We note that (6.32) and (6.33) are equivalent to

The amazing feature of the GN calculation is their explicit proof that the phonon velocity which enters into the pole of Gaj? is given precisely by the thermodynamic derivative which enters the compressibility sum rule. This is proven rigorously, within the calculational procedure previously outlined. In the process of obtaining the above results, GN also obtained the zero-frequency results given by (6.9), (6.10) and (6.11) using a direct diagrammatic analysis. In addition, they explicitly evaluated the vertex function Aa(Q,co) to obtain the rigorous result (at T = 0) .

,,,5,

to lowest order in co. Nepomnyashchii and Nepomnyashchii (1978) later gave a careful analysis of the infrared divergences discovered by GN. They concluded that the final results summarized by (6.31)—(6.35) were correct but that 212 (Q, CD) - 1 /In | co2 - c2Q2\ and hence (6.36)

These results required extensive changes in the original analysis of GN, which assumed that S n ( Q = 0,co = 0) was finite. The results in (6.36)

144 Response functions in the low-frequency, long-wavelength limit can be related to the thermodynamic derivatives (at T = 0) dfi

d/n

me2

dno

As noted by Nepomnyashchii and Nepomnyashchii (1978), (6.35)-(6.37) imply that A a (Q = 0,co) oc co and XnniQ^ = 0) = 0. This last result is consistent with the expression given in (6.34). It means that the condensate part of the density-response function makes no contribution to the compressibility sum rule (2.28) (see also Griffin, 1981). The physics behind (6.36) will be discussed shortly. It is useful to recall briefly the heuristic approach used by Hohenberg and Martin (1965) to derive (6.31) and (6.34). They first give general arguments (see HM (6.38)) that the longitudinal current correlation function Xn has a two-component form similar to (3.47). They write it in the form (see HM (5.16) and also Griffin, 1979a)

where the superfluid velocity correlation function is related to the singleparticle Green's function as in (6.24). Within the dielectric formalism, such an explicit two-component expression for XJJ is given in Section 8.3. The functions PJV(Q> 1) results entirely from their coupling to quantum field correlations as a result of Bose condensation. In an uncondensed Boson fluid this coupling is absent and Xnn is not sharply peaked for COT > 1." The analysis of the low-Q behaviour of Xnn(Q,co) given in Chapter 7 of Nozieres and Pines (NP, 1964, 1990) is also based on the same scenario, as shown by the statement "in the long wavelength limit, the only density fluctuations of importance are those produced by exciting a single quasiparticle from the condensate" (quote from p. 97 of NP). Since the origin of the sharp phonon resonance in Xnn was as a pole of Gajs, NP introduced the term "quasi-particle" sound to distinguish it from ordinary sound. At finite T, it was argued that Xnn m (3.47) would exhibit additional structure at low energies arising from the thermally excited quasiparticles. Above T ^ I K , the dominant contribution would be from rotons, while for T < 0.6 K, it would be from phonons. This thermal scattering was assumed to produce a broad background, such as given by (3.45), without any particular structure. In the analysis of NP (see also Griffin, 1979a), the two terms in (3.47) were identified with the condensate (or "superfluid") and normal (or "normal fluid") contributions. (For further discussion of this identification, see Section 8.3.) In the GN analysis, the role of any density fluctuations associated with the "normal" term Xnn w a s effectively suppressed in Xnn as a direct consequence of the infrared singularities. As a result, the resonances in Xnn as given by (3.47) are associated completely with the first term containing Gajg. Moreover, in the Q, co —• 0 limit, the poles of Gap are tied in closely with the fluctuations of the underlying Bose order parameter. In this limit, the long-wavelength density response of a Bose-condensed liquid at T = 0 is very similar to that in a dilute Bose gas as described by the Bogoliubov approximation (see Section 3.3) but for different reasons. In the latter case, there are essentially no non-condensate atoms while in a Bose liquid, the normal fluid dynamics is "suppressed" due to the infrared anomaly (as emphasized by Nepomnyashchy, 1992). We now turn to the question of the extent to which the GNP scenario described above is valid outside the small Q, co region at low temperatures. We have emphasized that (6.31) and (6.34) are rigorous results but have been only proved at T = 0. These expressions are in precise agreement with the zero-temperature limit of the two-fluid formulas (6.16) and (6.25).

150 Response functions in the low-frequency, long-wavelength limit In this limit, the weight of the second sound mode vanishes in both G^ and Xnn- Only the first sound pole remains, with a velocity u\ which is equivalent (at T = 0) to the compressional sound velocity c in (6.34). The fact that GN exhibited a phonon mode with the compressional sound velocity suggests that their results are, strictly speaking, valid only in the low-frequency domain. As shown in Fig. 6.2, there is a slight difference in magnitude between the first sound velocity c and the collisionless zero sound velocity u. Rather than c, the quasiparticle phonon velocity is given by u in the collisionless domain. We also note that the phonon dispersion curve WQ at low temperatures exhibits anomalous dispersion, with the slope being larger than the compressional sound velocity c up to Q = 0.55 A" 1 (see Fig. 7.2). This means that the expression in (6.29) would not be divergent at COQ for Q ^ 0.1 A" 1 . What the GNP scenario leaves out is that, in a Bose liquid, both G^ and xnn may exhibit excitation branches that have nothing to do with fluctuations of the condensate (for an early criticism of this kind, see Straley, 1972). This is certainly the case in the normal phase. We recall that, to a certain degree, collisionless density fluctuations are present in any liquid. In addition, the single-particle excitations Q2 /2m in a normal Bose gas are expected to be strongly renormalized but still present in a normal Bose liquid. A new scenario was formulated by Szepfalusy and Kondor (SK, 1974) in terms of understanding how the coupled spectra of Ga£ and Xnn in the superfluid phase uncoupled and smoothly merged with the spectra of the normal phase as the condensate vanished. The dielectric formalism exposes this structure in the most direct way. The interpretation of Glyde and Griffin (1990) is developed within this SK scenario (see Section 7.2). The GNP scenario given above is based on (3.47) or (5.24), while the SK scenario is based on (5.76). While these expressions are formally equivalent, they can lead to quite different physical pictures. In particular, the dielectric formalism results based on (5.76) show how Xnn and Gaj? can both be dominated by a phonon whose velocity is temperatureindependent. This is because the condensate and normal parts of xm add up coherently, as illustrated by (5.50). In the GNP analysis starting from two-component expressions like (3.47), in contrast, inclusion of the effect of the thermally excited quasiparticles (the "normal fluid") can lead to a strongly temperature-dependent phonon velocity (a difficulty which is illustrated by the results in Section 7.3 of NP and Appendix C of Hohenberg and Martin, 1965). The dielectric formalism of Chapter 5 naturally leads to a way of

6.3 The nature of phonons in Bose

fluids

151

interpreting the phonon excitation which is valid for both normal and superfluid Bose fluids (gas or liquid). In this interpretation, the phonon is viewed as a zero sound particle-hole excitation associated with two distinct kinds of effective fields. One is produced by the condensate atoms and is present in both gas and liquids. The other effective field is associated with the dynamics of the non-condensate atoms and may be thought of as the "normal fluid" effective field associated with thermally excited quasiparticles. This normal fluid mean field is well defined in a liquid but not in a gas. At T = 0, only the condensate mean field is important. However, as the temperature increases, the condensate field decreases in strength while the non-condensate mean field becomes increasingly important (in a liquid). Thus there is a smooth transition to the normal phase above Tx in the case of a Bose liquid. This zero sound interpretation of phonons in Bose-condensed fluids has its roots in the work of Pines (1963) and Ma, Gould and Wong (1971), although these authors only considered the low-temperature limit. The analysis of GN, like other analyses which we have been reviewing in this section, is strictly concerned with the excitation spectrum in the double limit of small Q and small co. We have seen that the HugenholtzPines theorem (3.32) requires that, in the Q -> 0 limit, there must be a zero-energy Goldstone mode associated with the breaking of gauge symmetry. The low-energy phonon spectrum exhibited by Gap and %nn, however, does not a priori exclude additional high-energy modes from appearing in the long-wavelength limit. Indeed, we know that in superfluid 4 He, the condensate-induced hybridization results in the high-energy two-roton spectrum having finite weight in Gap and Xnn even at small Q (see Chapter 10). GN note that their final result for Xnn given by (6.34) in the longwavelength limit is consistent with the compressibility sum rule as well as the /-sum rule (2.24). The fact that (6.34) exhausts the /-sum rule can be easily over-interpreted. It simply means that the phonon is the only low -energy mode at long wavelengths. A calculation which is limited to the small-co region can clearly say nothing about the existence of high-energy modes. One has the example of quantum crystals, where S(Q,co) can exhibit a lot of structure which makes no contribution to the /-sum rule (see Section 11.2). A useful sum-rule analysis of the single-phonon, multiphonon and interference contributions to S(Q, co) in the small-Q limit is given by Wong and Gould (1974) within the one-loop approximation (see pp. 292ff of WG). In the dielectric formalism of Chapter 5, a crucial role is played by the

152 Response functions in the low-frequency, long-wavelength limit regular single-particle Green's function Gap. As noted in Section 6.1, the regular self-energies £aj? obey the Hugenholtz-Pines theorem and thus Gap may also exhibit a low-frequency branch (probably acoustic) in the long-wavelength limit. As with Gajg, such a low-energy pole can be directly associated with the fluctuations of the Bose order parameter. However, Gap may also exhibit additional high-energy modes, corresponding to intrinsic single-particle excitations characteristic of the normal phase. These would not be expected to be modified by the appearance of a Bose condensate if they existed at relatively large energy. (In contrast, the two-roton multiparticle spectrum will only appear in Gap with finite weight in the condensed phase, as can be seen from the one-loop selfenergies in (5.52) and (5.53)). Thus in parameterizing the spectrum for Gaj5 for superfluid 4 He, one should allow for high-energy modes as well as a gapless low-energy excitation. These remarks will be the basis for the model spectrum we postulate in Section 7.2 in order to understand the S(Q,co) line-shape data (see Fig. 7.22). In the collisionless region, we have argued that the dielectric formalism naturally leads to viewing the phonon as a zero sound mode but one which is associated with the mean fields of both the condensate and non-condensate atoms. This generalized zero sound picture allows one to understand why the collisionless phonon velocity in superfluid 4 He is essentially unchanged as we go from T = 0 to above Tx. Different scenarios are possible within the dielectric formalism framework and, as noted above, Nepomnyashchy (1992) interprets the phonon mode as being essentially a single-particle (SP) excitation at all T below 7^, as in the GNP scenario described above. Further discussion of this alternative scenario is deferred to Section 12.1. In summary, we have seen in this section that the nature of the phonon excitations in Bose-condensed fluids is quite subtle. We have tried to distinguish between the phonons in the two-fluid domain (where they are intimately tied to the oscillations of the condensate) and those in the collisionless domain. We have argued that the only available rigorous microscopic calculations of the structure of Xnn and Ga^ are, in fact, mainly concerned with the long-wavelength, low-frequency two-fluid domain. It is really only in this "macroscopic" domain that phonons can be directly related to the dynamics of the condensate (or superfluid), a point also emphasized by Nozieres and Pines (1964, 1990). In contrast, the phonons which play the role of elementary excitations in thermodynamic and transport properties are in the collisionless region. These are most directly studied by inelastic neutron scattering, as we discuss in Chapter 7.

7 Phonons, maxons and rotons

In this chapter, we review the high-resolution neutron-scattering data for the dynamic structure factor S(Q,co) and suggest an interpretation within a unified picture of the excitations in liquid 4 He consistent with the ideas of Chapter 5. We argue that the phonons (0.1 < Q < 0.7 A" 1 ) in the collisionless region and rotons (Q ~ 1.9 A" 1 ) are really two separate branches of the density fluctuation spectrum in the superfluid phase which are hybridized by the condensate. The low-wavevector phonon is interpreted as a zero sound collective density fluctuation while the largeQ maxon-roton is interpreted as a strongly renormalized single-particle excitation. In the intermediate-wavevector region 0.8 < Q < 1.2 A" 1 , we argue that there is evidence that both excitation branches, a sharp single-particle (or atomic-like) maxon excitation and a broad high-energy zero sound phonon, are observed in S(Q,co). Within this scenario, the appearance of the sharp maxon-roton resonance (0.8 < Q <J 2.4 A" 1 ) in S(Q, (o) below the superfluid transition temperature Tx is direct dynamical evidence for the Bose broken symmetry and the associated Bose condensate in superfluid 4 He. In Section 7.1, we review the neutron-scattering intensity data for small, intermediate and large wavevectors. In Section 7.2, these results are interpreted starting from the assumption that superfluid 4 He is a Bose-condensed liquid. The condensate inevitably leads to a mixing of the single-particle spectrum described by Gaj5(Q,co) and the density fluctuation spectrum described by S(Q,co). In the scenario we have summarized in Section 5.5, the crucial question is: for a given wavevector Q, are we in the zero sound (ZS) or the single-particle (SP) regime? While our specific microscopic picture is new (Glyde and Griffin, 1990; Griffin, 1991), it is a natural development within the field-theoretic approach which we use in this book. We attempt to give a consistent 153

154

Phonons, maxons and rotons

picture of the entire density fluctuation spectrum over a wide range of Q, co and T. This scenario is still at an early stage of development and many specific aspects are still tentative. We hope our preliminary analysis will, however, be a stimulus and guide to future investigations. While this chapter contains many plots of experimental data, it should be emphasized that this book is primarily concerned with theory. Thus, in our selection of neutron data, our main interest is in deciding which of several possible theoretical scenarios is being realized in superfluid 4 He. We do not give any critical discussion of the experimental data as such. The recent time-of-flight data taken at IN6 at ILL (Andersen, Stirling et al.y 1991) is especially useful since it covers such a wide range of Q, co and T. Earlier high-resolution data taken using triple-axis spectrometers is less extensive but has the advantage of giving S(Q,co) in a more direct fashion than time-of-flight methods. The triple-axis data is nicely summarized by Svensson (1989, 1991).

7.1 <S(Q, co): neutron-scattering data

In Section 2.2, we have given a brief summary of the experimental results for S(Q,co) obtained from neutron-scattering studies over a wide range of Q and co, and described how the characteristic features change as we go from low temperatures (1 K) to Tx and above. Section 2.2 should be reviewed before reading the more detailed discussion of data given in the present section. At low temperatures, S(Q, co) exhibits an extremely sharp quasiparticle peak up to about 2.4 A" 1 . For example, the intrinsic phonon half-width at half-maximum TQ is of the order of 0.025 K at 1.2 K (Mezei and Stirling, 1983). Most plots of S(Q,co) in the literature have background scattering (empty-cell) removed but still include instrumental broadening, which depends on the particular wavevector Q being studied and the neutron spectrometer used. Since the intrinsic width of the quasiparticle peak is so small at low temperatures (T < 1 K), it can be ignored in view of the much larger instrumental resolution width. Thus the observed low-temperature quasiparticle peak in S(Q, co) can be fit to a (Gaussian) resolution function to find the instrumental half-width r G . The measured Sexp(Q,&>) is then viewed as the intrinsic density fluctuation spectrum S(Q, co) convoluted with this Gaussian resolution-function. In the recent study by Stirling and Glyde (1990) at Q = 0.4 A" 1 , for example, the Gaussian resolution half-width at half maximum is estimated to be TG = 0.022 THz (1.1 K) assuming that the width of the phonon peak at

7.1 S(Q,co): neutron-scattering data

155

Energy (THz) Fig. 7.1. The inelastic neutron-scattering intensity vs. frequency for Q = 0.4 A~*and T = 1.35 K. As usual, the background (empty-cell) scattering has been extracted. The line is a least-square fit to a convolution of (7.3) and (7.4) with a Gaussian resolution function, as described in the text [Source: Stirling and Glyde, 1990].

T = 1.35 K shown in Fig. 7.1 is entirely instrumental in origin. Following usual practice, both the intrinsic S(Q9co) and the resolution-broadened Sexp(Q?ft>) are referred to as S(Q,a>).

Low momentum: phonons In this section, we will focus our attention on neutron data showing how the phonon line shape in S(Q,co) changes with temperature for T ^ 1.3 K. However, we first make a few remarks about the phonon line width at lower temperatures. Phonon lifetimes have been measured with great accuracy by ultrasonic techniques in the temperature region T < 0.6 K, where one may ignore phonon interactions with thermally excited rotons.

Phonons, maxons and rotons

156

260

200

0.4 Wavevector (A"1)

0.6

0.8

Fig. 7.2. Phonon phase velocities co/Q for liquid 4He at 1.2 K and SVP. The quasiparticle frequency shows anomalous dispersion in the region 0.1 < Q < 0.55 A"1 [Source: Stirling, 1983]. This is a well studied, although complicated, subject (see Section 4 of Woods and Cowley, 1973; Maris, 1977). At sufficiently low temperatures, the main source of phonon damping is spontaneous decay via three-phonon processes (using the LandauKhalatnikov picture reviewed in Section 1.1). Unless one includes the finite width of the phonons, this three-phonon process is controlled by kinematics and only occurs because of "anomalous" dispersion. Anomalous dispersion refers to the fact that, as first suggested by Maris and Massey (1970), the phonon dispersion curve in Fig. 1.3 curves upward slightly before bending over, i.e., coQ = cQ(l

yQ2)

(7.1)

where y(T) is negative. This curvature has been studied by several groups using neutron scattering (for references, see p. 338 of Glyde and Svensson, 1987). The high-resolution data of Stirling (1983) at T = 1.2 K are shown in Fig. 7.2. As a result of this anomalous dispersion, a phonon can decay via the three-phonon process as long as its phase velocity is greater than c, i.e., as long as Q is less than some threshold wavevector Qc.

7.1 S(Q,co): neutron-scattering data

157

Stirling (1983) obtained the SVP value Qc = 0.55 A" ; Qc decreases with pressure (Svensson, Martel and Woods, 1975). Evidence for the resulting three-phonon decay has been reported by Mezei and Stirling (1983), who observed the expected abrupt decrease in the phonon width (at T = 0.95 K) in the region Q ~ 0.5-0.6 A" 1 . At such low temperatures, it is argued that phonons with Q > Qc can decay only by the much weaker four-phonon process involving thermally excited phonons. For temperatures above 1 K, in contrast, the main damping mechanism of phonons is scattering from thermally excited rotons; thus one expects that TQ will be approximately proportional to the number of rotons present

NR ~ Vfe~A/T

(7.2)

This prediction is borne out by finite-temperature studies on the width of phonons, as discussed in the remainder of this section (Cowley and Woods, 1971; Mezei and Stirling, 1983; Stirling and Glyde, 1990). From our perspective, however, the most striking feature of the phonon peak in the collisionless region of S(Q,co) is its persistence while the temperature increases, right through T*. This somewhat surprising result was first observed in a classic experiment by Woods (1965b) for Q = 0.38 A" 1 . More recent studies that confirm these original results include the work of Mezei and Stirling (1983) for 0.3 < Q < 0.7 A" 1 at temperatures up to 1.6 K, Stirling and Glyde (1990) for Q = 0.4 A" 1 over the temperature range 1.35 < T < 3.94 K, and the very complete ILL data of Andersen, Stirling et al. (1991). We concentrate on the data of Stirling and Glyde (SG), since it is the most extensive set of high-resolution data which has been analysed and published on the full temperature dependence of S(Q,co) for Q in the phonon region (see Fig. 1.4). As the temperature rises from 1.35 to 2.96 K, the phonon line at COQ ~ 0.16 THz (7.7 K) broadens and its peak intensity slowly decreases, but the peak position remains essentially constant (Fig. 1.4). In particular, there is no evidence of any qualitative change in the region around the superfluid transition at 2.17 K. There is also clear evidence (see Fig. 7.1) of a weak, high-frequency structure with an energy slightly larger than twice the roton energy 2A (i.e., it peaks at 0.40 THz (19.7 K) at 1.35 K). Evidence for this high-energy peak is also shown in Fig. 10.4. We defer analysis of such "multiparticle" contributions to Chapter 10. The original study by Woods (1965b) did not have sufficient resolution to measure the intrinsic phonon line width, but later studies by Cowley

158

Phonons, maxons and rotons

0.25

Temperature o 2.07 K A 2.12 K + 2.17 K x 2.26 K o 2.49 K v 2.76 K

0.2

0.3 0.4 Energy transfer (THz)

Fig. 7.3. Neutron-scattering intensity vs. frequency for Q = 0.6 A"1 in the region around Tk [Source: Andersen, Stirling et al., 1991].

and Woods (1971) for small Q observed that the widths increased fairly smoothly through the superfluid transition. The high-resolution neutron data at 0.4 A" 1 obtained by SG have confirmed these early results, with a more complete examination of the region around 7^. The only significant change in the peak position and line shape appears in the region from T = 2.96 K to 3.94 K, which may be connected to the fact that the Landau-Placzek ratio is steadily increasing with temperature (see Fig. 6.1) and becomes of order unity as we approach the gas-liquid critical point at 5.2 K. As discussed in Section 6.2, this means that the entropy fluctuations are coupled into the density fluctuations in the low-Q hydrodynamic domain; some residual effect of this may remain even for wavevectors as large as Q = 0.4 A" 1 . The line shape changes which are apparent in the high-temperature region 2.56 < T < 3.94 K certainly deserve more careful theoretical study. On the other hand, the data in Fig. 1.4 also clearly show that there is a temperature region extending from 1 K right up to about 2.5 K (i.e., well above Tx = 2.17 K) in which the low-g phonon line shape in S(Q,co) does not show much qualitative change. The same sort of line shapes are obtained for wavevectors up to about Q = 0.7 A" 1 . As illustration,

7.1 S(Q,co): neutron-scattering data 1

159

3 2

s N

i i

0.16

0.14

0.12

0.10

O Oo$f

0 08

oo o

}

2 T^ 3 Temperature (K)

, 4

Fig. 7.4. The fitting parameters COQ, TQ and Z(Q) as a function of the temperature, for Q = 0.4 A" 1 at SVP. These results are based on using (7.3) and (7.4). F G shows the Gaussian resolution width. Results are also shown for the total half-width obtained by Cowley and Woods (1971) and for the intrinsic half-width obtained by Mezei and Stirling (1983) at T < 1.7 K [Source: Stirling and Glyde, 1990].

160

Phonons, maxons and rotons 1

(2 = 0.7 A-1

Q = 0.5 A" (

Q = 0.9 A"1

1 -

i i

•

•• >

•

.1 i

3 O

Q =1.5 A"1

Q =1.3 A-1

Q= 1.1 A"1 1

1 -

1 1

1 2 co (meV) Fig. 7.5. Neutron-scattering line shapes vs. frequency for different wave vectors in normal liquid 3 He (SVP) at 0.12 K [Source: Scherm et al, 1987, 1989].

in Fig. 7.3, we show the ILL time-of-flight data (Andersen, Stirling et a/., 1991) at Q = 0.6 A" 1 in the region around Tx and above. SG found that the resolution-broadened data in Fig. 1.4 was quite well fitted using the simple ansatz = [JVM + l]{Z(Q)A(Q,co) + GM(Q,co)} ,

(7.3)

where the "one-phonon" peak is described by (see Section 2.2)

1 F

A(Q,a>) = K

To 2

(co + cog) + T

1 2

(7.4)

and the Gaussian GM describes the weak high-energy multiparticle contribution. SG determined the temperature-dependent Lorentzian parameters COQ(T) and TQ(T), as well as Z(Q,T) which determines the peak height, by convoluting (7.3) with the Gaussian resolution function discussed above. The fit parameters so obtained are shown in Fig. 7.4. We conclude that S(Q,a>) in both superfluid and normal liquid 4 He exhibits a collisionless phonon mode and that its existence appears to be independent of the presence of superfluidity or a Bose condensate. Contrary to the view expressed in most of the literature, we argue that it is only at T = 0 that this mode is usefully related to the Bogoliubov phonon excitation in a dilute Bose gas which is entirely associated with

7.1 iS(Q, co); neutron-scattering

data

161

0.5 1.0 Wavevector Q (k~l)

Fig. 7.6. Zero sound dispersion relation and width in normal liquid 3 He as a function of Q, at 0.12 K and SVP. The solid line gives the sound velocity. The crosses are the widths obtained by Hilton et ah (1980) [Source: Scherm et a/., 1987].

condensate fluctuations (see Sections 3.3, 3.4, 5.2 and 6.3). Pines (1963, 1966) first suggested that this phonon mode could be better interpreted as a collective zero sound mode, in analogy with zero sound propagation in normal liquid 3 He. In Fig. 7.5, we show the neutron-scattering intensity from liquid 3 He for various momentum transfers. The higher-energy peak position corresponds to zero sound, whose position and width are plotted in Fig. 7.6. Such collisionless modes arise from time-dependent self-consistent fields and are not very dependent on the quantum statistics obeyed by the atoms or the temperature. While such zero sound modes at low Q seem to be a general feature in many liquids (see the review

162

Phonons, maxons and rotons

by Copley and Lovesey, 1975), it is only in superfluid 4 He that they correspond to the "elementary excitations". Following our discussion in Chapter 5, this basic difference will be explained later as being a consequence of Bose condensation. At higher Q, such zero sound modes become increasingly damped. The self-consistent fields become weaker as Q increases (i.e., at shorter wavelengths) as a result of stronger p-h fluctuations, until the scattering intensity from this collective mode disappears entirely. Zero sound in normal liquid 3 He starts to broaden significantly at about ~ 0.7 A" 1 and has largely disappeared by about Q ~ 1.2 A" 1 (see Figs. 7.5 and 7.6). As another example, liquid Ne supports a zero sound mode at low g, but by Q ~ 0.8 A" 1 it disappears (Buyers et a/., 1975). A recent study by Dzugutov and Dahlborg (1989) shows that liquid Bi (a liquid metal) exhibits a collective mode in <S(Q,co), but only up to Q ~ 0.6 A" 1 . Further references are given in these papers and in the book by Hansen and McDonald (1986). Intermediate momentum: maxon region 1

At Q ~ 1.1 A" , the dispersion relation CDQ reaches its maximum energy ~ 0.30 THz (14.4 K). This is referred to as the "maxon", just as the Q ~ 2 A" 1 region of minimum energy is called the roton. We now turn our attention to this maxon wavevector region, defined roughly as 0.8 < Q < 1.5 A" 1 . There have been several high-resolution studies of S(Q,co) for Q ~ 1 A" 1 over a wide range of frequencies and temperatures through Tx. The original SVP work of Woods and Svensson (1978) made a detailed study at Q = 0.8, 1.13 (the maxon wavevector QM), 1.3 and 1.9 A" 1 . Talbot, Glyde, Stirling and Svensson (TGSS, 1988) have measured S(Q,co) for g = 1.13 A" 1 at a pressure of 20 bar. Very recently, Andersen, Stirling and coworkers (1991) at ILL have carried out further high-resolution experiments at SVP. In Fig. 7.7(b), we show their recent data at Q = 1.1 A" 1 as a function of co over a wide range of temperatures. In Fig. 7.7(a), we give an expanded view of the same data near Tx. The TGSS high-pressure data are shown in Fig. 7.8 at a series of temperatures below Tx. At low T, the intensity is seen to consist of a very sharp peak at 0.30 THz (14.4 K) and a broad distribution (at this resolution) centred at about ~ 0.5 THz (24 K). The sharp peak broadens while its scattering intensity decreases dramatically as T approaches Tx from below. Indeed, the sharp peak apparently disappears entirely at T ~ Tx, as the liquid passes from the superfluid to the normal phase

7.1 S(Q,co): neutron-scattering data

163

0.25

0.20

0.00

-0.10 0.0

0.1

0.3 0.4 0.5 Energy transfer (THz)

0.6

0.7

0.8

Fig. 7.7. S(Q,co) vs. frequency at SVP for Q = 1.1 A"1 at a series of temperatures above and below TA. The upper panel shows an expanded view of the region around Tx [Source: Andersen, Stirling et al, 1991; Stirling, 1991].

(see also Woods and Svensson, 1978). There is very little change in the remaining scattering intensity as T goes from 1.90 K to 3.94 K, as shown in the bottom panel of Fig. 2.4. In the high-pressure data in Fig. 7.8, this broad temperature-independent distribution, which is centred at an energy much higher than the sharp quasiparticle peak, appears to be present in the superfluid phase without much change. In particular, the multiparticle resonances so visible at low temperatures at SVP (see Fig. 7.7) are not so evident at higher pressures (see Fig. 7.8). The comparative data shown in Fig. 7.9 at 1.27 K also seem to be consistent with the "smearing" of the fine structure with pressure. However, recent high-resolution data at 1.25 K from ILL (see Fig. 7.10) does in fact show

Phonons, maxons and rotons

164 600

400

200

-0.2

0.2

0.4 0.6 v (THz)

0.8

1.0

1.2

Fig. 7.8. The net scattering intensity vs. frequency for Q = 1.13 A" 1 , Tx = 1.928 K, at p = 20 atm in the superfluid phase. Note the wide frequency range, up to 1.2 THz [Source: Talbot, Glyde, Stirling and Svensson, 1988].

7.1 S(Q,co): neutron-scattering data

165

Phonon

I I

r—^ 0.4

0.8 Energy (THz)

15 bars

° 6 SVP

_£-*

1.2

Fig. 7.9. The pressure dependence of the high-energy component of S(Q,co) at Q = 1.13 A-1 and T = 1.27 K [Source: Stirling, 1985]. a strong multiparticle resonance at ~ 0.45 THz at p = 15 bar. This suggests that the temperature-independent high-energy spectrum shown in Fig. 7.8 would probably exhibit considerably more structure as the temperature decreases in higher-resolution data. Above Tx, it seems difficult not to interpret the broad high-energy distribution at Q = 1.1 A" 1 as a strongly damped zero sound mode. Thus the collisionless density mode which was well defined at Q = 0.4 A" 1 (Fig. 1.4) is now spread over a large energy range due to damping when we reach Q ~ 1.1 A" 1 , but it is still present. Strong evidence for this interpretation is that a simple extrapolation of the phonon dispersion relation cQ at 20 bar up to Q = 1.13 A" 1 predicts a peak centred at ~ 0.6 THz (29 K). This prediction is in reasonable agreement with the observed peak at ~ 0.5 THz (24 K). As in liquid 3 He, at such a large value of Q , the zero sound energy would be expected to have "softened" because of the decreasing strength of the self-consistent fields at short wavelengths. In contrast to the high-pressure data, the SVP normal distribution (see Fig. 7.7) at Q = 1.1 A""1 peaks at considerably lower frequencies

Phonons, maxons and rotons

166

0.12 0.10 ~ 0.08 '3

M M I M M I M

0.14

% peak -

•§ 0.06 r

0.02

0 0 0

\ 0.00 i i 1. . . . 1 . . .. 1 . .M

-0.02

-0.1

0.0

0.1

0.2

0.3 0.4 0.5 Energy transfer (THz)

0.6

0.7

0.8

Fig. 7.10. Expanded view of the scattering intensity for Q = 1.1 A"1, at T = 1.24 K and p = 15 bar. The multiparticle peak at 0.45 THz is clearly visible [Source: Andersen, Stirling et al, 1991]. ~ 0.3 THz (14 K). This is again consistent with an extrapolation of the low-Q phonon dispersion relation to Q = 1.1 A" 1 using the reduced sound velocity at SVR This sensitivity to pressure of the normal distribution peak frequency is, of course, exactly what one would expect when dealing with a (zero) sound mode. The fact that the normal distribution is peaked just above the emerging maxon quasiparticle peak (see Fig. 7.7(a)) makes it more difficult to separate out the zero sound mode scattering intensity as the temperature is lowered below Tx at SVP than it is at high pressure. What happens to this zero sound mode at Q = 1.1 A" 1 as we go below Tx is somewhat complicated to follow because of the appearance of a multiparticle component (see Chapter 10) which overlaps the same frequency region. At low I (^ 1 K), this high-energy multiparticle structure is increasingly visible while the scattering intensity from the broad normal zero sound peak appears to have largely vanished (see Figs. 1.6 and 7.7). Turning this around, we can say that as the temperature increases towards Tx from below, scattering intensity associated with the zero sound mode steadily builds up, "submerging" the multiphonon resonances. In

7.1 S(Q,co): neutron-scattering data

167

Section 7.2, we argue that the zero sound intensity increases as a result of the decreasing strength of the condensate-induced hybridization, which is the origin of the weight of both the sharp quasiparticle peak and the pair-excitation spectrum in S(Q,co) below 7^. The key point we are making is that the high-energy component to S(Q,co) at intermediate values of Q ~ 1.1 A" 1 is some temperaturedependent mixture of multiphonon resonances associated with creating two quasiparticles (maxons and rotons) plus a broad zero sound component. The latter increasingly saturates the scattering intensity as T increases towards Tx from below. We note that the zero sound frequency will be expected to show much more dispersion (i.e., Q-dependence) than the multiphonon resonances, which may allow their contributions to be separated. The high-pressure maxon data in Fig. 7.8 appear to show the quasiparticle peak "growing out" of the normal distribution in a very clear fashion. As we have mentioned, the SVP data is less dramatic in this regard since the quasiparticle peak appears at a frequency just slightly below the normal distribution peak. However, careful study of the SVP data in Fig. 7.7(b) still shows a characteristic asymmetry of the rapidly growing quasiparticle peak which allows it to be distinguished from the normal distribution. This kind of asymmetry was first noted by Woods and Svensson (1978) in their pioneering study. It is also useful to compare the emerging line shape at 1.1 A" 1 in Fig. 7.7(a) with the line shape in the phonon region in Fig. 7.3. The qualitative difference between these two regions seems clear in such high-resolution data. In Section 7.2, we interpret the sharp maxon peak observed below Tx as a result of the neutron exciting a single atom out of the condensate, i.e., it is a pole of the single-particle Green's function G(Q, co). We recall from (3.47) and (5.24) that S(Q,co) can contain a term of the form A(Q,co)G(Q,co)A(Q,co), where the Bose broken-symmetry vertex function A(Q,co) depends on the condensate density no (see Section 7.2 for more details). This immediately leads to the single-particle excitation peak at COQ having a finite weight in S(Q,co) below Tx, with a weight A2 which increases with the value of no. We believe that these excitations exist above Tx, but that they would not show up as a distinct peak in S(Q,co) simply because the Bose vertex function A (no) has vanished with no. The sharp maxon peak at ~ 0.3 THz corresponds to the same excitation branch as the roton peak at Q ~ 2 A" 1 discussed below, but we shall argue that it is quite different from the zero sound branch which dominates the spectrum at low Q.

168

Phonons, maxons and rotons Roton region

From the earliest days of inelastic neutron scattering, there has been special interest in studying the roton dispersion relation around the minimum shown in Fig. 1.3. At SVP, this minimum occurs at QR = 1.925 A" 1 and the formula in (1.2) is an excellent fit in the region ±0.25 A" 1 around QR. The fact that the quasiparticle intensity Z(Q) has a strong peak at Q (see Figs. 2.2 and 2.8) means that the roton scattering cross-section is especially large, allowing higher instrumental resolution to be used. The first high-resolution study of S(Q,co) for Q ~ QR over a wide frequency range and at temperatures below and above TA was by Woods and Svensson (WS, 1978). New high-precision SVP data showing how the S(Q,co) line shape at QR changes near TA have been reported by Stirling and Glyde (1990). The most complete study is the time-of-flight data of Anderson, Stirling et al. (1991). In addition, there have been extensive studies of the effect of pressure on the roton spectrum (see Fig. 2.2). We mention, in particular, the older work of Dietrich, Graf, Huang and Passell (1972) and the more recent high-resolution study by Talbot, Glyde, Stirling and Svensson (TGSS, 1988) at p = 20 bar over the temperature range 1.29-2.97 K (at this pressure, Tk = 1.93 K, QR = 2.03 A" 1 and the roton energy A(T = 1.3 K) = 0.158 THz (7.56 K)). The SVP scattering intensity shown in Fig. 7.11 at 2.0 A" 1 clearly shows that there are qualitative changes in the line shape as one passes through TA. In Fig. 2.6, we show a contour plot of the scattering intensity in the roton wave vector region Q = 2.0 A" 1 . At low T, the observed intensity in Fig. 7.12 consists of a sharp peak at 0.16 THz (7.6 K). As T increases, the intensity in the sharp peak drops dramatically until it disappears at TA, just as in the maxon case Q = 1.13 A" 1 . As with the maxon, we interpret the sharp peak as scattering from a single-particle roton excitation COQ whose intensity in S(Q,co) is given by A 2 (g^,n 0 ). This single-particle excitation should also exist above TA but is no longer seen as a distinct peak in 5(Q,co) since HQ(T) = 0 in the normal phase. Above TA, we are left with a broad distribution, which is not very temperature-dependent (see Fig. 7.11). At very low temperatures (T ~ 1.3 K), we note that the high-pressure roton data of TGSS also exhibits a broad (at this level of resolution) multiparticle component peaked at - 0.5 THz (see Fig. 7.13). In the data shown in Figs. 7.11-7.13, as well as in Fig. 2.6, there is no evidence that normal liquid 4 He supports even a remnant of a highenergy collective zero sound mode at Q ~ 2 A" 1 . Further evidence for

7.1 <S(Q,co); neutron-scattering data

0.1 0.2 Energy transfer (THz)

0.3

169

0.4

0.5

Fig. 7.11. S(Q,co) vs. frequency at Q = 2.0 A"1 and SVP for temperatures well above and just below TA [Source: Andersen, Stirling et al, 1991]. the lack of zero sound at these wavevectors comes from a comparison with the S(Q,co) spectrum in normal 3 He, the only other liquid which exists in the temperature range 1-3 K. Normal liquid 3 He exhibits a well defined zero sound mode only for Q < 1 A" 1 (see Fig. 7.6). At substantially larger Q, RPA calculations of/ nn (Q,co) show that it can be well approximated by the incoherent particle-hole spectrum given by the Lindhard function y?nn{Q,co) of a Fermi gas of quasiparticles. The S(Q,co) spectra of normal liquid 4 He and 3 He are compared in Fig. 7.14. The broad distribution in normal 4 He for Q ~ 2 A" 1 is seen to be very similar to that in normal 3 He, as to both width and peak position. We also remark that these spectra are not that different from classical liquids at values of Q ^ 1 A" 1 , where there is no longer evidence for the collective behaviour which is apparent at lower wavevectors. As a crude estimate, one expects that the zero sound frequency will be given by coQ ~ (nV(Q)/m)l/2Q. Since the effective potential V(Q) starts to decrease when g ^ 1 A" 1 (see, for example, Section 9.2), so will the zero sound frequency. This phenomenon has been studied in detail in connection with zero sound in liquid 3 He (Pines, 1985). Such

170

Phonons, maxons and rotons

Temperature O = 1.24K A = 1.50K =1.96K X =2.12K O = 2.26 K

-0.2 -0.2

I . . . . . . . . I . . . . .... . . . . I . . . . I . . . . 1 . . . . I . . . . I . . . . I . . . . I .... I 0.1 0.0 0.1 0.2 0.3 Energy transfer (THz)

Fig. 7.12. S(Q,co) vs. frequency in the roton wavevector region at Q = 2.0 A" 1 and SVP for temperatures well below and just above Tk [Source: Andersen, Stirling et al., 1991].

a decrease in the peak position of the normal distribution of liquid 4 He was observed (Fig. 7.15) at 4.2 K by Woods, Svensson and Martel (1975) but at such high temperatures, the line shape has significant distortions (recall that 4.2 K is the boiling temperature of liquid 4 He at a pressure of 1 bar). In Fig. 7.16, we estimate the peak position and width of the normal distribution just above Tx using the recent ILL (Grenoble) results of Andersen, Stirling et al (1991). We note that the results in Figs. 7.15 and 7.16 have not had the effect of the detailed balance factor removed. We recall that at higher temperatures, the factor [N(co) + 1] in (2.20b) can lead to considerable difference between the observed S(Q,co) and the more fundamental Imxm(Q9co). If we could ignore instrumental broadening, the latter could be obtained by multiplying the S(Q,cw) data by (1 - e~Pco). Plots such as those in Figs. 7.15 and 7.16 are misleading, however. We believe there is a transition between the low Q < 1 A" 1 region (where zero sound arising from the near zero of the denominator of (5.80) dominates the scattering intensity) and the high-g region, where the

7.1 S(Q, (o): neutron-scattering data

-0.2

0.2

0.4

0.6

0.8

1.0

171

1.2

Fig. 7.13. S(Q,o) vs. frequency for Q = 2.03 A" 1 and p = 20 atm in the superfluid phase. The data at 1.29 K are shown on an expanded scale [Source: Talbot, Glyde, Stirling and Svensson, 1988].

Phonons, maxons and rotons

172 1

3.0

ift

•

1.0

0.0

0.2

0.4 v (THz)

0.6

0.8

Fig. 7.14. Comparison between S(Q,v) as a function of the frequency v at Q = 2.0 A" 1 for liquid 3 He at T = 15 mK (dots, Skold et a/., 1976) and liquid 4 He at 4.2 K [Source: Woods, private communication].

1.2 1.6 Q (A-i)

Fig. 7.15. The mean energy and the full width (at half maximum) of the peak in S(Q,co) sit 4.2 K and SVR It should be noted that, at this high temperature, the line shapes are quite asymmetrical [Source: Woods, Svensson and Martel, 1975].

7.1 S(Q,co): neutron-scattering data

173

0.7 0.6 0.5

0.3 h 0.2

0.1 0

X 1

0.5

1.0

1.5

Fig. 7.16. Frequency of the peak position of the normal scattering intensity (T = 2.26 K, SVP) as a function of wave vector Q, just above the superfluid transition temperature. The total width at half maximum (including instrumental resolution) is also indicated. See also Fig. 7.15. These results are based on an analysis of unpublished ILL data [Source: Andersen, Stirling et al, 1991].

contribution from the incoherent p-h spectrum (from the numerator of (5.80)) dominates. The fact that this transition is continuous hides the fact that the origin of the scattering intensity is completely different. Strong evidence for this transition is the result that, for Q ^ 1.3 A" 1 , the width of the broad normal distribution narrows (see Figs. 7.15 and 7.16). This decrease in the width would be very puzzling if S(Q,co) was describing the same collective mode for both small and large Q. (We also note that in superfluid 4 He below TA, there is no such anomalous behaviour of the width of the maxon-roton excitation in the region around Q ~ 1.5 A" 1 . This is consistent with our interpretation of the maxon-roton mode as a single-particle-like excitation which is quite distinct from zero sound branch.) In Fig. 7.17, we attempt to illustrate the scattering intensity from the two different kinds of density fluctuations in the normal phase. The same analysis is relevant to plots of the dispersion curve of the "peak" in S(Q,co) in normal liquid 3 He (see Fig. 7.18). The latter system has the advantage that one can clearly see how the scattering intensity from the p-h continuum crosses the remnant of the collective zero sound

174

Phonons, maxons and rotons

Fig. 7.17. A schematic illustration of the (zero sound) collective and the incoherent p-h contributions to S(Q,co) in the normal phase of liquid 4 He. The dots represent the width of the distributions and their density gives a rough measure of the scattering intensity. The anomalous Q-dependence of the widths shown in Figs. 7.15 and 7.16 is interpreted in terms of this transition between the coherent zero sound region at lower Q and the incoherent p-h region at high Q.

mode, simulating a roton-like dip. Our conclusion is that a pseudo"roton-like" depression (as in Fig. 7.15) in the dispersion relation of the peak position of S(Q,co) may well appear in a non-Bose-condensed liquid, but its physical origin is completely different from the roton observed in superfluid 4 He. The experimental data in the region just past the roton minimum, 2 < Q < 2.4 A" 1 , are especially interesting. In Fig. 7.19, we show recent SVP data from ILL (Fak and Andersen, 1991). As we have mentioned above, at Q ~ QR, the normal distribution is peaked at a frequency just slightly lower than the roton peak which appears when we go below Tx (see Fig. 7.11). However, we see that as Q approaches 2.4 A" 1 , the quasiparticle peak frequency rapidly increases towards the high-energy peak centred around 2 meV. We defer further analysis of this interesting intermediate-wavevector region to Section 7.2. The neutron-scattering line shape changes very dramatically once we go past Q ~ 2.4 A" 1 . As we discuss in Chapter 10, this change is caused by the strong mixing of the one- and two-roton spectra in this region, where they overlap in energy. In the region 2.5 < Q < 3 A" 1 , there is a low-intensity peak with an energy close to twice the roton energy (see Fig. 7.20). By Q ~ 3 A" 1 , however, most of the scattering intensity

7.1 S(Q,co): neutron-scattering data

175

Fig. 7.18. A plot of the peak positions in S(Q, co) vs. wavevector in liquid 3He at 0.12 K and SVP (see also Figs. 7.5 and 7.6). The zero sound and incoherent particle-hole spectrum are seen to merge, giving rise to an apparent "rotonminimum". For comparison, the peak position in superfluid 4He is shown. The wavevector Q* is given in dimensionless units such that the static structure factor S(Q) has its first maximum at Q* = 1 [Source: Scherm et a/., 1989]. has gone into a broad temperature-independent distribution centred at much higher energy. This is approximately peaked at the recoil energy = COP+Q—CO P is described by a modified "Lindhard function" for rotons, and thus we expect that the intensity will be concentrated at low energies co < COQ (see Fig. 7.17). If this is due to thermally excited rotons (which are the dominant excitations at temperatures above 1.2 K), one would expect the scattering intensity to decrease as the temperature is lowered. This feature is consistent with data such as in Figs. 7.11 and 7.12. In addition to the low-energy particle-hole contribution just described (which grows smoothly into the normal distribution above 7^), we expect the usual multiparticle contribution to Im Xnn m (7-8) which is associated with creating a pair of excitations (see third line of (3.45)). If these excitations involve rotons and maxons, this multiparticle contribution will be peaked at very high energies, as in Fig. 1.6. It can be expected to lose weight as the temperature increases, since it depends on the Boseinduced hybridization. This high-energy component is clearly visible in the data, concentrated at around ~ 2 meV up to about Q = 2.3 A" 1 (see Fig. 7.19). The data in the region 2.2 < Q < 2.8 A" 1 shown in Figs. 7.19 and 7.20 require some care in interpretation. What is happening is that the quasiparticle peak energy is steadily increasing; and when it becomes more or less degenerate with the multiparticle component (we estimate this to occur at Q ~ 2.4 A" 1 where the slope of the quasiparticle dispersion relation suddenly changes), we have the hybridization which occurs whenever two excitation branches cross. This is shown schematically in Fig. 7.21. The main point is that while at Q = 2.3 A" 1 the highenergy component is associated with the pair-excitation spectrum, by the time we reach Q = 2.8 A" 1 it should be interpreted as due to the p-h spectrum discussed above. At even higher Q, this goes over to the impulse-approximation line shape described in Chapter 4. It would be very useful to have data such as in Figs. 7.19 and 7.20 at a series of temperatures, including above T^. We expect the p-h spectrum to increase steadily with temperature below Tx with a dependence related to the

180

Phonons, maxons and rotons

Fig. 7.21. The dashed line represents schematically the bare single-particle spectrum such as assumed in the analysis of Ruvalds and Zawadowski (1970). As discussed in Section 10.2, the condensate-induced hybridization of this sharp single-particle excitation with the broad pair-excitation branch (here represented by the two-roton continuum) leads to renormalized branches which transform into each other. The lower branch (solid line) is the (observed) maxon-roton dispersion relation, which goes over into the broad pair-excitation branch for Q J> 2.5 A" 1 . Similarly the upper pair-excitation branch transforms into the particle-hole branch, peaked at the recoil or single-particle energy Q2 /2m at high Q. The shaded areas (centred at the black dots) are a schematic representation of this renormalized upper band [Source: adapted from Bedell, Pines and Zawadowski, 1984].

number of rotons present. In contrast, the weight of the pair-excitation spectra in S(Q,co) should decrease with the condensate density no(T). An adequate theory of the momentum region 2.3 < Q < 2.6 A" 1 will involve extending the kind of analysis given by Glyde and Griffin (1990) to include the effects of the pair-excitation branch shown in Fig. 7.21, i.e., the Pitaevskii-Ruvalds-Zawadowski analysis discussed in Section 10.2. It follows that any discussion of this wavevector region is inadequate if it ignores the existence of the two-roton branch. In Fig. 9 of by Glyde (1992a) (see also Fig. 3 of Fak and Andersen, 1991), the high-energy two-roton peak at Q = 2.3 A" 1 (shown in Fig. 7.19) is identified with the low-temperature remnant of the normal scattering intensity above Tx. In Fig. 10 of Glyde (1992a), the low-energy two-roton peak at

7.2 S(Q,a>): theoretical interpretation

181

Q = 2.5 A" (shown in Fig. 7.20) is interpreted as the single-particle branch in the Glyde-Griffin picture. It is clear that in our SP scenario, the expression in (7.8) is the analogue of the impulse approximation (4.2) which becomes valid at much higher momentum. We argue that the strong roton resonance which appears in S(Q,co) as we go below Tx is precisely the elusive "single-particle peak" which has been the object of so many experimental studies in the high-Q region (see Chapter 4). While the simple impulse approximation is not valid at values of Q as low as 2 A" 1 , (7.8) can be viewed as a generalized version of it. One reason that the roton resonance is so visible is that, as T goes below Tx, it becomes extremely sharp. Moreover, we note that the instrumental line width typically increases linearly with Q and, as a result, the single-particle resonance no longer appears as a distinct peak sitting on a broad (Doppler-broadened) background. See Section 4.3 for further discussion. A separate reason why the roton is so visible is that its weight Z(QR) in S(Q,co) is of order unity at low temperatures (Fig. 2.2). In contrast, at even larger values of Q ;> 3 A" 1 , the single-particle peak in S(Q,a>) is predicted to have a weight given by no/n (see (4.2)). This reduction in the scattering intensity by a factor of 10 adds to the experimental difficulties of detecting the single-particle excitation in <S(Q,co) at larger wavevectors. The essentials of this SP scenario for the roton data were first suggested in the classic paper by Miller, Pines and Nozieres (MPN, 1962), although their analysis was at T = 0. Indeed, MPN already suggested that the weight of the roton resonance which appears in <S(Q, co) in the superfluid phase might be a direct measure of the condensate density no. (This idea was later extended by Hohenberg and Platzman, 1966, to much larger values of g, as we discuss in Chapter 4.) MPN argued that the ratio f(Q) = Z(Q)/S(Q)

(7.10)

should be a measure of the probability of exciting an atom "out of the condensate." At Q ~ 2.5 A" 1 , this expression should reduce to no/n, since at such large values of Q and co, various Bose coherence factors should have disappeared (i.e., UQ = 1,VQ = 0). MPN suggested that the experimental value f(Q) ~ 0.1 at Q ~ 2.5 A" 1 (see Fig. 2.8) was consistent with the value of no/n at low temperatures predicted by Penrose and Onsager (1956). Strong hybridization effects with the two-roton spectra which start at about 2.4 A" 1 mean that the ratio in

182

Phonons, maxons and rotons

(7.10) loses meaning at wavevectors larger than this (see Fig. 7.21 and Section 10.2). A subject well worth further study is the rapid change in Z(Q) shown in Fig. 2.1 as one goes from Q = 1 A" 1 to 2 A" 1 . In the scenario we have been discussing, based on the hybridization of single-particle excitations and density fluctuations, it is clear that the quasiparticle weight shown in Fig. 2.1 is describing a fairly complicated situation. Earlier, we noted that in the roton region, the quasiparticle weight Z(Q) is related to the Bose broken-symmetry vertex function by (7.9). This relation is based on the validity of (7.8). In the maxon intermediate-wavevector region, the condensate-induced hybridization would seem to preclude the validity of any simple relation like (7.9). Fig. 12 of Svensson (1991) shows a schematic decomposition of Z(Q) into zero sound and single-particle components, based on the GlydeGriffin picture. Relative to the free-atom energy Q2/2m, the maxon is a high-energy excitation and the roton is a low-energy excitation. Griffin and Svensson (1990) have used these facts to argue that Z(QM) ~ n$/n while Z(QR) ^ 1 (which is consistent with the data in Fig. 2.1). However their arguments are qualitative and more detailed studies are needed. The main point to remember is that the Bose vertex functions play a crucial role in determining the strength of the single-particle excitations in iS(Q,co) and this strength depends very much on both no(T) and the wavevector region of interest. Phonon-maxon region In Section 7.1, we noted that the scattering intensity for intermediate values of Q ~ 1 A" 1 contained a broad distribution both above and below Tx. We argued that this was partly associated with a strongly damped zero sound mode. This broad component was in addition to a sharp resonance which only appeared below Tx. These data can be naturally understood on the basis of (7.6), the sharp resonance being associated with an intrinsic pole of Gajg(Q,co) overlapping with a zero sound mode arising from a zero of the real part of eR(Q, co) in (7.7). This is the remnant of the sharp zero sound phonon mode which dominates S(Q, a*) at lower values of Q. In our present scenario, the maxon or intermediate-Q region is especially interesting since it is a transition region between the ZS collective region at low Q (where both Gaj? and Xnn are dominated by a sharp zero sound co\ pole) and the single-particle region at high Q (where both

7.2 S(Q,(o): theoretical interpretation

183

Gap and Xnn are dominated by a renormalized single-particle roton co2 excitation). We recall from the general discussion in Chapter 5 (for a summary, see Section 5.5) that if both Ga^ and %nn have well defined poles above Tx (0)2 and cd\, respectively), these excitations will be hybridized below Tx and both will appear in G^ and Xnn with finite weight. Section 5.2 illustrates this with a dilute Bose gas calculation. The frequencies of these two renormalized modes (co\ and coj) are given by the zeros of the function C(Q,a>) in (5.22). As is evident from the discussion in Section 7.1, a zero sound mode exists in the normal phase scattering intensity at low and intermediate Q values (estimated to be < 1.3 A" 1 ). In this case, the two terms in (7.6) or (5.24) are strongly coupled through the Bose vertex functions since Aa = Ka/eR. It is then more useful to work with the alternative (but equivalent) form (see (5.2) or (5.76))

where

+ x^(Q9a)) .

(7.12)

As discussed in Section 5.1, one can prove that if no ^ 0, the poles of Ga^(Q,co) are also given by the zeros of e(Q,co), with a weight governed by the Bose vertex functions Aa. Here Gap(Q,co) is the singleparticle Green's function without the reducible self-energy contributions s a£ = V{Q)Khp/eR, as given in (5.75). Above Tx, we have Gajg = Gap. In physical terms, our scenario can be summarized as follows. The phonon and roton regions of the quasiparticle spectrum involve quite distinct excitation branches. The low-energy phonon is a zero sound mode involving the effective fields associated with the condensate as well as the thermally excited quasiparticles. The high-energy roton, in contrast, is viewed as a renormalized atomic-like excitation associated with the normal liquid. This picture inevitably implies that the intermediate maxon region will involve some sort of hybridization where the phonon and roton branches cross, producing the observed continuous quasiparticle spectrum. In order to illustrate this scenario (first suggested by Glyde and Griffin, 1990), we now consider a schematic parameterization of the regular functions in (7.11) and (7.12). Our interest is to introduce a simple model which incorporates the essential physics of the above scenario. Suitably generalized, such parameterizations may be useful as a basis for fits to the S(Q,co) line-shape data.

184

Phonons, maxons and rotons

In order to calculate S(Q, co) using (7.11) and (7.12), we recall from (5.79) that Im Xnn is directly proportional to Im (1/e), where e(Q,G>) = 1 - V(Q)?m(Q,co) - K(Q)x* (Q,o) .

(7.13)

Thus the resonances in S(Q,co) will be directly related to the zeros of Ree(Q,co). In order to proceed, we must introduce approximations for Ga£, Aa and Xnn m (7.12) which are consistent with the underlying microscopic constraints discussed in Chapter 5 and Section 6.3. We start with a two-pole ansatz for G^p in (7.12), given schematically by G^

CO^ CO — CO^ 1

\ Z\ \ \

CO1 1 — CO

l (D(D s2

•

(7.14)

Here coS2 is a high-energy single-particle excitation which is assumed to be present in both the normal and superfluid phases of liquid 4 He. At large Q, this excitation is expected to be quite close to the observed maxonroton quasiparticle frequency COQ. At lower Q < 1 A" 1 , nothing is really known about the dispersion relation of this high-energy excitation. The first term in (7.14) describes the low-energy, long-wavelength excitation cbs\ in Gap which is explicitly associated with the Bose broken symmetry, as discussed in Section 6.3. This first term arises due to the appearance of a condensate mean field, which increasingly loses its effectiveness at larger wavevectors as well as higher temperatures. Little is known about the behaviour of the first term in (7.14) outside the \ow-Q, low-T region. Hypothetical dispersion relations of the single-particle branches cosi and cbS2 are sketched in Fig. 7.22. Using the simple ansatz for Gap given in (7.14), one is led to parameterizing the condensate contribution %n in (7.12) and (7.13) by V(Q)?m(Q,co) =

Acl

_2 + 2Ac2_2 . co — co;x co2 - cozs2 2 2

(7.15)

The weights Aci and A& are both proportional to the square of the Bose vertex function Aa and hence vanish as T —• 7^. For simplicity, we leave out any explicit reference to damping in (7.14) and (7.15). The contribution of the first term in (7.15) was not included by Glyde and Griffin (1990). It will be seen to be crucial in understanding why the phonon mode at low Q has a temperature-independent velocity in superfluid 4 He. Finally, we turn to the second or "normal" contribution in (7.12) and

7.2 S(Q,a>): theoretical interpretation u.o

i i/

0.5

0.4

185

l _

20 bar

' - - ^ /

0.3

\N(226K)

) 0.2

0.1

/ f..

0.5

1.0

Q (A-1)

1.5

2.0

2.5

Fig. 7.22. A sketch of one model for the excitation branches associated with G(Q,co) and Xnn(Q,u>) in the normal phase. The frequencies of the peaks in S(Q,a>) at 1.24 K and 2.26 K are taken from ILL data at SVP (Andersen, Stirling et a/., 1991). The normal distribution at 20 bar is schematic since the only data point is at 1.1 A" 1 (Talbot et al, 1988). As indicated in Figs. 7.17-7.19, S(Q,w) in the normal phase is quite broad. The high-Q maxon-roton singleparticle excitation cbs2 associated with the normal phase is extrapolated to low Q "by hand." The lower energy cosi acoustic branch associated with the condensate is also indicated (see Section 6.3). The high-energy multiparticle branches (see Fig. 7.21) are not shown here.

introduce the traditional approximation used in discussing zero sound, AT,

co +

(7.16)

In the normal phase (where xcnn = 0), the zero of (7.13) is given by CD2

= A

(7.17)

where the /-sum rule (see (2.24) and Section 8.1) gives (7.18)

186

Phonons, maxons and rotons

Thus in the normal phase, (7.16) leads to a zero sound phonon o>o — of half-width r 0 , (719)

The physics behind this zero sound mode is the usual one, namely it is a collisionless mode associated with the mean field of the normal p-h excitations. An expression like (7.19) is assumed to describe the resonance in S(Q,a>) reasonably well up to about Q ~ 1 A" 1 in the normal phase of liquid 4 He. Combining (7.15) and (7.16), we can try to understand the excitations in superfluid 4 He in the phonon and maxon region on the basis of e(Q,a>) = 1 - ^

CO1

- - ^ ^

CD1 — (Dlsl

- ^

CD1

(7.20)

For simplicity, we ignore the damping r 0 in (7.16) and we also assume we can neglect cbs\ in the denominator of the first term in (7.15). This latter assumption is valid if the relevant solutions of e(Q, co) = 0 occur at frequencies much larger than cosi (see Fig. 7.22). With the model spectrum shown in Fig. 7.22, e(Q,co) = 0 will have two solutions. For small wavevectors, hybridization effects will be small and the excitation branches are given by

While we expect Ac\ and AR to be individually quite temperaturedependent in the Bose-condensed phase, their sum should be relatively temperature-independent. This can be seen using the /-sum rule, which requires that A* +AR = c2Q2 - Ac2 ,

(7.22)

where c is defined in (7.18). The observed phonon velocity is essentially constant in the superfluid and normal phases of superfluid 4 He. Within the dielectric formalism, this fact is naturally understood in terms of the phonon mode being zero sound arising from contributions from both the condensate (%n) and regular (x%n) parts of Xnn- As the temperature increases, the decreasing weight of the condensate associated with Ac\ is compensated by the increasing strength of the normal fluid mean field AR. We have already alluded to this picture in Section 5.2 (see (5.50)). The results of Gavoret and Nozieres (1964) reviewed in Section 6.3

7.2 S(Q, co): theoretical interpretation

187

imply that, at T = 0 and in the limit of small Q, both AR and AC2 vanish. In this case, (7.22) reduces to Ac\ = c2Q2 and the coS2 mode in (7.20) would have zero weight in S(Q,co). Thus it would seem that the high-energy peak shown in Figs. 7.1 and 7.2 is not evidence for the coS2 mode in (7.20) but is rather related to a pair excitation (as discussed in Chapter 10). The precise nature of this high-energy peak at low Q deserves further study. Within the context of the preceding discussion, one might attempt a unified analysis of the phonon and maxon line-shape data based on

) line-shape data. Glyde (1992a) has done this starting from an expression for the dielectric function of the form (A is not the roton energy, as in the rest of this book) (7.2S) All six parameters in (7.28) depend on Q. Setting Tsp — 0, this gives a scattering intensity S(Q,a>) oc -Imxm(Q,

co)

-ImK(Q) e(Q,co)

190

Phonons, maxons and rotons a

2coT0(o}2 — a2) (co2 - cv2N)(co2 - &2p) -Aa]2+

[Imtodo2

- cb2p)]2 ' (7.29)

where cb2N = Q 2 + a . Above Tx (where A = 0), (7.29) simplifies to

If the mean energy Q of the particle-hole excitations is set to zero, (7.29) reduces to (7.24). The fitting procedure used by Glyde is based on the key simplifying assumption that only A changes with temperature (through its dependence on the condensate no(T)) in the superfluid phase. The other four parameters a>sp, Q, a and r 0 are assumed to be temperature-independent. In the normal phase (T > Tx), line-shape fits to (7.31) determine (bN and To. The fact that the numerator of S(Q,co) in (7.29) has a minimum at co = cbsp can be used to estimate a>sp from the low-temperature data. Finally, at low T, line-shape fits are used to determine the best values for A and a. At intermediate temperatures, A(T) is then allowed to vary in magnitude in order to simulate the condensate-induced hybridization. The results shown in Fig. 5 of Glyde (1992a) illustrate in a concrete way the scenario suggested by Glyde and Griffin (1990) in the maxon region Q ~ 1.1 A" 1 to explain data such as in Fig. 7.8. Unfortunately, the simplifying assumption that only A(T) varies with temperature below Tx appears to lead to unphysical values for some of the parameters in (7.28). In particular, for Q = 1.1 A" 1 at both SVP and 20 bar, the parameter fits which Glyde obtained imply that: (a) a is negative. It is easily verified from (7.29) and (7.31) that this would require that V(Q) < 0, otherwise S(Q,co) would be negative. However, a negative V(Q) in the maxon region is inconsistent with its positive value in the zero sound phonon region at slightly lower values of Q. (b) The resonance frequency Q is extremely large - in particular, Q ~ I.ICOJV and Q > cbsp. In contrast with the cosp pole of %n in (7.28), the physical origin of such a high-energy resonance in Xnn *s unclear.

7.2 S(Q,co): theoretical interpretation

191

Moreover, in this parameterization, the peak in S(Q,co) above Tx (see (7.31)) would have its origin in this assumed resonance in Xnn> rather than as a zero sound mode as in the Glyde-Griffin picture. This is most dramatically shown by the fact that at T > Tx, the density fluctuation associated with Xnn by itself is given by

So(Q, co) oc -Im ** (Q, co) = -£When Q ~

I.ICOJV,

J?^°

^ , 2 • (7.32)

V(Q) (co2 - Q2)2 + (2cor0)2 this is almost identical to the expression in (7.31).

The above discussion illustrates some of the difficulties in using (7.28) and (7.29) as the basis for a fit to experimental line shapes. It shows that one must allow other parameters besides A(T) to vary with temperature. In any fit to experimental data, one should also ensure that known constraints are satisfied: (a) S(Q, co) should satisfy the /-sum rule at all T, as the expression in (7.23) does. (b) The static structure factor S(Q), defined in (2.11), should be independent of temperature. (c) At low T (~ 1 K), the intensity of the normal zero sound distribution appears to vanish, all of its spectral weight shifting into an extremely sharp single-particle mode at COQ (with S(Q,co < COQ) = 0) as well as pair excitations at higher energies. One simple way of proceeding would be to start with some reasonable assumption about the temperature dependence of the various parameters in (7.28), such as A(T) = a(l-N)

a(T) = co 0 2 -A(T) ,

Here the temperature-dependent parameter N goes from N = 1 at T > Tx to N = 0 at T = 0. In an approximate way, one may think of N as a measure of the "normal fluid" excitation density, with N = 0.1 corresponding to T ~ 1 K. Further studies are clearly needed to find simple parameterized forms for S(Q,co) such as (7.24) and (7.29) which fit the data and incorporate the general structure implied by the dielectric formalism. We note that it may be more advantageous to develop approximate forms for the

192

Phonons, maxons and rotons

regular longitudinal current response function XJJ (Q5°>) a n d use (5.81) to find e(Q,co). The precise nature of the hybridization in the maxon wavevector region is complicated by the fact that we are dealing with strongly damped excitation branches at all but the lowest temperatures. In this connection, we note that the last term in (7.23) or (7.28) doesn't really describe Landau damping of normal zero sound (decay into two single-particle excitations). The width r 0 in (7.28) describes the halfwidth of particle-hole excitations and Landau damping will still occur even if r 0 = 0. The scattering intensity in the roton region does not exhibit any evidence (see Fig. 7.12) for the kind of asymmetrical line shape associated with hybridization which is evident in the maxon region. This is consistent with our previous argument that in the roton region, (7.8) is a better starting point than (7.11) or (7.12). By the same token, expressions like (7.23) and (7.28) do not appear to be the appropriate starting points for understanding the line shapes in the roton wavevector region. In our analysis, the simple expression (7.23) is introduced not as an ad hoc formula for data analysis but within the general conceptual picture summarized in Sections 5.5 and 6.3. A key role is played by the poles of Gap in (7.14). In our analysis, we were naturally led to assume that there may be two (bsp branches: a long-wavelength, low-energy mode cosi associated with the condensate (see Section 6.3), and a short-wavelength, high-energy mode coS2 already present in normal liquid 4 He and identified with the maxon-roton. Both branches are sketched in Fig. 7.22. Line-shape results predicted by (7.23) clearly ignore any low-energy structure associated with the low-energy symmetry-restoring a>si mode. As we discuss in Section 6.3, the precise dispersion relation of this mode and its continuation outside the long-wavelength two-fluid region is not that well understood at present, especially at finite temperatures (T > 1 K). Further theoretical studies are needed. In this connection, we note that Svensson and Tennant (1987) have looked for scattering intensity on the low-frequency side of the maxon resonance at low temperatures. Although they could have detected scattering as low as 10~3 of the maxon intensity at Q = 1.1 A" 1 , they found nothing. Svensson and Tennant have also noted that 5(Q,co) can exhibit spurious peaks in this low-frequency region at the free 4 He atom energy. These originate from higher-order Bragg-scattering contamination when the energy of the scattered neutron is measured.

7.2 S(Q,co); theoretical interpretation

193

Summary To summarize this chapter, we have shown how the neutron-scattering line shape exhibited by superfluid 4 He over a wide range of Q,co and T can be naturally understood in terms of a zero sound phonon at low Q and a single-particle maxon-roton at high Q, the latter appearing in S(Q,co) through the hybridizing effect of the Bose condensate. At some intermediate Q value, say Qc ~ 0.7 — 0.8 A" 1 , we predict that the sharp zero sound mode at low Q will go smoothly over into the maxonroton branch, with the usual level repulsion and cross-over behaviour characteristic of hybridization when two modes cross (see Fig. 7.23). The dielectric formalism thus provides a natural microscopic basis for the continuous phonon-maxon-roton dispersion curve first postulated by Landau (1947), a curve which has always been puzzling given that the physical nature of phonons and rotons has been long recognized as being quite different (see also Section 12.1). The work of Glyde and Griffin (1990) provides a new scenario built on microscopic theory which anchors the phonon-maxon-roton quasiparticle to the existence of a Bose condensate. In this picture, however, the maxon-roton excitation is viewed as essentially a normal phase elementary excitation. In addition, the zero sound oscillations in the non-condensate atoms play a crucial role in renormalizing the condensate fluctuations, leading to a temperature-independent phonon velocity. In an alternative scenario (see Sections 6.3 and 12.1), the entire phononmaxon-roton quasiparticle spectrum in superfluid 4 He is viewed as a single branch associated with the oscillations of the condensate. While theoretically possible, this more traditional scenario does not seem to be compatible with the observed line-shape changes in superfluid 4 He as the temperature increases towards Tx. We have also introduced a simple model (7.23) which illustrates some of the features expected from the microscopic field-theoretic analysis. This model expression was constructed keeping the experimental data on S(Q,a>) in mind, especially as concerns the temperature-dependent changes. Clearly there are many ramifications of this new picture, which will require much further experimental and theoretical work to both confirm and/or elucidate. In particular, the simple parameterization in (7.23) of the two terms in (7.13) is meant only as an illustration (see Fig. 7.23) of the kind of temperature-dependent structure we might expect in <S(Q, co) due to the condensate coupling of SP and ZS modes. We emphasize, however, that this kind of hybridization is an inevitable

194

Phonons, maxons and rotons

consequence of the Bose broken symmetry and the dielectric formalism, as summarized in Section 5.5. Moreover, a proper parameterization at finite temperatures most naturally starts with the two terms in \ nn given by (7.13). Attempting to parameterize directly the two contributions in (7.6) is more difficult, since the physics of the hybridization is much less clearly exhibited. The Woods-Svensson (1978) parameterization (summarized in Section 8.2) suffers from this difficulty as it is based on (7.6) rather than (7.11) and (7.12). Thus it is incapable of reproducing the hybridization in the maxon region predicted by (7.24) and illustrated in Fig. 7.23. In further studies, a fruitful avenue of research would be to develop simple parameterizations such as given in (7.24) and (7.29) in order to see how many of the features exhibited by <S(Q,co) can be fitted. One needed generalization at higher energies would be to include a contribution to Xnn m (7.12) from the two-excitation continuum discussed in Chapter 10. In particular, this latter extension is crucial before one can discuss the line shapes in the interesting region around 2.4 A" 1 (see Figs. 7.19-7.21).

8 Sum-rule analysis of the different contributions to S(Q,<x>)

The dielectric formalism gives the general structure of various correlation functions in a Bose-condensed fluid. However, quantitative calculations of the "regular" functions discussed in Chapter 5 are difficult even for a dilute Bose gas, let alone a Bose liquid like superfluid 4 He. As an aid to analysing neutron data and also in developing parameterizations of the dielectric formalism expressions (such as we discussed at the end of Section 7.2), frequency-moment sum rules specific to a Bose-condensed fluid are of considerable interest. In Section 8.1, we discuss several rigorous /-sum rules for the condensate and non-condensate parts of Xnn(Q,o)) based on (5.24). These were originally derived by Wagner (1966), Hohenberg (1967) and Wong and Gould (1974) for soft-core interatomic potentials. In Section 8.2, we critically review the phenomenological two-fluid formula for S(Q,CD) of Woods and Svensson (1978). This was the first attempt to describe the fact that the maxon-roton quasiparticle peak appeared to disappear at Tx. In Section 8.3, we use the long-wavelength, zero-frequency limit of the longitudinal current-response function XJJ(Q,CO) to derive what one might call "superfluid" and "normal fluid" /-sum rules. The usefulness of these results and those in Section 8.1 is briefly discussed.

8.1 The Wagner-Wong-Gould /-sum rules The Placzek or longitudinal /-sum rule (2.24) is a direct consequence of the high-frequency behaviour given by (2.26), nO2 lim Xnn(Q, OJ) = lim xm(Q, ©) = - ^ + ... .

CO—KX)

CO—>-00

Wlto

195

(8.1)

196

Sum-rule analysis of the different contributions to S(Q,co)

This leads to the first frequency moment -

i r° dco

r°

- / —-co Im Xnn(Q,co) = / dco coS(Q,co) = sQ , n J_oo 2n J_O0

(8.2)

valid for arbitrary wavevectors. We will now show that in any Bosecondensed fluid, the two components of S(Q,co) which are given by (5.24) each satisfy their own "/-sum rule." A sum rule can be derived for Si in (3.18) by using the kind of result first discussed by Wagner (1966). Expanding Gap in (3.29) in powers of 1/co, one finds lim V Gap(Q,co) = lim Y GajS =

co—•oo *-^

—[SQ+B^Q)]

COZ

co—•oo *-^

(8.3)

where we have defined floo(Q) = Zn(e,G) -^ oo) -l>n(Q9co -+co)-ti

(8.4)

This means that Si (Q, co) satisfies the high-frequency sum rule /»OO

dco coS{(Q,co) = —SQ + — Bao(Q) .

(8.5)

In the co —• 00 limit, the only self-energy diagrams which are left in (8.4) are those which are frequency-independent. For a "soft" potential such that V(q) is non-singular, these are the Hartree-Fock diagrams given by the first terms in (5.52) and (5.53). In this case, (8.4) reduces to BaoiQ) = BHF(Q) = nV(q = 0) - p + / - ^ J

F(p + Q)(np - mp) . (8.6)

(2TC)

The sum rule (8.5) is closely related to the one given in (4.5). Both of these sum rules can be generalized to potentials with a hard core, by introducing the ^-matrix for free-atom scattering (Griffin, 1984). For an arbitrary interatomic potential, one can show that lim (8.7) where the vertex function defined by

m l ^ q ' ) ^ /(q,q') + J - ^

]

^_ »a ^

(8.8)

is the high-frequency limit of Eq. (22.3) of Fetter and Walecka (FW, 1971). The scattering amplitude /(q,q') for two 4 He atoms in free space

8.1 The Wagner-Wong-Gould f-sum rules

197

is given by the solution of the integral equation in Eq. (11.14) of FW. The scattering amplitude / is well defined even for a potential with a hard core and hence so is Too in (8.8) (in Eqs. (17) and (19) of Griffin, 1984, one should replace /(±q,q) by raFo^+^q)). It is interesting to note that for a non-singular potential with a well defined Fourier transform, one can show that Foo(q,qO a s defined in (8.8) reduces to the bare potential V(q — q'). This follows from the third equation on p. 131 of FW. Thus we see (somewhat surprisingly) that in this case (8.7) does reduce to the expression which led to (4.5), namely

Anp[7(q

= O)

+ 7(p + Q)] .

(8.9)

Suitably generalized using (8.7), one can view B^Q) in (8.5) as, in principle, known. One can numerically calculate the diagonal and offdiagonal momentum distributions hp and mp for realistic interatomic potentials (see, for example, McMillan, 1965). It would be useful to have #oo (2) evaluated as a function of Q and tabulated in the literature. An additional /-sum rule can be derived using the exact Ward identities (5.17)—(5.19). In the high-frequency limit, only the lowest-order diagram in (5.55) contributes to A£. Using this in (5.17), one obtains to leading order

lim Aa(Q,co) = ^

CO-+CO

[l - sgna ^ ® 1 . L

(8.10)

Using results analogous to (8.3) in combination with (8.10) gives (Wong and Gould, 1974; Talbot and Griffin, 1983)

(8.11) Thus one finds / dco coSc(Q,co) = - % - -B^Q) , (8.12) J-oo n * n where Sc is the dynamic structure factor associated with fnn = AGA in (5.24). An immediate consequence of (8.1) and (8.12) is that the second term in (5.24) satisfies 5

dco coSR(Q,co) = -sn + — BaoiQ) , n

(8.13)

where n = n — no. In summary, when no ^ 0, the usual /-sum rule (8.2)

198

Sum-rule analysis of the different contributions to S(Q,co)

splits into two separate sum rules (8.12) and (8.13). These are rigorous results, valid at all wavevectors. The same sum rules also hold for %n

and xL

As we have mentioned, #00 (g) should be calculable by Monte Carlo numerical methods. For large enough g, it is clear from (8.6) (or its equivalent for hard-core potentials) that #00 (g) will become g-independent. In particular, it will become negligibly small compared to the free-atom energy SQ = Q2/2m. In this large-g limit, the Bose fluid /-sum rules (8.12) and (8.13) reduce to OO

dco coSc(Q,co) = ^sQ

lim / Q^oo J_o0

dco coSR(Q,co)

= ^8Q .

(8 14)

)

The two components of S(Q,co) given by the high-g impulse approximation in (4.2) satisfy these sum rules. The Wagner-Wong-Gould sum rules can be recast into analogous ones for cross-correlation functions such as (p(Q, £)S_Q(0)). These are of special interest since they give a direct measure of how the density and single-particle fluctuations are mixed via the condensate. Recalling the formal definitions in (5.4) and (5.8), we have (here a represents the 4 He atom destruction operator) (8.15) XnAQ>v) = Xn,i(Q,co) = A1G12 + A2G22 . Using (5.6), (8.10) and the high-frequency expansions of Gap (see also (8.3)), COZ

(8.16) we finally obtain the remarkably simple high-frequency expansion lim z»,a(Q,«) = lim XnAQ,(o) = - ^ co—>-oo

co—•oo

+^ s CO

(8.17)

Recalling the spectral representation such as in (2.25), (8.17) immediately gives the frequency-moment sum rules (compare with (8.2))

(8.18) J-ao

In contrast with (8.5) and (8.12), these sum rules do not involve

8.1 The Wagner-Wong-Gould f-sum rules

199

on the right hand side. This difference can be traced to the fact that Xn,a involves the total density p(Q) rather than PQ (see (3.13)). Stringari (1991, 1992) and Giorgini and Stringari (1990) have recently given an elementary derivation of various frequency moment sum rules for response functions involving combinations of SQ,SQ and PQ, using equal-time commutation relations. These authors have emphasized the usefulness of particle density sum rules such as (8.18) as a way of determining the overlap between single-particle and density fluctuations (see also remarks at the end of Section 9.1). Stringari (1992) has also derived additional sum rules analogous to those of Wagner in (4.5) and Wong and Gould in (8.12) but with the detailed balance factor [N(co) + 1] in the frequency integral. This factor effectively gives more weight to the low-frequency region and thus these new sum rules can be useful in studying low-energy, long-wavelength excitations. In contrast, the sum rules (4.5) and (8.12) give more weight to high frequencies. This spectral region is much more dependent on the details of the hard-core potential (see also Wong and Gould, 1974). On the other hand, it is the high-energy spectral region of various response functions which is of most interest in relation to the predicted hybridization effects resulting from the maxon excitation overlapping the zero sound particle-hole mode (see discussion at end of Section 7.2). In addition, at around Q ~ 2.4 A" 1 , we expect similar hybridization effects when the roton branch crosses the two-roton branch (see Section 10.2). High-frequency tail ofS(Q,co) The /-sum rules (8.5) and (8.12) are direct consequences of the highfrequency behaviour of the various terms contributing to Xnn in (5.24). A related subject is the high-frequency behaviour of S(Q,co) itself. This has been the subject of several studies related to the so-called "deep inelastic scattering" behaviour of quantum fluids. It is argued that at high enough energy transfers, S(Q, co) should be independent of the nature of the elementary excitations, quantum statistics, etc. This deep inelastic region is sometimes defined as when the energy transfer is much larger than the free-particle energy (co > SQ). This deep inelastic region has been mainly studied at T = 0 (Wong and Gould, 1974; Bartley and Wong, 1975; Family, 1975; Wong, 1977). One finds the asymptotic behaviour is given by (Kirkpatrick, 1984)

^ £

(8.19)

200

Sum-rule analysis of the different contributions to S(Q,co) i

400

•

1 -

i \

300

200 -

100

Fig. 8.1. Fit of the high-frequency tail of S(Q,ca) to the predicted co~1/2 line shape, for Q = 0.8 A"1, T = 1.2 K. The neutron data are from Woods et al., 1972 [Source: Wong, 1977]. for a quantum fluid of hard spheres with diameter a (where n = 4 for Qa • 1). The coefficient in (8.19) is proportional to the second derivative of the pair distribution function g"(r = a). In Fig. 8.1, we show a fit of the high-energy tail to co~7/2 for Q = 0.8 A" 1 and T = 1.2 K (Wong, 1977). It would be useful to have more systematic studies of the high-frequency tail of S(Q, co) over a wide spectrum of Q values, as well as theoretical predictions based on a more realistic interatomic potential. While no calculations at T ^ 0 have been reported, it is expected that the high-frequency tail of S(Q, co) will be almost temperature-independent since it apparently only depends on the pair distribution function. As noted by Talbot and Griffin (1984a) and Talbot et al. (1988), this temperature independence is in agreement with the 5(Q,co) data in the range 0.8 < Q < 2 A" 1 . As shown by the examples in Figs. 1.5 and 2.3, the high-frequency tail appears to be the same at all of temperatures for co ^ 0.6 THz (30 K). For further discussion, see Svensson (1991). 8.2 Woods-Svensson two-fluid ansatz for 5(Q, co) The first systematic high-resolution study of S(Q,co) as a function of the temperature was by Woods and Svensson (1978). Their results strongly indicated that the intensity of the maxon-roton peak went to zero as one approached 7^. In an attempt to explain their observations, Woods

8.2 Woods-Svensson two-fluid ansatz for S(Q,co)

201

and Svensson introduced an ad hoc two-component expression which divided S(Q, GO) into a "superfluid" part and a "normal" part, allowing one to extract out a "quasiparticle" resonance from the former. As with any parameterization, there are two questions about the WS formula: (a) Does it give a reasonable, consistent fit to the data? (b) Does the parameterization have a well defined microscopic basis? Both aspects have been extensively discussed in the literature (Glyde and Svensson, 1987; Griffin, 1987; Talbot, Glyde, Stirling and Svensson (TGSS), 1988; Svensson, 1989). On the experimental question as to how good the fit is, the detailed analysis by TGSS at p = 20 bar suggests serious problems. We also recall from Chapter 7 that the temperature dependence of the S(Q,co) line shape is qualitatively different in the phonon, maxon and roton wavevector regions. Even in the Q ;> 0.8 A" 1 region, the analysis presented in Section 7.2 gives no reason to expect that there will be a single "universal formula" for S(Q,co) which describes both the maxon and roton regions. However, we think it is still worthwhile to review the WS scenario, as historically the first attempt at the kind of finitetemperature parameterization we are looking for. Moreover, a similar parameterization of Xnn instead of Xnn may be a promising direction in analysing data, as we discuss in Sections 7.2 and 8.3. The WS ansatz divides S(Q, co) below Tx into two parts, S = Ss + SN ,

(8.20)

where the "superfluid" component is Ss(Q,co) =

Z(Q9T)[N(G})

+ 1]A(Q9;T

= 1 K) (8.21)

and the "normal" component is ^

;

)

(8.22)

Effectively, S(Q,co) is assumed to be a weighted sum of three separate contributions: a quasiparticle Lorentzian peak (of frequency COQ and width TQ) described by A(Q,co) as in (7.4); a multiparticle part SM', and a normal component S#. Within this scheme, one makes the key simplifying assumptions that SM can be described by its low-temperature value (~ 1 K) with a weighting factor ps/p', and SN can be approximated by its value just above 7^, weighted by PN/P- Using (8.20)-(8.22), one can then isolate the quasiparticle contribution in the S(Q,co) data and determine the temperature dependence of Z() which only describes the quasiparticle resonance at COQ. In Fig. 8.2, we show the relative contribution of S\ and Sn to the ordinary /-sum rule at T = l . l K (Cowley and Woods, 1971). As noted by Wong (1979), the sum rules for Sc and SR are analogous to the Ambegaokar-Conway-Baym (ACB) sum rules in anharmonic crystals. As we discuss in Section 11.2, while useful, the ACB /-sum rules in quantum crystals still do not allow one to cleanly isolate the phonon peak in S(Q,co) data on solid Helium at large g, where interference effects are significant. We now derive another set of Bose /-sum rules, based on the lowfrequency, long-wavelength limit of longitudinal current-response func-

8.3 Superfluid and normal fluid f-sum rules

205

tions. We recall that, using (5.16a) and (5.16b), Xnn(Q,)

f ? / J J ( Q ) m coz

Imxm(Q,co) = ^

( 8 < 2 9)

As with Xnm one can split Xn m t o the s u m of t w ° terms, one of which vanishes in the normal phase (Talbot and Griffin, 1984b) Xn = Xn + Xn ,

(8.30)

with Xn = — XjJ ' Xjn

(8.31) R Xjn •>

where we find it convenient to define a new vertex function

Ai = Ai + Aa^®f/n.

(8.32)

The correlation function Xjn i s defined in (5.13b) and (5.6). Similarly, the regular part xfjj 1S given by (see also (5.13b)) j&=jft + j$>

(8-33)

with the proper condensate contribution

The zero-frequency limit of %JJ(Q,CO) is given by (6.2), - = p ,

(8.35)

valid for arbitrary wavevectors. The zero-frequency, long-wavelength limits of the condensate and regular contributions in (8.30)-(8.34) can be summarized by Km - /n(Q,co = 0) = -fjj(Q,co

= 0) = pN(T) ,

Km - /fj(Q,a> = 0) = -Xn(Q,co =0)= ps(T) .

(8-36)

These results follow from (6.2), (6.4) and the fact that Xjj[(Q,a) = 0) = 0.

206

Sum-rule analysis of the different contributions to S(Q, co)

Writing (8.36) in the form of spectral representations as in (8.35), we immediately obtain

and the analogous expressions for XJCJ and ~yf/jResults such as (8.37) immediately suggest that at low g, it is useful to divide S(Q,co) into "superfluid" and "normal" parts as follows:

(8.38) nn mzcoz With these definitions, the sum rules in (8.37) lead to the following new long-wavelength /-sum rules: lim/ —oo 6~*0 J—oo

lim /

dcoa>Ss(Q, 0 limit and do not offer a microscopic basis for the Woods-Svensson two-component ansatz introduced to explain line-shape data in the region Q ^ 0.8 A" 1 (see Section 8.2). Several comments are needed concerning the significance of the results in (8.39). First of all, we note that these superfluid and normal fluid /-sum rules are a consequence of taking the Q, co —> 0 limit of the components of the longitudinal current-response function. In contrast, the /-sum rules in (8.12) and (8.13) involve the high-frequency limit of the components of the density response function and are, formally, valid at arbitrary values of Q. The results we have found seem physically reasonable, namely that the high-frequency limit /-sum rules of Section 8.1 depend on no and h while the low-frequency limit sum rules of this section involve ps and pN. The sum rules in (8.12) and (8.13) are thus not equivalent to those

8.3 Superfluid and normal fluid f-sum rules

207

in (8.39). In particular, one cannot make a naive identification based on (8.29) and assume that

im fm= im

[wrong]

•

(84o)

This would lead to the equalities Ss = Sc and S^ = SR, which are not satisfied in general (the right hand sides of (8.12) and (8.39a) are not equal). For further discussion of these differences, see Wong (1979) and Talbot and Griffin (1983). The preceding analysis also shows that there are two equally valid ways of splitting Im inn into two components, one component being proportional to the single-particle Green's function. Depending on whether we start from (5.24) for inn or (8.30) for XJJ> the part of xnn containing Gap as a separate factor involves quite different Bose broken-symmetry vertex functions. This means that the particle-hole and multiparticle spectral contributions (both contained in Aa and A£) would modulate how the structure of Gap appears in S(Q,co) in different ways. As a result, the apparent weight of the quasiparticle peak could be different in the two formulations. A partial resolution of this apparent arbitrariness lies in the fact that there may be cancellations between the two components of expressions such as (8.30) and (5.24). It was for this reason that we argued in Section 7.2 that (5.24) and (3.47) are only useful in the large-g limit (£ 1.5 A" 1 ). In conclusion, while we have derived formally exact two-component expressions for inn and %JJ, there does not appear to be a simple way of experimentally distinguishing the resulting two components in S(Q,co) data. The same difficulty occurs in solid 4 He, as discussed in Section 11.2. On the other hand, the analogous two-component forms for inn and ijj, and the various /-sum rules they satisfy, should be useful in developing simple parameterized expressions for these regular functions. The preliminary analysis given at the end of Section 7.2 worked in terms of xcnn and x£r An improved analysis might start with simple approximations for fjj and Jjj for use in (5.81). The application of the Bose sum rules of the kind discussed in this chapter clearly is in its infancy and further studies are needed. For a systematic but elementary discussion of rigorous frequency-moment sum rules associated with Bose broken symmetry, we recommend the recent papers by Pitaevskii and Stringari (1991) and Stringari (1992).

9 Variational and parameterized approaches

In Chapter 5, we argue that in an interacting fluid having a Bose condensate, a sharp peak in S(Q,co) could have two origins. It could be either a collective zero sound (ZS) density fluctuation or a renormalized single-particle (SP) excitation. In Section 7.2, we use this scenario to argue that the low-g phonon peak observed in superfluid 4 He was a collective ZS mode but the high-Q maxon-roton peak was associated with an SP excitation. However, we also saw that while this qualitative picture is plausible in terms of the microscopic theory for a Bose-condensed fluid (see summary in Section 5.5 and discussion in Section 7.2), at the present time we have few quantitative calculations of the regular functions Xnn> Aa and £a£ which are relevant to liquid 4 He. These include results at arbitrary temperatures in the case of a dilute Bose gas (see Sections 5.2 and 5.3) and the T = 0 calculations of Gavoret and Nozieres (1964) in the long-wavelength limit (discussed in Section 6.3). In addition, we have rigorous limits at very low and very high frequencies (see Chapters 6 and 8). In this chapter, we discuss some alternative approaches which have been used to calculate (mainly at T = 0) the phonon-maxon-roton dispersion curve as well as the S(Q,a>) line shape. In Section 9.1, we review what is perhaps the most successful method, that based on direct numerical calculations of the many-body wavefunctions and energies using variational techniques. The ground-state wavefunction is described by a Jastrow-Feenberg type form which can incorporate the strong short-range correlations present in a quantum liquid. (The same kind of wavefunction is also used in quantum solids, as we discuss in Chapter 11.) The variational excited states are then built up from this ground state by considering density fluctuations coupled into backflow processes, as first done by Feynman and Cohen (1956). These methods (often referred to as the correlated-basis-function or CBF 208

9.1 Variational theories in coordinate space

209

approach) are now able to obtain impressive results for the phononmaxon-roton quasiparticle dispersion relation at T = 0, as well as for its weight in S(Q,co) using a direct evaluation of the expression (2.14). While these methods can be used to evaluate the single-particle density matrix (and hence the 4 He atomic momentum distribution and condensate fraction, as discussed in Chapter 4), they do not deal with the single-particle Green's functions. Consequently, the dynamical role of the condensate is not very clear. In addition, using variational methods to deal with the effects of finite temperature and damping on time-dependent correlation functions is always difficult and somewhat ad hoc. In Section 9.2, we review the polarization potential theory introduced by Aldrich and Pines (1976). This approach (which so far has been worked out only at T = 0) may be viewed as a phenomenological parameterization of the density response function %nw(Q,co), which attempts to include the correct many-body physics. We discuss its consistency with the general structure of Xnn(Q,&>) implied by the dielectric formalism (as summarized in Section 5.5). The memory function formalism sketched in Section 9.3 has many of the same goals as polarization potential theory. In existing formulations, however, we do not believe that broken symmetry unique to a Bose-condensed fluid has been adequately incorporated. As a general comment on these alternative methods, we note that they all focus on the density-response function Xnn{Q, co) directly, rather than on the more fundamental single-particle Green's function Gajg(Q,co) which lies at the basis of more complicated two-particle correlation functions like Xnn- Consequently these approaches, useful as they are, are quite different from the field-theoretic methods used in the rest of this book.

9.1 Variational theories in coordinate space Beginning with the pioneering work of Bijl (1940) and Feynman (1954), there have been many studies of superfluid 4 He based on ab initio calculations of the many-body wavefunctions using a variational approach. To the extent that one can find good approximations to the ground state and low-lying excited states, one can directly obtain the excitation energies and also evaluate S(Q,co). In this approach, the variational wavefunctions are given in a coordinate-space representation, depending on the positions of all the atoms (ri,r 2 ,...,rjv). A key requirement for Bose fluids is that these wavefunctions be symmetric with respect to the interchange of any two particles. Matrix elements taken with respect to these

210

Variational and parameterized approaches

wavefunctions can then be reduced to quantities which are dependent on various static distribution functions (the radial two-body distribution function g(r) in (2.10) being the simplest one). A nice introduction to this approach is given in Chapter 10 of Mahan (1990). Extensive reviews of this method have been given by Feenberg (1969), Woo (1976) and Campbell (1978). More recent developments are summarized by Clark and Ristig (1989) and Campbell and Clements (1989). The work of Manousakis and Pandharipande (1984, 1986) gives the results of state-of-the-art calculations using correlated-basis-function methods for both the quasiparticle excitation energy COQ and the dynamic structure factor S(Q,co). Up to the present time, these variational methods have been developed mainly at T = 0 (see, however, Senger et al, 1992). Campbell and Clements (1989) review attempts to extend this approach to finite temperatures, building on the pioneering work by Penrose (1958) and Reatto and Chester (1967). Typically, trial variational approximations to excited states are given in terms of coordinate-space operators acting on the ground state |Oo). The Jastrow-type ansatz |Oo) = O ( n , r 2 , . . . , r N ) = Hfirtj)

(9.1)

incorporates the short-range two-body correlations (r,7 = |r,—r;|) and the conditions which the ground-state wavefunction for a Bose system must satisfy. Since two atoms cannot get closer than the hard-core diameter a, f(r) must vanish for r < a. One usually takes f(r)=e-u(r)/2

( 9 2 )

with u(r) = (b/r)5, where b is determined variationally. The long-range correlations induced by the condensate discussed in Section 4.1 can be shown to imply that

A generalized Jastrow-Feenberg ansatz for |Oo) which includes threebody correlations ("triplets") gives a ground-state energy (Oo|//|Oo) = £o, which is within a few per cent of the value obtained by direct Green's function Monte Carlo (GFMC) methods. The associated atomic momentum distribution np and condensate fraction can also be calculated

9.1 Variational theories in coordinate space

211

from |Oo) using the single-particle density matrix (Clark and Ristig, 1989) Pl

(r-r') = N f dx2 f dr3 • • • f drN®*0(r, r2, • • r^)O0(r/, r2,• • • rN) , (9.4)

with

j * T )

(9.5)

As we discussed in Section 1.2, the well known Feynman-Bijl approximation to the normalized excited state is given by (1.4) and (1.5),

Within this approximation, we note that such excited states are pure density fluctuations of wavevector Q. The excitation energy defined by (&FB{Q)\H

~ EO\®FB(Q))

= CDFQB

(9.7)

can be expressed in terms of the radial distribution function g(r) or, more precisely, its Fourier transform S(Q). As indicated by (1.7), the spectral weight of S(Q,co) is exhausted by this mode, which has a weight given by S(Q). Feynman and Cohen (1956) introduced an improved approximation for the excited states in an attempt to include "backflow" processes. In their original coordinate-space description, the FC wavefunction is given by (9.8)

with the generator given by

Q = Z e i Q T j f1 + l Z

At large r, it can be shown that n(r) = A/r3, which describes dipolar backflow. In recent variational calculations based on (9.8) and (9.9), however, rj(r) is determined variationally so as to minimize the energy of the state (9.8). In momentum space, one can rewrite (9.9) in the form (Miller, Pines and Nozieres, 1962)

= P+(Q) X

+ +

One finds that Aq& vanishes as Q —• 0 and hence FQ generates a pure density fluctuation. However the Feynman-Cohen excited state (9.10) increasingly deviates from a pure density fluctuation as Q increases.

212

Variational and parameterized approaches

While they were originally introduced to describe backflow processes, the wavefunctions in (9.9) and (9.10) are much more general when rj and A are determined variationally. In the original Feynman-Bijl approximation, it is implicitly assumed that all the important correlations between atoms are already contained in the ground-state wavefunction |O0). The excited states do not involve any new correlations and thus \Q>FB) simply involves multiplying |O0) by a one-body operator, as in (1.4) or (9.6). This is not the case with \

(9 24)

with the momentum-dependent effective mass being defined by m*Q = m + nfQ (Aldrich and Pines, 1976). With (9.24), the zero sound dispersion relation given by the zero of the denominator of (9.23) reduces to !2 •

(9.25)

This result implies that the sound velocity is directly related to the long-wavelength limit of fg, YYIC

l™ So. = —

•

(9-26)

It is convenient to define the scalar potential fq in terms of a realspace soft potential / s (r). For r > rc (the core radius), / s (r) is assumed to be identical to the attractive long-range potential of the bare 4 He- 4 He interaction. For r < rc, / s (r) is parameterized in PP theory by the ansatz (9.27) In liquid 4 He, it is found that the core radius is a constant rc ~ 2.68 A, at all densities. The only remaining parameter in (9.27) is the soft-core strength a and this is uniquely determined through (9.27) and (9.26) by the (pressure-dependent) sound velocity c. It is found that a = 49 K in (9.27) from a fit to SVP data, and increases with pressure. We note that the parameterized / s (r) leads to a fsQ which goes through zero at

9.2 Polarization potential theory

221

Q ~ 1.85 A" 1 at all pressures. The determination of fg is somewhat less direct, but a procedure for obtaining it will be discussed shortly. In applying (9.23) to superfluid 4 He at T = 0, Aldrich and Pines (1976) and subsequent papers (Pines, 1985, 1987) have been based on the model approximation for the screened response function s c X n n(Q,co)

nQ2/rn0 = aQ-f—^ + XmP(Q, co) . CO —

(9.28)

Physically the first term (of weight OLQ) describes the density fluctuation produced by exciting a single quasiparticle e*Q = Q2/2m*Q out of the condensate. The second or "multiparticle" term describes the excitation of two (or more) such quasiparticles out of the condensate and becomes increasingly important as Q increases. This contribution is peaked at high energies (Hess and Pines, 1988). As we have noted above, PP theory is based on separating Xnn into two parts, the polarization fields (whose strength is determined by fsQ and /£) and the screened responses to these fields. The main effect of pressure is expected to be on renormalizing the values of fsQ and / £ , rather than on changes in %%n. Confirmation of this basic idea was obtained by Aldrich and Pines (1976) in the following way. They fitted the observed dispersion curve COQ at T ~ 1 K and SVP to the PP expression given by (9.23), (e,co) .

(9.29)

This fit was based on the value of fsQ discussed above in conjunction with the ansatz (9.28), the multiparticle part being approximated by its zerofrequency limit, —nAq. The values of ag, AQ and fq are then adjusted to give the best fit to the SVP data for coQ up to Q ~ 2 A" 1 . The resulting values of OLQ and AQ are assumed to be independent of pressure and are shown in Fig. 9.3. Since COQ does not change with pressure at Q = 1.85 A" 1 , AP use (9.29) in conjunction with the ansatz (9.28) to determine the fvQ at this particular value of Q for different pressures (assuming that OCQ and AQ are pressure-independent). The value o f / Q at Q = 0 can be obtained from the pressure-dependent effective mass mj in liquid 3 He at the same density (at SVP, this gives n/g ^ 3.2m). Finally, JQ at intermediate values of Q is determined by a smooth extrapolation between Q = 0 and 1.85 A" 1 . Armed with the pressure-dependent values of fsQ and / £ (the SVP values are shown in Fig. 9.4) and the pressure-independent values of OCQ and AQ, Aldrich and Pines (1976) then used (9.28) and (9.29) to predict

222

Variational and parameterized approaches 0.020

0.0

Fig. 9.3. The momentum dependence of the relative weight of the single-particle (OLQ) and the multiparticle (nAQ = —/ mp(Q,co — 0)) contributions to the screened density-response function given by the ansatz in (9.28) [Source: Aldrich and Pines, 1976; Pines, 1985].

how COQ changes with pressure. They found excellent agreement up to the maximum possible pressure of 25 bar. We view this agreement as justification of the idea that the main pressure dependence of %««(Q, co) comes through fsQ and / £ , rather than the precise form of xsncn given by (9.28). In trying to relate the PP expression (9.23) to the general results of the dielectric formalism of Chapter 5, we encounter a problem: the soft potential fsQ corresponds to some sort of t-matrix in the manybody analysis. By analogy with similar discussions in liquid 3 He, where multiple scattering has been extensively discussed, V(Q) in (5.2) can be replaced by a soft-core potential such as fsQ, i.e.

i-/eX«n(Q^)

(9.30)

If we compare this with the PP expression (9.23), we obtain Xnn(Q,CO) =

1 - fn

(9.31)

9.2 Polarization potential theory

223

2.0

Fig. 9.4. The momentum dependence of the polarization potential scalar and vector (backflow) pseudopotentials at SVP [Source: Aldrich and Pines, 1976, 1978].

The two-component form (5.1 \b) for Xnn = Xnn+Xnn which results from the existence of the Bose condensate will be reflected in the ansatz one makes for xSnn- The PP expression given in (9.28) may be viewed as appropriate at T = 0, the first term being an approximation to %n — AaG^Ap in the dielectric formalism. In this context, (5.50) suggests a natural extension of (9.28) to finite temperatures which includes the effect of the "normal fluid" associated with thermally excited quasiparticles. For completeness, in Fig. 9.5 we plot the single-particle excitation energy s*Q used in the polarization potential theory (see (9.28)). It would be useful to have further studies of ways to formulate the dielectric formalism results in terms of a suitably parameterized PP-type theory.

Variational and parameterized approaches

224

0.1 -

0.5

1.0

1.5

2.0

2.5

Q (A-1) Fig. 9.5. The dispersion relation of the bare single-particle energy E*Q = Q2/2m*Q used in the PP expression (9.28). The effective mass is m*Q = m + nfp, where the vector polarization potential is given in Fig. 9.4. The "zero sound" frequency is denned by (9.25).

In Section 3.4, we outlined a mean-field derivation of the simplest version of the dielectric formalism results. It is a straightforward process to extend that analysis to include the linear response due to a vector potential. This generalization gives a version of the polarization potential theory which explicitly includes the dynamical effect of the Bose order parameter. One is thus able to exhibit the effects of backflow, which have been left out of the correlation functions given in (3.52)—(3.55). After a somewhat lengthy calculation, one finds that Xnn is once again given by (9.23) (and hence Xnn is given by (9.31)), with #^(Q,co) being equal to the density-response function #j}n(Q,co) of a non-interacting Bose gas. Thus the key PP expression (9.23) is seen to be valid even when one explicitly includes the effect of the Bose order parameter. In this extended PP theory, one can also discuss how backflow modifies the single-particle Green's functions Ga^(Q,co) due to Bose broken symmetry (Griffin, Wu and Lambert, 1992). The arguments leading to (9.23) are also valid for liquid 3 He (Pines, 1985). Equally important is that, to a good first approximation, xSnn *s given by the Lindhard function for a quasiparticle gas with an effective

9.3 Memory function formalism

225

mass m*0 ~ 3m - in contrast with superfluid He, where we as yet do not have a simple model expression for %%n. 9.3 Memory function formalism Another method of evaluating the density response function xWn(Q,<w) is to use the memory function formalism. The method can be expressed in terms of projection operators which lead to a systematic (continued fraction) expansion for the memory (or polarization) function M(Q,co) defined below. (For a general introduction to this method, see Chapter 9 of Hansen and McDonald, 1986.) In most applications to quantum liquids, various simplified forms are introduced for M(Q,co). As with the polarization potential method discussed in Section 9.2, the basic physical assumptions concerning the nature of the excitations are then contained in the choice of model approximation for M(Q,co). To some extent, "what goes in is what comes out." An advantage of the memory function approach is that one can treat both classical and quantum liquids in the same way. In addition, the hydrodynamic low-co limit can be properly described, which is not the case with the variational approach discussed in Section 9.1. The application of the memory function formalism to the calculation of S(Q,co) in classical liquids is reviewed by Copley and Lovesey (1975). For a discussion of quantum liquids using the memory formalism, we recommend Gotze and Lucke (1976) and Lucke (1980). In the memory formalism, the usual "starting" expression for the density-response function or susceptibility is an expression of the kind

where the characteristic frequency Q(Q) is defined as (933)

This definition ensures that Xnn(Q,u>) automatically has the correct co = 0 limit, namely %nn(Q,co = 0) = — x(Q), where x(Q) *s the static susceptibility discussed in Section 2.2 and satisfies the compressibility sum rule (2.28). In (9.32), M(Q,a>) plays a role somewhat analogous to the selfenergy function in the single-particle Green's function. In the derivation of (9.32), projector techniques are used to separate the motions into a coherent one (characterized by the frequency Q ( 0 ) and incoherent fluctuations described by M(Q,a>). To the extent that this separation can be

226

Variational and parameterized approaches

done, M(Q,co) will be a fairly smooth function of co; it is thus a better candidate for approximations than Xnn(Q,a)). One may also treat (9.32) as a definition of M(Q,co), since Q(Q) in (9.33) is given in terms of an experimentally determined quantity x(Q). At such a phenomenological level, one may attempt to extract M(Q,co) directly from neutron scattering data. To the extent that M(Q, co) is non-singular, this is a very convenient way of presenting experimental data. For example, expanding (9.32) for large co and using (2.26), we find lim coM(Q, co) = ^a> = 0). One may easily verify that (9.32) can be written in terms of xr as follows:

*nQ'fr(r.

X«n(Q,a>) = -

, ,

1 - V(Q,co)xr{Q,co) where a frequency-dependent potential has been introduced: F(Q,co) = -y-1 =

z(Q)

(9-37)

— h ^ [M(Q, co) - M r (Q, co)] . (9.38) x (Q) "Q2 r

9.3 Memory function formalism

227

(In the literature - see, for example, Yoshida and Takeno, 1987 - such a reference system is often denoted by a "zero" superscript.) An approach based on (9.37) is only useful if one can incorporate enough of the dynamics into the reference system /(Q,co), leaving V(Q,co) in (9.38) to play the role of a relatively simple "effective" interaction. The similarity of (9.37) to the polarization potential expression (9.23) is clear. In normal liquid 3 He, the natural choice of xr is that of a non-interacting Fermi gas xQnn, i.e., the Lindhard function. The appropriate form for the "model response function" of superfluid 4He is not so obvious, as we noted in Section 9.2. Gotze and Liicke (1976) and Liicke (1980) have given a detailed analysis of S(Q,co) for superfluid 4He at T = 0 based on the assumption that the coherent mode Q(g) is self-consistently coupled into the twomode spectrum. In their mode-mode coupling approximation model, the imaginary part of the memory function is (we refer to the original paper for details)

x - [sgn co' + sgn(co - co')k™(P, co)/ n'n(Q - p, a> - co') , (9.39) where the double primes denote the imaginary parts. It is important that the pair modes are described self-consistently by true spectral densities x'nn associated with the response functions in (9.32). They find the interaction vertex function in (9.39) to be given by

S(Q-p)

[

where the static structure factors <S(p) and S(Q — p) are taken as inputs. As Gotze and Liicke note, the memory function analysis based on (9.32) and (9.39) is reminiscent of the work carried out earlier by Jackson (1969, 1973) within a variational scheme where one- and two-density fluctuations are coupled (see Section 9.1). One should be able to work out an improved version of the Gotze-Liicke theory which is built on the Feynman-Cohen states. It would also be useful to have the ManousakisPandharipande variational calculation (see Section 9.1) formulated via the Gotze-Lucke approach.

228

Variational and parameterized approaches

We have seen in earlier chapters that the existence of a condensate has the crucial effect of coupling the density fluctuations of a normal Bose liquid to the single-particle excitations. The resulting hybridization leads to the characteristic phonon-maxon-roton spectrum (see Section 7.2) observed in S(Q,co). A similar condensate-induced hybridization couples the one- and two-quasiparticle spectrum, as discussed in Section 10.1. The memory formalism with its continued-fraction expansion of M(Q, co) is general enough to incorporate explicitly these effects of the Bose broken symmetry. It would be useful to extend the work of Gotze and Lucke (1976) in this direction. Finally, we sketch the bare bones of the continued-fraction representation of Xnn(Q,a>). Working with the Laplace transform = //

dte-ztxm{Q,t)

(9.41)

one finds Xm(Q,z) = A0(Q)[zM0(Q,z) - 1] ,

(9.42)

with the recursion relation

The wave vector-dependent functions An(Q) are defined in terms of frequency moments of Xnn(Q, co)- For further details of how one derives these results, we refer to the literature (pp. 64ff of Boon and Yip, 1980; p. 311 of Yoshida and Takeno, 1989; Chapter 9 of Hansen and McDonald, 1986). Using (9.43), one can express Mo(Q,z) in (9.42) as a continued fraction as follows:

Mo(Q, z) =

i— Z+

(9.44)

z+A 2 (0M 2 (Q,z) TT^

•

(945)

A2(6) z + A 3 (6)M 3 (Q,z) Thus one can formally express #wn(Q,z), in terms of the frequency moments An(Q), in the form of an infinite-order continued fraction. One may verify using (9.42) and (9.44) that Xm

(Q z) = -Ao(0A1(0_ 'Z z 2 + A i ( 0 + A 2 (0zM 2 (Q,z)

9.3 Memory function formalism

229

This is the analogue of (9.32), with z = ico. Using the fact that zM2(Q,z) = 0 for z = 0, one concludes from (9.46) that A0(g) = x(Q). Similarly the high-frequency limit

l,mz,.(Q,z) = - ^ M > ,

(9.47,

z-»oo

in conjunction with the /-sum rule, shows that A o (g)Ai(0 = nQ2/m. Thus we find Ai(g) = Q 2 (g), where Q(Q) is defined in (9.33). Approximations enter when we truncate the expansion in (9.45) at some finite order. One can show that keeping terms up to order n in (9.45) means that Mo(Q, t) will be given correctly up to order t2n. If we truncate the expansion in (9.45) using z2

(9.48)

then (9.46) reduces to (z = ico) (9.49) CB2-A3(Q)

This result is equivalent to that obtained by Yoshida and Takeno (1987) for the time dependence of xnn. The poles are given by the solution of Eq. (3.10) of YT. Clearly this form can be interpreted as a density mode of frequency Q(Q) = y/A\(Q) being coupled (or hybridized) via A2(Q) with another mode of frequency Q 3 ( 0 = y/A^Q). From this point of view, we conclude that the peculiar oscillatory time dependence of the memory function obtained by YT (1987) is most naturally interpreted as a condensate-induced hybridization of quasiparticles with the two-quasiparticle continuum. (Note, however, that the ideal Bose gas response function used by YT as a "reference system" is not valid for superfluid 4He.) This same problem was originally studied in frequency space by Ruvalds and Zawadowski (1970) using Green's function techniques (see Section 10.2). It would be interesting to use (9.49), or something similar, to treat the hybridization of zero sound and the maxon-roton excitations which is discussed in Section 7.2. A phenomenon similar to the hybridization of zero sound and the single-particle maxon-roton excitations in pure superfluid 4 He also occurs in 3 He- 4 He mixtures. As discussed by Lucke and Szprynger (1982), when we use the memory function formalism generalized to deal with a twocomponent 3 He- 4 He mixture, the 4 He quasiparticle memory function is

230

Variational and parameterized approaches

found to have a resonance associated with the 3 He particle-hole modes. The resulting hybridization of the 4 He quasiparticle with the 3 He p-h states is often referred to as mode-mode coupling (for further discussion, see Hsu, Pines and Aldrich, 1985). The dielectric formalism for 3 He- 4 He mixtures is briefly reviewed in Section 12.2.

10 Two-particle spectrum in Bose-condensed fluids

At many points in this book, we have mentioned the high-frequency scattering intensity which appears in the S(Q,co) data. This high-frequency component (see Fig. 1.6) is usually identified with the spectrum of two excitations (with total momentum Q) and is thus referred to as the multiphonon or multiparticle component. In addition to inelastic neutron scattering, this two-excitation spectrum can be more directly probed by inelastic Raman light scattering, but only at Q = 0. In this chapter, we briefly review the microscopic theory of such pair excitations and how they show up in both neutron and Raman scattering cross-sections. Raman light scattering in superfluid 4 He has been extensively studied both theoretically and experimentally, especially with regard to the possible formation of bound states involving roton-roton, roton-maxon and maxon-maxon pairs (Ruvalds and Zawadowski, 1970; Iwamoto, 1970). High-resolution Raman experiments over a wide range of pressure and temperature are reviewed by Greytak (1978) and more recently by Ohbayashi (1989, 1991). An excellent theoretical introduction at a phenomenological level is given by Stephen (1976). Our emphasis will be on the role of the Bose broken symmetry. In earlier chapters, we have carefully distinguished the single-particle Green's function Gi(Q, co) (which may be a 2 x 2 matrix) and the densityresponse function Xnn(Q,<x>)> The latter gives the dynamic structure factor measured by neutron scattering. In the present chapter, we introduce several additional correlation functions which are needed to describe the pair-excitation spectrum and Raman scattering. The two-particle Green's function G2(Q,co) describing the propagation of two atoms of total momentum hQ and energy hco is discussed in Section 10.1. We follow the analysis of Pitaevskii (1959) as well as Ruvalds and Zawadowski (1970). We also review how the pair spectrum is hybridized into the single231

232

Two-particle spectrum in Bose-condensed fluids

particle spectrum described by Gi(Q,co). This mixing occurs through the effect of the Bose condensate and is analogous to the hybridization of the p-h spectrum of Xnn(Q,&>) with Gi(Q,co), as described in Chapters 5-7. In Section 10.2, we discuss the pair-excitation spectrum as exhibited in S(Q,co). In Section 10.3, we briefly consider the Raman light-scattering crosssection h(Q ~ 0, co) and its relation to a correlation function involving four density operators (in contrast to the two density operators in S(Q, co) involved in the Brillouin light scattering and in neutron scattering). In the usual analysis of this four-point correlation function, the density fluctuations are treated as elementary excitations and hence h(Q ~ 0, co) is viewed as describing the propagation and interaction of two such density fluctuations. In contrast, our analysis views the pair excitation in the context of the field-theoretic dielectric formalism of Chapter 5, in which the quantum field operators (rather than the density fluctuation operators) are the starting point. We indicate the crucial role of the Bose broken symmetry in coupling the pair-excitation spectrum into both Ga^ and inn. 10.1 Two-particle Green's function We first outline the T = 0 calculations of Ruvalds and Zawadowski (1970) and Zawadowski, Ruvalds and Solana (ZRS, 1972), which are a development of those by Pitaevskii (1959). In contrast to more phenomenological approaches to be discussed in later sections, the work of ZRS is grounded in a field-theoretic analysis of Bose-condensed liquid, such as we have used in earlier chapters. The pair-excitation spectrum is discussed in terms of the two-particle Green's function G2(Q,co), the Fourier transform of

G2(r - r\ r - t1) = -(7>(r, Ovfc OvV, O v V , 0) • +

(10.1)

Here the quantum field operators tp ,tp are defined as in (3.10) and describe the creation and destruction of non-condensate 4 He atoms. Clearly G2(Q,co) describes the propagation of two atoms with total (centre-of-mass) momentum hQ and total energy hco. ZRS calculate G2(Q, co) in the standard ladder approximation used for discussing bound states, which is described by a Bethe-Salpeter integral equation. The existence of a bound state (for an attractive interaction) shows up as a resonance below the two-particle continuum of two non-interacting excitations.

10.1 Two-particle Green's function

233

Before considering the approximate calculations of ZRS and Pitaevskii, it is useful to recall the general analysis of two-particle Green's functions Klp(p',p;Q) given by Gavoret and Nozieres (1964) and summarized in Section 5.4. Using the notation of that section, we found that Ky^ as defined in (5.60) is given by the sum of a condensate term (5.63) (see Fig. 5.12) C

KW^

Q) = Z e^(p'» Q)GPAQ)Ql3(p> Q) p,a

and a regular term (5.65), written schematically as (see Fig. 5.13) R

K = GiGi + GiGi + GiGirGiGi .

(10.3)

Finally the generalized interaction vertex F is given by the Bethe-Salpeter equation (5.68), written schematically as (see Fig. 5.14) r = / + ^/GiGir .

(10.4)

These are all matrix equations. We note that Ky^ in (5.59) is defined in terms of the total field operators tp,tp+ and not just the non-condensate parts \p,xp+. Thus G2 in (10.1) is associated with the regular part RK++(p',p;Q), as given by (10.3) and (10.4). We see, however, that because the Gfs are 2 x 2 matrix propagators, the Bethe-Salpeter equation (10.4) for T\\(p\p\Q) will be coupled into several of the other 15 functions F ^ . The functions K^ and r£t satisfy various symmetry relations (see GN) which can be used to simplify the resulting coupled integral equations. Such calculations are illustrated by the analysis of Cheung and Griffin (1971b) at T ^= 0 and also Nepomnyashchii and Nepomnyashchii (1974) at T = 0. At this basic level, of course, the single-particle excitations, the particle-hole excitations and the two-particle excitations are all coupled into each other due to the Bose condensate. Apart from the asymptotic region of Q, co —» 0 (see Section 6.3), there are essentially no theoretical studies which work out the details of such a complete microscopic theory (while ensuring that all correlation functions share the same poles). Rather than carry out a full calculation of the type sketched in the preceding paragraph, we make use of a simplified scheme following ZRS and Pitaevskii. First of all, we note that all components of G^ share the same poles in a Bose-condensed fluid. Moreover, if we are interested in the spectrum associated with creating two excitations, we can concentrate on the positive-energy poles of Ga^ and effectively schematize the structure

234

Two-particle spectrum in Bose-condensed fluids

of the equations of motion (Pitaevskii, 1959; pp. 237ff of Abrikosov, Gor'kov and Dzyaloshinskii, 1963). This is especially the case when one is considering a region (Qc,(oc) where a single excitation can decay into two excitations, both of which come from (2, oo) regions far removed from (Qc,a>c)' More specifically, one can also argue that at large wavevectors the numerator of the anomalous propagator Gi2(Q, co) is much smaller than that of Gn(Q, co) because of the decreased importance of the Bose coherence factors (this is illustrated by the Bogoliubov approximation results in Section 3.2, with UQ —> 1 and VQ —• 0 at large Q). All these remarks set the stage for the ZRS calculations, based on solving (5.65) and (5.68) for RK++ keeping only the Gn (or G ++ ) diagonal component of the 2 x 2 matrix Green's functions. Moreover, the latter is approximated by Gn(Q,co)-

-=r- ,

CO —

(10.5)

where the "unhybridized" single-particle excitation EQ is assumed to be given approximately by the observed maxon-roton dispersion relation in the region 1 < Q < 2.4 A" 1 . As Q increases, EQ goes over smoothly to Q2/2m (see the dashed line in Fig. 7.21). Following Ruvalds and Zawadowski (1970), we distinguish EQ from the observed spectrum OQ since the former will be strongly hybridized when it overlaps with the pair-excitation spectrum. As we have noted above, this approach attempts to isolate the problem of finding the pair spectrum and ignores the tricky self-consistency problem which requires that all two-particle correlation functions K^ and one-particle functions Gap share the same poles. This kind of procedure may still give reasonable results for the two-particle energy spectrum. We recall, for example, that a conserving approximation for Xnn m&y be built on the Hartree-Fock single-particle Green's functions (Cheung and Griffin, 1971b). In Section 10.2, we discuss how the two-particle spectrum we obtain here shows up in the spectrum of Gap and Xnn using the dielectric formalism results of Chapter 5. In the context of the above kind of simplified analysis, we now discuss the evaluation of G2(Q,co) in (10.1) in a slightly more direct manner. It is convenient to introduce a general four-point two-particle Green's function for Bosons: G 2 (l,2; l',2') = -RQR/4TZQ (and again we recall that E = w — 2A). These results are only valid for intermediate values of Q. As with the Q = 0 case discussed above, we see that (10.13) reduces to (g'4 = gipoiQ)) Pi(Q,^

= J^

for co < 2A. Thus, as long as g4 is attractive (< 0), we have a roton bound state with energy (^\

(10.23)

238

Two-particle spectrum in Bose-condensed fluids
), given by (10.10) and (10.11) in the ladder-diagram approximation. In this language, the pair-excitation spectrum in (3.45) corresponds to using G2 = 2Fo, where Fo is defined in (10.11). Examining (3.45), one sees that the pair excitation has its origin in the fact that the single-particle Green's functions have poles at ±COQ, the negative-energy pole having a finite weight due to the Bose broken symmetry (see (3.36)). The resulting cross-terms in (3.44) lead to the two-particle contributions of energy (op + (Op+Q. This two-particle continuum contribution of energy S(Q, co) has a coherence factor involving products of the up and vp amplitudes and hence vanishes if vp does (i.e., if no = 0). Needless to say, (3.45) involves

10.2 Two-excitation spectrum in neutron scattering

243

15 Energy (K)

Fig. 10.4. Scattering intensity vs. frequency for Q = 0.3 A^and T = 1.2 K. The unlabelled arrow is at 2A = 17.3 K. The lines are only a guide to the eye [Source: Woods, Svensson and Martel, 1972]. the same kind of integration as discussed in Section 10.1. The dominant contributions come from the roton and maxon regions because of high density of states. This multiparticle contribution should not, however, be thought of as the excitation of two atoms out of the condensate. Rather it arises from the fact that in a Bose-condensed system, creating (or destroying) an atom with finite momentum is a coherent process involving both creation and destruction of quasiparticles of that momentum. In an analogous way, <S(Q,a>) in the Gavoret-Nozieres formalism of Section 5.4 is the sum of the two contributions CK±+ and RK±+ given by (10.2) and (10.3), respectively. f*ckushima and Iseki (1988) have made a careful analysis of the pair spectrum (including the maxon-roton part left out of (10.28)) based on the GN formalism, but only consider the R K±+ contribution, as given by (5.65) and (5.68). The contribution given by the analogue of (10.31) was not included. The first detailed study of the multiparticle distribution for low Q values (0.3 A" 1 and 0.8 A" 1 ) was by Woods, Svensson and Martel

244

Two-particle spectrum in Bose-condensed fluids

0.8 Energy (THz)

Fig. 10.5. The pressure dependence of the scattering intensity vs. frequency at T = 1.27 K, Q = 1.5 A" 1 . This should be compared with similar data at 1.13 A" 1 shown in Fig. 7.9 [Source: Stirling, 1985].

(1972), whose data are shown in Figs. 10.4 and 8.1 for T = 1.2 K. As do Stirling and Glyde (1990), they find the multiparticle peak to be quite symmetric for Q = 0.3 A" 1 , although at Q = 0.8 A" 1 , it has a high-energy tail extending to <x> ~ 70 K. This high-energy tail seems characteristic of all data at high Q (see further remarks at the end of Section 8.1). The low-temperature 5(Q,co) data always show a broad distribution centred at - 20-25 K for 0.3 < Q < 2.5 A" 1 (see Fig. 1.6). There seems little doubt that such a broad, high-energy component can arise from processes involving the creation of two quasiparticles. High-resolution neutron studies (especially in the maxon region) show detailed structure at frequencies close to that expected for the creation of two rotons (2A), a roton and a maxon (A + AM) and two maxons (2AM)- The roton and maxon regions make the dominant contribution to the two-quasiparticle

10.2 Two-excitation spectrum in neutron scattering

400

200

245

10 15 Energy (K)

Fig. 10.6. Scattering intensity vs. energy loss for Q = 0.3 A"1, at a temperature 2.3 K just above TA. An expanded view of the high-energy region shows no evidence for the resonance which is visible in the low-temperature data in Fig. 10.4 [Source: Svensson, Martel, Sears and Woods, 1976].

scattering because of the high density of states at QM and QR. Such fine-scale structure is strikingly evident in recent ILL data of Stirling and coworkers, as shown in Fig. 10.5 as well as in Figs. 7.7, 7.9, and 7.10. The preceding analysis gives a natural explanation of the origin of the high-frequency multiparticle component in terms of the two-quasiparticle spectrum. This spectrum is coupled (or mixed) into the density fluctuation spectrum only because of the effect of the Bose order parameter. As a counterexample, S(Q, co) in normal liquid 3 He exhibits a well defined zero sound peak for Q < 1 A" 1 but there is no evidence for any high-energy multiparticle structure of the kind that arises in superfluid 4 He in this low-momentum region. Apart from our conclusion that the multiparticle component at high frequencies should disappear with the condensate fraction (i.e., it should be absent in normal liquid 4 He), at the present little is known about its precise temperature dependence. It deserves more study in its own right, especially in the region of low Q where it appears to separate out clearly as a distinct symmetric peak, without any high-frequency tail (see Figs. 7.1 and 10.4). Svensson (1989) has argued that the Q = 0.3 A" 1 data of

246

Two-particle spectrum in Bose-condensed fluids 1

One-phonon -

160 • 120 Multi-phonon

40 1 0.20

1 0.60

1 1.00 Frequency (THz)

1 1.40

1.80

Fig. 10.7. Scattering neutron intensity distribution at Q = 2.9 A"1 for T = 1.1 K, and a pressure of 24.3 atm. This can be compared with the SVP data in Fig. 7.20 [Source: Smith, Cowley, Woods, Stirling and Martel, 1977]. Svensson, Martel, Sears and Woods (1976) at 1.2 and 2.3 K (shown in Figs. 10.4 and 10.6, respectively) are consistent with the disappearance of this peak above 7A, in agreement with the preceding remarks. Further high-resolution studies at low Q would be very useful, at a series of temperatures. A careful study was carried out by Smith, Cowley, Woods, Stirling and Martel (1977) in the region 2.9 < Q < 3.3 A" 1 , at both low and high pressures, looking for evidence of a two-roton bound state. Some of their data are shown in Figs. 10.7 and 10.8. The curves shown in Fig. 10.8 are based on the Ruvalds-Zawadowski (1970) expression for Gi(Q,co) in conjunction with (10.31), S(Q,o>) oc n0 Im where G2(Q,a>) is given by (10.28). The hybridization coupling strength g3 = y/nog4, po(6) = nA/Q and the energy cutoff D in (10.28) were treated as fit parameters. The overall agreement is found to be quite reasonable, but this must be viewed in the context of the remarks we made following (10.31). Many calculations (at T = 0) of the scattering due to the creation of two quasiparticles have been carried out (see, for example, Jackson, 1973,

10.2 Two-excitation spectrum in neutron scattering

247

300

Q = 2.7 A-1 200

100

o U 0 50

•fi=3.lA-i

Q = 3.3 A-1

0.52 0.36 0.44 Frequency (THz) Fig. 10.8. Scattered neutron distributions (same temperature and pressure as in Fig. 10.7) for momentum transfers in the range 2.7 < Q < 3.3 A" 1 . The solid lines are fits based on the RZ expressions in (10.27) and (10.28), convoluted with instrumental resolution [Source: Smith et a/., 1977]. 0.20

0.28

1974; Gotze and Liicke, 1976) using semi-phenomenological approaches. These studies involve the evaluation of the equivalent of the expression in the third line of (3.45). The most detailed study using the correlatedbasis-function approach is by Manousakis and Pandharipande (1986), who evaluated the contributions to S(Q,co) due to the creation of one and two quasiparticles. Over a wide range of wavevectors, their calculations exhibited structure from roton-phonon, roton-maxon, maxon-maxon and roton-phonon pair spectrum. However, as we noted in Section 9.1, MP ignore the possibility of bound states. In the context of this chapter, (10.10) is approximated by G2 = 2F 0 .

248

Two-particle spectrum in Bose-condensed fluids

In concluding this section, it is clear that current theoretical discussions of the pair spectrum exhibited by S(Q,co) are still very crude, although promising. Several extensions are needed before one can hope to make a serious comparison with the data. We list a few areas for future theoretical work: (a) We need a calculation of S(Q, co) which includes the pair contribution from both terms in (3.47) or its analogue. Expressions like (10.31) cannot be expected to give a quantitative fit to the data. (b) In using expressions like (10.28), we have limited ourselves to the structure related to exciting two rotons, co ~ 2A. It is straightforward to evaluate p^Q^co) m (1014) to obtain a result which is valid at all Q in the whole spectral region 2A < co < 2AM- This allows one to include the spectrum (and possible bound states) associated with maxon-roton and maxon-maxon pairs. The pair-excitation structure exhibited by lmFo(Q,co) is strongly modified in ImG2(Q,co), which in turn is further modified by hybridization with the single-particle excitation described by Im Gi(Q, co). The results are very dependent on the values chosen for the effective two-particle interaction g4 and the condensate-induced hybridization strength. For further details, see Juge and Griffin (1993). (c) At finite temperatures T ^ 1.2 K, one must include the quasiparticle width EQ — iTq, where the temperature dependence of the half-width TQ(T) is given by (7.2). A simple way of doing this is to use (Ruvalds, Zawadowski and Solana (RZS), 1972)

F0(Q,co) = 2 jdco' J

^'l

co — co' +

r IITQ

(10.32)

in (10.13) with p\ as defined by (10.14). The resulting generalization of (10.27) and (10.28) is given by Eq. (4.16) of ZRS for co ~ 2A. In addition, one can take into account the changing condensate-induced hybridization by allowing the parameter flo(T) in (10.27) to decrease as the temperature increases. (d) In this and the preceding section, we assume that the input phononmaxon-roton spectrum EQ (see (10.5)) is a single quasiparticle branch. A major extension would be to incorporate the Glyde-Griffin scenario developed in Section 7.2. This generalization is necessary if one hopes to understand the high-frequency maxon region in any detail. In this region, the particle-hole and two-particle spectrum can be expected to overlap in energy and hybridize.

103 Raman scattering from superfluid 4He

249

10.3 Raman scattering from superfluid 4 He Second-order inelastic light scattering was first suggested as a useful probe of superfluid 4 He by Halley (1969). Up to a constant which is of no interest here, the Raman-scattering rate or extinction coefficient for an isotropic liquid is given by (si9 8f are the polarization vectors of the incident and scattered light)

h(a>) = /s(co)(e, • ef)2 + Id{a>) \1 + i(g, • ef)A

(10.33)

where hco = Q, — Q/ is the energy transferred from the photons to the liquid and

If((o)=(J^\

j dqAntf j dq[An) but not on the shape of the two-particle spectral density p2(Q = 0,co) (for the case 2 ^ 0 , see Figs. 10.1 and 10.2). In many-body language, propagator renormalization effects are included but not vertex corrections (the effect of interactions between the excitations, which may lead to bound states). The second comment concerns the interpretation of the results in Figs. 10.10 and 10.11 near and above Tx in terms of the "roton" width and energy. In Section 7.2, we argued that a careful analysis of the S(Q,Q)) line shape at Q ~ 2 A" 1 in the region near Tx (see Fig. 7.11) was consistent with the idea that the roton peak intensity was vanishing, being replaced by a broad particle-hole spectrum (associated with thermally excited rotons) which characterizes the normal phase. If one models the resulting changes in the S(Q9co) line shape in terms of a single Lorentzian with a rapidly increasing width near Tx, one will naturally be led to results of the kind shown in Figs. 10.10 and 10.11. One cannot, however, interpret such fits as giving information about rotons near and above Tx. As we discussed in connection with neutron-scattering data in Section 7.2, a meaningful extraction of information about the

254

Two-particle spectrum in Bose-condensed fluids

quasiparticle spectrum at temperatures above about 1.7 K requires a rather sophisticated theoretical input. Even at low temperatures, the simple convolution approximation (10.38) has several deficiencies: (a) It ignores what one might call the excluded-volume effect due to the 4 He atom hard core. In the coordinate space form given in (10.36), it is clear that there can be no contribution from the regions |r3 — r4|, 1*1 —r2| < a, where a is the hard-core diameter. The simple decoupling (10.38) does not handle this constraint properly (for further discussion and references, see Halley, 1989). (b) As we mentioned above, by approximating (10.37) in terms of two non-interacting rotons, all the effects discussed in Section 10.1 have been ignored. In particular, we cannot discuss how h(co) will show the presence of a two-roton bound state which may arise when there is an attractive roton-roton interaction g4. (c) In approximations such as (10.38), the key step lies in treating the density fluctuations, rather than the field operators, as the fundamental variables. All available calculations of S2(q,qr;o>) which include the effect of two-roton bound states are based on identifying the density fluctuations as the elementary excitations. That is, something like p(q) = y/Z(q) [a+ + a_q]

(10.42)

is used, where a+ is the creation operator of an excitation (see Section 9.1). In this type of approach, the correlation function S2(q9q';co) involving four density operators is effectively reduced to a two-particle Green's function, such as G2(Q = 0, co) defined in (10.1). To this extent, the description of two-roton bound states can be taken over from the analysis given in Section 10.1. This approach of relating S2(q,q';&>) directly to G2(Q = 0,co) involves the same sort of approximation as taking the density-response function Xnn to be directly proportional to the single-particle Green's function, as in (10.31). Halley and Korth (1991) have extended the analysis based on (10.42) to include backflow by calculating S2(q,qr;co) starting from (10.37) using the correlated-basisfunction approach of Manousakis and Pandharipande (1986) discussed in Section 9.1. What one would like to see is a calculation of S2(q,q''9co) based on treating it as a true four-particle Green's function (involving eight quantum field operators). In earlier chapters, we saw the importance of keeping the

103 Raman scattering from superfluid 4He

255

30 Energy shift (K)

Fig. 10.12. Raman intensity vs. energy shift &>, measured at 0.65 K and SVP. The instrumental FWHM is 0.75 K. Weak structure above the two-roton peak at 2A is clearly evident in the expanded part of the high-energy data [Source: Ohbayashi, 1991].

distinction clear between inn and the more fundamental single-particle Green's functions Gap. The analogous investigation of S2(q,qr;co) has not been carried out in the literature to date. Writing (10.37) in the form (10.43)

one sees that reducing this expresion to products of pairs of single-particle operators (i.e., single-particle Green's functions) results in additional terms which are not included in (10.38) even in a normal Bose fluid. Moreover, when there is a Bose condensate present, we have a whole

256

Two-particle spectrum in Bose-condensed fluids

new class of pairings involving the off-diagonal averages, (a£(t)a*k) and (ak(t)a-k). We thus conclude that S2(q,q';co) describes dynamical correlations in superfluid 4 He which are not expressed simply in terms of density correlation functions, as in (10.38). This is a complication, but it also suggests that the Raman-scattering intensity may yield unique information not available from neutron-scattering experiments. Further studies of S2(q,q';c0) based on the field-theoretic analysis appropriate to a Bosecondensed fluid are clearly needed, as are studies of the two-particle Green's functions discussed in Section 10.1. As we have reviewed in Section 10.1, there is good evidence from the low-temperature Raman data that, at low pressure, there is a two-roton bound state. This was first observed by Greytak and Yan (1969) and has been confirmed by increasingly high-precision studies. Combining the Raman data of Murray et al. (1975) with the roton energy given by neutron scattering, the two-roton binding energy is estimated to be 0.27 + 0.04 K (Woods et al., 1977). More recent work is summarized by Ohbayashi (1989, 1991), who has also found evidence for additional fine-scale structure at higher energies (see Fig. 10.12). The latter may be due to higher-order resonances associated with maxon-maxon states as well as to bound states involving combinations of three or more maxons and rotons (for further discussion, see Iwamoto, 1989).

11 Relation between excitations in liquid and solid 4 He

In the early 1970's, attention was drawn to the remarkable similarity between the excitation spectra exhibited by S(Q,co) in solid 4 He and superfluid 4 He at low temperatures (Werthamer, 1972; Horner, 1972a; Glyde, 1974), as shown very dramatically in the theoretical results of Figs. 11.1 and 11.2. While various suggestions have been made as to the origin of this similarity, it remains an unresolved and intriguing problem. In this brief chapter, we compare the theoretical description of excitations in a quantum solid with those of a Bose-condensed liquid. While we review the key ideas, we assume that the reader has some familiarity with an introductory account of quantum crystals. (The modern theory of excitations in quantum crystals was essentially completed in the early 1970's. For background and a more detailed discussion of solid 4 He than we give in this chapter, we recommend the review by Glyde, 1976.) In both condensed phases, it is important to distinguish clearly between the elementary excitations and the density fluctuations. We argue that the phonons in solid 4 He are the natural analogue of the single-particle excitations in liquid 4 He. In Section 11.1, defining the phonons as the poles of the displacement correlation function, we briefly review theories which start with the self-consistent harmonic (SCH) approximation or something similar. In Section 11.2, we discuss the relation between the displacement-displacement and the density-density correlation functions in solid 4 He. This relation is based on the well known Green's function analysis of anharmonic crystals initiated by Ambegaokar, Conway and Baym (1965). Finally, in Section 11.3, we compare the expressions for S(Q,(o) in solid and superfluid 4 He. We do not discuss the interesting possibility of finding a Bose condensate in a quantum solid. For references, see Meisel (1992). 257

258

Relation between excitations in liquid and solid 4He

co (meV)

Fig. 11.1. Theoretical results for S(Q,co) as a function of Q (along the (111) direction) and co, in bcc solid 4 He. The peak intensity is seen to follow a "phonon-maxon-roton" type dispersion curve (dark line), with the shifting of spectral weight to free-particle-like behaviour (dashed line) at high Q [Source: Horner, 1974a]. 11.1 Phonons as poles of the displacement correlation function

The usual Hamiltonian describing an anharmonic crystal is obtained by expanding the interatomic potential energy in powers of the atomic displacements from the (Bravais) equilibrium sites. The degrees of freedom are described by the displacement field and thus the elementary excitations (phonons) correspond to the poles of the displacementdisplacement correlation function (usually called the phonon propagator). The lowest-order harmonic approximation consists of neglecting the cubic and higher-order anharmonic force constants. Expressing the atomic displacement of the / -th atom in terms of the usual phonon creation and annihilation operators, we have

2mNa>°qX

(11.1)

where AqX = aq)i+a*x and e x is the polarization vector of the X phonon branch. The Fourier transform of the (retarded) one-phonon Green's

11.1 Phonons as poles of the displacement correlation function

259

2 (A"1) Fig. 11.2. Theoretical dispersion curves (co vs. Q) for one-phonon and multiphonon structure in S(Q,co) in bcc solid 4 He (in three different directions). The heavy lines and shaded region give the mean energy and width, respectively, of the multiphonon scattering. The dash-dot line gives the free-atom recoil frequency. The heavy dots represent the peaks in the S(Q,co) data (Kitchens et al, 1972), with a width given by the dashed lines [Source: Horner, 1974b].

260

Relation between excitations in liquid and solid 4He

function Dfat) = -i9(t)([Aq,(t)J^(0)])

(11.2)

is given in a (self-consistent) harmonic approximation by Dx(q9co)=(

—0

^-0-) .

(11.3a)

The corresponding one-phonon spectral density is Ai(q,co) = -2 Im D^(q,co + «y)

= 2TT [5(o> - G)JA) - S(co + co,0,)] .

(11.36)

The effect of including higher-order anharmonic terms in the Hamiltonian can be treated using the usual diagrammatic perturbation methods of many-body theory (see, for example, Kwok, 1967; Horner, 1967). In such discussions, the phonon Green's function corresponding to some sort of harmonic approximation is the "building block", analogous to the single-particle Green's function for a non-interacting gas in discussions of quantum liquids. The effect of the anharmonic interactions (cubic, quartic, ...) gives rise to a phonon self-energy which leads to frequency shifts and damping of the original harmonic phonons. Extensive calculations of this kind are available in the literature (see, for example, Maradudin and Fein, 1962; Cowley, 1968; Glyde, 1971). In trying to formulate an analogous theory for solid 4He, one is immediately faced with the problem that small displacements about the equilibrium sites are unstable and thus the standard harmonic approximation is not a good starting point. Since the associated phonon frequencies are imaginary, self-consistent phonon (SCP) theories have been developed, based on renormalized force constants. The latter incorporate the effects of large zero-point fluctuations and lead to well defined phonon frequencies. These methods are based either on field-theoretic resummations of higher-order effects or on variational approximations to the many-body wavefunctions. In the latter approach, the physics is clear, since a wavefunction which is Gaussian in the atomic displacements is equivalent to some effective harmonic Hamiltonian. (For a lucid summary of how this procedure is carried out, we refer to Section II of Gillis, Werthamer and Koehler, 1968.) The problem is that if the strong short-range correlations (SRC) are included by an additional Jastrow-Feenberg-type function, we do not necessarily find any simple equivalence to an SC harmonic Hamiltonian. In particular, many theories which include SRC no longer guarantee that the excitations still

11.1 Phonons as poles of the displacement correlation function

261

correspond to poles of the displacement correlation function. More or less satisfactory self-consistent phonon theories with short-range correlations were developed in the early 1970's. A Green's function approach is given by Horner (1974c). One such theory of phonon excitations can be elegantly formulated in variational terms using the general approach of correlated basis functions (Feenberg, 1969). As we review in Section 9.1, this kind of approach has been extensively developed as a description of liquid 4 He at T = 0. Thus it is useful to discuss the analogous approach for solid 4 He. Here we briefly sketch the work of Koehler and Werthamer (KW, 1971), who have applied the CBF philosophy to the determination of the excited states of solid Helium in a consistent manner. Earlier work by Koehler (1967, 1968) treated the ground state in terms of a variational many-body wavefunction |Oo) which was a product of a Jastrow function describing the short-range correlations (SRC) and a Gaussian function |G) automatically leads to the diagonalization of the Hamiltonian in the one-phonon-state subspace. Several authors have emphasized that the poles of the displacement correlation function are not necessarily exhausted by the phonon fre-

262

Relation between excitations in liquid and solid 4He

quencies given by SCP theory. A useful comparison can be made between SCH theories of quantum crystals and Hartree-Fock theories of quantum liquids. The Hartree-Fock approximation gives the best renormalized single-particle states but completely ignores collective modes. The SCH approximation, in contrast, gives the best collective modes but ignores single-particle excitations. Most SCP theories implicitly assume from the beginning that the elementary excitations can be classified in terms of some equivalent harmonic lattice. This assumption is built into the form used for the excited-state wavefunctions. In this connection, we note that the precise relation between collective vs. single-particle theories of excitations in solid 4 He has never really been resolved (Fredkin and Werthamer, 1965; Gillis and Werthamer, 1968; Werthamer, 1969). Some aspects of this relation are touched on by the numerical work of Horner (1972b), whose results for the spectral density Ax(Q,co) show the clear transition from phonon-like excitations at low Q and co to more single-particle-like excitations at high Q and co. The same transition is shown by the results for S(Q, co) given in Figs. 11.1 and 11.2. One should also recall that when anharmonic corrections are included in such theories, the phonon propagator spectral density is not usually strongly peaked as in (11.3b). As Q increases, the spectral density spreads out over a larger frequency region and develops a high-energy tail, as shown by the example in Fig. 11.3. Such self-energy effects inevitably arise when one corrects the SCH spectral density to include anharmonic interactions. Thus, the peak positions in Fig. 11.3 are quite different from the line centres (denoted by the arrows) defined as the normalized first frequency moment of Ax(Q,co). One is faced with how to define an appropriate mean or average phonon energy for given Q, X (Horner, 1972b). Indeed, one is led to question the usefulness of any such simplified description in calculating the thermodynamic properties of a quantum (or highly anharmonic) crystal in place of the full displacement-field spectral density A^(Q,co). Finally, we note that for small Q , the phonons in a solid can be also usefully classified as either hydrodynamic or collisionless, following the terminology of Section 6.2. Typically phonons studied by ultrasonics or Brillouin light scattering are in the low-energy, hydrodynamic domain and are often referred to as first sound. Solid 4 He also exhibits second sound, which may be viewed as an oscillation in the local number density of phonons. As Kwok and Martin (1966) discuss, both first sound and second sound appear as poles of the displacement correlation function of an anharmonic crystal in the long-wavelength limit (see also Kwok,

11.2 Phonons vs. density fluctuations in solid 4He

0.2

0.4 0.6 Frequency (THz)

0.8

263

1.2

Fig. 11.3. The one-phonon spectral density A(Q,co) as a function of Q and co, for a longitudinal mode in bcc solid 4He along the (1,0,0) direction. The phonon self-energy due to a single "bubble" has been computed self-consistently. In comparing these results with Fig. 3 of Horner (1972b), see ref. 20 in McMahan and Beck (1973) [Source: McMahan and Beck, 1973]. 1967). In contrast, the low-Q phonons in solids which are excited by neutron scattering are in the high-energy collisionless region. For this reason, such excitations are often referred to as zero sound phonons, in analogy to the collective modes in quantum liquids.

11.2 Phonons vs. density fluctuations in solid 4He Since the middle 1960's, it has been realized that, in general, the dynamic structure factor S(Q,co) of anharmonic crystals can be a complicated function of the underlying phonon excitations. The key simplifying feature of a crystal is that the dynamics involve displacements of the atoms with respect to a Bravais lattice, r/ = R/ + u/. Substituting (2.5) into the structure factor (2.7) gives (11.5)

Relation between excitations in liquid and solid 4He

264

Expanding (11.5) in powers in the atomic displacements, we obtain S(Q,co) = SBmgg(Q)S{co) + Si(Q,o>) + Sint(Q,co) + Smp(Q,co) ,

(11.6)

where

^2YiQiRR'f)Q'(ul(t)u,(0))-Q

(11.7)

is the "one-phonon" contribution involving the displacement-displacement correlation function DiV{t) = (M,(*)K/'(0)). In (11.7), d2(Q) = exp[-2W(Q)] is the Debye-Waller factor. The additional terms in (11.6) describe twophonon (and higher) contributions plus interference terms. The higherorder phonon contributions which arise in (11.6) complicate the situation considerably in a highly anharmonic quantum crystal like solid 4 He. These give rise to both a broad multiphonon continuum in S(Q,(o), and interference terms which contribute within the one-phonon region and considerably modify the contribution of Si (Q, co) to the total dynamic structure factor at large Q. It is now recognized that in solid 4 He, there is an important difference between the density fluctuation spectrum (peak positions in S(Q, co)) and the underlying phonon excitations (peaks in Ax{Q9o)) or Sx(Q9(o)). The many-body theory of the density-response function XnniQ, co) in solid 4 He can be formulated in terms of Dx(Q, co) following the analysis of Ambegaokar, Conway and Baym (ACB, 1965). The final result can be written schematically in the form Xm(Q, « )

Particle - hole or density excitations

Av Single-particle excitations

211

Neutron scattering

X G2(Q, co) Two-particle or pair excitations

Raman scattering

Fig. 12.1. A block diagram showing how the condensate couples the various kinds of excitations in a Bose-condensed fluid.

excitations are all coupled into each other (Fig. 12.1). This sharing of excitations is the key dynamical consequence of a Bose condensate, as was first emphasized by Gavoret and Nozieres (1964) and Hohenberg and Martin (1965). The formal relations between the single-particle Green's function Ga^(Q,co), the particle-hole density-response function Xnn(Q,co) and the two-particle Green's functions G2(Q, co) have been discussed at length in Chapters 5 and 10. The precise implications of these coupled equations of motion, however, require the solution of very complex equations for self-energies, vertex functions, polarization parts etc. This has only been successfully carried out in the limit of small Q and co at T = 0 (as reviewed in Section 6.3). Even in the absence of rigorous calculations, however, the general structure of the coupled correlation functions made evident using the dielectric formalism (Chapter 5) does allow one to develop various scenarios concerning the excitations in a Bose fluid like superfluid 4 He. A quote from p. 23 of Bogoliubov (1947) applies equally well here: "All we can require from a molecular theory of superfluidity, at least at the first stage of investigation, is to be able to account for the qualitative picture of this phemonenon being based on a certain simplified scheme." In the interpretation of phonons, maxons and rotons developed in Chapter 7, the phonon is a zero sound (ZS) mode while the maxonroton is a single-particle (SP) excitation which has weight in S(Q,co) only because of the Bose broken symmetry. Physically, these two modes are interpreted as quite different excitation branches which are hybridized through the condensate to produce the observed phonon-maxon-roton dispersion curve (Glyde and Griffin, 1990). In this section, we review how

272

The new picture: some unsolved problems

this picture grew out of earlier studies. We also comment on what this picture assumes and implies about the nature of phonons and rotons. The standard microscopic view of excitations in superfluid 4 He was formulated in the classic paper by Miller, Pines and Nozieres (MPN, 1962) as well as by Nozieres and Pines in their 1964 monograph (see Section 7.3 of NP, 1964, 1990). This early work is based on Hugenholtz and Pines (1959), who first pointed out that in the presence of a Bose condensate, S(Q, co) would have a term directly proportional to the Beliaev single-particle Green's function Ga^(Q,co). Working to all orders in perturbation theory, Gavoret and Nozieres (1964) proved that, at small Q and co and at T = 0, both of these correlation functions exhibit the same phonon pole. (This behaviour already shows up in a dilute, weakly interacting Bose gas.) While explicit calculations were limited to the low 8 K) and exist only at relatively large wavevectors (^ 0.8 A" 1 ), and thus it seems physically reasonable that they would not be modified very much when the liquid goes from the normal to the superfluid phase (or for that matter, to the solid phase discussed in Chapter 11). The idea that a roton is physically quite different from a phonon has a long history. We recall that Feynman (1954) originally argued that the roton corresponds to a high-energy excited state involving the motion of a single atom in the potential well of its neighbors. Miller, Pines and Nozieres (1962) also viewed the roton as an atomic-like excitation modified by backflow. Chester gave several suggestive arguments that indicated that phonons and rotons were physically quite distinct excitations, concluding that (to quote from p. 65 of Chester, 1963): From Q = 0 up to Q — 1 A"1 the motions in the fluid consist of longitudinal density waves .... On the other hand, for Q ~ 2 A"1 we find that there is essentially no local density fluctuation and we have a single particle propagating fast through the liquid. The increase in effective mass of this atom comes from a large scale hydrodynamic backflow - with little or no change in the local fluid density. The region between Q = 1 A"1 and 2 = 2 A"1 must be interpreted as a rather complicated region in which we pass from a high frequency phonon mode at 1 A"1 to a single particle propagation at 2 A"1. This is clearly a possible transition in a liquid with an open structure and with low potential barriers. Our present analysis gives a reasonable scenario which allows for realiza-

12.1 Comments on the development of the new picture

275

tion of this kind of picture and extends these ideas to higher temperatures, to Tx and above. One of the most interesting implications of this new picture is that it suggests one should be able to understand the properties of normal liquid 4 He just above Tx in terms of a gas of roton-like excitations. Indeed, models based on excitations with a roton energy gap were used with partial success in the early literature (see, for example, Henshaw and Woods, 1961). Within the Glyde-Griffin picture, the "roton liquid" theory developed by Bedell, Pines and Fomin (1982) might also be appropriate above Tx. We recall that this phenomenological theory attempts to describe the thermodynamic properties of liquid 4 He in the region just below Tx in terms of an interacting gas of rotons, using a self-consistent field approach modelled after that used in the Landau Fermi liquid theory description of quasiparticle interactions in normal liquid 3 He. If the roton excitations are indeed intrinsic poles of the single-particle Green's functions (with an energy largely unaffected by the appearance of a condensate), the roton liquid theory formulation of Bedell, Pines and Fomin might be as relevant in the region immediately above Tx as it is below. (Of course, singularities in the specific heat and other thermodynamic quantities in the vicinity of Tx, specifically associated with the critical behaviour at a second-order phase transition, are not described by such a theory.) The relation of our scenario to the well known analysis of the excited states in a Bose liquid by Feynman (1954) is not completely clear. We recall that Feynman's analysis showed how Bose statistics was the key in ensuring that there were no other low-energy, long-wavelength excitations, apart from a phonon-like density fluctuation which by itself is not dependent on the nature of the quantum statistics involved. This work, however, made no reference to any role of a Bose condensate. The following remarks may be useful in understanding Feynman's work. In Section 1.1, we briefly reviewed the ideas behind the variational calculation of Feynman (1954). While Feynman (1953b, 1954) was motivated by a detailed microscopic examination of the various possible excited states in a Bose liquid, the variational theory he was led to is now viewed as a generic "single-mode approximation" for the densityresponse function. That is, if one starts with the assumption that S(Q,co) is dominated at all Q by a single undamped excitation of frequency COQ, the dispersion relation is given by the Feynman formula (1.6). This result is thus equally valid for normal liquid 4 He and liquid 3 He (for further discussion, see p. 915 of Mahan, 1990). The "dip" that such a disper-

276

The new picture: some unsolved problems

Fig. 12.2. The pair distribution function g(r) as a function of distance at 2.0 K, SVP. The solid line represents high-precision neutron data (Svensson et al, 1980) and the circles are path-integral Monte Carlo results. At this temperature, the calculated condensate fraction is 8%. When Bose symmetrization is not done, the only change in g(r) is shown by the dashed line [Source: Ceperley and Pollock, 1986].

sion relation exhibits occurs at the wavevector Q ~ 2 A l where the static structure factor S(Q) has its first maximum. Such a "roton dip" is thus a consequence of the short-range spatial correlations between atoms and has little to do with the quantum statistics the atoms obey. This fact is further emphasized by the path integral Monte Carlo simulations of Ceperley and Pollock (1986). As shown in Fig. 12.2, the short-range behaviour of the pair distribution function g(r) in superfluid 4 He is essentially unaffected when one does not carry out the Bose symmetrization. However, we recall the arguments given in Section 7.2 (see Fig. 7.17) against the assumption that the peak in S(Q,co) in normal liquids is associated with a single continuous excitation branch valid at both small and large Q. On the other hand, Bose statistics do play a crucial (but indirect) role in understanding why there are no other low-lying states in liquid 4 He besides the one generated by (1.4) (or the more modern versions such as (9.10)), at least at small wavevectors. By way of contrast, in normal liquid 3 He, at small wavevectors there are, in addition to the zero sound phonon mode, many other low-energy modes corresponding to the singleparticle Fermi quasiparticle excitations. It is Bose symmetrization which forces the excitations involving the motion of a single atom to have finite energy in the Feynman analysis. In turn, it is the Bose condensate that

12.2 Dielectric formalism for superfluid 3He-4He mixtures

277

allows the density fluctuation spectrum (phonons) to have a significant overlap on the field fluctuation spectrum described by the single-particle Green's function G(Q,co), justifying the implicit assumption of Feynman and Landau. Needless to say, by itself, our interpretation of the roton as an intrinsic single-particle excitation of normal liquid 4 He does not shed any light on why it has the precise energy spectrum hypothesized in Fig. 7.23. In this connection, Feynman's explicit discussion of the motion of a single atom in liquid 4 He is of great interest and deserves careful study. Very recently, Stringari (1992) has suggested a promising way of studying this question in a quantitative manner (see remarks at end of Section 9.1). 12.2 Dielectric formalism for superfluid 3 He- 4 He mixtures In this book, the dynamics of superfluid 4 He has been discussed starting from the microscopic theory of Bose-condensed liquids. The various scenarios we have developed concerning the nature of the excitations have the common feature that the Bose condensate plays a pivotal role. In order to test these ideas, we have seen that neutron-scattering experiments done at varying temperatures and pressures have been of special importance. Indeed the recent picture put forward by Glyde and Griffin (1990) was developed in an attempt to understand the temperature-dependent line shapes as the liquid goes from the superfluid to normal phase (see Section 7.2). In developing these ideas, the study of superfluid 3 He- 4 He mixtures is particularly promising since the 3 He concentration gives a new parameter which can be varied, in addition to the pressure and temperature. In particular, recent high-momentum-transfer neutron-scattering data (see Chapter 4) indicate that the condensate fraction in a superfluid 3 He- 4 He solution of 12% is considerably increased (Wang, Sosnick and Sokol, 1992). This suggests that hybridization effects induced by the 4 He condensate may be much stronger in mixtures than in pure superfluid 4 He. Talbot (1983) and Talbot and Griffin (1984c) have worked out the formal extension of the dielectric formalism given in Section 5.1 to a fully interacting system of Fermions and Bosons, with due allowance for a Bose condensate. This formalism can be applied to superfluid 3 He- 4 He mixtures. Just as in the case of pure liquid 4 He, the Bose broken symmetry leads to a coupling of the various correlation functions which is made most manifest by working in terms of irreducible, proper diagrams. The formal analysis is more complicated than that given in

278

The new picture: some unsolved problems

Section 5.1 by the necessity to include a whole new class of response functions involving the 3 He atoms. There is already a considerable body of literature on the elementary excitations and collective modes in superfluid 3 He- 4 He mixtures. A useful introduction to the literature is given in the review by Glyde and Svensson (1987). Fak et al. (1990) give a detailed analysis of recent neutron scattering data over a wide range of temperatures (0.07-1.5 K), with extensive references. The phenomenological quasiparticle theory of dilute 3 He- 4 He mixtures is discussed in detail by Baym and Pethick (1978), Ruvalds (1978) as well as in Chapter 24 of Khalatnikov (1965). Much of the work on mixtures to date has been concerned with low 3 He concentrations at low temperatures, with an emphasis on using the system as a way of studying a dilute Fermi gas of quasiparticles. In contrast, our major interest is to see how the phonon-maxon-roton excitation associated with the 4 He atoms is modified by the 3 He atoms. With this end in mind, it would be useful to have more neutron-scattering data at high 3 He concentrations, over a wide range of temperature (including above the superfluid transition). Besides the dielectric formalism we review in this section, the variational and phenomenological approaches reviewed in Chapter 9 have all been extended to superfluid 3 He- 4 He mixtures at T = 0. In connection with neutron-scattering data, the polarization potential approach of Section 9.2 is discussed by Hsu, Pines and Aldrich (1985) while the memory function formalism has been used by Lucke and Szprynger (1982) and Szprynger and Lucke (1985). As in the case of pure superfluid 4 He, these approaches can be brought into contact with the dielectric formalism by suggesting useful parameterizations of the "regular" functions in the latter theory. The coherent (spin-independent) neutron-scattering differential crosssection is proportional to (see Section 2.1)

)

(12.1)

where b3 and b* are the coherent neutron-scattering lengths for 3 He and He atoms, respectively, and x3 is the molar concentration of 3 He (see Fak et al., 1990, for further details). The dynamic structure factors are related to various number density response functions in the usual way:

Sab(Q,co) =

^=[N(co)

+ 1] Im Xab(Q,co) ,

(12.2)

12.2 Dielectric formalism for superfluid 3He-4He mixtures

279

where XabiQ, co) can be related to the Bose Matsubara Fourier components T

(pa(Q,T)pb(-Q))

(12.3)

(see (3.41) and (5.4) for analogous definitions for pure 4 He). Here a, b can stand for 3 He or 4 He and pa(Q) is the Fourier component of the number density operator of 3 He (a = 3) and 4 He (a = 4), defined as in (2.5) and (3.12). In the interpretation of neutron-scattering data on dilute mixtures, the cross-term S34 in (12.1) is often neglected, in which case the resonances in the total scattering intensity can be identified with renormalized 4 He and 3 He excitations associated with S44 and 533, respectively. Since 3 He and 4 He atoms have the same electronic polarizability, the cross-section for Brillouin light scattering (Stephen, 1976) can be expressed in terms of the dynamic structure factor associated with the total number density n = n^ + «4, Sab(Q,a/) .

(12.4)

a,b

As we discuss in Section 6.2, Brillouin light scattering probes the hydrodynamic domain. In superfluid 3 He- 4 He mixtures, this region is especially interesting since the intensity of second sound in Snn(Q,co) is greatly enhanced by the presence of the 3 He concentration fluctuations (Khalatnikov, 1965). The hydrodynamic spectrum of superfluid 3 He- 4 He mixtures has been thoroughly mapped out by light scattering (see, for example, Stephen, 1976; Rockwell, Benjamin and Greytak, 1975). It would be interesting to use the two-fluid equations for mixtures (Khalatnikov, 1965) to derive the single-particle Green's functions for the 4 He atoms in the asymptotic region of low Q and co (the analogous calculation for pure 4 He is sketched in Section 6.2). We now briefly summarize the results of Talbot and Griffin (1984c). These formally rigorous expressions are valid for all Q, co and T and are very relevant to inelastic neutron-scattering studies. In terms of irreducible contributions xab, one finds 133=

(12.5)

280

The new picture: some unsolved problems

where we define e(Q,oj) = 1 - 7(6)1X44 + 2X34 + Z33] , 1 > A(Q,CO) = Z447(0X33 - B 4 7 ( 0 X 4 3 • J

(12.6)

We make use of the fact that the bare interatomic potentials between the Helium atoms are identical, Vab = 7, for a, b = 3,4. As expected, all four response functions Xab in (12.5) share the same poles, which are given by the zeros of the dielectric function e(Q,a>) (Bartley et al, 1973). We also note that Xab = Z T7^ , (12.7) ajb

* ™

where xm = ^Xab- The simplest MFA approximation (see Section 3.4) aj>

corresponds to using the free-gas response functions for xah, i.e., xAA = X44, X 33 = X33

and

X34 = X34 = 0.

The crucial step in the dielectric formalism for Bose-condensed systems is to split irreducible contributions into proper (regular) and improper (condensate) parts, as discussed in Section 5.1. We first define (a = 3 or 4) the mixed density-field correlation functions

A= J ^ ^ ( / U Q , T ) S p ,

(12.8)

the generalization of (5.4). One finds that the irreducible parts are given by (see (5.11a)) lav(Q, iojn) = AflAI(Q, icon)G$(Q,

Uon) ,

(12.9)

where G$ is the matrix 4 He single-particle Green's function which only contains proper self-energies £j$ and the summation convention is used for repeated Greek subscripts. Thus in mixtures, we have two Bose symmetry-breaking vertex functions A4/l and A3^, which play a key role in that they determine the hybridization of the poles of Gj$ and XabThe irreducible contribution to the response function separates into two parts, lab = tab + fab >

(12-10)

with (see (5.11b)) ^K

(12.11)

12.2 Dielectric formalism for superfluid 3He-4He mixtures

281

The improper 4 He self-energy is given by ^

^ v 4

+ Av3) ,

(12.12)

where (see (5.10)) eR = 1 - V(Q)tn •

(12.13)

Similarly, one can derive the equivalent of (3.47) or (5.24) for the response functions Xab in mixtures, lab = \aG($Kb

+ XRab,

(12.14)

where A ^ = A^a/^R and eR is defined in (12.13). This is the key result of the dielectric formalism for mixtures since it shows how the spectrum of the 4 He single-particle Green's function G$ is coupled into the numberdensity-response functions Xab- The zeros of e(Q,co) in (12.6) determine the poles of both Xab and Gjfj. In contrast, the spectrum of the 3 He single-particle Green's function G(3)(Q,co) is not determined by the zeros of e(Q, co). This difference can already be seen by the fact that G^ and Xab involve Bose Matsubara frequencies while G(3) involve Fermi Matsubara frequencies. Talbot and Griffin (1984c) have derived the equivalent of the Ward identities given in Section 5.1 and used them to obtain various exact results in the zero-frequency limit. As one example, one can show that the vertex function A3^(Q, a> = 0) = 0. At the present time, this formalism is mainly useful in showing how the Bose broken symmetry leads to a certain inevitable coupling between the poles of Xab and G$ in mixtures. The equivalent of the analysis given in Section 7.2 for pure superfluid 4 He requires further high-resolution neutron-scattering data as a function of temperature, especially in the region below and above the superfluid transition. This seems like one of the most promising areas of research in the near future. Most phenomenological studies of the zeros of e(Q,a>) defined in (12.6) approximate ^33 as the Lindhard function of a non-interacting gas of Fermi quasiparticles and take %44 in a single-mode approximation appropriate to low temperatures (see, for example, Bartley et al, 1973; Hsu, Pines and Aldrich, 1985; Szprynger and Lucke, 1985). One immediately sees that the resulting response functions Xab will exhibit complicated hybridization effects between the Fermi quasiparticle p-h spectrum and the Boson pole of £44. In the polarization approach, for example, £44 is modelled by an expression similar to (9.28). In neutron-scattering studies,

282

The new picture: some unsolved problems

we need to know the 3 He Lindhard function at relatively large values of Q (say 1-1.5 A" 1 ). The original Landau-Pomeranchuk (LP) spectrum for a single 3 He atom in bulk liquid 4 He is Q2/2m*3, with an effective mass m\ ~ 2.4ra3. With this LP quadratic spectrum, the quasiparticle particle-hole spectrum is a band described by Q2/2m\ + Qvp. For a dilute solution, this is a very narrow band centred on the LP 3 He quasiparticle branch (at X3 = 0.06, we have kF ~ 0.3 A" 1 ). At such low concentrations, there is no collective zero sound mode associated with the 3 He atoms (in contrast to pure liquid 3 He, shown in Figs. 7.5, 7.6 and 7.18). Thus if we use the LP single-particle spectrum, the only 3 He density fluctuation branch in dilute mixtures is a narrow band which would cross the maxon-roton 4 He excitation at around Q ~ 1.7 A" 1 and co ~ 10 K. This would be expected to produce strong hybridization effects between these two branches in this cross-over region and a strong-level repulsion of the p-h branch (Bartley et al, 1973). This cross-over region has been extensively discussed in the literature (Hsu et a/., 1985; Szprynger and Lucke, 1985) with the conclusion that it is crucial to include the lower density of 4 He atoms in mixtures and also that the LP single-particle spectrum must be modified at the large values of Q probed in the neutron-scattering experiments. In a formal sense, these hybridization effects are analogous to the coupling between the single-particle maxon-roton spectrum and the 4 He p-h branch which we considered in the case of pure 4 He (see Section 7.2). A natural extension of the model calculations given at the end of Section 7.2 (see (7.23)) would be to add in the contribution X33 to describe the p-h spectrum of the 3 He atoms. The neutron-scattering line shapes from superfluid mixtures are clearly a rich area for future studies, especially at higher temperatures and concentrations where the collective dynamics of both 3 He and 4 He are important. The dielectric formalism should give a rigorous basis for understanding and describing the various hybridization effects which occur. 12.3 Suggestions for future research In this concluding section, we pull together a few suggestions as to where further work would be especially useful. Many of these suggestions have already been made in the earlier chapters. Our selection of topics for further research is influenced by what we believe is the most interesting feature of superfluid 4 He and 3 He- 4 He mixtures, namely the unique

12.3 Suggestions for future research

283

dynamical structure of various correlation functions which arises because of the Bose condensate.

Theoretical (1) Develop parameterized forms for <S(Q, co), as a function of temperature, which are consistent with the general structure induced by the Bose condensate. The preliminary analysis given at the end of Section 7.2 must be extended to include the two-excitation spectra before one can successfully deal with the very interesting maxon region Q ~ 1 A" 1 . (2) Work out the finite-temperature version of the Zawadowski, Ruvalds and Solana (1972) analysis of the condensate-induced hybridization between the one-roton and two-roton branches (Section 10.2). (3) The polarization potential approach (see Section 9.2) seems to give a useful way of parameterizing the general structure of correlation functions as predicted by the dielectric formalism. The PP approach should be extended to finite temperatures (> 1 K) by including the thermally excited particle-hole excitations. The high-energy multiparticle excitations have been discussed (at T = 0) by Hess and Pines (1988). (4) Formulate the Glyde-Griffin scenario at T = 0 in terms of a variational many-body wavefunction (see Section 9.1). The recent work of Stringari (1992) seems promising in this connection. (5) Make use of the first frequency-moment sum rules specific to Bosecondensed fluids (see Section 8.1 and 8.3). Extension of these to third frequency-moment sum rules would be of interest. (6) Work out the energy of single-particle excitations in a Bose liquid from first principles (Feynman, 1953b; Stringari, 1992). (7) The interaction V(Q) which appears in the final formulas of the dielectric formalism is assumed to be renormalized to some appropriate t-matrix when one includes multiple scattering. It would be useful to carry out this procedure more explicitly, summing up all contributions to regular quantities which involve two isolated propagator lines (see Section 5.4). (8) The one-loop approximation to the dielectric formalism (see Section 5.3) should be worked out in detail for superfluid 3 He- 4 He mixtures discussed in Section 12.2.

284

The new picture: some unsolved problems Experimental

(1) Measure the temperature dependence of the S(Q,co) line shape at high energies in the cross-over region around Q ~ 2.4 A" 1 (see Figs. 7.19 and 7.20). This should be done at a series of temperatures above and below 7^. (2) Measure the temperature dependence and dispersion of the highenergy peak which shows up in the low-Q neutron data (see Figs. 7.1 and 10.4). Is this a two-roton bound state (as usually assumed) or is it a remnant of a single-particle excitation, as suggested in Figs. 7.22 and 7.23? (3) High-resolution experiments on the temperature dependence of the S(Q,co) line shape at a series of different pressures would be very useful, especially in the intermediate maxon region. As shown very dramatically by Fig. 7.8, at a pressure of 20 bar, the normal distribution is peaked at a much higher energy than the maxon peak. As a result, S(Q,co) under high pressure exhibits the Glyde-Griffin scenario in a clearer fashion than the SVP data (where the superfluid and normal distributions are peaked at very similar energies). (4) Measure the pressure and temperature dependence of the neutronscattering line shapes in 3 He- 4 He mixtures in the vicinity of the tricritical point (i.e., at high concentrations and temperatures). (5) Develop new experimental probes of the existence and properties of rotons above the superfluid transition temperature. In the GlydeGriffin picture, rotons (more generally, maxon-rotons) are intrinsic single-particle excitations of normal liquid 4 He. The rotons develop finite weight in 5(Q,co) only as a result of the Bose broken symmetry. Thus neutron scattering is not a useful probe of rotons above T^. Quantum evaporation studies would seem one way of studying singleparticle excitations above and below Tx (see Caroli et al, 1976; Maris, 1992).

References

Abrikosov, A.A., Gor'kov, L.P. and Dzyaloshinskii, I.E. (1963), Methods of Quantum Field Theory in Statistical Physics (Prentice-Hall, Englewood Cliffs, N.J.). Ahlers, G. (1978), in Quantum Liquids, ed. by J. Ruvalds and T. Regge (North-Holland, New York), p. 1. Aldrich III, C.H., Pethick, C.G. and Pines, D. (1976), Phys. Rev. Lett. 37, 845. Aldrich III, C.H. and Pines, D. (1976), Journ. Low Temp. Phys. 25, 673. Aldrich III, C.H. and Pines, D. (1978), Journ. Low Temp. Phys. 32, 689. Allen, J.F. and Misener, A.D. (1938), Nature 141, 75. Ambegaokar, V., Conway, J. and Baym, G. (1965), in Lattice Dynamics, ed. by R.F. Wallis (Pergamon, New York), p. 261. Andersen, K.H., Stirling, W.G., Scherm, R., Stunault, A., Fak, B., Dianoux, A.J. and Godfrin, H. (1991), to be published. Anderson, P.W. (1966), Rev. Mod. Phys. 38, 298. Anderson, P.W. (1984), Basic Notions of Condensed Matter Physics (Benjamin/Cummings, Menlo Park, Calif.). Bartley, D.L., Robinson, J.E. and Wong, V.K. (1973), Journ. Low Temp. Phys. 12, 71. Bartley, D.L. and Wong, V.K. (1975), Phys. Rev. B12, 3775. Baym, G. (1969), in Mathematical Methods in Solid State and Superfluid Theory, ed. by R.C. Clark and G.H. Derrick (Oliver and Boyd, Edinburgh), p. 121. Baym, G., and Kadanoff, L.P. (1961), Phys. Rev. 124, 287. Baym, G. and Pethick, C.J., (1978), in The Physics of Liquid and Solid Helium, ed. by K.H. Bennemann and J.B. Ketterson (Wiley, New York), Vol. II. Bedell, K., Pines, D. and Fomin, I. (1982), Journ. Low Temp. Phys. 48, 417. Bedell, K., Pines, D. and Zawadowski, A. (1984), Phys. Rev. B29, 102. Beliaev, S.T. (1958a), Sov. Phys. - JETP 7, 289. Beliaev, S.T. (1958b), Sov. Phys. - JETP 7, 299. Bendt, P.J., Cowan, R.D. and Yarnell, J.H. (1959), Phys. Rev. 113, 1386. Bijl, A. (1940), Physica 7, 869. Bogoliubov, N.N. (1947), J. Phys. U.S.S.R. 11, 23. Bogoliubov, N.N. (1963, 1970), Lectures on Quantum Statistics (Gordon and Breach, New York), Vol. 2. Boon, J.P. and Yip, S. (1980), Molecular Hydrodynamics (McGraw-Hill, New York). Brown, G.V. and Coopersmith, M.H. (1969), Phys. Rev. 178, 327. 285

286

References

Brueckner, K.A. and Sawada, K. (1957), Phys. Rev. 106, 1117. Buyers, W.J.L., Sears, V.F., Lonngi, P.A. and Lonngi, D.A. (1975), Phys. Rev. All, 697. Campbell, C.E. (1978), in Progress in Liquid Physics, ed. by C.A. Croxton (Wiley, New York), p. 213. Campbell, C.E. and Clements, B.E. (1989), in Elementary Excitations in Quantum Fluids, ed. by K. Ohbayashi and M. Watabe (Springer-Verlag, Berlin), p. 8. Caroli, C, Roulet, B. and Saint-James, D. (1976), Phys. Rev. B13, 3875. Carraro, C. and Koonin, S.E. (1990), Phys. Rev. Lett. 65, 2792. Carraro, C. and Rinat, A.S. (1992), Phys. Rev. B45, 2945. Ceperley, D.M. and Pollock, E.L. (1986), Phys. Rev. Lett. 56, 351. Ceperley, D.M. and Pollock, E.L. (1987), Can. J. Phys. 65, 1416. Chester, G.V. (1963), in Liquid Helium, ed. by G. Careri (Academic Press, New York), p. 51. Chester, G.V. (1969), in Lectures in Theoretical Physics (Vol. XI-B): Quantum Fluids and Nuclear Matter, ed. by K.T. Mahanthappa and W.E. Britten (Gordon and Breach, New York), p. 253. Chester, G.V. (1975), in The Helium Liquids, ed. by J.G.M. Armitage and I.E. Farquhar (Academic Press, London), p. 1. Cheung, T.H. (1971), Ph.D. Thesis, University of Toronto, unpublished. Cheung, T.H. and Griffin, A. (1970), Can. Journ. Phys. 48, 2135. Cheung, T.H. and Griffin, A. (1971a), Phys. Lett. A35, 141. Cheung, T.H. and Griffin, A. (1971b), Phys. Rev. A4, 237. Clark, J.W. and Ristig, M.L. (1989), in Momentum Distributions, ed. by R.N. Silver and PE. Sokol (Plenum, New York), p. 39. Cohen, M. and Feynman, R.P. (1957), Phys. Rev. 107, 13. Copley, J.R.D. and Lovesey, S.W. (1975), Rep. Prog. Phys. 38, 461. Cowley, R.A. (1968), Rep. Prog. Phys. 31, 123. Cowley, R.A. and Buyers, W.J.L. (1969), J. Phys. C: Solid State Phys. 2, 2262. Cowley, R.A. and Woods, A.D.B. (1968), Phys. Rev. Lett. 21, 787. Cowley, R.A. and Woods, A.D.B. (1971), Can. J. Phys. 49, 177. Cummings, F.W., Hyland, G.J. and Rowlands, G. (1970), Phys. Condensed Matter 12, 90. Dietrich, O.W, Graf, E.H., Huang, C.G. and Passell, L. (1972), Phys. Rev. A5, 1377. Donnelly, R.J. (1991), Quantized Vortices in Helium II (Cambridge University Press, New York). Dzugutov, M. and Dahlborg, U. (1989), Phys. Rev. 40A, 4103. Einstein, A. (1925), Sitz. Berlin Preuss. Akad. Wiss. , p. 3. Fak, B. and Andersen, K.H. (1991), Phys. Lett. A160, 468. Fak, B , Guckelsberger, K., Korfer, M., Scherm, R. and Dianoux, A.J. (1990), Phys. Rev. B41, 8732. Family, F. (1975), Phys. Rev. Lett. 34, 1374. Feenberg, E. (1969), Theory of Quantum Fluids (Academic Press, New York). Ferrell, R.A., Menyhard, N , Schmidt, H., Schwabl, F. and Szepfalusy, P. (1968), Ann. Phys. (New York) 47, 565. Fetter, A.L. (1970), Ann. Phys. (New York) 60, 464. Fetter, A.L. (1972), Journ. Low Temp. Phys. 6, 487. Fetter, A.L. and Walecka, J.D. (1971), Quantum Theory of Many Particle Systems (McGraw-Hill, New York). Feynman, R.P (1953a), Phys. Rev. 91, 1291.

References

287

Feynman, R.R (1953b), Phys. Rev. 91, 1301. Feynman, R.R (1954), Phys. Rev. 94, 262. Feynman, R.R and Cohen, M. (1956), Phys. Rev. 102, 1189. Forster, D. (1975), Hydrodynamic Fluctuations, Broken Symmetry and Correlation Functions (W.A. Benjamin, Reading, Mass.). Fredkin, D.R. and Werthamer, N.R. (1965), Phys. Rev. 138A, 1528. f*ckushima, K. and Iseki, F. (1988), Phys. Rev. B38, 4448. Gavoret, J. (1963), These de Doctorat es Sciences Physiques, Paris, Annales de Physique 8, 441. Gavoret, J. and Nozieres, P. (1964), Ann. Phys. (New York) 28, 349. Gay, C. and Griffin, A. (1985), Journ. Low Temp. Phys. 58, 479. Gersch, H.A. and Rodriguez, LJ. (1973), Phys. Rev. A8, 905. Gillis, N.S. and Werthamer, N.R. (1968), Phys. Rev. 167, 607. Gillis, N.S., Werthamer, N.R. and Koehler, T.R. (1968), Phys. Rev. 165, 951. Ginzburg, V.L. (1943), J. Exp. Theor. Phys. (U.S.S.R.) 13, 243. Giorgini, S., Pitaevskii, L. and Stringari, S. (1992), Phys. Rev. B46, 6374. Giorgini, S. and Stringari, S. (1990), Physica B 165-166, 511. Girardeau, M. and Arnowitt, R. (1959), Phys. Rev. 113, 755. Glyde, H.R. (1971), Can. Journ. Phys. 49, 761. Glyde, H.R. (1974), Can. Journ. Phys. 52, 2281. Glyde, H.R. (1976), in Rare Gas Solids, ed. by M.L. Klein and J.A. Venables (Academic Press, New York), vol. 1, p. 382. Glyde, H.R. (1984), in Condensed Matter Research Using Neutrons, ed. by S.W. Lovesey and R. Scherm (Plenum, New York) p. 95. Glyde, H.R. (1992a), Phys. Rev. B45, 7321. Glyde, H.R. (1992b), to be published. Glyde, H.R. and Goldman, V.V. (1976), Journ. Low Temp. Phys. 25, 601. Glyde, H.R. and Griffin, A. (1990), Phys. Rev. Lett. 65, 1454. Glyde, H.R. and Svensson, E.C. (1987), in Methods of Experimental Physics, ed. by D.L. Price and K. Skold (Academic Press, New York), vol. 23, ch. 13 . Gotze, W. and Lucke, M. (1976), Phys. Rev. B13, 3825. Greytak, T.J. (1978), in Quantum Liquids, ed. by J. Ruvalds and T. Regge (North-Holland, Amsterdam), p. 121. Greytak, T.J. and Kleppner, D. (1984), in New Trends in Atomic Physics, ed. by G. Grynberg and R. Stora (North-Holland, New York), Vol. II, p. 1127. Greytak, T.J. and Yan, J. (1969), Phys. Rev. Lett. 22, 987. Greytak, T.J. and Yan, J. (1971), Proc. of the 12th International Conference on Low Temperature Physics, Kyoto, 1970, ed. by E. Kanda (Keigaku Publ. Comp., Tokyo), p. 89. Griffin, A. (1972), Phys. Rev. A6, 512. Griffin, A. (1979a), Phys. Rev. B19, 5946. Griffin, A. (1979b), Phys. Lett. 71A, 237. Griffin, A. (1980), Phys. Rev. B22, 5193. Griffin, A. (1981), Journ. Low Temp. Phys. 44, 441. Griffin, A. (1984), Phys. Rev. B30, 5057. Griffin, A. (1985), Phys. Rev. B32, 3289. Griffin, A. (1987), Can. Journ. Phys. 65, 1368. Griffin, A. (1988), Physica C156, 12. Griffin, A. (1989), in Elementary Excitations in Quantum Fluids, ed. by K. Ohbayashi and M. Watabe (Springer-Verlag, Berlin), p. 23.

288

References

Griffin, A. (1991), in Excitations in 2D and 3D Quantum Fluids, ed. by A.F.G. Wyatt and H J . Lauter (Plenum, New York), p. 15. Griffin, A. and Cheung, T.H. (1973), Phys. Rev. A7, 2086. Griffin, A. and Payne, S. (1986), Journ. Low Temp. Phys. 64, 155. Griffin, A. and Svensson, E.C. (1990), Physica B165-166, 487. Griffin, A. and Talbot, E. (1981), Phys. Rev. B24, 5075. Griffin, A., Wu, W.C. and N. Lambert (1992), to be published. Halley, J.W. (1969), Phys. Rev. 181, 338. Halley, J.W. (1989), in Elementary Excitations in Quantum Fluids, ed. by K. Ohbayashi and M. Watabe (Springer-Verlag, Berlin), p. 106. Halley, J.W. and Korth, M.S. (1991), in Excitations in 2D and 3D Quantum Fluids, ed. by A.F.G. Wyatt and H.J. Lauter (Plenum, New York), p. 91. Hansen, J.P. and McDonald, I.R. (1986), Theory of Simple Liquids, Second Ed. (Academic Press, London). Harling, O.K. (1970), Phys. Rev. Lett. 24, 1046. Harling, O.K. (1971), Phys. Rev. A3, 1073. Henshaw, D.G. and Woods, A.D.B. (1961), Phys. Rev. 121, 1266. Herwig, K.W, Sokol, P.E., Snow, WM. and Blasdell, R.C. (1991), Phys. Rev. B44, 308. Herwig, K.W, Sokol, RE., Sosnick, T.R., Snow, WM. and Blasdell, R.C. (1990), Phys. Rev. B41, 103. Hess, D.W and Pines, D. (1988), Journ. Low Temp. Phys. 72, 247. Hilton, P.A., Cowley, R.A., Scherm, R. and Stirling, WG. (1980), J. Phys. C: Solid State Phys. 13, L295. Hohenberg, PC. (1967), Phys. Rev. 158, 383. Hohenberg, PC. (1973), Journ. Low Temp. Phys. 11, 745. Hohenberg, PC. and Martin, PC. (1964), Phys. Rev. Lett. 12, 69. Hohenberg, PC. and Martin, PC. (1965), Ann. Phys. (New York) 34, 291. Hohenberg, PC. and Platzman, P.M. (1966), Phys. Rev. 152, 198. Horner, H. (1967), Z. Physik 205, 72. Horner, H. (1971), Z. Physik 242, 432. Horner, H. (1972a), Phys. Rev. Lett. 29, 556. Horner, H. (1972b), Journ. Low Temp. Phys. 8, 511. Horner, H. (1974a), in Proceedings of the 13th International Conference on Low Temperature Physics, ed. by K.D.T. Timmerhaus, WJ. O'Sullivan and E.F. Hammel (Plenum Press, New York), vol. 1, p. 3. Horner, H. (1974b), in Proceedings of the 13th International Conference on Low Temperature Physics, ed. by K.D.T. Timmerhaus, WJ. O'Sullivan and E.F. Hammel (Plenum Press, New York), vol. 1, p. 125. Horner, H. (1974c), in Dynamical Properties of Solids, ed. by G.K. Horton and A.A. Maradudin (North-Holland, Amsterdam), vol. I, p. 451. Hsu, W, Pines, D. and Aldrich III, C.H. (1985), Phys. Rev. B32, 7179. Huang, K. (1964), in Studies in Statistical Mechanics, ed. by J. De Boer and G.E. Uhlenbeck (North-Holland, Amsterdam), vol. II. Huang, K. (1987), Statistical Mechanics, Second Edition (John Wiley, New York). Huang, K. and Klein, A. (1964), Ann. Phys. (New York) 30, 203. Hugenholtz, N. and Pines, D. (1959), Phys. Rev. 116, 489. Inkson, J.C. (1984), Many-Body Theory of Solids (Plenum, New York). Iwamoto, F. (1970), Prog. Theoret. Phys. (Kyoto) 44, 1135.

References

289

Iwamoto, F. (1989), in Elementary Excitations in Quantum Fluids, ed. by K. Ohbayashi and M. Watabe (Springer-Verlag, Berlin), p. 117. Jackson, H.W. (1969), Phys. Rev. 185, 186. Jackson, H.W. (1973), Phys. Rev. A8, 1529. Jackson, H.W. (1974), Phys. Rev. A10, 278. Juge, K.J. and Griffin, A. (1993), to be published. Kadanoff, L.P. and Baym, G. (1962), Quantum Statistical Mechanics (Benjamin, New York). Kadanoff, L.P. and Martin, P.C. (1963), Ann. Phys. (New York) 24, 419. Kalos, M.H., Lee, M.A., Whitlock, PA. and Chester, G.V. (1981), Phys. Rev. B24, 115. Kapitza, P. (1938), Nature 141, 74. Khalatnikov, I.M. (1965), An Introduction to the Theory of Superfluidity (Benjamin, New York). Kirkpatrick, T.R. (1984), Phys. Rev. B30, 1266. Kirkpatrick, T.R. and Dorfman, J.R. (1985), Journ. Low Temp. Phys. 58, 399, 301; 59, 1. Kitchens, T.A., Shirane, G., Minkiewicz, VJ. and Osgood, E.B. (1972), Phys. Rev. Lett. 29, 552. Koehler, T.R. (1967), Phys. Rev. Lett. 18, 654. Koehler, T.R. (1968), Phys. Rev. 17, 942. Koehler, T.R. and Werthamer, N.R. (1971), Phys. Rev. A3, 2074. Koehler, T.R. and Werthamer, N.R. (1972), Phys. Rev. A5, 2230. Kondor, I. and Szepfalusy, P. (1968), Ada Phys. Hung. 24, 81. Kwok, P.C. (1967), in Solid State Physics, ed. by F. Seitz, D. Turnbull and H. Ehrenreich (Academic Press, New York), vol. 20, p. 306. Kwok, P.C. and Martin, P.C. (1966), Phys. Rev. 142, 495. Landau, L.D. (1941), J. Phys. U.S.S.R. 5, 71. Landau, L.D. (1947), J. Phys. U.S.S.R. 11, 91. Landau, L.D. (1949), Phys. Rev. 75, 884. Landau, L.D. and Khalatnikov, I.M. (1949), Zh. Eksp. Teor. Fiz. 19, 637. Landau, L.D. and Lifsh*tz, E.M. (1959), Statistical Physics (Pergamon, Oxford). Lee, T.D., Huang, K. and Yang, C.N. (1957), Phys. Rev. 106, 1135. Leggett, A.J. (1975), Rev. Mod. Phys. 47, 331. Lifsh*tz, E.M. and Pitaevskii, L.P. (1980), Statistical Physics, Part 2, (Pergamon, Oxford). London, F. (1938a), Nature 141, 643. London, F. (1938b), Phys. Rev. 54, 947. London, F. (1950), Superfluids, Vol. 1: Macroscopic Theory of Superconductivity (Wiley, New York). London, F. (1954), Superfluids, Vol. 2: Macroscopic Theory of Superfluid Helium (Wiley, New York). Lucke, M. (1980), in Modern Trends in the Theory of Condensed Matter, Proc. of the 16th Karpacz Winter School of Theoretical Physics, ed. by A. Pekalski and J. Przystawa (Springer-Verlag, Berlin), p. 74. Lucke, M. and Szprynger, A. (1982), Phys. Rev. B26, 1374. Ma, S.K. (1971), Journ. Math. Phys. 12, 2157. Ma, S.K., Gould, H. and Wong, V.K. (1971), Phys. Rev. A3, 1453. Ma, S.K. and Woo, C.W (1967), Phys. Rev. 159, 165. Mahan, G. (1990), Many-Particle Physics, Second Edition (Plenum, New York).

290

References

Manousakis, E. (1989), in Momentum Distributions, ed. by R.N. Silver and RE. Sokol (Plenum, New York) p. 81. Manousakis, E. and Pandharipande, V.R. (1984), Phys. Rev. B30, 5064. Manousakis, E. and Pandharipande, V.R. (1986), Phys. Rev. B33, 150. Manousakis, E., Pandharipande, V.R. and Usmani, Q.N. (1985), Phys. Rev. B33, 7022. Maradudin, A.A. and Fein, A.E. (1962), Phys. Rev. 128, 2589. Maris, H.J. (1977), Rev. Mod. Phys. 49, 341. Maris, H.J. (1992), Journ. Low Temp. Phys. 87, 773. Maris, H.J. and Massey, W.E. (1970), Phys. Rev. Lett. 25, 220. Marshall, W. and Lovesey, S.W. (1971), Theory of Thermal Neutron Scattering (Oxford University Press, New York). Martel, P., Svensson, E.C., Woods, A.D.B., Sears, V.F. and Cowley, R.A. (1976), Journ. Low Temp. Phys. 32, 285. Martin, PC. (1965), in Low Temperature Physics - LT9, ed. by J.G. Daunt, D.O. Edwards, F.J. Milford and M. Yaqub (Plenum, New York), p. 9. Martin, PC. (1968), in Many-Body Physics, ed. by C. De Witt and R. Balian (Gordon and Breach, New York), p. 37. McMahan, A.K. and Beck, H. (1973), Phys. Rev. A8, 3247. McMillan, WL. (1965), Phys. Rev. 138, A442. Meisel, M.W. (1992), Physica B178, 121. Mezei, F. and Stirling, W.G. (1983), in 75th Jubilee Conference on Helium-4, ed. by J.G.M. Armitage (World Scientific, Singapore), p. 111. Miller, A., Pines, D. and Nozieres, P. (1962), Phys. Rev. 127, 1452. Mineev, V.R (1980), Sov. Phys. - JETP Lett. 32, 489. Minkiewicz, V.J., Kitchens, T.A., Shirane, G. and Osgood, E.B. (1973), Phys. Rev. A8, 1513. Mook, H.A. (1974), Phys. Rev. Lett. 32, 1167. Mook, H.A. (1983), Phys. Rev. Lett. 51, 1454. Mook, H.A., Scherm, R. and Wilkinson, M.K. (1972), Phys. Rev. A6, 2268. Murray, R.L., Woerner, R.L. and Greytak, T.J. (1975), J. Phys. C: Sol. State Phys. C8, L90. Nakajima, S. (1971), Prog. Theoret. Phys. (Kyoto) 45, 353. Nelson, D.R. (1983), in Phase Transitions and Critical Phenomena, ed. by C. Domb and J.L. Lebowitz (Academic Press, London), Vol. 7, p. 1. Nepomnyashchii, YA. (1983), Sov. Phys. - JETP 58, 722. Nepomnyashchy, YA. (1992), Phys. Rev. B46, 6611. Nepomnyashchii, YA. and Nepomnyashchii, A.A. (1974), Sov. Phys. - JETP 38, 134. Nepomnyashchii, YA. and Nepomnyashchii, A.A. (1978), Sov. Phys. - JETP 48, 493. Nozieres, P. (1964), Theory of Interacting Fermi Systems (Benjamin, New York). Nozieres, P. (1966), in Quantum Fluids, ed. by D.F. Brewer (North-Holland, Amsterdam), p. 1. Nozieres, P. (1983), Quantum Liquids and Solids, lecture notes, College de France. Nozieres, P. and Pines, D. (1964, 1990), Theory of Quantum Liquids, Vol. II: Superfluid Bose Liquids (Addison-Wesley, Redwood City, Calif.). O'Connor, J.T., Palin, C.J. and Vinen, WF. (1975), J. Phys. C: Solid State Phys. 8, 101. Ohbayashi, K. (1989), in Elementary Excitations in Quantum Fluids, ed. by K. Ohbayashi and M. Watabe (Springer-Verlag, Berlin), p. 32.

References

291

Ohbayashi, K. (1991), in Excitations in 2D and 3D Quantum Fluids, ed. by A.F.G. Wyatt and H.J.L. Lauter (Plenum, New York), p. 77. Ohbayashi, K., Udagawa, K., Yamash*ta, H., Watabe, M. and Ogita, N. (1990), Physica B165-166, 485. Osgood, E.B., Kitchens, T.A., Shirane, G. and Minkiewicz, V.J. (1972), Phys. Rev. A5, 1537. Palevsky, H., Otnes, K., Larsson, K.E., Pauli, R. and Stedman, R. (1957), Phys. Rev. 108, 1346. Parry, W.E. and ter Haar, D. (1962), Ann. Phys. (New York) 19, 496. Payne, S.H. and Griffin, A. (1985), Phys. Rev. B32, 7199. Penrose, O. (1951), Phil. Mag. 42, 1373. Penrose, O. (1958), in Proceedings of the International Conference on Low Temperature Physics, ed. by J.R. Dillinger (University of Wisconsin Press, Madison), p. 117. Penrose, O. and Onsager, L. (1956), Phys. Rev. 104, 576. Pike, E.R. (1972), Journ. de Physique 33, Cl, 25. Pike, E.R., Vaughan, J.M. and Vinen, W.F. (1970), J. Phys. C: Sol. State Phys. 3, L40. Pines, D. (1963), in Liquid Helium, ed. by G. Careri (Academic Press, New York), p. 147. Pines, D. (1965), in Low Temperature Physics - LT9, ed. by J.G. Daunt, D.O. Edwards, F.J. Milford and M. Yaqub (Plenum, New York), p. 61. Pines, D. (1966), in Quantum Fluids, ed. by D.F. Brewer (North-Holland, Amsterdam), p. 257. Pines, D. (1985), in Highlights of Condensed Matter Theory, ed. by F. Bassani, F. Fumi and M.P Tosi (North-Holland, New York), p. 580. Pines, D. (1987), Can. Journ. Phys. 65, 1357. Pines, D. (1989), Physics Today 42, 61. Pines, D. and Nozieres, P. (1966), Theory of Quantum Liquids, Vol. I: Normal Fermi Liquids (Addison-Wesley, Redwood City, Calif.). Pitaevskii, L.P. (1959), Sov. Phys. - JETP 9, 830. Pitaevskii, L.P. (1987), Sov. Phys. - JETP Lett. 45, 185. Pitaevskii, L.P. and Fomin, LA. (1974), Sov. Phys. - JETP 38, 1257. Pitaevskii, L.P. and Stringari, S. (1991), Journ. Low Temp. Phys. 85, 377. Placzek, G. (1952), Phys. Rev. 86, 337. Platzman, P. and Tzoar, N. (1965), Phys. Rev. 139, A410. Pollock, E.L. and Ceperley, D.M. (1987), Phys. Rev. B36, 8343. Popov, V.N. (1965), Sov. Phys. - JETP 20, 1185. Popov, V.N. (1983), Functional Integrals in Quantum Field Theory and Statistical Physics (Reidel, Dordrecht). Popov, V.N. (1987), Functional Integrals and Collective Excitations (Cambridge University Press). Popov, V.N. and Serendniakov, A.V. (1979), Sov. Phys. - JETP 50, 193. Puff, R.D. (1965), Phys. Rev. 137, 17406. Puff, R.D. and Tenn, J.S. (1970), Phys. Rev. Al, 125. Rahman, A., Singwi, K.S. and Sjolander, A. (1962), Phys. Rev. 126, 986. Reatto, L. and Chester, G.V (1967), Phys. Rev. 155, 88. Rickayzen, G. (1980), Green's Functions and Condensed Matter (Academic Press, New York). Rinat, A.S. (1989), Phys. Rev. B41, 4247.

292

References

Rockwell, D.A., Benjamin, R.F. and Greytak, TJ. (1975), Journ. Low Temp. Phys. 18, 389. Ruvalds, J. (1978), in Quantum Liquids, ed. by J. Ruvalds and T. Regge (North-Holland, New York), p. 263. Ruvalds, J. and Zawadowski, A. (1970), Phys. Rev. Lett. 25, 233. Scherm, R., Dianoux, A.J., Fak, B., Guckelsberger, K., Korfer, M. and Stirling, W.G. (1989), Physica B156 & 157, 311. Scherm, R., Guckelsberger, K., Fak, B., Skold, K., Dianoux, A.J., Godfrin, H. and Stirling, W.G. (1987), Phys. Rev. Lett. 59, 217. Schrieffer, J.R. (1964), Theory of Superconductivity (W.A. Benjamin, Reading, Mass.). Sears, V.F. (1969), Phys. Rev. 185, 200. Sears, V.F. (1981), Can J. Phys. 59, 555. Sears, V.F. (1985), Can J. Phys. 63, 68. Sears, V.F., Svensson, E.C., Martel, P. and Woods, A.D.B. (1982), Phys. Rev. Lett. 49, 279. Senger, G., Ristig, M.L., Campbell, C.E. and Clark, J.W. (1992), Ann. Phys. (N.Y.) 218, 160. Silver, R.N. (1988), Phys. Rev. B37, 3794. Silver, R.N. (1989), Phys. Rev. B38, 2283. Silver, R.N. and Sokol, P.E., eds. (1989), Momentum Distributions (Plenum, New York). Skold, K. and Pelizzari, C.A. (1980), Phil. Trans, of the Roy. Soc. of London B290, 605. Skold, K., Pelizzari, C.A., Kleb, R. and Ostrowski, G.E. (1976), Phys. Rev. Letters 37, 842. Smith, A.J., Cowley, R.A., Woods, A.D.B., Stirling, W.G. and Martel, P. (1977), J. Phys. C: Solid State Phys. 10, 543. Snow, WM. (1990), Ph.D. thesis, Harvard University, unpublished. Snow, WM., Wang, Y. and Sokol, P.E. (1992), Europhys. Lett. 19, 403. Sokol, P.E. (1987), Can. J. Phys. 65, 1393. Sokol, P.E. and Snow, WM. (1991), in Excitations in 2D and 3D Quantum Fluids, ed. by A.F.G. Wyatt and H.J. Lauter (Plenum, New York), p. 47. Sokol, P.E., Sosnick, T.R. and Snow, WM. (1989), in Momentum Distributions, ed. by R.N. Silver and P.E. Sokol (Plenum, New York), p. 139. Sosnick, T.R., Snow, WM., Silver, R.N. and Sokol, P.E. (1991), Phys. Rev. B43, 216. Sosnick, T.R., Snow, WM. and Sokol, P.E. (1990), Phys. Rev. B41, 11185. Stephen, M.J. (1969), Phys. Rev. 187, 279. Stephen, M.J. (1976), in The Physics of Liquid and Solid Helium, ed. by K.H. Benneman and J.B. Ketterson (Wiley, New York) Vol. 1, p. 307. Stirling, W.G. (1983), in 75th Jubilee Conference on Helium-4, ed. J.G.M. Armitage (World Scientific, Singapore), p. 109. Stirling, W.G. (1985), in Proc. 2nd International Conference on Phonon Physics, ed. by J. Kollar, N. Kroo, N. Menyhard, and T. Siklos (World Scientific, Singapore), p. 829. Stirling, W.G. (1991), in Excitations in 2D and 3D Quantum Fluids, ed. by A.F.G. Wyatt and H.J. Lauter (Plenum, New York), p. 25. Stirling, W.G. and Glyde, H.R. (1990), Phys. Rev. B41, 4224. Stirling, W.G., Talbot, E.F., Tanatar, B. and Glyde, H.R. (1988), Journ. Low Temp. Phys. 73, 33.

References

293

Straley, J.R (1972), Phys. Rev. A5, 338. Stringari, S. (1991), Journ. Low Temp. Phys. 84, 279. Stringari, S. (1992), Phys. Rev. B46, 2974. Svensson, E.C. (1989), in Elementary Excitations in Quantum Fluids, ed. by K. Ohbayashi and M. Watabe (Springer-Verlag, Heidelberg), p. 59. Svensson, E.C. (1991), in Excitations in 2D and 3D Quantum Fluids, ed. by A.F.G. Wyatt and H.J. Lauter (Plenum, New York), p. 59. Svensson, E.C, Martel, P., Sears, V.F. and Woods, A.D.B. (1976), Can. J. Phys. 54, 2178. Svensson, E.C, Martel, P. and Woods, A.D.B. (1975), Phys. Lett. 55A, 151. Svensson, E.C, Sears, V.F., Woods, A.D.B. and Martel, P. (1980), Phys. Rev. B21, 3638. Svensson, E.C and Tennant, D.C (1987), Japan Journ. Applied Phys. 26, (Supp. 26-3), 31. Szepfalusy, P. (1965), Acta Phys. Hungarica 19, 109. Szepfalusy, P. and Kondor, I. (1974), Ann. Phys. (New York) 82, 1. Szprynger, A. and Lucke, M. (1985), Phys. Rev. B32, 4442. Talbot, E.F. (1983), Ph.D. Thesis, University of Toronto, unpublished. Talbot, E., Glyde, H.R., Stirling, WG. and Svensson, E.C. (1988), Phys. Rev. B38, 11229. Talbot, E. and Griffin, A. (1983), Ann. Phys. (New York) 151, 71. Talbot, E. and Griffin, A. (1984a), Phys. Rev. B29, 2531. Talbot, E. and Griffin, A. (1984b), Phys. Rev. B29, 3952. Talbot, E. and Griffin, A. (1984c), Journ. Low Temp. Phys. 56, 141. Tanatar, B , Talbot, E.F. and Glyde, H.R. (1987), Phys. Rev. B36, 8376. Tarvin, J.A, Vidal, F. and Greytak, T.J. (1977), Phys. Rev. B15, 4193. Tisza, L. (1938), Nature 141, 913. Tserkovnikov, Yu. A. (1965), Sov. Phys. - Doklady 9, 1095. van Hove, L. (1954), Phys. Rev. 95, 249. Vinen, W E (1969), in Superconductivity, ed. by R.D. Parks (M. Dekker, New York) Vol. II, p. 1167. Vinen, W.F. (1971), J. Phys. C: Solid State Physics 4, L287. Vollhardt, O. and Wolfle, P. (1990), The Superfluid Phases of Helium 3 (Taylor and Francis, London). Wagner, H. (1966), Z. Physik 195, 273. Wang, Y, Sosnick, T.R. and Sokol, RE. (1992), to be published. Weichman, RB. (1988), Phys. Rev. B38, 8739. Werthamer, N.R. (1969), Amer. Journ. Phys. 37, 373. Werthamer, N.R. (1972), Phys. Rev. Lett. 28, 1102. Werthamer, N.R. (1973), Phys. Rev. A7, 254. West, G.B. (1975), Phys. Rep. 18C, 263. Whitlock, RA. and Panoff, R.M. (1987), Can. J. Phys. 65, 1409. Wilks, J. (1967), The Properties of Liquid and Solid Helium (Clarendon, Oxford). Wong, VK. (1977), Phys. Lett. 61A, 455. Wong, V.K. (1979), Phys. Lett. 73A, 398. Wong, V.K. and Gould, H. (1974), Ann. Phys. (New York) 83, 252. Wong, V.K. and Gould, H. (1976), Phys. Rev. B14, 3961. Woo, C.W. (1976), in The Physics of Liquid and Solid Helium, ed. by J.B. Ketterson and K.H. Bennemann (Wiley, New York) Vol. I, p. 349. Woods, A.D.B. (1965a), in Inelastic Scattering of Neutrons (IAEA, Vienna), p. 191.

294

References

Woods, A.D.B. (1965b), Phys. Rev. Lett. 14, 355. Woods, A.D.B. and Cowley, R.A. (1973), Rep. Prog. Phys. 36, 1135. Woods, A.D.B., Hilton, P.A, Scherm, R. and Stirling, W.G. (1977), J. Phys. C: Solid State Physics 10, L45. Woods, A.D.B and Sears, V.F. (1977), Phys. Rev. Lett. 39, 415. Woods, A.D.B. and Svensson, E.C. (1978), Phys. Rev. Lett. 41, 974. Woods, A.D.B., Svensson, E.C. and Martel, P. (1972), in Neutron Inelastic Scattering (IAEA, Vienna), p. 359. Woods, A.D.B., Svensson, E.C. and Martel, P. (1975), in Low Temperature Physics - LT14, ed. by M. Krusius and M. Vuorio (North-Holland, Amsterdam), Vol. 1, p. 187. Woods, A.D.B., Svensson, E.C. and Martel, P. (1976), Phys. Lett. 57 A, 439. Woods, A.D.B., Svensson, E.C. and Martel, P. (1978), Can. J. Phys. 56, 302. Yang, C.N. (1962), Rev. Mod. Phys. 34, 694. Yarnell, J.L., Arnold, G.P., Bendt, P.J. and Kerr, E.C. (1959), Phys. Rev. 113, 1379. Yoshida, F. and Takeno, S. (1987), Prog. Theor. Phys. (Kyoto) 77, 864. Yoshida, F. and Takeno, S. (1989), Phys. Reports C173, 301. Zawadowski, A. (1978), in Quantum Liquids, ed. by J. Ruvalds and T. Regge (North-Holland, New York), p. 293. Zawadowski, A., Ruvalds, J. and Solana, J. (1972), Phys. Rev. A5, 399.

Author index

Abrikosov, A.A. 233, 241 Ahlers, G. 69 Aldrich III, C.H. 209, 218-24, 230, 278, 281, 282 Allen, J.F. 2 Ambegaokar, V. 24, 204, 257, 264 Andersen, K.H. 6, 12, 23, 38, 154, 157-60, 162-3, 166, 168-70, 173-6, 180, 185 Anderson, RW. 1, 17, 51, 53 Arnold, G.P. 5 Arnowitt, R. 60, 116

Conway, J. 24, 204, 257, 264 Coopersmith, M.H. 59 Copley, J.R.D. 162, 225 Cowan, R.D. 5 Cowley, R.A. 3, 7, 13, 33-5, 41, 45, 78, 134, 156-61, 175, 204, 216, 246-7, 260, 264 Cummings, F.W. 69

Bartley, D.L. 199, 280-2 Baym, G. 24, 60, 70, 128, 204, 235, 257, 264, 278 Beck, H. 263 Bedell, K. 180, 236-9, 242, 275 Beliaev, ST. 17, 22, 51, 53, 60, 116 Bendt, RJ. 5 Benjamin, R.F. 279 Bijl, A. 9, 209 Blasdell, R.C. 83, 85 Bogoliubov, N.N. 2, 8, 16-18, 49-53, 56, 72, 130, 134, 271 Boon, J.P. 228 Brown, G.V. 59 Brueckner, K.A. 59 Buyers, WJ.L. 162 Caroli, C. 284 Campbell, C.E. 10, 210, 214 Carraro, C. 45, 83 Ceperley, D.M. 15-18, 48, 69, 81-5, 88, 128, 276 Chester, G.V. 6, 17, 72, 210, 217, 2 7 3 ^ Cheung, T.H. 19, 60, 93, 100, 107, 116, 140, 233-4, 272 Clark, J.W. 74,210-11,218 Clements, B.E. 210, 214 Cohen, M. 5, 10, 116,208,211

Dahlborg, U. 162 Dianoux, AJ. 6, 12, 23, 38-9, 154, 157-62, 166-70, 173, 175, 185, 278 Dietrich, O.W. 168 Donnelly, R.J. 21 Dorfman, J.R. 133 Dzugutov, M. 162 Dzyaloshinskii, I.E. 234, 241 Einstein, A. 48 Fak, B. 6, 12, 23, 38-9, 154, 157-63, 166-70, 173-6, 180, 185, 278 Family, F. 199 Feenberg, E. 210, 212, 261 Fein, A.E. 260 Ferrell, R.A. 72 Fetter, A.L. 32, 47, 55, 58-9, 62, 73, 131, 196, 249 Feynman, R.P. 5-6, 8-11, 15, 20, 48, 50, 116,208-11,273-5,283 Fomin, I. 239, 275 Forster, D. 51, 134, 138 Fredkin, D.R. 262 f*ckushima, K. 243 Gavoret, J. 18, 22, 51, 54, 64, 71-2, 92-3, 101, 105, 116-30, 141, 147-8, 186, 208, 233, 271-2 Gay, C. 109, 111, 133, 136 Gersch, H.A. 43, 45, 78, 83 Gillis, N.S. 260-2

295

296

Author index

Ginzburg, V.L. 135 Giorgini, S. 72, 148, 199 Girardeau, M. 60, 116 Glyde, H.R. 11, 20-3, 33-8, 69, 79, 85, 88, 150, 153-62, 164, 168, 171, 175-7, 180, 183-94, 200-1, 244, 252, 257, 260, 265-6, 270-5, 277-8 Godfrin, H. 6, 12, 23, 38, 154, 157-8, 161-3, 169-70, 173 Goldman, V.V. 265 Gor'kov, L.P. 234, 241 Gotze, W. 225-8, 247 Gould, H. 19, 23, 66, 93-105, 110-16, 125, 151, 195-9, 217, 266 Graf, E.H. 168 Greytak, T.J. 109, 127, 136-8, 140, 231, 252, 256, 279 Griffin, A. 19-23, 60-4, 69-72, 88-9, 93, 100-2, 104, 107-16, 129-33, 136-40, 144, 146, 149-50, 153, 178, 180-4, 196-7, 200-7, 224, 233-4, 242, 248, 270-81 Guckelsberger, K. 160-1, 175, 278

Kleppner, D. 109 Koehler, T.R. 260-1, 268 Kondor, I. 19, 73, 93, 103, 107, 111, 150, 272 Koonin, S.E. 45, 83 Korfer, M. 161, 175, 278 Korth, M.S. 254 Kwok, RC. 260-3

Halley, J.W. 249, 254 Hansen, J.R 26, 162, 225, 228 Hading, O.K. 78 Henshaw, D.G. 5, 275 Herwig, K.W. 83, 85 Hess, D.W. 218, 221, 283 Hilton, RA. 5, 161, 256 Hohenberg, RC. 14, 18, 45, 50-1, 59-60, 64, 68, 72, 100, 110, 125, 128-30, 135-7, 140, 144, 148-50, 181, 195, 271 Homer, H. 257-68 Hsu, W. 218, 230, 278-82 Huang, K. 17, 50, 130 Huang, C.G. 168 Hugenholtz, N. 18, 51-3, 57-9, 62, 105, 272 Hyland, G.J. 69

Ma, S.K. 19-21, 51, 66, 93, 97, 110, 125, 133, 151, 272 Mahan, G. 33, 40, 54, 59, 63, 70, 139, 210, 275 Manousakis, E. 10, 74-5, 81-2, 210-15, 247, 254 Maris, H.J. 35, 156, 284 Marshall, W. 25, 27 Martel, P. 27, 33-5, 39, 43, 45, 70, 75, 78-9, 86-8, 134, 157, 170, 172, 175, 200, 243-7, 276 Martin, RC. 18-19, 50-1, 59-60, 64, 72, 100, 110, 125, 128-30, 134-6, 140, 144, 148-50, 262, 271 Massey, W.E. 156 McDonald, I.R. 26, 162, 225, 228 McMahan, A.K. 263 McMillan, W.L. 15, 197 Meisel, M.W. 257 Menyhard, N. 72 Mezei, F. 36, 154, 157, 159 Miller, A. 6, 10, 33, 67, 115-16, 178, 181, 211,216,272-4 Mineev, V.R 202 Minkiewicz, V.J. 259, 265 Misener, A.D. 2 Mook, H.A. 78, 88 Murray, R.L. 256

Iseki, F. 243 Iwamoto, F. 231, 249, 256 Jackson, H.W. 78, 214, 246 Juge, K.J. 248 Kadanoff, L.P. 60, 134, 235 Kalos, M.H. 17 Kapitza, P. 2 Kerr, E.C. 5 Khalatnikov, I.M. 3-4, 8, 22, 35, 127, 132-3, 136, 278-9 Kirkpatrick, T.R. 133, 199 Kitchens, T.A. 259, 265 Kleb, R. 172 Klein, A. 130

Lambert, N. 224 Landau, L.D. 2-8, 16, 35, 273 Larsson, K.E. 5 Lee, M.A. 17 Lee, T.D. 17 Leggett, A.J. 1 Lifsh*tz, E.M. 3, 8, 21, 28, 30, 49, 55, 72, 117-18, 138,241 London, F. 1, 4-7, 14-17, 48 Lonngi, D.A. 162 Lonngi, P.A. 162 Lovesey, S.W. 25-7, 162, 225 Lucke, M. 225-30, 247, 278, 281-2

Nakajima, S. 249 Nelson, D.R. 132 Nepomnyaschii, A.A. 127, 143-4, 147-8, 233

Author index Nepomnyashchii, Y.A. 127, 143-4, 147-9, 152, 233, 272 Nozieres, P. 1, 6, 10, 14, 18-23, 28, 32-3, 51-4, 64, 68, 71-2, 92-3, 100-1, 105, 115-30, 133, 141-2, 147-52, 178, 181, 186, 206, 208, 211, 216, 233, 239, 271-4 O'Connor, J.T. 137 Ogita, N. 251-4 Ohbayashi, K. 231, 251-5 Onsager, L. 12, 49-50 Osgood, E.B. 259, 265 Ostrowski, G.E. 172 Otnes, K. 5 Palevsky, H. 5 Palin, C.J. 137 Pandharipande, V.R. 10, 74, 81-2, 210-15, 247, 254 Panoff, R.M. 12, 75, 81-4 Parry, W.E. 59 Passell, L. 168 Pauli, R. 5 Payne, S.H. 107-11, 130, 187 Pelizzari, C.A. 172 Penrose, O. 12, 16, 49-50, 181, 210 Pethick, C.G. 218, 278 Pike, E.R. 134-5 Pines, D. 1, 6, 9, 14, 18-23, 28, 32-3, 51-3, 57-9, 62, 66-7, 105, 110, 115-6, 128, 133, 148, 161, 169, 178-81, 206, 209, 211, 216, 218-24, 226, 230, 236, 238, 242, 272-5, 278-83 Pitaevskii, L.P. 8, 19, 21, 24, 28, 30, 49, 56, 72-3, 104, 117-18, 148, 207, 231-3, 239, 241 Placzek, G. 25, 30 Platzman, P. 14, 45, 68, 77, 181 Pollock, E.L. 15-18, 48, 69, 81-5, 88, 128, 276 Popov, V.N. 108, 127, 132-3, 139, 145 Puff, R.D. 30, 78, 226 Rahman, A. 30-1, 41 Reatto, L. 72, 210 Rickayzen, G. 54, 58 Rinat, A.S. 45, 83 Ristig, M.L. 74, 210-11,218 Robinson, J.E. 280-2 Rockwell, D.A. 279 Rodriguez, L.J. 43, 45, 78, 83 Roulet, B. 284 Rowlands, G. 69 Ruvalds, J. 19, 24, 104, 180, 214, 229, 231-48, 278, 283 Saint-James, D. 284

297

Sawada, K. 59 Scherm, R. 5-6, 12, 23, 38, 78, 154, 157-63, 166-75, 185, 256, 278 Schmidt, H. 72 Schrieffer, J.R. 53 Schwabl, F. 72 Sears, V.F. 27, 33, 35, 39, 42-5, 68, 70, 75, 78-9, 86-8, 162, 175, 245, 276 Serendniakov, A.V. 127 Shirane, G. 259, 265 Silver, R.N. 43-5, 69, 78, 83 Singwi, K.S. 30-1, 41 Sjolander, A. 30-1, 41 Skold, K. 160-1, 172 Smith, A.J. 246-7 Snow, W.M. 14, 44, 78-91 Sokol, RE. 14, 22, 42-4, 69, 78-89, 277 Solana, J. 232, 238-9, 242, 248, 283 Sosnick, T.R. 44, 78-89, 277 Stedman, R. 5 Stephen, M.J. 127, 134, 137, 231, 249, 279 Stirling, W.G. 5-6, 11-12, 23, 34-9, 154-66, 168-77, 185, 200-1, 244-7, 252, 256, 273 Straley, J.R 150 Stringari, S. 23, 72, 148, 199, 207, 217-18, 277, 283 Stunault, A. 6, 12, 23, 38, 154, 157-8, 162-3, 166-70, 173, 185 Svensson, E.C. 23, 27, 33-9, 42, 45, 69-70, 75, 78-9, 86-8, 134-6, 154-7, 160-4, 167-72, 175, 178, 182, 185, 192-4, 200-3, 243-6, 273, 276-8 Szepfalusy, P. 19, 72, 93, 103-4, 107, 111, 150, 272 Szprynger, A. 229, 278, 281-2 Takeno, S. 227-9 Talbot, E.F. 23, 34, 37-8, 62, 95, 100, 102-5, 112, 115, 129-31, 139, 146, 160, 164, 168, 171, 175, 177-8, 185, 197, 200-1, 205-7, 270, 273, 277-81 Tanatar, B. 175, 177 Tarvin, J.A. 137^0 Tenn, J.S. 78 Tennant, D.C. 192 ter Haar, D. 59 Tisza, L. 3-5 Tserkovnikov, Yu. A. 107, 110 Tzoar, N. 77 Udagawa, K. 251-3 Usmani, Q.N. 74, 81-2 van Hove, L. 9, 25-6 Vaughan, J.M. 134 Vidal, F. 137-40

298

Author index

Vinen, W.F. 53, 134, 137 Vollhardt, O. 53, 122, 133 Wagner, H. 23, 70, 195-6 Walecka, J.D. 32, 47, 55, 58-9, 73, 196 Wang, Y. 90, 277 Watabe, M. 251-3 Weichman, P.B. 59, 105, 146-7 Werthamer, N.R. 257, 260-2, 268 West, G.B. 43 Wilkinson, M.K. 78 Whitlock, P.A. 12, 17, 75, 81-4 Wilks, J. 4, 9 Woerner, R.L. 256 Wolfle, P. 53, 122, 133 Wong, V.K. 19, 23, 66, 93-105, 110-16, 125, 151, 195-200, 204, 207, 217, 266, 280-2

Woo, C.W. 10, 19-21, 51, 93, 97, 210, 272 Woods, A.D.B. 3-7, 11-13, 23, 27, 33-6, 39-45, 70, 75, 78-9, 86-8, 104, 134-6, 156-7, 159, 162, 167-72, 175, 178, 194, 200-4, 216, 243-7, 256, 264, 275-6 Yamash*ta, H. 251-3 Yan, J. 252, 256 Yang, C.N. 17, 49 Yarnell, J.L. 5 Yip, S. 228 Yoshida, F. 227-9 Zawadowski, A. 19, 24, 104, 172, 180, 214, 229, 231-48, 283

Subject index

angular momentum rotating liquid Helium 7, 128 transverse perturbation 128 two-roton bound state 236, 239, 249 anharmonic (quantum) crystal 258-66 ACB sum rule 24, 204 definition of a phonon 260-2 anomalous dispersion 5, 150, 156 anomalous propagator Gn 54-7, 76, 156, 234 atomic-like excitations maxon-roton in superfluid 4 He 9, 88, 179, 183, 273-7 normal liquid 4 He 152, 183, 275, 284 polarization potential theory 221, 224 solid 4 He 268 spurious free-particle peaks 192 atomic motions 9-10, 26, 46, 68, 76, 219, 268, 274, 277 atomic polarizability 134, 249, 279 atomic scattering cross-section 45 attractive interaction (two-roton) 237-9 backflow 10, 114, 211-12, 216, 219, 223, 254 Beliaev matrix single-particle Green's function 55, 233-4 Beliaev second-order approximation 60-1, 116 Bethe-Salpeter integral equation 120-2, 232-3, 237 Bijl-Feynman wavefunction 9, 20, 211-13, 217, 275-6 Bogoliubov approximation 22, 56-61, 114 as description of superfluid 4 He 59, 64 dynamic structure factor 47, 61-3 effect of non-condensate atoms 59, 62, 108, 110 quasiparticle energy 57-8 zero sound interpretation 66, 160

Bogoliubov-Hohenberg-Martin sum rule 72, 130, 148 Bogoliubov prescription (for condensate) 104, 146 Born approximation (neutron scattering) 26, 278 Bose broken symmetry 1, 11, 18-21, 49-51, 64, 69, 104 Bose condensate direct evidence 69, 86-90, 181 as mean field 64-6, 151-2, 184 as reservoir 50, 104, 107 transitions (into or out of) 52-3, 181 Bose distribution function 7, 70-1 identities 32, 63 Bose-Einstein condensation 1, 49 dynamics of condensate atoms 19, 32, 104, 146-7 experimental evidence in liquid 4 He 14-15, 77-90 gas vs. liquid 22, 49 in a solid 257 Monte Carlo results 15, 81-2 relation to superfluidity 4, 15, 50, 128-31 Bose gas (ideal) 16, 47-8 density response 32, 106 Bose gas (weakly interacting), see WIDBG Bose liquid comparison to Bose gas 8, 22, 59 definition of condensate 49 Feynman treatment 9, 275 Bose order parameter 14 phase and amplitude 22, 49-50, 69, 91, 138 fluctuations 72, 132, 138-9, 147, 152 symmetry breaking 50 Bose statistics 10, 178, 274-6 Bose symmetrization, effect on excitation spectrum 10

299

300

Subject index

Feynman-Cohen state 217 momentum distribution 82 pair distribution function 276 bound states (roton) 19, 236-9, 256 Brillouin light scattering 23, 132-5, 137, 279 broadening collisional 44-5, 77, 177 see also final state effects, instrumental resolution broken symmetry (non-Bose) 14, 50, 141 broken-symmetry vertex functions 96-8, 100-3, 118, 123, 143, 177-84 high-frequency limits 197-8 key role 167, 178-9, 183-4, 187-9, 193 low-frequency limits 129, 143, 281 3 He- 4 He mixtures 280-1 Brueckner-Sawada theory 59 bubble diagrams 66, 119, 122, 179 central frequency moments 31, 45, 85 Chalk River neutron data (4He) key papers 33, 78, 154 momentum distribution (analysis) 78-9 multiparticle contribution 36-7, 40, 200, 246-7 phonon-maxon-roton dispersion curve 7, 35, 136 chemical potential 27, 47 determination of condensate fraction 105 Hugenholtz-Pines sum rule 57, 130 classical liquids 161-2, 174, 225 momentum distribution 46, 68-9 coherence factors (Bogoliubov) 57, 63, 76, 181, 235, 242 collisional broadening 44-5, 77, 177 collision-dominated, see hydrodynamic region collisionless regime 132, 149 relation to hydrodynamic region 23, 152, 263 commutation relations 104 compressibility sum rule 31, 143-5, 206, 225-6 compressional sound velocity 61, 71, 111, 150 Compton profile 43-4, 83-7 condensate atoms 51-2, 104 dynamics 146 condensate current 131, 139 condensate fraction ho appearance in sum rules 129-30, 196-8 effect on momentum distribution 69-75, 87 extraction from neutron data 14, 46, 86-91

magnitude and temperature dependence 15, 67, 90-1 Monte Carlo calculations 15-17, 81-4, 86, 88, 90-1 relation to pair-distribution 69 at superfluid-solid transition 91 variational calculations 75, 81-2 in 3 He- 4 He mixtures 91, 277 condensate-induced changes momentum distribution 69-75, 88 weight of maxon-roton in S(Q, co) 23, 69, 178-81, 189, 193 see also dielectric formalism, Glyde-Griffin, hybridization condensate wavefunction 49 conservation laws 60, 99, 104, 125 conserving approximations 60, 107, 116 continuity equation 50, 99, 104, 114, 125, 128, 131 continuum particle-hole (density) 32, 170-3, 240 particle-particle (pair) 63, 180, 187, 240, 242-3 thermally excited rotons 149, 179, 251 two-phonon 63, 115 Cooper pairs 1, 53, 133 correlated-basis-functions (CBF) 209-15 Feynman-Cohen basis 10, 74, 211-12 physical meaning 212, 217 solid 4 He 261, 268 role of condensate 20, 74 correlation (response) functions 95 current-current 95, 98, 114, 125, 192, 206 current-single particle 95 density-current 95, 98 density-density 29, 61, 64, 93, 101, 123 density-single-particle 95, 102, 198 four-point 234 3 He- 4 He mixtures 279-81 order parameter 146 single-particle 28, 54, 139 superfluid velocity 139, 144 correlation length 138 critical mode see Goldstone mode, second sound critical point (gas-liquid), 2 critical region (near 7^), 21 Bose order parameter 14, 91 correlation functions 69, 137-8 critical velocity 3, 16 cross-over phenomena collective and single-particle 20, 193 in 3 He- 4 He mixtures 230, 282 first and zero sound 71, 134, 136 zero sound and incoherent p-h excitations 170-5

Subject index cross-section 4 He- 4 He scattering 45 neutron- 4 He scattering 25-6 neutron- 3 He scattering 278 Raman scattering 249 current density operator 94, 125 Debye-Waller factor 264 analogue of Bose order parameter 266 decay of quasiparticles (stability) 19, 22, 156-7 deep-inelastic scattering 43, 199 density fluctuations (3He) 160-2, 172, 175, 275 density fluctuations (superfluid 4 He) as eigenstates 10, 211 coupling to single-particle excitations 20, 51, 100, 123, 193, 271 coupling to two-particle excitations 194, 240, 271 density matrix ODLRO 49 single-particle 49, 211, 218 two-particle 218 density operator 9, 26 second quantized 27, 51 density response function (Bose) 29, 93, 94, 101 Bogoliubov approximation 62 differences above and below Tx 102, 148-52 ideal gas approximation 32 RPA 106 relation to S(Q,co) 29, 61, 124 sum rules 30-1, 196-7 depletion of condensate 50, 143 Bose gas 73 superfluid 4 He 110 detailed-balance factor 30-2, 170, 199 diagrammatic perturbation theory (Bosons) key papers 18-19, 51, 93 treatment of condensate 51-2, 104-5 see also dielectric formalism, Gavoret-Nozieres diagrams 93-103 bubble 62, 119, 122 irreducible 94 ladder 58-9 one-loop 114 proper 94, 97-8, 129 regular 94, 98, 118 self-energy 57-8, 97-8 vertex function 96-8 zero-loop 105-7 dielectric formalism 21-2, 64, 92-105 condensate component 102-3, 143

301

dielectric functions 93, 96, 100, 102, 109, 115, 124-6, 184-9 Feynman diagrams 93, 95 generalized Ward identities 99, 1 0 3 ^ , 281 3 He- 4 He mixtures 24, 277-82 key papers, 19, 93 normal (non-condensate) contribution 97-8, 102-3, 115, 143, 149, 151, 184-5 relation to mean field 64-6, 279-80 summaries 92, 103, 123-6 two-component expressions 64, 96-8, 101-3, 123, 144, 150-2, 177-83 displacement field, see phonon Green's function divergences Popov treatment 139, 145 see also infrared divergences Doppler-broadening 46, 67, 181 oscillations 175-7 dynamic structure factor in liquids (general) 26, 124, 278-9 eigenstate representation 27 general properties 29-31 impulse-approximation 44, 67 sum rules 30-1 symmetric and antisymmetric components 78 dynamic structure factor in liquid 4 He 13, 33-9, 154-77 correlated-basis-functions 212-13 basic features 63 Feynman approximation 10 multi-particle contribution 13, 24, 33, 179 quasiparticle term 34, 178 relation to liquid 3 He 161-2, 169, 173-5 relation to solid 4 He 24, 266-9 temperature dependence of line shape 11, 158, 163-4, 169-71, 191 two-component forms 33, 64, 123, 177-9 two-fluid region 135 see also impulse approximation (high Q) dynamic structure factor in solid 4 He ACB sum rule 204, 266 Ambegaokar-Conway-Baym (ACB) formula 264 comparison to superfluid 4 He 12, 24, 258-9, 266-8 interference contribution 13, 264, 267 neutron scattering data 263-5 two-phonon contribution 244, 267 Dyson-Beliaev equations of motion 55-7, 99, 103 effective fields, 64, 219-20

302

Subject index

effective mass 3 He- 4 He mixtures 282 liquid 3 He 161 polarization potential theory 220, 224 energy gap 9, 60, 275 energy resolution, see instrumental resolution expansion parameter (gas) 58 Fermi liquids 3 He- 4 He mixtures 281-2 microscopic 118, 142, 224 normal Fermi liquid (Landau) 8, 218 particle-hole continuum 28, 32, 162 quasiparticle vs. zero sound 53 zero sound 160-2 Fermi wavevector ( 3 He- 4 He) 282 Feynman-Bijl excited state 9-10, 211 Feynman-Cohen excited state backflow 10,211,216 momentum space representation 211 relation to two-phonon states 10, 212 Feynman wavefunction relation to single-particle excitation 217, 274 role of Bose condensate 275-7 as single-mode approximation to

S(Q,w)275

variational aspects 9-10, 211 field-theoretic approach for Bose fluids field operators 49 key papers 18-19, 51, 93 history 17-20, 93, 272 field-theoretic approach for quantum solids definition of phonons 260, 262, 268 two-component form for S(Q,co) 264 final state effects (FSE) 45, 68 lifetime broadening 77 neutron scattering 46, 78, 80-6 normal liquid 4 He 46, 68, 78, 83 non-Lorentzian behaviour 45, 78, 85 solid 4 He 46-8 theories 45, 78, 83 finite-size effects 15, 74 finite-temperature perturbation theory 51, 55 first sound 5, 8, 135, 140, 150 fluctuations of order parameter, see Bose order parameter four-point correlation function 249-50 backflow 254 non-interacting approximation 250, 253 frequency moments compressibility sum rule 31, 143-5, 206 /-sum rules 30, 115, 195-207 third-moment sum rules 30, 226, 283 functional differentiation

conserving approximations 60 generation of Gi from G\ 60, 116 gapless spectrum (phonon) 57, 60, 147, 151, 184-5 gauge broken-symmetry 50, 53, 146 associated phonon mode 184 Gaussian approximation impulse approximation for <S(Q, co) 42 momentum distribution 43, 46, 68, 73, 85-86 Gavoret-Nozieres analysis (T = 0) 18, 23, 61, 105, 116-22, 141-3,233 Bethe-Salpeter equation 120, 233 common phonon pole 18, 61, 143 relation to dielectric formalism 121-2, 125 relation to two-fluid model 144-5 two-component expressions 64, 101, 143 Gavoret-Nozieres-Pines (GNP) scenario role of "normal fluid" component 149-52 summaries 125-6, 148-9, 272 zero temperature region 141-3 glory oscillations 177 Glyde-Grifrin scenario 178-94 development 2 7 1 ^ experimental evidence 164, 167, 183 key ideas 20, 153, 178, 183, 189, 193, 273 parameterizations 183-94 relation to other approaches 209, 215, 223, 272-3 relation to Ruvalds-Zawadowski 189, 248 Goldstone mode (symmetry-restoring) 14, 17, 50, 108, 147, 151, 184 Green's function Monte Carlo (GFMC), see Monte Carlo ground state wavefunction 9, 50, 210, 261 Hamiltonian (Bose) 52, 58 symmetry-breaking 51-2, 64, 146 hard-sphere Bosons 50, 58, 71, 197 Hartree-Fock approximation (Bose) self-consistent 60, 116, 196 self-energy 57, 113, 196-7 normal fluids 58, 113, 262 Helium atoms, see atomic polarisability, interatomic potential Helium-I vs. Helium-II 2 3 He- 4 He superfluid mixtures 24, 277-82 high-energy region 23 long-wavelength single-particle excitation 9, 151-2, 184-6, 245, 284 maxon 162-7 zero sound 165-7 see also pair excitations

Subject index high-frequency expansions 30, 196-8 tail in S{Q,co) 33, 199-200, 245 high-momentum neutron scattering 67-90 Chalk River analysis 78-9 condensate peak 67-9 impulse approximation 67 Sokol-Sosnick-Snow analysis 80-4 Hugenholtz-Pines theorem 57, 60-1, 71, 130, 147, 151 hybridization (condensate-induced) 178-94, 233, 271 one- and two-particle spectra 19, 24, 178-81, 189, 234-48 single-particle and density fluctuations 20, 23, 64-6, 97, 100, 109, 123, 151, 167-8, 183, 189, 192-3, 229 hydrodynamic region (Bose) 132-41 behaviour of response functions 114, 140-1, 145, 279 sum rules 31, 71, 129-30 two-fluid region (Landau) 23, 135, 139-40 vs. collisionless 132, 152 imaginary-time formulation (Matsubara) 29, 51, 54 impulse approximation 39-46, 67, 75-7, 198 additive corrections 78-9 dynamic structure factor 46, 67, 181 final-state corrections 45, 79, 83 neutron scattering data 14, 44, 77-87 incoherent scattering cross-section 26-31, 41 infrared (low OJ>) divergences 127, 142-8 Institut Laue-Langevin (ILL) neutron data (4He) maxon region 6, 12, 37-9, 163-4 past the roton 176 pressure dependence 165-6, 244 phonon region 11, 154-60 roton region 38, 169-71 instrumental resolution broadening 68, 78-80, 83-5, 154, 181 interaction vertex function 60, 120, 227 interatomic potential in 4 He 70 bare 4 He- 4 He interaction 45 effective interaction in liquid 4 He 177, 220-3, 235, 248, 283 hard-core 58, 71, 197, 199-200 interference contributions anharmonic crystals 201, 264, 267 superfluid 4 He 53, 183, 203, 242 WIDBG 62-4, 115 intermediate scattering function 26, 41 irreducible diagrams 93-103 irreducible response functions

303

two-component expressions 97-8, 123, 183^ Jastrow(-Feenberg) wavefunction 72, 208-10, 260-1 kinetic energy of atoms 73, 108 kinetic equations 122, 133 Kosterlitz-Thouless scenario (2D) 132 ladder diagrams (t-matrix approximation) 58-9, 122, 214, 222, 283 lambda transition (TA) region 8, 14-16, 20-1, 48, 67-9, 92, 102, 107, 113, 124-8, 134, 140-1, 152, 166-7, 178-82, 245, 253, 272-5 Landau-Feynman scenario 12, 20, 61, 272 Landau-Khalatnikov theory 8, 22, 133 Landau-Placzek ratio 134-7, 140, 158 lattice vibrations 258-62 lifetimes (thermal) 132 Lindhard function Bosons 32, 142, 179, 229 Fermions 169, 224, 227, 281-2 linear response theory 64, 146, 219 liquid-gas transition 2, 135 liquid 3 He 53, 77, 142, 160-1, 218, 221 Longitudinal current response function (Bose) 95 continuity equation 99, 125, 128, 192 zero-frequency limit 105, 128-9, 144 longitudinal susceptibility (Bose) 147 long-range order (LRO) 132 long-range spatial correlations Bose-condensed fluid 49, 128-9 relation to phonons 72 variational wavefunctions 210 Lorentzian approximation quasiparticle peak 34, 77, 160 Raman scattering 251 low-density Bose gas finite temperature 108-11 zero temperature 57-61 see also WIDBG low-energy, long-wavelength behaviour (Bose) effect of infrared divergence 145-8 Gavoret-Nozieres 18, 61, 141-3, 148-9 Goldstone mode 50, 151-2 Hugenholtz-Pines theorem 61 hydrodynamic vs. collisionless 150 Popov analysis 139, 145 regular functions 103, 184 sum rules 31, 128-31, 199 two-fluid region 132-40 macroscopic region (Bose), see two-fluid description

304

Subject index

macroscopic wavefunction 49 phase coherence 16, 50 relation to superfluidity 13-16 see also Bose order parameter Matsubara formalism (imaginary frequencies) analytic continuation 55, 61 Bose 29, 55, 147, 281 frequency sums 63, 113, 235 maxon region (intermediate Q) hybridization effects 167, 182-3, 187-90, 193, 203 neutron data 5-7, 163-7 maxon-roton branch hybridization with pair excitations 189 hybridization with phonons 20, 23, 182-9, 193 intrinsic to normal liquid 4 He 20, 126, 178, 183, 193, 273-5, 284 intrinsic to superfluid 4 He 125, 272 weight in S{Q,co) 23, 69, 178, 189, 284 mean-fields 108, 165 due to condensate 64-6, 151, 224 in 3 He- 4 He mixtures 280 zero-loop approximation 107 memory function formalism for S(Q,co) 23, 225-30 in 3 He- 4 He mixtures 229, 278 mode-mode coupling 227, 230, 281-2 momentum current response, see current response momentum distribution function (liquid 4 He) condensate peak 67, 81 data from neutron scattering 14, 43, 46, 77-85 extraction from 5(Q,co), theory 69, 78-9 Gaussian 46, 68, 79, 84-5 high-momentum behaviour 85, 88, 99 low-momentum behaviour 69-75, 79-81, 84, 88-9, 130, 148 Monte Carlo calculations 74, 81-5 non-condensate (normal) part 87-9 normalization 67, 84, 87 normal liquid 4 He 69, 79, 83, 88 pressure dependence 91 relation to single-particle spectral density 70-4 rigorous results 71, 75, 89, 148 temperature dependence 73 two-Gaussian fit 86-90 variational calculations 74, 81-2, 210, 214 Monte Carlo calculations Green's function (GFMC) 74, 81, 84, 86, 90-1, 210

path integral (PIMC) 15-18, 48, 69, 74, 81-5, 90, 128, 276 multiparticle spectrum neutron scattering data 13, 33, 155, 163-4, 167, 171, 244-7 overlap with density fluctuation 221, 244 overlap with single-particle excitations 13, 19, 240, 248 multiphonon see bound states, multiparticle spectrum multiple scattering (t-matrix), see ladder diagrams negative-energy poles in spectral density 57, 63, 72-3, 242 neutron scattering (inelastic) 26 dilute 3 He- 4 He mixtures 278, 281-2 liquid 3 He 160-2 liquid 4 He 33-40, 154-77 role of Bose condensate 19-21, 92, 160, 181, 276 solid 4 He 259, 267 see also dynamic structure factor neutron scattering lengths 26, 278 non-condensate atoms 5, 52, 56, 59-61, 73, 87 density response 32, 62, 115, 123, 149-51, 273 effect on SP spectrum 108 "normal" component in dielectric formalism 100-2, 109, 116, 123, 149-52, 179, 189, 197, 223 normal fluid density as measure of quasiparticle density 3, 5, 8, 110, 150, 191,223 PIMC results 15, 18 quasiparticle formula 7, 15, 131 relation to zero-frequency response functions 15, 18, 110, 128 temperature dependence 4, 8, 18 normal liquid 4 He 150, 268, 275 density response 11, 37, 46, 64, 79, 134, 158, 163-74 number density 28, 50-1 ODLRO 49 one-loop approximation 112-16, 151 in 3 He- 4 He mixtures 283 zero-frequency limit (sum rules) 131 order parameter (Bose) 49, 69 c-number approximation 104, 146 dynamics 146 macroscopic wavefunction 15-16, 49 see also Bose condensation, phase fluctuations

Subject index order parameter (non-Bose) 14, 50, 53, 109, 146 pair-distribution function 27, 45, 210 effect of Bose statistics 276 relation to condensate 69 pair excitation spectrum (Q j=0) 19, 180, 234-9, 243-7, 271 Bogoliubov approximation 63 coupling into S(Q,co) 212-14, 239 coupling to single-particle spectrum 19, 23, 240-8 temperature dependence 179-80, 213-15, 245 two-particle Green's functions 116-22, 271 particle-hole excitations (non-interacting) in 3 He- 4 He mixtures 281-2 liquid 3 He 28, 224 particle-hole excitations (liquid 4 He) 28 Doppler-broadened peak 168, 174-7 incoherent 32, 169, 170-5 Lindhard function (gas) 32, 179 see also zero sound particle-hole propagator, see density response function path integral Monte Carlo (PIMC), see Monte Carlo phase coherence (Bose) 16, 50 phase diagram for 4 He 2 phase fluctuations in superfluid 4 He 50, 72, 139, 146-8 phase shift (s-wave) 58 phonons in Bose-condensed fluids (theory) 141-52 WIDBG (T = 0) 57-61 W I D B G ( T ^ O ) 108-11 phonon Green's function (crystal) 260 analogue of single-particle Green's function 257, 260, 268 relation to S{Q,co) 264, 268 phonons in quantum crystals 258-63 collective vs. single-particle 262, 268 effect of short-range correlations (SRC) 260-1 as pole of displacement correlation function 24, 261, 268 relation to S{Q,co) 263-6 phonon scattering processes (superfluid 4 He) 8, 156-7 phonons in superfluid 4 He (experiment) 7 vs. maxon-roton branch 20, 153, 167 vs. normal liquid 4 He 11, 134, 158, 160 region around T;t 157-9 relation to zero sound 160-2 temperature independence (above 1 K) 11, 150, 157, 159, 184

305

widths 22, 155-7 see also anomalous dispersion Pitaevskii plateau 19, 180, 241 polarization potential theory 23, 218-24, 281-2 pressure dependence (liquid 4 He) 137, 164-8, 247 condensate fraction 91 effective polarization potentials 221-2 proper (and improper) diagrams 94, 97-8, 129 quantum evaporation 284 quantum hydrodynamics 8, 22, 216 Popov derivation 139, 145 quantum kinetic equations 122, 133 quasiparticle dispersion relation in liquid 4 He 7, 35 impulse-approximation region 174-7 maxon region 162-7 past the roton region 174, 180 phonon region 11, 158-60 roton region 168-74 single branch vs. two branches 183-93, 273 variational calculations 212-14 quasiparticles 33 Bogoliubov model 57 dilute 3 He- 4 He mixtures 281-2 interpretation 8, 12, 20, 148-53, 193, 272-6 normal liquid 4 He 12, 20, 277 see also phonons in quantum crystals, single-particle Green's function Raman light scattering 249 data analysis near TA 251-3 evidence for two-roton bound state 256 non-interacting pair approximation 250 s- and d-wave components, 236-7, 249 random phase approximation (RPA) 106-7, 218-19 Rayleigh central peak 134 recoiling atom energy 31-2 impulse-approximation 41-2, 67-8, 78, 175 regular diagrams 94-8, 114, 118, 280 regular single-particle Green's function G^ 96, 103, 124, 178, 183, 223, 280 long-wavelength 130, 151-2, 184 multiparticle spectrum induced by condensate 152 sum rules 130, 196-8 regularity assumption 116, 143—4 renormalization group 21, 59

306

Subject index

restricted ensemble 49-50 see also Bose broken-symmetry rotating superfluid 4 He 4, 7, 128 roton-liquid theory 275 rotons (experiment) 5 condensate-induced weight in S(Q,co) 178-82, 240-1, 273-4 existence above Tx 168, 178, 275, 284 lineshape 168-71, 284 overlap with p-h excitations 169, 172-3, 179 overlap with two-roton spectrum 19, 179-80 Ruvalds-Zawadowski theory 19, 104, 180, 214, 232^8 screened response functions in 3 He- 4 He mixtures 280-2 mean-field analysis 65 polarization potential theory 219-23 second sound Bose gas 133, 136 in 3 He- 4 He mixtures 279 quantum solid 262 in superfluid 4 He 5, 8, 133-40 weight in single-particle Green's functions 139-41 weight in S{Q,co) 135-7 self-consistent fields, see mean-fields self-consistent harmonic (SCH) phonon theory 260-1 self-consistent ^-matrix 59 self-energies (Bose fluids) 56-60 irreducible 97-8, 242 one-loop 112-15 relation to chemical potential 61, 130, 196 zero-frequency limit 130, 142-3, 147 zero-loop 106, 110, 114-15 shielded potential approximation (SPA) 65-6, 105-11, 113 short-range spatial correlations (SRC) liquid 4 He 45, 210, 219 solid 4 He 260-1, 268 short-time behaviour 30, 229 single-particle density matrix 49, 74, 211, 218 single-particle excitations in quantum crystals neutron data at high Q 259 relation to phonons 262, 268 single-particle excitations (spectrum) coupling into density 51 coupling into S(Q,w) 18-22, 60, 64, 178-83, 246, 284 at finite T 108, 110, 148-52

high-energy modes at long wavelengths 151, 184 single-particle Green's functions (Bose) 53-6 Bogoliubov approximation 57-9 in conserving approximations 60 low-frequency behaviour 18, 61, 148, 184 physical interpretation 148, 272 relation to S(Q,co) 18, 22, 29, 148-50, 177, 242-8, 272 spectral representation 28, 139 single-particle (SP) operator 217-18 single-particle spectral density (Bose) 28 Bogoliubov approximation 72 hydrodynamic (two-fluid) region 71, 139^0 relation to momentum distribution 70-4 rigorous sum rules 70, 196, 217 singular contributions, see diagrams, Gavoret-Nozieres analysis sound velocity 5 first sound 8, 111, 150 second sound 111, 135, 262 soft modes 109-11 thermodynamic (compressional) 31, 61, 111, 150 zero sound 111, 150, 263 specific heat near T-A 15-16 spin-polarized atomic Hydrogen 109 static structure factor S {Q) 27, 41, 211, 214, 226-7 quasiparticle contribution Z(Q) 34, 41, 181-2 Stokes (anti-Stokes) component 30, 32 strong-coupling limit 109-11 sum rules for Bose fluids Bogoliubov-Hohenberg-Martin 71, 130 compressibility 31, 145, 206 density-single-particle 198-9, 217 /-sum rule 30, 99, 115, 151, 185-6, 191, 196 Hugenholtz-Pines 57, 60-1, 71, 130, 151 third order frequency-moments 30, 226, 283 Wagner 70, 196, 198 Wong-Gould 197 superconductivity (BCS) 1, 53 supercurrents, relation to condensate 131, 139 superfluid density appearance in sum rules 129-30, 206 Landau two-fluid model 3, 5 relation to condensate 4, 22, 50, 131 relation to current-response functions 15, 128-9 temperature dependence near Tx 91 superfluid Fermi liquids 53, 122, 133

Subject index superfluidity, relation to Bose condensate 48, 50, 129, 131 superfluid-solid phase transition 90 superfluid velocity field 2, 49, 139 susceptibility dynamic 29 static 31, 225 symmetry-breaking perturbation 51-4, 104, 146 symmetry relations broken-symmetry vertex functions 97-8 dynamic structure factor 29-30 single-particle matrix Green's functions 55, 74 two-particle Green's functions 95 Taylor expansions at small Q, co Gavoret-Nozieres results 142-3 Nepomnyaschii-Nepomnyaschii results 143-4, 147-8 regularity assumption 116, 143-^ temperature fluctuations, see Brillouin light scattering, Landau-Lifsh*tz ratio, second sound thermal scattering processes as background 149 dilute Bose gas 32, 63 superfluid 4 He 142, 150, 179 thermodynamic derivatives ground state energy 61, 105 self-energies 61, 142-4, 196 thermodynamic potential 104-5 thermodynamic properties 4-7, 135-7 three-body correlations 210 Mnatrix, see ladder diagrams transport coefficients 4, 7, 133 transverse response functions 15, 95, 98, 128-9 tricritical point 281, 284 two-component forms for S(Q,co) dielectric formalism 64, 101, 123-6, 177-8, 183, 207, 223, 279-81 Gavoret-Nozieres {T = 0) 64, 143, 148-50 Glyde-Griffin 126, 183-94, 273 hydrodynamic (two-fluid) 135, 140-1 Szepfalusy-Kondor scenario 109-11, 150, 272 Woods-Svensson ansatz 23, 200-3 two-dimensional Bose fluids 132 two-fluid description equations of motion 4, 133, 279 Landau formulation 3, 5 3 He- 4 He mixtures 279 microscopic basis 18, 50, 128-32 Popov derivation 139, 145

307

at T = 0 limit 144-5, 149 vs. collisionless 152 two-particle density matrix 118 two-particle Green's functions (G2) 60, 116-22,232-9 two-particle interaction g4 (pair spectrum) 235, 248 two-phonon terms quantum crystal 261, 264, 267 Bose gas (WIDBG) 63, 115 two-roton spectrum in S(Q, co) 239-48 ultrasonic studies liquid 4 He 134, 155 solid 4 He 262 variational wavefunctions, see wavefunctions (many-body) velocity field (superfluid) 3, 139 vertex corrections 214, 253 viscosities of normal fluid 3, 133—4 vorticity in superfluid 4 He 21, 134 Wagner sum rule 70, 196-8 Ward identities (generalized) continuity equations 99, 104 high-frequency limit 197-8 importance 100, 103, 122, 129 one-loop approximation 112 zero-frequency limit 129-31, 281 zero-loop approximation 105-7 wavefunctions (many-body) 82, 209-14 Bose phase coherence 216 Bose statistics 9, 217 Feynman 9-10, 20 Feynman-Cohen 74, 213 Gaussian 260-1 hard-sphere Bosons 50 spatial correlations 72-4, 260-1 see also correlated-basis-functions (CBF) weak-coupling limit 63, 108, 111 WIDBG (weakly interacting dilute Bose gas) 1-2, 5 16, 20-1, 47, 56, 59, 66, 73, 92, 103-16 Woods-Svensson ansatz 23, 194, 200-3, 206 Y-scaling 42-4 zero-frequency limits 128-31, 143, 147 zero-loop approximation 105-11 zero-momentum atoms 48-50, 104, 273 zero sound in classical fluids 161-2, 183 zero sound in 3 He 160-1, 165, 169

308

Subject index

zero sound (ZS) in Bose fluids coupling to single-particle excitations 124-6, 183-9 damping 159 due to condensate mean field 66, 110, 151, 186, 273 at large Q 165, 168, 178

liquid 4 He (normal) 134, 165-8, 178, 183-5 liquid 4 He (superfluid) 20-2, 158-9, 182-3 normal fluid component 110, 126, 149, 151, 166-7, 185-9 superfluid vs. normal 151-2, 273