Adilet Imambekov - Strongly Correlated Phenomena With Ultracold Atomic Gases

Strongly Correlated Phenomena with Ultracold Atomic
Gases
A dissertation presented
by
Adilet Imambekov
to
The Department of Physics
in partial fulllment of the requirements
for the degree of
Doctor of Philosophy
in the subject of
Physics
Harvard University
Cambridge, Massachusetts
June 2007
c _2007 - Adilet Imambekov
All rights reserved.
Thesis advisor Author
Eugene Demler Adilet Imambekov
Strongly Correlated Phenomena with Ultracold Atomic Gases
Abstract
In this thesis we investigate strongly correlated phenomena in the eld of ultra-
cold atomic gases. Chapter 2 addresses a question of the insulating phases of cold spin-one
bosonic particles with antiferromagnetic interactions, such as
23
Na, in optical lattices. Mag-
netic properties of the ground state in the insulating regime are studied using various tech-
niques. Chapter 3 considers a one dimensional interacting Bose-Fermi mixture with equal
masses of bosons and fermions, and with equal repulsive interactions between Bose-Fermi
and Bose-Bose particles. Properties of such mixture are studied using exact Bethe-ansatz
techniques. Chapter 4 deals with certain phenomena which appear in the experiments with
imbalanced fermionic mixtures in strongly anisotropic traps. Chapter 5 gives a comprehen-
sive review of interference phenomena, analyzing eects which contribute to the reduction
of the interference fringe contrast in matter interferometers.
iii
Contents
Title Page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Citations to Previously Published Work . . . . . . . . . . . . . . . . . . . . . . . vii
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
1 Introduction 1
2 Spin-one bosons in optical lattices 8
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Derivation of Bose-Hubbard model for spin-one particles . . . . . . . . . . . 11
2.3 Insulating state with an odd number of atoms . . . . . . . . . . . . . . . . . 14
2.3.1 Eective spin Hamiltonian for small t . . . . . . . . . . . . . . . . . 14
2.3.2 Phase diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Insulating states with two atoms per site . . . . . . . . . . . . . . . . . . . 19
2.4.1 Two site problem: exact solution . . . . . . . . . . . . . . . . . . . . 19
2.4.2 Eective spin Hamiltonian for an optical lattice . . . . . . . . . . . . 21
2.4.3 Phase diagram from the mean-eld calculation . . . . . . . . . . . . 22
2.4.4 Quantum uctuations corrections for the spin singlet state . . . . . . 27
2.4.5 Spin wave excitations in the nematic phase . . . . . . . . . . . . . . 29
2.4.6 Magnetic eld eects . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5 Large number of particles per site . . . . . . . . . . . . . . . . . . . . . . . 34
2.5.1 Mean eld solution without magnetic eld . . . . . . . . . . . . . . . 37
2.5.2 Magnetization plateaus . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.6 Global phase diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.6.1 Two and three dimensional lattices . . . . . . . . . . . . . . . . . . . 43
2.6.2 One dimensional lattices . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.7 Detection of spin order in insulating phases . . . . . . . . . . . . . . . . . . 46
2.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3 Exactly solvable case of a one-dimensional Bose-Fermi mixture 52
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2 Bethe ansatz solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3 Numerical solution and analysis of instabilities . . . . . . . . . . . . . . . . 59
iv
Contents v
3.4 Local density approximation and collective modes . . . . . . . . . . . . . . . 62
3.5 Zero-temperature correlation functions in Tonks-Girardeau regime . . . . . 70
3.5.1 Factorization of spin and orbital degrees of freedom . . . . . . . . 71
3.5.2 Bose-Bose correlation function . . . . . . . . . . . . . . . . . . . . . 72
3.5.3 Fermi-Fermi correlation function . . . . . . . . . . . . . . . . . . . . 79
3.5.4 Numerical evaluation of correlation functions and Luttinger parameters 82
3.6 Tonks-Girardeau regime at low temperatures . . . . . . . . . . . . . . . . . 84
3.6.1 Low energy excitations in Tonks-Girardeau regime. . . . . . . . . . . 86
3.6.2 Density proles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.6.3 Fermi-Fermi correlations . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.6.4 Bose-Bose correlation function . . . . . . . . . . . . . . . . . . . . . 94
3.7 Experimental considerations and conclusions . . . . . . . . . . . . . . . . . 96
4 Breakdown of the local density approximation in interacting systems of
cold fermions in strongly anisotropic traps 102
5 Fundamental noise in matter interferometers 109
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.1.1 Interference experiments with cold atoms . . . . . . . . . . . . . . . 109
5.1.2 Fundamental sources of noise in interference experiments with matter 115
5.2 Interference of ideal condensates . . . . . . . . . . . . . . . . . . . . . . . . 117
5.2.1 Interference of condensates with a well dened relative phase . . . . 117
5.2.2 Interference of independent clouds . . . . . . . . . . . . . . . . . . . 123
5.3 Full counting statistics of shot noise . . . . . . . . . . . . . . . . . . . . . . 124
5.3.1 Interference of two independent coherent condensates . . . . . . . . 128
5.3.2 Interference of independent clouds in number states . . . . . . . . . 129
5.3.3 Clouds with a well dened relative phase . . . . . . . . . . . . . . . 131
5.4 Interference of low-dimensional gases . . . . . . . . . . . . . . . . . . . . . . 131
5.4.1 Interference amplitudes: from high moments to full distribution func-
tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
5.4.2 Connection of the fringe visibility distribution functions to the parti-
tion functions of Sine-Gordon models . . . . . . . . . . . . . . . . . . 140
5.4.3 Non perturbative solution for the general case . . . . . . . . . . . . . 143
5.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
5.5.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
5.5.2 Some experimental issues . . . . . . . . . . . . . . . . . . . . . . . . 151
5.5.3 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
A Appendix to Chapter 2 154
A.1 Derivation of the eective magnetic Hamiltonian for insulating states with
odd number of atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
A.1.1 Normalization of the states . . . . . . . . . . . . . . . . . . . . . . . 155
A.1.2 Calculation of
0
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
vi Contents
A.2 Derivation of the eective magnetic Hamiltonian for the insulating state with
two atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
A.3 Mean eld solution for the case of two bosons per site . . . . . . . . . . . . 162
A.4 Large N expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
B Appendix to Chapter 3 165
C Appendix to Chapter 5 167
C.1 Expansion to order (1/K)
2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
C.2 General properties of (1/K)
m
terms, and expansion to order (1/K)
5
. . . . . 171
C.3 Properties of the K distribution . . . . . . . . . . . . . . . . . . . . . 174
C.4 D=2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
Bibliography 177
Citations to Previously Published Work
Publications relating to the particular chapters are as follows:
Chapter 2:
Adilet Imambekov, Mikhail Lukin, Eugene Demler,Spin-exchange interactions of
spin-one bosons in optical lattices: singlet, nematic and dimerized phases, Phys.
Rev. A 68, 063602 (2003), also as cond-mat/0306204.
Adilet Imambekov, Mikhail Lukin, Eugene Demler, Magnetization plateaus for spin-
one bosons in optical lattices: Stern-Gerlach experiments with strongly correlated
atoms, Phys. Rev. Lett. 93, 120405 (2004), also as cond-mat/0401526.
Chapter 3:
Adilet Imambekov, Eugene Demler, Exactly solvable case of a one-dimensional Bose-
Fermi mixture, Phys. Rev. A 73, 021602(R) (2006), also as cond-mat/0505632.
Adilet Imambekov, Eugene Demler, Applications of exact solution for strongly inter-
acting one dimensional Bose-Fermi mixture: low-temperature correlation functions,
density proles and collective modes, Annals of Physics 321, 2390 (2006), also as
cond-mat/0510801.
Chapter 4:
Adilet Imambekov, C.J. Bolech, Mikhail Lukin and Eugene Demler, Breakdown of
the local density approximation in interacting systems of cold fermions in strongly
anisotropic traps, Phys. Rev. A 74, 053626, available as cond-mat/0604423.
Chapter 5:
Adilet Imambekov, Vladimir Gritsev, Eugene Demler, Distribution functions of in-
terference contrast in low-dimensional Bose gases, submitted to Phys. Rev. Lett.,
available as cond-mat/0612011.
Adilet Imambekov, Vladimir Gritsev, Eugene Demler, Fundamental noise in matter
interferometers, to be published in the Proceedings of the 2006 Enrico Fermi Sum-
mer School on Ultracold Fermi gases, organized by M. Inguscio, W. Ketterle and
C.Salomon (Varenna, Italy, June 2006), available as cond-mat/0703766.
vii
Acknowledgments
First of all, I would like to thank my advisor, Eugene Demler. His constant support
and encouragement, as well as the freedom he gave me made my years at graduate school
enjoyable and rewarding.
I would also like to thank Misha Lukin, who was a co-advisor on some of the
projects I did. His unique approach to formulating complicated problems in a simple lan-
guage is inspiring. I am also grateful to John Doyle for accepting the burden of serving on
the thesis committee.
During the later stages of my thesis work, I enjoyed working with postdocs Carlos
Bolech and, especially, Vladimir Gritsev. Vladimirs attitude to science always reminds
me why I chose to be a scientist in the rst place. There have been many other postdocs
and fellow graduate students with whom I have been lucky to interact. Among them are
Vincenzo Vitelli, Ryan Barnett, Bob Cherng, Daniel Podolsky, Gil Refael, Alexey Gorshkov,
Ehud Altman, Anatoli Polkovnikov, Daw-Wei (Charles) Wang, and Anton Burkov.
I owe a special trubute to my dear friends, Dima Abanin and Itay Yavin, for being
like a family to me during my years at graduate school. Itaychik taught me that being
able to laugh at ones shortcomings is the best way to overcome them. His example always
inspires me to try out new things. Dimochka was a great friend during last 10 years, and
his open-mindness is a constant source of fun. I am also thankful to Tom Hunt, for being
a great oce mate during the rst year; Pavel Petrov, for sharing with me his masterful
control of Russian language; George Gosha Brewster, Ilya Tatar and Alexey Dynkin, for
their company during numerous adventures outdoors and seless driving.
Id also like to thank administrative sta of the department, especially Sheila
Ferguson, for making it such a great place. Her care for students and their needs makes the
department feel like home.
My parents, Dzhanat and Onlasyn, my brother Akniet and my sister Akbota have
always been with me in my heart, even though most of the time we were on the opposite
parts of the globe. I am grateful to my father for getting me interested in science in
childhood. Last, but not the least, I would like to thank Aigerim for adding a whole new
dimension to my life.
viii
Dedicated to my father, Onlasyn Imambekov,
and my mother Dzhanat Imambekova.
ix
x
Chapter 1
Introduction
In this thesis we will discuss strongly correlated phenomena in the eld of ultracold
atomic gases. As has been concisely formulated by J. R. Anglin and W. Ketterle in 2002
[1], Our eld is now at a historic turning point, in which we are moving from studying
physics in order to learn about atom cooling to studying cold atoms in order to learn about
physics. The remarkable experimental progress in the eld of ultracold atoms in the last
decade has reached the stage at which interactions between dilute gases cannot be described
using a picture of weakly interacting quasi-particles. Such regime is characteristic of the
physics of strongly correlated systems. Problems in which strongly correlated phenomena
appear are notoriously hard to treat theoretically, with high temperature superconductivity
being a prominent example. The simplest model proposed to study high temperature su-
perconductivity, single band 2D fermionic Hubbard model, is intractable analytically. Even
if a solution of this problem were available, it is clear that there is a variety of properties
of high temperature superconductors which are not contained in 2D fermionic Hubbard
model. Thus in a majority of traditional solid-state systems which exhibit strong correla-
tions, one can at most hope to have a qualitative agreement between theory and experiment.
Systems of ultracold gases, on the other hand, provide a unique example for which micro-
scopic Hamiltonians are usually known from rst principles, hence a detailed quantitative
comparison between experiment and theory is possible. On the top of that, the remarkable
degree of control achieved in experiments can be used to tune the interactions, thus driving
quantum phase transitions. The most outstanding achievement up to date in this direction
is the observation of the superuid-insulator transition [2] for ultracold bosons in optical
lattices [3, 4], which will be illustrated below. In the rest of this introductory chapter we
will give a brief overview of current experimental situation and will provide references to
more comprehensive review articles and books.
Experiments on cooling and trapping of neutral atoms have started in 70s, and
culminated in the achievement of Bose Einstein Condensate (BEC) in 1995 [5, 6, 7]. Fig. 1.1
illustrates the typical energy and length scales involved in the atomic cooling. Experimental
techniques used to cool and trap atoms at such low temperatures are nicely summarized
in a book by H. J. Metcalf and P. van der Straten [9]. The limit of quantum degeneracy
is reached when atoms de Broglie wavelength
dB
= h/
2mk
B
T becomes comparable to
characteristic interparticle separation n
1/3
, where n is the atomic density. For non inter-
1
2 Chapter 1: Introduction
Figure 1.1: Typical energy scales in atomic cooling and trapping. Reprinted with permission
from Macmillan Publishers Ltd: Nature ([8]), copyright (2002).
Chapter 1: Introduction 3
Figure 1.2: Observation of Bose-Einstein condensation by absorption imaging. The data was
taken after the gas had been allowed to freely expand for several milliseconds. The images
were taken at three dierent temperatures: one just above the transition temperature,
one just below the transition temperature, and one well below the transition temperature.
Image courtesy of W. Ketterle and Dallin S. Durfee, MIT.
acting bosons, condensate appears at a precise temperature, controlled by n
3
dB
= 2.612 (see
e.g. [10]). Most experiments reach quantum degeneracy with temperatures 500 nK2 K
and densities 10
14
10
15
cm
3
. Fig. 1.2 illustrates density proles measured by absorption
imaging after the gas had been allowed to freely expand for several milliseconds. These
time of ight images show the velocity distribution of the atoms. Above the transition
temperature, the velocity distribution is a spherical gaussian. But as the transition line
is crossed, there is a sudden change. The distribution becomes bimodal, with two sepa-
rate contributions from excited states and from the ground state. In the third picture, the
temperature is low enough so that most of the atoms are in the condensate.
For such low temperatures under non-resonant conditions, only scattering in s
wave channel is important [11], and collisions are characterized by a single parameter, scat-
tering length a. It can have dierent signs, with positive sign corresponding to repulsion
and negative sign corresponding to attraction between atoms. Scattering length is sensitive
to the details of the interatomic molecular potentials, and currently numerical methods
are available to calculate scattering lengths from rst principles [12]. Dimensionless com-
bination, which controls the strength of interactions, is given by so-called gas parameter,
na
3
. The remarkable success of the theory of weakly interacting Bose gases relies on the
smallness of the gas parameter. Suciently below the condensation temperature, dilute
BECs are essentially pure condensates, with all atoms occupying the same macroscopic
state. Static and dynamic properties of such condensates are quantitatively described by a
nonlinear Schrodinger equation, the celebrated Gross-Pitaevskii equation [13, 14]. In this
case many-body physics is reduced to a single-particle description, with interactions pro-
viding additional density-dependent eective potential. By adding small uctuations on
the top of Gross-Pitaevskii state, one recovers the familiar Bogoliubov picture of weakly
interacting quasi-particles.
Major experimental advances in the studies of weakly interacting BECs include
observation of interference of two independent Bose clouds [15], optical trapping of spinor
BECs [16, 17], cooling of Fermi gases and Bose-Fermi mixtures to quantum degeneracy
[18, 19], demonstration of long-range coherence [20] and observation of vortices and vortex
lattices [21, 22]. Excellent reviews of this area of research have been presented in review
articles [23, 24, 25] and in more recent books by C.J. Pethick and H. Smith [26] and L.
Pitaevskii and S. Stringari [27].
In the past several years, the physics which is accessible by ultracold gases has
been enormously enlarged by two major developments, the ability to tune interactions by
magnetic Feshbach resonances [28], and the possibility to generate periodic potentials for
atoms using optical lattices [4, 29, 30, 31, 32]. The idea of Feshbach resonances dates back
to work in nuclear physics in 50s [33]. Feshbach resonance in a two particle collision ap-
pears whenever a bound state in a closed channel is resonant with the energy of scattering
particles in an open channel. Particles then are temporarily captured in the quasi-bound
state, which results in a Breit-Wigner type resonance [11] in the scattering cross-section.
Feshbach resonances are particularly useful in the scattering of cold atoms, since the mag-
netic moments of closed and open channels are dierent. Hence the position of the resonant
energy level can be controlled by uniform external magnetic eld [34]. More detailed in-
troduction to Feshbach resonances can be found in review articles [35, 36, 37]. Optical
lattices [4, 29, 30, 31, 32] have become another standard tool to control the interactions in
recent years. The basic mechanism which produces an external potential is the polarization
of atoms due to external ac-Stark shift in o-resonant light eld. Induced dipole moment
interacts with oscillating electric eld, creating trapping potential
V (r) = dE(r) (
L
)[E(r)[
2
.
Here () denotes the polarizability of an atom and [E(r)[
2
I(r) is proportional to
the intensity of the laser eld. A periodic potential can be formed by overlaping two
counter-propagating beams. By interfering several laser beams, one can create 1D, 2D or
3D lattices, as illustrated in Fig. 1.3. By varying the intensity of laser elds, one can
control the tunneling and interactions of atoms. The regime of strong interactions can be
reached not by increasing the atom-atom scattering length, but by reducing other relevant
energy scales, such as kinetic energy of atoms. In the lattice the latter is controlled by the
tunneling, and can be made small.
Figure 1.3: Optical lattice potentials formed by superimposing two or three orthogonal
standing waves. a, For a 2D optical lattice, the atoms are conned to an array of tightly
conning 1D potential tubes. b, In the 3D case, the optical lattice can be approximated by
a 3D simple cubic array of tightly conning harmonic oscillator potentials at each lattice
site. Reprinted with permission from Macmillan Publishers Ltd: Nature Physics ([32]),
copyright (2005).
Figure 1.4: Absorption images of multiple matter wave interference patterns. For weak
optical lattices interference peaks get sharper, while at critical lattice strength they disap-
pear, signaling the superuid - Mott insulator transition. Reprinted with permission from
Macmillan Publishers Ltd: Nature ([4]), copyright (2002).
The natural Hamiltonian which describes bosons in optical lattice is given by [3, 38]
H
BH
= t
ij
(a
i
a
j
+a
j
a
i
)
i
n
i
+
U
2
i
n
i
( n
i
1). (1.1)
Here t is the tunneling, U is the onsite interaction and is the chemical potential. Ratio
t/U may be controlled by varying the intensity of laser beams, so one can go from the
regime in which kinetic energy dominates (weak periodic potential, t U), to the regime
where interaction energy is the most important part of the Hamiltonian (strong periodic
potential, t U). For integer llings (number of atoms per lattice site), the two regimes
have superuid and Mott insulating ground states, respectively, as can be obtained from
the mean-eld analysis of the Bose-Hubbard Hamiltonian [3, 2]. In the superuid phase,
atoms are delocalized in the lattice, uctuations in the number of atoms on each site are
strong, and there is a phase coherence between dierent sites. In the insulating state,
atoms are localized, uctuations in the particle number at each site are suppressed, and
there is a gap to all excitations. Such an insulating state represents a correlated many body
state of bosons, where strong interactions between atoms result in a new ground state of
the system. Quantum phase transition between these two states has been demonstrated in
experiments of Greiner et al. [4] via analysis of time of ight images, as shown in Fig. 1.4.
For superuid phase, long range coherence results in interference peaks at the reciprocal
lattice wave vectors. For weak optical lattices, interference peaks get sharper, since Wannier
wave functions at individual lattice sites are getting more localized. However, interference
disappears at a critical lattice strength, signaling the superuid - Mott insulator transition.
One of the new directions opened by ultracold gases is the studies of low-dimensional
Bose systems. Following the suggestion by Olshanii [39], the exactly solvable Lieb-Liniger
model [40] of bosons was realized in 1D. In the regime of strong interactions, this model ex-
hibits eective fermionization, realizing the famous Tonks-Girirdeau [41] gas of strongly
interacting bosons [42, 43]. As suggested by Wilkin and Gunn [44], physics of quantum
Hall eect with bosons can be explored in fast rotating quantum gases [45]. Feshbach reso-
nances with ultracold fermions can be used to experimentally study [46] the crossover from
molecular BEC of paired fermions to BCS superuid of Cooper pairs. In particular, one
can create imbalanced Fermi mixtures [47, 48] and explore physical regimes not accessible
in traditional condensed matter systems. More comprehensive review of the current state
of aairs in the eld of strongly correlated phenomena with ultracold atoms can be found
in recent reviews [49, 50].
In the rest of this thesis, we will discuss the original work by the author on several
topics of current interest. Chapter 2 discusses the insulating phases of cold spin-one bosonic
particles with antiferromagnetic interactions, such as
23
Na, in optical lattices. We study
magnetic properties of the ground state in the insulating regime. In chapter 3 we consider a
one dimensional interacting Bose-Fermi mixture with equal masses of bosons and fermions,
and with equal repulsive interactions between Bose-Fermi and Bose-Bose particles. Proper-
ties of such mixture are studied using exact Bethe-ansatz techniques. In chapter 4 we discuss
certain phenomena which appear in the experiments with imbalanced fermionic mixtures
in strongly anisotropic traps. In chapter 5 we give a comprehensive review of interference
phenomena, analyzing eects which contribute to the reduction of the interference fringe
contrast in matter interferometers.
Chapter 2
Spin-one bosons in optical lattices
In this chapter we consider insulating phases of cold spin-one bosonic particles with
antiferromagnetic interactions, such as
23
Na, in optical lattices. We show that spin exchange
interactions give rise to several distinct phases, which dier in their spin correlations. In two
and three dimensional lattices, insulating phases with an odd number of particles per site
are always nematic. For insulating states with an even number of particles per site, there is
always a spin singlet phase, and there may also be a rst order transition into the nematic
phase. The nematic phase breaks spin rotational symmetry but preserves time reversal
symmetry, and has gapless spin wave excitations. The spin singlet phase does not break
spin symmetry and has a gap to all excitations. In the presence of magnetic eld we nd a
series of quantum phase transitions between states with xed magnetization (magnetization
plateaus) and a canted nematic phase. In one dimensional lattices, insulating phases with
an odd number of particles per site always have a regime where translational symmetry is
broken and the ground state is dimerized. We discuss signatures of various phases in Bragg
scattering and time of ight measurements.
2.1 Introduction
Modern studies of quantum magnetism in condensed matter physics go beyond
explaining the details of particular experiments on cuprate superconductors, heavy fermion
materials, organic conductors, or related materials, and aim to develop general paradigms
for understanding complex orders in strongly interacting many body systems [51, 52, 53,
54, 55, 56, 57, 58, 59, 60]. Spinor atoms in optical lattices provide a novel realization
of quantum magnetic systems that have several advantages compared to their condensed
matter counterparts, including precise knowledge of the underlying microscopic models, the
possibility to control parameters of the eective lattice Hamiltonians, and the absence of
disorder.
Degenerate alkali atoms are generally considered as a weakly interacting gas due
to the smallness of the scattering length compared to the interparticle separation [26]. The
situation may change dramatically either when atomic scattering length is changed by means
of Feshbach resonances [8], or when an optical potential created by standing laser beams
connes particles in the minima of the periodic potential and strongly enhances the eects
8
Chapter 2: Spin-one bosons in optical lattices 9
of interactions. In the latter case the existence of the nontrivial Mott insulating state of
atoms in optical lattices, separated from the superuid phase by the quantum superuid-
insulator (SI) phase transition, was demonstrated recently in experiments [30, 4, 61]. Low
energy (temperature) properties of spinless bosonic atoms in a periodic optical potential
are well described by the Bose-Hubbard Hamiltonian [3]
H
BH
= t
ij
(a
i
a
j
+a
j
a
i
)
i
n
i
+
U
0
2
i
n
i
( n
i
1), (2.1)
The parameters of (2.1) may be controlled by varying the intensity of laser beams, so one can
go from the regime in which kinetic energy dominates (weak periodic potential, t U
0
), to
the regime where interaction energy is the most important part of the Hamiltonian (strong
periodic potential, t U
0
). For integer llings (number of atoms per lattice site), the two
regimes have superuid and Mott insulating ground states, respectively, as can be obtained
from the mean-eld analysis of the Bose-Hubbard Hamiltonian [3, 2]. In the superuid
phase, atoms are delocalized in the lattice, uctuations in the number of atoms on each site
are strong, and there is a phase coherence between dierent sites. In the insulating state,
atoms are localized, uctuations in the particle number at each site are suppressed, and
there is a gap to all excitations. Such an insulating state represents a correlated many body
state of bosons, where strong interactions between atoms result in a new ground state of
the system.
In conventional magnetic traps, spins of atoms are frozen so eectively that they
behave like spinless particles. In contrast, optically trapped atoms have extra spin degrees
of freedom which can exhibit dierent types of magnetic orderings. In particular, alkali
atoms have a nuclear spin I = 3/2. Lower energy hyperne manifold has three magnetic
sublevels and a total moment S = 1. Various properties of such condensate in a single trap
were investigated [66, 63, 62, 64, 65, 67]. For example, for particles with antiferromagnetic
interactions, such as
23
Na, the exact ground state of an even number of particles in the
absence of a magnetic eld is a spin singlet described by a rather complicated correlated
wave function [66]. However, when the number of particles in the trap is large, the energy
gap separating the singlet ground state from the higher energy excited states is extremely
small, and for the experiments of Ref. [16], the precession time of the classical mean-
eld ground state is of the order of the trap lifetime. So, experimental observation of the
quantum spin phenomena in such systems is very dicult. To amplify quantum spin eects
one would like to have a system with smaller number of particles and stronger interactions
between atoms. Hence it is natural to consider an idea of spin-one atoms in an optical
lattice, in which one can have a small number of atoms per lattice site (in experiments of
Ref. [4] this number was around 1-3) and relatively strong interactions between atoms.
In this chapter we study bosonic spin-one atoms in optical lattices with spin sym-
metric conning potentials and antiferromagnetic interaction between atoms. We demon-
strate that spin degrees of freedom result in a rich phase diagram by establishing the ex-
istence of several distinct insulating phases, which dier from each other by their spin
correlations.
In the insulating state of bosons in an optical lattice uctuations in the particle
number on each site are suppressed but not frozen out completely. Virtual tunneling of
10 Chapter 2: Spin-one bosons in optical lattices
-0.5 0.5 1 1.5 2 2.5 3
1
2
3
4
5
6
N=1
N=2
N=3
N=4
N=5
Nematic
Nematic
Nematic
Nematic
Nematic
Singlet
Singlet
POLAR SUPERFLUID
tunnelling
chemical potential
Figure 2.1: General phase diagram for spin-one bosons in 2D and 3D optical lattice. Detailed
discussion of the phase diagram, including explicit expressions for various phase boundaries,
is given in section 2.6.
atoms between neighboring lattice sites gives rise to eective spin exchange interactions
that determine the spin structure of the insulating states (spin exchange interactions for
S = 1/2 bosons in optical lattices were discussed previously in [68, 69]).
We will show that in two and three dimensional lattices, insulating states with
an odd number of atoms per site are always nematic, whereas insulating states at even
llings are either singlet or spin nematic [70], depending on the parameters of the model.
In one dimensional systems, even more exotic ground states should be realized, including
the possibility of a spin singlet dimerized phase that breaks lattice translational symmetry
[71, 72]. The 2D and 3D general phase diagram, including singlet, nematic and superuid
phases, is shown in Fig. 2.1. The extended version of this diagram, including discussion of
various transition lines, is presented in section 2.6.
It is useful to point out that the lattice model for spin-one bosons, which we
analyze here, is very general and may also be applicable to systems other than cold atoms
in optical lattices. For example, triplet superconductors in strong coupling limit may be
described by a similar Hamiltonian, and some of the phases discussed in this chapter may
correspond to non-BCS states of such superconductors [73].
The chapter is organized as follows. In section 2.2 we provide a derivation of the
Hubbard-type Hamiltonian for spin-one bosons in optical lattices starting from microscopic
interactions between atoms, and describe some general properties of our model. In section
2.3 we derive an eective spin Hamiltonian which is valid for any odd number of atoms
per site, N, in the limit of small tunneling between sites. We demonstrate equivalence
between our system and a Heisenberg model for S = 1 spins on a lattice with biquadratic
interactions, and argue that the ground state is a nematic in two and three dimensions
and is a dimerized singlet in 1D. In section 2.4 we derive eective spin Hamiltonian for a
system with N = 2 atoms per site, valid deep in the insulating regime, and use mean-eld
approximation to determine the phase boundaries between singlet and nematic phases. We
consider the phase diagram in the presence of external magnetic eld. In section 2.5 we
derive eective spin Hamiltonian for the limit of large number of particles per site N >> 1
and small tunneling, and discuss singlet-nematic transition for even N. In the presence
of magnetic eld we nd a series of quantum phase transitions between states with xed
magnetization (magnetization plateaus) and a canted nematic phase. In section 2.6 we
summarize our results and review the global phase diagram for spin-one bosons in optical
lattices. Finally, in section 2.7, we discuss approaches to experimental detection of singlet
and nematic insulating phases of spin-one bosons. We also describe how magnetization
plateaus can be detected. The technical details of calculations are presented in Appendices
A.1-A.4.
2.2 Derivation of Bose-Hubbard model for spin-one particles
At low energies scattering between two identical alkali atoms with the hyperne
spins S = 1 is well described by the contact potential [26]
V (r
1
r
2
) = (r
1
r
2
)(g
0
P
0
+g
2
P
2
), (2.2)
g
F
= 4h
2
a
F
/M. (2.3)
Here P
F
is the projection operator for the pair of atoms into the state with total spin
F = 0, 2, a
S
is the s-wave scattering length in the spin F channel, and M is the atomic
mass. When writing Eq. (2.2) we used the fact that s-wave scattering of identical bosons
in the channel with total spin F = 1 is not allowed by the symmetry of the wave function.
Interaction (2.2) can be written using spin operators as
V (r
1
r
2
) = (r
1
r
2
)(
g
0
+ 2g
2
3
+
g
2
g
0
3
S
1
S
2
). (2.4)
For example, in the case of
23
Na, g
2
> g
0
, and we nd eective antiferromagnetic interac-
tion, as was originally discussed in Refs. [62, 63].
Kinetic motion of ultracold atoms in the optical lattice is constrained to the lowest
Bloch band when temperature and interactions are smaller than the band gap (this is the
limit that we will consider from now on). Atoms residing on the same lattice site have
identical orbital wave functions and their spin wave functions must be symmetric. If we
introduce creation operators, a
i
, for states in the lowest Bloch band localized on site i and
having spin components = 1, 0, 1, we can follow the approach of Ref. [3] and write
the eective lattice Hamiltonian as
H = t
ij,
(a
i
a
j
+a
j
a
i
) +
U
0
2
i
n
i
( n
i
1) +
U
2
2
i
(
S
2
i
2 n
i
)
i
n
i
, (2.5)
where
n
i
=
i
a
i
(2.6)
is the total number of atoms on site i, and
S
i
=
a
i
(2.7)
is the total spin on site i (
are the usual spin-one operators). The rst term in Eq.

(2.5) describes spin symmetric tunneling between nearest-neighbor sites, the second term
describes Hubbard repulsion between atoms, and the third term penalizes non-zero spin
congurations on individual lattice sites for antiferromagnetic interactions. The origin of
this spin dependent term is the dierence in scattering lengths for F = 0 and F = 2 channels
as was discussed in Ref. [66]. Finally, the fourth term in Eq. (2.5) is the chemical potential
that controls the number of particles in the system.
Hamiltonian (2.5) carries important constrains on possible spin states of the sys-
tem. The rst of them derives from the fact that the total spin of a system of N spin-one
atoms cannot be larger than N, so for each lattice site we have
S
i
N
i
. (2.8)
The second constraint is imposed by the symmetry of the spin wave function on each site
S
i
+N
i
= even. (2.9)
Optical lattices produced by far detuned lasers with wavelength
i
= 2/[
k
i
[ create
an optical potential V (r) =
i
V
i
sin
2

k
i
r, with

k
i
being the wave vectors of laser beams.
Using various orientations of beams, one can construct dierent geometries of the lattice.
For the simple cubic lattice, parameters of Hamiltonian (2.5) can be estimated as
U
2
=
2
2
3
E
R
a
2
a
0
x
3/4
,
U
0
=
2
2
3
E
R
a
0
+ 2a
2
x
3/4
,
t =
4
E
R
x
3/4
e
2x
1/2
,
where E
R
= h
2
k
2
/2M is the recoil energy and x = V
0
/E
R
. Note that the ratio U
2
/U
0
is
xed by the ratio of scattering lengths, a
2
/a
0
, for all lattice geometries. Scattering lengths
for
23
Na given in [74] are a
2
= (52 5)a
B
and a
0
= (46 5)a
B
, where a
B
is the Bohr
radius. This corresponds to 0 < a
2
a
0
<< 2a
2
+ a
0
, so the spin dependent part of the
interaction is much smaller than the spin independent one. Throughout this chapter we
will always assume 0 < U
2
<< U
0
. While applying results of this chapter for the case of
23
Na, one should note that errors in the estimation of the exact value of U
2
/U
0
are very
large. While considering the spin structure of Mott insulating phases, we will assume that
8 10 12 14 16
0.5
1
1.5
2
kHz
t
U
2
V
0

E
R
Figure 2.2: U
2
and t for
23
Na atoms in the simple cubic optical lattice created by three
perpendicular standing laser beams with = 985nm. V
0
is the strength of the optical
potential and E
R
= h
2
k
2
/2M is the recoil energy. The ratio of the interaction terms in
(2.5), U
2
/U
0
, is xed by the ratio of the scattering lengths and is independent of the nature
of the lattice (U
2
/U
0
0.04 for
23
Na).
U
2
/U
0
is small enough to see the interplay between tunneling and spin dependent U
2
term
before the superuid-insulator transitions take place. The positions of superuid-insulator
transitions and the validity of this assumption will be discussed in detail in section 2.6. We
will use the value U
2
/U
0
= 0.04 to make estimates of various phase boundaries. In Fig.
2.2 we show U
2
/h and t/h as a function of the strength of the optical potential for a three
dimensional cubic lattice produced by red detuned lasers with = 985nm.
Superuid-insulator transition is characterized by a change in uctuations of par-
ticle numbers on individual lattice sites. When the spin dependent interaction (U
2
) is much
smaller than the usual Hubbard repulsion (U
0
), the superuid - insulator transition is de-
termined mostly by U
0
. The spin gap U
2
term, however, is important inside the insulating
phase, where it competes with the spin exchange interactions induced by small uctuations
in the particle number, and an interesting spin structure of the insulating states appears as
a result of such competition. The spin structure of the insulating phases of spin-one bosons
in optical lattices will be the main subject explored in this chapter.
In what follows we will often nd it convenient to use particle creation operators
that transform as vectors under spin rotations. Such representation may be constructed as
a
z
= a
0
, a
x
=
(a
2
, a
y
= i
(a
+a
2
. (2.10)
Operators a
{x,y,z}
satisfy the usual bosonic commutation relations, and they can be used to
construct spin operators as
S
ia
= ie
abc
a
ib
a
ic
,

S
2
i
= (
bn
bm
n
)a
b
a
n
a
m
. (2.11)
We can verify the transformation properties of a
{x,y,z}
by noting that
[S
a
, a
b
] = [ie
apc
a
p
a
c
, a
b
] = ie
abc
a
c
, (2.12)
[S
a
, a
b
] = [ie
apc
a
p
a
c
, a
b
] = ie
abc
a
c
. (2.13)
Using these operators, we can rewrite the hopping term in the Hamiltonian (2.5)
as
t
<ij>,p{x,y,z}
(a
ip
a
jp
+a
jp
a
ip
),
and it is invariant under global spin rotations. We will use this property later to simplify
calculations and classify eigenstates of eective interactions by the total spin.
2.3 Insulating state with an odd number of atoms
2.3.1 Eective spin Hamiltonian for small t
We start with the insulating state of Hamiltonian (2.5) with an odd number (N =
2n + 1) of bosons per site in the limit t = 0. The number of particles on each site is xed,
and the bosonic symmetry of the wave function requires that the spin in each site is odd.
The interaction term U
2
is minimized when the spins take the smallest possible value S
i
= 1.
In this limit the energy of the system does not depend on the spin orientations on dierent
sites. When t is nite but small, we expect that we still have spin S
i
= 1 in each site, but
that boson tunneling processes induce eective interactions between these spins. In this
section we will compute such interactions in the lowest (second) order in t. We will also
discuss conditions for which our eective Hamiltonian provides an adequate description of
the system.
In the second order perturbation theory in t, we generate only pairwise interactions
between atoms on neighboring sites, so we can write the most general spin Hamiltonian for
S
i
= 1 that preserves spin SO(3) symmetry as
H = J
0
J
1
ij
S
i
S
j
J
2
ij
(
S
i
S
j
)
2
. (2.14)
Here ij) labels near neighbor sites on the lattice. The absence of the higher order terms,
such as (
S
i
S
j
)
3
, follows from the fact that the product of any three spin operators for S = 1
can be expressed via the lower order terms.
To nd the exchange constants J
0,1,2
, we need to consider virtual processes that
create a state with N
i
= 2n, N
j
= 2n + 2, and N
i
= 2n + 2, N
j
= 2n. The dierence in
energy between the intermediate state and low energy S
i
= S
j
= 1 subspace is of order U
0
.
Since our subspace is much lower in energy, the second order perturbation theory is valid.
It is convenient to rewrite the Hamiltonian (2.14) as
H =
0
ij
P
ij
(0) +
1
ij
P
ij
(1) +
2
ij
P
ij
(2), (2.15)
0
= 4J
2
+ 2J
1
J
0
,
1
= J
2
+J
1
J
0
,
2
= J
2
J
1
J
0
. (2.16)
Here P
ij
(S) is a projection operator for a pair of spins on near neighbor sites i and j into
a state with total spin S
i
+S
j
= S (S = 0, 1, 2). The equivalence of Eqs. (2.14) and (2.15)
can be proven by noting simple operator identities for two spin one particles
1 = P
ij
(0) +P
ij
(1) +P
ij
(2),
(
S
i
+

S
j
)
2
= 4 + 2
S
i
S
j
= 2P
ij
(1) + 6P
ij
(2),
(
S
i
+

S
j
)
4
= 16 + 16
S
i
S
j
+ 4(
S
i
S
j
)
2
= 4P
ij
(1) + 36P
ij
(2). (2.17)
Note that states [S
i
= 1, S
j
= 1; S
i
+S
j
= S) have only trivial degeneracy corresponding to
possible projections of total spin S on a xed quantization axis D
S
= 2S + 1.
Since we know a general form of our eective Hamiltonian, we can compute
0,1,2
by
calculating the expectation values of energy for arbitrary states in the appropriate subspaces
S
= t
2
|m
[ < m[(a
ip
a
jp
+a
jp
a
ip
)[S
i
= 1, S
j
= 1; S
i
+S
j
= S > [
2
E
m
E
0
. (2.18)
Here E
0
= 2U
2
is the energy of the conguration with N = 2n + 1 bosons in each of the
two sites, and E
m
is the energy of the intermediate (virtual) states, [m), that have 2n and
2n+2 bosons in the two sites, respectively. Both energies should be computed in the zeroth
order in t.
It is useful to note that the tunneling Hamiltonian is spin invariant; therefore,
intermediate states in summation over [m) in Eq. (2.18) should also have the total spin S.
Another constraint on the possible states [m) comes from the fact that the tunneling term
can only change the spin on each site by 1, since in a Hilbert space of each site operators
a
ip
, a
ip
act as vectors, according to their transformational properties (2.12)-(2.13).
Direct calculations in Appendix A.1 give
0
=
4t
2
(n + 1)(2n + 3)
3(U
0
2U
2
)

16t
2
n(5 + 2n)
15(U
0
+ 4U
2
)
, (2.19)
1
=
4t
2
n(5 + 2n)
5(U
0
+ 4U
2
)
, (2.20)
2
=
28t
2
n(5 + 2n)
75(U
0
+ 4U
2
)

4(15 + 20n + 8n
2
)
15(U
0
+U
2
)
. (2.21)
Combining Eqs. (2.15)-(2.21) we nd
J
0
t
2
=
4(15 + 20n + 8n
2
)
45(U
0
+U
2
)

4(1 +n)(3 + 2n)
9(U
0
+ 2U
2
)
+
128(5 + 2n)
225(U
0
+ 4U
2
)
,
J
1
t
2
=
2(15 + 20n + 8n
2
)
15(U
0
+U
2
)

16(5 + 2n)n
75(U
0
+ 4U
2
)
,
J
2
t
2
=
2(15 + 20n + 8n
2
)
45(U
0
+U
2
)
+
4(1 +n)(3 + 2n)
9(U
0
2U
2
)
+
4n(5 + 2n)
225(U
0
+ 4U
2
)
. (2.22)
1 2 3 4
n
0.7
0.8
0.9
J
1

J
2
Figure 2.3: Ratio J
1
/J
2
for the eective spin Hamiltonian (2.14) for an odd number of
bosons N = 2n + 1 per site. U
2
= 0.04U
0
.
It will turn out that the ratio between J
1
and J
2
determines magnetic ground
state, and the dependence of this ratio on n is quite strong, as shown in Fig. 2.3.
We now discuss limitations of the Hamiltonian (2.14) with parameters given by Eq.
(2.22). In the insulating state with exactly one boson per site, near neighbor interactions
always have the form (2.14). Explicit expressions for the Js given in Eq. (2.22) only
apply in the limit t << U
0
. When t becomes comparable to U
0
(but we are still in the
insulating phase), higher order terms become important, including the possibility of spin
coupling beyond the near neighbor sites. In the insulating state with more than one boson
per site (N = 2n+1, n > 0), we have an additional constraint: we should be able to neglect
congurations with spins on individual sites higher than 1. Matrix elements for scattering
into such states are of the order of (Nt)
2
/U
0
(see Eq. (2.22)), and their energy is set by
U
2
. Therefore, the Hamiltonian (2.14) applies only when Nt << (U
0
U
2
)
1/2
, which is well
within the insulating state when U
2
<< U
0
(SI transition takes place for Nt U
0
).
2.3.2 Phase diagram
To understand the nature of the Hamiltonian (2.14) in the relevant regime of
parameters J
2
> J
1
> 0, it is useful to start by considering a two-site problem
H
12
= J
1
S
1
S
2
J
2
(
S
1
S
2
)
2
(2.23)
with S
1
= S
2
= 1. Eigenstates of Eq. (2.23) can be classied according to the value of
the total spin S
tot
, and their energies may be computed using 2
S
1
S
2
= S
tot
(S
tot
+ 1) 4.
Two spin one particles can combine into S
tot
= 0, 1, and 2. The J
1
term in Eq. (2.23)
favors maximizing

S
1
S
2
by polarizing

S
1
along the direction of

S
2
, so that S
tot
= 2. By
contrast, the J
2
term favors maximizing (
S
1
S
2
)
2
by forming a singlet state S
tot
= 0 (see
Table 2.1). So, the latter term acts as an eective antiferromagnetic interaction for this
spin one system, and it dominates for J
2
> J
1
. If we go beyond a two site problem and
consider a large lattice, we see that each pair of near neighbor sites wants to establish a
singlet conguration when J
2
> J
1
. However, because one cannot form singlets on two
S
tot

S
1
S
2
(
S
1
S
2
)
2
Energy
0 -2 4 2J
1
4J
2
1 -1 1 J
1
J
2
2 1 1 J
1
J
2
Table 2.1: Eigenstates of a two site problem (2.23).
dierent bonds that share the same site, some interesting spin order, whose precise nature
will depend on the lattice and dimensionality, will appear.
Phase diagram for D = 1
From the discussion above we see the conict intrinsic to the Hamiltonian (2.14):
each bond wants to have a singlet spin conguration, but singlet states on the neighboring
bonds are not allowed. There are two simple ways to resolve this conict:
A) Construct a state that mixes S = 0 and S = 2 on each bond but can be repeated on
neighboring bonds;
B) Break translational symmetry and create singlets on every second bond.
At the mean-eld level, solution of the type A is given by
[N) =
i
[S
i
= 1, m
i
= 0). (2.24)
This can be established by noting that for any neighboring pair of sites we indeed have a
superposition of S = 0 and S = 2 states
[S
i
= 1, m
i
= 0)[S
j
= 1, m
j
= 0) =
1
3
[S
tot
= 0) +
_
2
3
[S
tot
= 2, m
tot
= 0). (2.25)
State (2.24) describes a nematic state that has no expectation value of any component of
the spin S
x,y,z
i
= 0), but spin symmetry is broken since (S
x
i
)
2
) = (S
y
i
)
2
) = 1/2 and
(S
z
i
)
2
) = 0. It is useful to point out the similarity between wave function (2.25) that mixes
singlet and quintet states on each bond, and a classical antiferromagnetic state for spin
1/2 particles that mixes spin singlets and triplets on each bond. Colemans theorem [75]
(the quantum analog of Mermin-Wagner theorem) forbids the breaking of spin symmetry
in D = 1, even at T = 0. However, a spin singlet gapless ground state that has a close
connection to the nematic state (2.24) has been proposed in Refs. [77, 76] for J
2
close to
J
1
.
The simplest way to construct a solution of type B is to take
[D) =
i=2n
[S
i
= 1, S
i+1
= 1, S
i
+S
i+1
= 0). (2.26)
Such a dimerized solution has exact spin singlets for pairs of sites 2n and 2n +1, but pairs
of sites 2n and 2n 1 are in a superposition of S = 0, 1, and 2 states.
According to the variational wave functions (2.24) and (2.26), the dimerized so-
lution becomes favorable over a nematic one only for J
2
/J
1
> 3/2 in D = 1. However,
numerical simulations [78] showed that for J
2
> J
1
, the ground state is always dimerized.
It is a spin singlet and has an energy gap to all spin excitations. This means that the
variational wave function (2.26) may only be taken as a caricature of the true ground state,
although it captures such key aspects of it, such as broken translational symmetry and the
absence of spin symmetry breaking.
Phase diagram for D = 2, 3
The nematic state for the Hamiltonian (2.14) in a simple cubic lattice (D = 3) for
J
2
> J
1
has been discussed using mean-eld approximations [79], semiclassical approaches
[80], and numerical methods [81]. Finally, recent work of Tanaka et.al [82] provided a
rigorous proof of the existence of the nematic order at least in some part of this region,
which satises 2.66J
1
> J
2
2J
1
. The variational state for the nematic order may again
be given by equation (2.24) and its mean eld energy is E
MF
N
= 2J
2
. It is important to
emphasize, however, that the actual ground state is suciently dierent from its mean-eld
version (2.24). It is possible to write down dimerized states with energy expectation lower
than 2J
2
; however, numerical results [81] suggest that the ground state doesnt break
translational symmetry. A way to obtain a more precise ground state wave function is to
include quantum uctuations near the mean eld state, as was done in Ref. [80]. Hence, the
mean-eld wave function (2.24) does not provide a good approximation of the ground state
energy of the nematic state. Nevertheless, it is useful for the discussion of order parameter
and broken symmetries of the nematic state.
In the nematic state, spin space rotational group O(3) is broken, though time
reversal symmetry is preserved. The order parameter for the nematic state is a tensor
Q
ab
= S
a
S
b
)

ab
3
S
2
). (2.27)
In the absence of ferromagnetic order S
a
S
b
) = S
b
S
a
); hence, Q
ab
is a traceless symmetric
matrix. The minimum energy of (2.14) is achieved for Q
ab
that has two identical eigenvalues,
which corresponds to a uniaxial nematic [83].
Then, the tensor Q
ab
can be written using a unit vector

d as
Q
ab
= Q(d
a
d
b
1
3
ab
). (2.28)
Vector

d is dened up to the direction (i.e.
d are equivalent) and corresponds to the

director order parameter [83]. For the mean-eld state (2.24), the director

d can also be
dened from the condition that locally our system is an eigenstate of the operator

d
S with
eigenvalue zero. However, such a denition may not be applied generally.
The nematic phase behaves in many aspects as antiferromagnetic [84], the direction
of

d being analogous to staggered magnetization. Namely in weak magnetic elds,

d aligns
itself in the plane perpendicular to magnetic eld, and spin-wave excitations have linear
dispersion [76], with velocity
c =
_
2zJ
2
(J
2
J
1
).
Nematic phases for spin-one lattices have been considered before in literature [77, 82, 81,
78, 80, 84, 79, 76], so we will not discuss them here more extensively.
2.4 Insulating states with two atoms per site
In this section we consider an insulating state of two bosons per site. Possible spin
values for individual sites are S = 0 and S = 2. In the limit t = 0, the interaction part of
the Hamiltonian (the U
2
term) is minimized when S = 0. The amplitude for creating S = 2
states, as well as the exchange energy of the latter, is of the order of t
2
/U
0
. So, when t is
of the order of (U
0
U
2
)
1/2
or larger, we may no longer assume that we only have singlets
in individual sites, and we need to include S = 2 congurations in our discussion. This
regime is still inside the insulating phase for small enough U
2
/U
0
(the superuid-insulator
transition takes place for zt U
0
). In this section we will assume that U
2
/U
0
is small
enough, so that S = 2 becomes important in the insulating phase, before the transition
to superuid. More careful consideration of superuid transition line and comparison with
the case of
23
Na will be presented in section 2.6. In section 2.4.1 we exactly solve the
problem for two sites. In section 2.4.2 we derive an eective Hamiltonian that takes into
account competition between spin gap of individual sites, that favors S = 0 everywhere,
and exchange interactions between neighboring sites that favor proliferation of S = 2 states.
Mean eld solution of the eective magnetic Hamiltonian is considered in section 2.4.3 and
we nd rst order quantum phase transition from isotropic to nematic phase. We discuss
collective excitations in sections 2.4.4 and 2.4.5 and the eects of magnetic eld in section
2.4.6. We note that the state with N = 2 has an advantage over states with higher N from
an experimental point of view since it has no three-body decays.
2.4.1 Two site problem: exact solution
To construct an eective magnetic Hamiltonian for this system, we note that in
the second order in t it can be written as a sum of interaction terms for all near neighbor
sites (identical for all pairs of sites). These pairwise interactions can be found by solving
a two site problem and nding the appropriate eigenvalues and eigenvectors in the second
order in t.
The Hilbert space for two sites with two atoms at each site is given by the direct
sum of the following subspaces:
[E
1
> = [N
1
= 2, N
2
= 2, S
1
= S
2
= 0, S
1
+S
2
= 0 >,
[E
2
> = [N
1
= 2, N
2
= 2, S
1
= S
2
= 2, S
1
+S
2
= 0 >,
[E
3
> = [N
1
= 2, N
2
= 2, S
1
= 0, S
2
= 2, S
1
+S
2
= 2 >,
[E
4
> = [N
1
= 2, N
2
= 2, S
1
= 2, S
2
= 0, S
1
+S
2
= 2 >,
[E
5
> = [N
1
= 2, N
2
= 2, S
1
= S
2
= 2, S
1
+S
2
= 2 >,
[E
6
> = [N
1
= 2, N
2
= 2, S
1
= S
2
= 2, S
1
+S
2
= 1 >,
[E
7
> = [N
1
= 2, N
2
= 2, S
1
= S
2
= 2, S
1
+S
2
= 3 >,
[E
8
> = [N
1
= 2, N
2
= 2, S
1
= S
2
= 2, S
1
+S
2
= 4 > . (2.29)
Hopping term in Eq. (2.5) conserves total spin; therefore, energy in each subspace
doesnt depend on the z component of S
1
+ S
2
, and states [E
6
>, [E
7
>, and [E
8
> form
orthogonal subspaces that do not mix with any other states. We can then use formula
analogous to Eq. (2.18) to calculate corrections to the energies of these states in the second
order in t (see Appendix A.2 for details):
6
= 6U
2
4
t
2
U
0
,
7
= 6U
2
4
t
2
U
0
,
8
= 6U
2
12
t
2
U
0
. (2.30)
In Eq. (2.30) we used U
2
<< U
0
and neglected U
2
relative to U
0
in denominators of the
exchange terms.
Boson tunneling can connect two subspaces, [E
1
> and [E
2
>, and three subspaces,
[E
3
>, [E
4
>, and [E
5
> (only states with the same component of S
z
have tunneling matrix
elements). Thus, the energies and eigenstates should be found by diagonalizing the matrix
H
= E
0
t
2
|m
< [(a
1p
a
2p
+a
2p
a
1p
)[m >< m[(a
1p
a
2p
+a
2p
a
1p
)[ >
E
m
E
0
. (2.31)
Here E
0
is the energy of the state in the zeroth order in t: E
01
= 0, E
02
= 6U
2
,E
03
=
3U
2
,E
04
= 3U
2
, E
05
= 6U
2
. Summation over [m) covers intermediate states that have 1 and
3 bosons in the two sites, have the same total spin as states [) and [), and have spins in
individual sites that dier from spins of states [) and [) by 1. The explicit calculation
presented in Appendix A.2 gives the matrix H
and its eigenvalues. Energies of the states

with total spin zero are
1
= 3U
2
6t
2
U
0
(3U
2
6t
2
U
0
)
2
+ 40
U
2
t
2
U
0
,
2
= 3U
2
6t
2
U
0
+
(3U
2
6t
2
U
0
)
2
+ 40
U
2
t
2
U
0
. (2.32)
Energies of the states with S = 2 are

3
= 3U
2
4
t
2
U
0
,

4
=
1
2
_
9U
2
12t
2
+
144
t
4
U
2
0
+ 40
t
2
U
0
U
2
+ 9U
2
2
_
,

5
=
1
2
_
9U
2
12t
2
144
t
4
U
2
0
+ 40
t
2
U
0
U
2
+ 9U
2
2
_
, (2.33)
and each of these states is vefold degenerate. Energies
1
-
8
are shown in Fig. 2.4.
0.05 0.1 0.15 0.2
t
-0.4
-0.3
-0.2
-0.1
0.1
0.2
Energy
e1
e2
e3
e4
e8
e6=e7
e5
Figure 2.4: Eigenstates of the eective spin Hamiltonian for a two site problem with two
atoms per site. Energy and t are measured in units of U
0
, and we assumed U
2
= 0.04U
0
.
The lowest energy states correspond to total spin S = 0 (e1), S = 2 (e5), and S = 4 (e8).
For U
2
> 0 the lowest energy state is a total spin singlet that has some mixture of
S = 2 states on individual sites when t is nonzero. The next favorable state has total spin
2, [E
5
). When the value of t is increasing, the ferromagnetic state [E
8
) becomes the third
low lying state. At this point, we have solved the problem for two sites, taking into account
competition between the hopping and the spin dependent interaction (overall Hilbert space
for two sites is 36 dimensional).
2.4.2 Eective spin Hamiltonian for an optical lattice
In the previous subsection we used perturbation theory in tunneling t to study the
problem of two sites with two atoms at each site. If we label the two sites 1 and 2, in the
second order in t the eective Hamiltonian can be written as
H
12
= 3U
2
[P(S
1
= 2) +P(S
2
= 2)] +[ >
1
[ >
2
J
,;,
< [
1
< [
2
, (2.34)
Here P(S
{1,2}
= 2) are projection operators into states with spin S = 2 on sites 1 and 2 and
J
,;,
gives exchange interactions that arise from virtual tunneling processes into states
with particle numbers (n
1
= 1, n
2
= 3) and (n
1
= 3, n
2
= 1). The second term of (2.34)
includes all initial states ([)
1
and [)
2
for sites 1 and 2, respectively) and all nal states
([)
1
and [)
2
).
Generalization of the eective spin Hamiltonian (2.34) for the case of optical lattice
is obviously
H = 3U
2
i
P(S
i
= 2) +
<ij>
[ >
i
[ >
j
J
,;,
< [
i
< [
j
, (2.35)
This Hamiltonian is linear in U
2
and, therefore, can be written as a sum of the bond terms
H =
ij
H
ij
,
H
ij
=
3U
2
z
[P(S
i
= 2) +P(S
j
= 2)]
+ [ >
i
[ >
j
J
,;,
< [
i
< [
j
. (2.36)
Individual terms

H
ij
dier from (2.34) only by rescaling U
2

U
2
z
, where z is the coordina-
tion number of the lattice. Here we did not give explicit expressions for J
,;,
in the basis
of eigenstates of individual spins S
i
and S
j
but in the basis of eigenstates of the total spin
of the pair (2.29), expressions for J
,;,
can be obtained from eigenstates and eigenvalues
of Eq. (2.34) (see Eqs. (2.30) -(2.33) and Appendix A.2, with rescaled U
2
). Therefore, we
can write
H
ij
=
,,S
z
[E
ij
, S
z
>

H
< E
ij
, S
z
[, (2.37)
where states [E
ij
, S
z
> have been dened in equations (2.29). Expressing [E
ij
, S
z
> via
states [N
i
= 2, S
i
= 0, 2, S
iz
= S
i
...S
i
> using known Clebsch-Gordon coecients
[E
ij
, S
z
>=
iz
S
jz
C
S
,S
z
S
i
,S
iz
;S
j
,S
jz
[N
i
= 2, S
i
, S
iz
> [N
j
= 2, S
j
, S
jz
>,
we can write the Hamiltonian (2.35) as
H =
<ij>
[ >
i
[ >
j
H
,;,
< [
i
< [
j
, (2.38)
where states [) - [) belong to the set S = 0, S = 2, S
z
= 2, ..., 2, and H
,;,
is
given by proper rotation of

H
,
.
2.4.3 Phase diagram from the mean-eld calculation
In this section we study the phase diagram of the system described by the Hamil-
tonian (2.38) using translational invariant variational wave functions. Such mean-eld ap-
proach gives correct ferromagnetic and antiferromagnetic states for Heisenberg Hamiltonians
in d 2, so we expect it to be applicable in our case. We think that this approach suc-
cessfully captures the main features of the system: rst order transition between the spin
gapped and the nematic phases, the nature of the order parameter in the nematic phase,
and elementary excitations in both phases. However, we cannot rule out the possibility
of more exotic phases that fall outside of our variational wave functions, for example, the
dimerized phase discussed in Ref. [71]. Numerical calculations are required to study if such
phases will actually be present.
As we saw in the previous section, the energy of the two-site problem is minimized
when total spin is 0. However, energy on all bonds cannot be minimized simultaneously, so
we cannot solve a problem exactly for a lattice. We use a mean eld approach to overcome
this diculty, taking variational wave function
[ >=
i
(c
0,0
[N
i
= 2, S = 0, S
z
= 0 > +
m=2,...,2
c
2,m
[N
i
= 2, S = 2, S
z
= m >), (2.39)
[c
0,0
[
2
+
m
[c
2,m
[
2
= 1. (2.40)
Now we can evaluate expectation value of energy over variational state (2.39) and nd the
ground state numerically.
We parameterize conditions (2.40) as
c
0,0
= cos , c
2,m
= sina
m
, (2.41)
m=2,...,2
[a
m
[
2
= 1.
In Appendix A.3 we demonstrate that for which minimizes the mean-eld energy, up to
SU(2) rotations [a
m
] has the form
1
,
[a
m
] = (0, 0, 1, 0, 0)
T
. (2.42)
Mean eld energy does not depend on

d and we nd in the region of interest the
energy per lattice site to be
E[] = 3U
2
sin
2
+
zt
2
12U
0
(51 + 4 cos 2 + 7 cos 4 8
2 sin 2 + 4
2 sin 4). (2.43)

One can immediately see that if we try to expand this expression near = 0, there is no
linear term, but second and third order terms are present. This indicates that by changing
the parameters of the Hamiltonian, we will have a rst order quantum phase transition, at
which the value of that minimizes the energy changes discontinuously. This is typical for
ordinary nematics [83] since in Landau expansion third order terms are not forbidden by
d symmetry. The reason why our transition is of the rst order can be traced back
to the fact that mean eld energy has terms which mix c
0,0
and c
2,m
in odd powers, i.e.
c
0,0
c
2,2
(c
2,1
)
2
, so overall U(1) symmetry doesnt prohibit odd powers of in functional
(2.43).
Since phase transition is of the rst order, the transition is characterized by several
regimes. First, when t is small, global energy minimum is at = 0 and there are no other
local minima, i.e. we have spin singlets in all individual sites. Then, when condition
zt
2
/(U
0
U
2
) 0.4928 is satised, a local minimum appears at
0.25, see Fig. 2.5. As

we continue increasing t, the minimum at nonzero becomes deeper, and eventually at
zt
2
c
/(U
0
U
2
) = 1/2 the global minimum of functional (2.43) is reached for sin
c
= 1/3, (see
Fig.2.6). However, there is still a local minimum at = 0. If we keep increasing t, the
1
Expression (2.42) describes an eigenstate of S
z
with eigenvalue zero. After an SU(2) rotation we will
have [am] that is an eigenstate of

d
S with zero eigenvalue. Vector

d corresponds to the direction of uniaxial
nematic.
0.1 0.2 0.3 0.4 0.5 0.6
-3.284
-3.282
-3.278
-3.276
-3.274
-3.272
Energy
Figure 2.5: Dependence of the energy functional (2.43) on when zt

2
/(U
0
U
2
) 0.4928
(energy is given per lattice site in units of U
2
).
minimum at = 0 becomes completely unstable at zt
2
+
/(U
0
U
2
) = 9/16 and there is only
one minimum at
+
0.5 (see Fig. 2.7). As we increase t further, sin
+
continues to grow,
approaching the value sin
= (2/3)
1/2
. It is useful to point out that when t is changed
in experiments (e.g. by changing the strength of the optical potential [85]), we expect that
the system will not switch between the singlet and nematic phases at t
c
, but will remain
in the appropriate metastable local minimum until it becomes completely unstable. So,
in experiments with increasing t, the transition from the singlet to the nematic states will
occur at t
+
, and in experiments with decreasing t, the transition from the nematic to the
singlet state will take place at t
. We note, however, that the dierence between dierent

t is quite small.
In Fig. 2.9 we show the phase diagram for the insulating phase with N = 2,
including a true rst order transition line at t
c
and limits of metastability at t
and t
+
.
The Superuid-Insulator transition line is shown as an eye guide for the case of small enough
U
2
/U
0
; its exact position will be presented in section 2.6. It is useful to point out that in
the discussion above we used canonical ensemble (xed number of particles) rather than
grand canonical ensemble (xed chemical potential) to discuss the singlet to nematic phase
transition. However, intermediate states that contribute to exchange interactions always
involve one particle and one hole. Hence, their energy does not depend on the chemical
potential. This explains why the singlet to nematic phase boundary in Fig. 2.9 does not
depend on . It is consistent with our physical intuition that insulating states have a certain
number of particles, but their chemical potential is not well dened as long as is inside
the Mott gap. In the discussion presented in this section, we assumed that the system
remains deep in the insulating phase and the superuid to insulator transition does not
preempt the isotropic to nematic transition inside the insulating lobe. Precise conditions
under which this is justied will be given in Section 2.6.
0.1 0.2 0.3 0.4 0.5 0.6
-3.332
-3.328
-3.326
-3.324
-3.322
Energy

2
/(U
0
U
2
) = 1/2 (energy
is given per lattice site in units of U
2
).
0.1 0.2 0.3 0.4 0.5 0.6
-3.88
-3.86
-3.84
-3.82
-3.78
-3.76
Energy

2
/(U
0
U
2
) = 9/16 =
0.5625 (energy is given per lattice site in units of U
2
).
Increasing t
global
transition
decreasing t
0.66 0.68 0.7 0.72 0.74 0.76 0.78 0.8
t
0.1
0.2
0.3
0.4
0.5
Sin
Figure 2.8: Dependence of sin in equation (2.43) on t, measured in units of
_
U
0
U
2
z
.
t t
c
t
+
0.5 1 1.5 2
t
1.2
1.4
1.6
1.8
2
Singlet Nematic
First order phase transition
Figure 2.9: Mott phase diagram for N = 2. Chemical potential is measured in units
of U
0
, t is measured in units of
_
U
0
U
2
z
. Critical tunneling t
c
marks the actual rst order
phase transition, while t
and t
+
correspond to limits of metastability. Superuid-Insulator
transition line is presented as an eye guide, and will be discussed in section 2.6.
2.4.4 Quantum uctuations corrections for the spin singlet state
For small enough t, mean-eld analysis of the previous section predicts the singlet
ground state that does not depend on t. Now we will consider quantum uctuations near
this state to obtain more accurate wave function and excitation spectra. We can rewrite
Eq. (2.38) via Hubbard operators
A
i
= [)
i
[
i
. (2.44)
Here [)
i
and [)
i
belong to the set S = 0, S = 2, S
z
= 2, ..., 2. Commutation relations
between A
i
are very simple:
[A
i
, A
i
] =
i
. (2.45)
Now we introduce boson operators b
i
, that create states with S
i
= 2, S
iz
= , and c
, that
creates a singlet on ith site. Our physical subspace is smaller than a generic Fock space of
these bosons and should satisfy the condition
c
i
c
i
+
i
b
i
= 1 (2.46)
on each site.
One can easily check that if we set A
= b
for spin S = 2 states and similar

substitution with c bosons when one of the states is a singlet state (which we will denote as
s), then commutation relations (2.45) are satised. Since for small enough t only a singlet
state is occupied in mean eld approximation, we can resolve constraint (2.46) using an
analog of Holstein-Primako representation near c
c = 1 state [80], which is given by

A
i
= b
i
b
i
, A
s
i
= (1 b
i
b
i
)
1/2
b
i
, (2.47)
A
ss
i
= 1 b
i
b
i
, A
s
i
= b
i
(1 b
i
b
i
)
1/2
. (2.48)
Now we expand our initial Hamiltonian in terms of now independent operators b
i
up to the
second order:
H
(2)
=
<ij>
(H
;ss
b
i
b
j
+H
ss;
b
i
b
j
+H
s;s
b
i
b
j
+H
s;s
b
j
b
i
) +
+
z
2
i
(H
s;s
b
i
b
i
+H
s;s
b
i
b
i
+H
ss;ss
(1 b
i
b
i
b
j
b
j
)). (2.49)
Calculation of matrices H
;
gives necessary matrix elements
H
s;s
= H
s;s
=
(
20
3
t
2
U
0
+
3U
2
z
),
H
ss;ss
=
20
3
t
2
U
0
,
H
s;s
= H
s;s
=
8
3
t
2
U
0
,
H
;ss
= H
ss;
=
8
3
t
2
U
0
_
_
0 0 0 0 1
0 0 0 1 0
0 0 1 0 0
0 1 0 0 0
1 0 0 0 0
_
_
.
We rewrite our Hamiltonian in terms of Fourier transforms of b
i
operators, b
k
, b
k
:
H
(2)
= E
0
+
z
2
k
(H
;ss
b
k
b
k
+H
ss;
b
k
b
k
+H
s;s
b
k
b
k
+H
s;s
b
k
b
k
)+
+2(H
s;s
H
ss;ss
)b
k
b
k
, (2.50)
where E
0
is classical energy and
k
=
1
z
e
e
ike
. (2.51)
Now we will use canonical Bogoliubov transformations to diagonalize this Hamiltonian.
Since most of the terms are diagonal in , subspace, it is easy to see that required
transformation mixes operators b
0k
, b
0k
, b
1k
, b
1k
, b
1k
, b
1k
, b
2k
, b
2k
, and
b
2k
, b
2k
.
Transformation that mixes the rst pair is
b
0k
= cosh
k
0k
+ sinh
k
0k
,
b
0k
= cosh
k
0k
+ sinh
k
0k
,
and its complex conjugates. Substituting this transformation into Eq. (2.50) and requiring
that terms with
0k
0k
and
0k
0k
vanish, we obtain the equation for
k
tanh 2
k
=
g
k
f
k
=
8
3
zt
2
U
0
k
3U
2
8
3
zt
2
U
0
k
,
where
g
k
=
8
3
zt
2
U
0
k
,
f
k
= 3U
2
8
3
zt
2
U
0
k
.
Energy of this excitation equals
E(k) =
_
f
2
k
g
2
k
=
U
2
(9U
2
16
zt
2
U
0
k
). (2.52)
Equation (2.52) suggests that the rst instability appears at k = 0 and gives the
phase boundary that agrees with the metastability line t
+
found in the previous subsection.
0.2 0.4 0.6 0.8 1
t
0.5
1
1.5
2
2.5
3
Gap
Figure 2.10: Dependence of excitation gap (measured in units of U
2
) on t, measured in units
of
_
U
0
U
2
z
.
However, the results of the previous subsection suggest that the phase transition is of the
rst order and takes place before the mode softening at k = 0. The rst order transition
may also be obtained with the formalism presented in this section by noting that expansion
of Eq. (2.48) allows third order terms cb
+ c.c..
We can use similar analysis to discuss excitations with other spin quantum num-
bers. For example, excitations with S
z
= +1, 1 are diagonalized by analogous Bogoli-
ubov transformations with
k

k
, and excitations with S
z
= +2, 2 are diagonalized
with transformations with the same
k
. As required by the spin symmetry of the singlet
state, all of these excitations have the same energy.
Now we can discuss the approximations made while expanding over b
k
, b
k
. While
transformation (2.48) is the exact resolution of the constraint (2.46), expansion to the second
order adds states with higher boson occupation numbers and changes Hilbert space (this is
completely analogous to usual antiferromagnet spin-wave theory). However, if a posteriori
we can verify that only states with occupation numbers n
i
= 0, 1 are present in the
ground state, then expansion of the constraint (2.46) up to the second order was justied.
The parameter that controls such expansion is
b
i
b
i
=
1
N
k
b
k
=
_
sinh
2
k
d
d
k
(2)
d
.
Calculation of this quantity while the singlet state is still a global maximum for D = 3
gives numerical values < 0.001; therefore, our expansion is much more precise than for
Heisenberg antiferromagnet, where this quantity is not much smaller than 1 and one needs
the condition S 1 to justify the spin wave theory.
2.4.5 Spin wave excitations in the nematic phase
Now we will consider excitations for the states with nematic order. For the states
described by Eqs. (2.39) - (2.42), there are no expectation values of the spin operators
S) = 0, but there is a nematic order (2.27) when ,= 0. For example, when

d is pointing
along z we nd
Q
ab
= sin
2
_
_
1 0 0
0 1 0
0 0 2
_
_
(2.53)
In the singlet phase, expectation values of both the spin operators are
S) = 0 and Q
ab
) = 0.
In the singlet phase the system has a gap to all excitations of order U
2
, while nematic
phases have gapless spin-wave excitations that originate from the breaking of the continuous
symmetry. The general form of the state with minimum energy is expressed via Euler angles
of order parameter

d as
U
z
()U
y
()U
z
()0, 0, 1, 0, 0
T
,
where U() are nite angle rotation matrices. From Eqs. (2.53) and (2.28) we can express
the nematic order parameter for such a state as
Q
ab
= 3 sin
2
(d
a
d
b
1
3
ab
).
Goldstone theorem tells us that low lying modes are given by uctuations of the direction
of

d, and that there will be two degenerate modes. We can utilize the approach used in
the previous subsection to consider excitations in the nematic phase. For that we should
make a generalized Holstein-Primako expansion near the nematic state. First, we make a
unitary transformation in Hilbert subspace of each site, which is given by
_
_
_
_
_
_
_
_
[0)
[1)
[2)
[3)
[4)
[5)
_
_
_
_
_
_
_
_
=
_
_
_
_
_
_
_
_
_
cos 0 0 sin 0 0
0 0
1
2
0
1
2
0
0 0
1
2
0
1
2
0
sin 0 0 cos 0 0
0
1
2
0 0 0
1
2
0
1
2
0 0 0
1
2
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
[S = 0)
[S
z
= 2)
[S
z
= 1)
[S
z
= 0)
[S
z
= 1)
[S
z
= 2)
_
_
_
_
_
_
_
_
. (2.54)
Making appropriate transformation on H
;
, we can write our Hamiltonian as
H =
<ij>
[ >
i
[ >
j

H
,;,
< [
i
< [
j
, (2.55)
where states [) - [) belong to the set [0) [5). After that, we proceed exactly as in
the previous subsection, expanding near [0) state. Since dependence on t
2
/(U
0
U
2
) is
determined by the minimization of the energy, linear terms in b
k
and b
k
are absent.
Quadratic terms have exactly the same form as in Eq. (2.49), and all matrices become
diagonal due to the proper basis choice (2.54). Now we can use Bogoliubov transformation
to diagonalize the quadratic part. For excitations to states [1) and [2), we obtain energies
E
2
1
(k) = E
2
2
(k) =
1
36
(16
2
k
z
2
t
4
U
2
0
(4 cos 2 +
2 sin2)
2
+
0.72 0.74 0.76 0.78 0.8 0.82 0.84
t
2.8
2.85
2.9
2.95
3.05
3.1
Velocity
M
e
t
a
s
t
a
b
l
e
Figure 2.11: Dependence of the spin wave velocity (measured in units of U
2
/
z) on t,
measured in units of
_
U
0
U
2
z
.
(9
zt
2
U
0
18
k
zt
2
U
0
+ 9U
2
+ (2(
k
1)
zt
2
U
0
+ 9U
2
) cos 2 7
zt
2
U
0
cos 4
+4
2(1
k
)
zt
2
U
0
sin 2 4
2
zt
2
U
0
sin 4)
2
),
where
k
was dened in Eq. (2.51), and dependence of on zt
2
/(U
0
U
2
) is shown in Fig.
2.8. We nd that for k = 0, energies of these excitations are zero, as expected for nematic
waves from Goldstone theorem. These excitations create states with S
z
= 1. For small

k,
the energy of excitations depends linearly on [
k[, and dependence of the spin wave velocity

on the parameters of the lattice is shown in Fig. 2.11.
Let us now consider gapped excitations for the nematic phase. Excitation to the
state [3) corresponds to longitudinal uctuations in the value of , and the energy of such
excitations becomes zero at t
since at this point uctuations of are not suppressed.

Excitations to the states [4) and [5) correspond to the creation of S
z
= 2 states and they
are degenerate. For all of these excitations, energies are minimized for

k = 0. Dependence
of the gap on parameters is shown in Fig. 2.12.
2.4.6 Magnetic eld eects
Let us now consider the eects of a magnetic eld, H U
2
, on our system. For
U
2
in the range of kHz (see Fig. 2.2), this corresponds to magnetic elds of the order of
1mG. This eld is small enough that it does not change the scattering lengths due to the
energy level shifts inside of atoms (Feshbach resonances). Since all atoms have the same
gyromagnetic ratio, interaction with external magnetic eld depends only on the total spin,
and the internal structure of the states is not important.
Let us rst consider the case of small magnetic elds, H << U
2
. In the case of a
nematically ordered insulating state, the ground state energy doesnt have any contributions
linear H (this follows from < 0[HS[0 >= 0), and the second order contribution depends on
the relative orientation of the nematic order parameter

d and magnetic eld

H. Suppose
0.72 0.73 0.74 0.75
t
1
2
3
4
Gap
M
e
t
a
s
t
a
b
l
e
longitudinal excitation
S
z
=2 excitation
Figure 2.12: Energy of the gapped excitations in the nematic phase of N = 2 atoms per
site at zero wave vector. The energy gaps and t are measured in units of U
2
and
_
U
0
U
2
z
,
respectively.
d is directing along the z axis and

H lies in x, z plane. In the second order perturbation
theory, energy correction to the ground state is always non positive:
E
(2)
=
En
< 0[HS[n >< n[HS[0 >
E
n
E
0
=
En
(H
x
)
2
< 0[S
x
[n >< n[S
x
[0 >
E
n
E
0
< 0;
this quantity is of order (H
x
)
2
/t
2
. Since for the state with

d[[
H it is again zero and

the system doesnt benet from magnetic eld, energy is minimized when

d lies in a plane
perpendicular to

H (this is completely analogous to antiferromagnets). Using this property,
one can distinguish a nematic phase from a singlet phase. One should apply a small magnetic
eld in z direction to x the plane in which

d lies, release the trap, and let the atoms fall
in the gravitational eld with some magnetic gradient (to separate the states with dierent
S
z
). Then, one should measure quantities of each spin component. These values will have
a sharp change when we cross the rst order phase transition line. Since we know how
to express spin states via original boson operators, we can calculate expectation values of
dierent spin components to be:
n
1
= n
1
=
2
3
cos[]
2
+
2
3
cos[] sin[] +
5
6
sin[]
2
,
n
0
=
2
3
cos[]
2
2
3
cos[] sin[] +
1
3
sin[]
2
.
Using known expressions for dependence of on t, we can make mean-eld predictions on
occupation numbers, shown in Fig. 2.13
An interesting question to consider now is the eect of magnetic elds of the order
of H U
2
on the phase diagram. We have found out earlier that singlet - nematic transition
is weakly rst ordered, and mode softening analysis is quite reliable. Thus stability region
of the singlet phase can be estimated by looking at the mode softening in the presence of
0.5 1.5 2 2.5
t
0.2
0.4
0.6
0.8
1
n
1
=n
-1
n
0
Metastable transitions
Figure 2.13: Dependence of occupation numbers n
0
and n
1
= n
1
on t for an insulating
state with two bosons per site, N = 2. Tunneling t is measured in units of
_
U
0
U
2
z
.
magnetic elds. Energies of the excitations out of singlet phase have been calculated earlier,
see Eq. (2.52). Out of these excitations, the mode with spin S
z
= +2 becomes unstable
rst, which denes the boundary of the singlet phase as
H =
1
2
U
2
(9U
2
16
zt
2
U
0
).
We note that mode softening analysis is exact for t = 0.
If t = 0 and H > 3U
2
/2, the global ground state is ferromagnetic. Using the same
approach, we can also nd the instability of the ferromagnetic phase. Ferromagnetic state
is the exact eigenstate of the eective Hamiltonian, and in the second order of Holstein-
Primako expansion terms like b
+
i
b
+
j
dont appear, so one particle excitation spectra is
exact. Calculation analogous to what was done before shows that the mode which becomes
unstable rst is the linear combination of S = 0, S
z
= 0 and S = 2, S
z
= 0, with the
critical magnetic eld given by
H =
1
2
_
_
3U
2
2
4
zt
2
U
0
+
1
2
9U
2
2
+ 16U
2
zt
2
U
0
+ 64(
zt
2
U
0
)
2
_
_
.
Results of these calculations together with the results for small magnetic elds suggest the
following picture for the global ground state in the presence of magnetic elds. For zero
magnetic eld, there are two phases, singlet and nematic. When small magnetic eld is
applied, nematic order parameter lies in the plane perpendicular to a magnetic eld, and
there is a small ferromagnetic component along the axis of the magnetic eld. As magnetic
0.5 1 1.5 2
H
U
2
0.2
0.4
0.6
0.8
1
1.2
1.4
zt
2
U
0
U
2
Singlet
Canted
Nematic Ferromagnet
Figure 2.14: Magnetic phase diagram for N = 2 insulating phase. On the mean eld level
singlet state is the product of S = 0 states at each site. Transition out of the singlet state
is of the rst order. Transition from insulating phase to the superuid takes place at the
values of y axis U
0
/(zU
2
), and for small enough U
2
/U
0
all the features of the diagram are
present.
eld goes up, magnetization gradually increases, and saturates when the instability region
of the ferromagnet is crossed. If we start from a singlet phase, there is a region of canted
nematic phase between singlet and ferromagnet, and this phase is stabilized by magnetic
eld. Magnetization changes continuously inside of this phase. This phase diagram is shown
in Fig. 2.14.
2.5 Large number of particles per site
In this section we discuss the case N 1 for both parities of N. We show how one
can separate variables describing angular momentum and the number of particles at each
site [86], and derive an eective Hamiltonian which is valid under conditions U
2
, Nt U
0
,
which is less restrictive than in section 2.3 for N >> 1.
When we have N spin-one bosons localized in the same orbital state, their total
spin may take any value that satises constrains
S +N = even, (2.56)
S N. (2.57)
We dene pure condensate wave functions as
[N, n) =
1
F
(n
x
a
x
+n
y
a
y
+n
z
a
z
)
N
[0), (2.58)
which minimize the U
2
interaction energy at the Gross-Pitaevskii (mean-eld) level at each
given site [64]. Here F = [2(N 1)!]
1/2
is a normalization factor, which is calculated in
Appendix A.4.
Now we can construct states as
[)
N
=
_
n
(n)[N, n), (2.59)
where
_
n
stands for
_
dn/(4).
Condition (2.56) corresponds to the symmetry of the states (2.58)
[N, n) = (1)
N
[N, n). (2.60)
Hence, we need to consider only wave functions that satisfy (n) = (1)
N
(n).
Now we can consider how a spin rotation operator acts on the wave function (n):
e
iS
[)
N
=
_
n
(n)e
iS
[N, n) =
_
n
(e
iS
n
)[N, n
) =
_
n
(n
))[N, n
). (2.61)
Expanding the last expression for small we nd
L
= i
, (2.62)
where we used L
rather than S
to show that it acts on the wave function . Therefore,

operator L is an angular momentum operator for n. If we want to construct some spin state,
we should take (n) to be a usual spherical harmonic. We note that the S = 0 result in
Ref. [64] is just a special case of our general statement. The most general form of the state
at each site can be expanded as
(n) =
S,|m|S
c
S,m
Y
Sm
(n),
where S satises conditions (2.56)-(2.57).
Up to this point, what we have done is valid not only for large N, but for all
N. This representation is particularly suitable for N 1 since in this limit states that
correspond to dierent ns are orthogonal to each other (see Appendix A.4)
N, n
1
[N, n
2
) =
N
(n
1
n
2
). (2.63)
The delta function is dened from the condition
_
n
1
_
n
2
f
1
N
(n
1
)f
2
N
(n
2
)
N
(n
1
n
2
) =
_
n
f
1
N
(n)f
2
N
(n) (2.64)
for the functions that satisfy f
N
(n) = (1)
N
f
N
(n).
We show in the Appendix A.4 that
a
[N, n) = (N + 1)
1/2
n
[N + 1, n),
a
[N, n) = N
1/2
n
[N 1, n) (2.65)
after projecting into the pure condensate wave functions.
This allows us to represent a
, a
as products of two operators, which act in

dierent spaces. For each trap we dene the particle creation and annihilation operators
that change the number of particles N but not the direction of n
2
b
i
[N
i
, n
i
) = (N
i
+ 1)
1/2
[N
i
+ 1, n
i
),
b
i
[N
i
, n
i
) = N
1/2
i
[N
i
1, n
i
). (2.66)
The number of particles in each trap may be expressed using b operators as
N
i
= b
i
b
i
. (2.67)
Hamiltonian (2.5) can now be represented as
H = t
ij,
n
i
n
j
(b
i
b
j
+b
j
b
i
)
N
i
+
U
0
2
N
i
(

N
i
1) +
U
2
2
i
(
L
2
i
2

N
i
), (2.68)
where
L
i
= i n
i

n
i
. (2.69)
If the system is in the Mott insulating phase, we can easily derive the eective
Hamiltonian for (n). Using the second order perturbation theory, we nd the eective
Hamiltonian on the sphere to be
H =
i
_
U
2
2
L
2
i
HL
zi
_
2N
2
t
2
U
0
ij
n
ia
n
ib
n
ja
n
jb
,
(n) = (1)
N
(n). (2.70)
We note that this Hamiltonian corresponds to a lattice of quantum rotors that interact
via quadrupolar moments. For completeness we also included magnetic eld H. Angular
momentum at each site satises the following constrains:
L
i
+N
i
= even, L
i
N
i
. (2.71)
2
The operator b applied directly to |N takes us outside of the physical Hilbert space since it does not
change the symmetry of the (n) wave function simultaneously with changing the number of particles by
one. The physical Hamiltonian, however, will always have this operator b in combination with some odd
function of n and will preserve the physical Hilbert space.
2.5.1 Mean eld solution without magnetic eld
Now we can nd a mean eld ground state of (2.70). We consider the case when
the quadrupolar interaction term is much larger than the kinetic term. We show that in this
case the ground state is a uniaxial nematic and nd its energy. Comparing the energy of
this state to that of a singlet, we estimate the phase boundary for nematic-singlet transition
for even N.
Our general mean eld anzats has the form
[ >=
i
(
i
,
i
),
_
n
[(, )[
2
= 1. (2.72)
Expectation value of Eq. (2.70) per site over the wave function (2.72) equals
U
2
2
<

L
2
i
> J < n
>< n
>,
where
J =
zN
2
t
2
U
0
, (2.73)
L
2
i
=
1
sin
(sin

)
1
sin
2
2
. (2.74)
As in the case of N = 2, all of the states that can be transformed into each other by global
rotation have the same energy. Therefore, we can impose three additional conditions. The
most convenient choice is to require symmetric real matrix < n
i
n
j
> to be diagonal, and
without the loss of generality we can choose < n
2
z
> to be the largest eigenvalue. In such
gauge the interaction between neighboring sites equals J(< n
2
x
>
2
+ < n
2
y
>
2
+ < n
2
z
>
2
).
Since we have the extra constraint < n
2
x
> + < n
2
y
> + < n
2
z
>= 1, it is now obvious that
interaction energy is minimized when
< n
2
z
>= cos
2
1.
However, states with sin
2
0 have higher angular moments, and the ground state is
determined by the competition of these two factors. We write mean eld equations to
determine the ground state as
_
U
2
2
1
sin
(sin

)
1
sin
2
2
2J(n
2
x
< n
2
x
> +n
2
y
< n
2
y
> +n
2
z
< n
2
z
>)
(, ) = (, ), (2.75)
where is Lagrange multiplier.
Simple form of the solution to Eq. (2.75) may be obtained for J U
2
. In this case
interaction energy dominates and we expect sin
2
0. Hence, the wave function becomes
localized along z and z directions. We can solve the problem by expanding only near = 0
and then taking a symmetric or antisymmetric combination to satisfy condition (2.70). If
we expand the kinetic part of Eq. (2.75) up to the rst non vanishing order in , then we
will get a two dimensional Laplace operator
nx,ny
, and our problem becomes equivalent
to a harmonic oscillator. The eective parameters are expressed as
m =
1
U
2
,
2
x,y
= 4JU
2
< n
2
z
n
2
x,y
> . (2.76)
Since we have already neglected higher order terms in <
2
> while obtaining a
harmonic Hamiltonian from (2.75), with the same accuracy we can set
x
=
y
=
4JU
2
in Eq. (2.76), i.e. the ground state is a uniaxial nematic. Since we know wave functions, we
can calculate the expectation value of the energy. We will use the fact that for a harmonic
oscillator the expectation value of kinetic energy is the same as that of potential energy.
Energy of the ground state becomes
E =
3
8
(
x
+
y
) J< n
2
z
>.
Quantum uctuations of the direction of n equal
< n
2
x
>=< n
2
y
>=
U
2
2
=
1
4
_
U
2
J
, (2.77)
and expectation value of the ground state energy is
E =
3
4
+
JU
2
J = 2
_
U
2
J J. (2.78)
Symmetrization or antisymmetrization of the wave function introduces exponentially small
shifts in energy, so in limit J U
2
, energy doesnt feel the parity of N. Though in this
subsection we explicitly started from variational ansatz (2.72), now we can justify it in the
limit J U
2
since in this case quantum uctuations of the direction of n are small and
given by Eq. (2.77).
From section 2.3 we know that for small nonzero t there is a uniaxial nematic state
for odd N. Since in the opposite limit there is also a nematic state, we expect that for all
N = 2n + 1 1, the insulating state will be nematic. For the case of even N, for small
enough J the true ground state is a singlet. Since for J U
2
ground state is nematic, for
large enough U
0
/U
2
there should be a rst order singlet-nematic transition.
For N 1 one can also calculate the location of the transition out of the sin-
glet phase (see next subsection, Fig. 2.16). Extrapolation of this result to N = 2 gives
zt
2
c
/U
0
U
2
= 0.625, which is close to the value 0.5625 obtained in section 2.4.
2.5.2 Magnetization plateaus
In this subsection we consider eects of the magnetic eld for N 1. We show,
that the magnetic eld phase diagram of (2.5) in the insulating phase has magnetization
plateaus, which are analogous to Mott lobes of usual spinless Boson Hubbard model [2].
Models of quantum rotors, that exhibit magnetization plateaus, have been considered in
literature before [87, 88]. The model (2.70) has an advantage of being experimentally
realizable with atoms in an optical lattice.
Let us rst consider the case with t = 0 (deep optical lattice). Hamiltonian doesnt
mix the states from dierent sites, and exact ground state factorizes:
[
0
) =
i
[L = S, L
z
= S), (2.79)
where S is an integer number that satises
(2S + 3) 2
H
U
2
(2S 1)
together with (2.71). Unless there is a degeneracy between states with spin S and S + 2,
this state has a gap.
When nonzero t is introduced, we cant solve the problem exactly. However, we
can make some exact statements about the magnetization in the vicinity of the t = 0.
The reasoning goes analogous to the case of Boson Hubbard model, as discussed in Ref.
[87]. Hamiltonian (2.70) is rotationally invariant, therefore total z component of the spin
operator
L
tot
z
=
i
L
iz
commutes it. As we already know, for t = 0, the exact ground state has a gap. Therefore,
when small nonzero t is introduced, ground state will adiabatically change without undergo-
ing any level crossings. Since interaction between sites commutes with L
tot
z
, magnetization
stays pinned at the same value. Now, assuming that translational symmetry is not broken,
we obtain that for small enough t
L
iz
) = S. (2.80)
Therefore, this quantized ferromagnet (QF) phase is analogous to the Mott phase of the
Bose-Hubbard model. When t is increased, at some point it becomes favorable to have a
mixture of the states with dierent L
z
at each site, and the ground state becomes analogous
to the superuid phase of Boson Hubbard model. We will see later that similar to N = 2 case
(see Fig. 2.14), this phase is topologically connected to nematic phase for small magnetic
elds, so we will also call it canted nematic.
Analogous to the treatment of Bose-Hubbard model [2, 87], we can use mean
eld approximation to calculate the phase diagram. In mean eld approach, we substitute
Hamiltonian (2.70) with the sum of single site Hamiltonians
H
MF
=
i
U
2
2
L
2
i

H

L
i
2zN
2
t
2
U
0
n
ia
n
ib
n
a
n
b
)
+
zN
2
t
2
U
0
n
a
n
b
)n
a
n
b
). (2.81)
Matrix elements of the operators n
a
between states with given L and L
z
can be
easily calculated, using formulas from Ref. [11]. Operators n
x
and n
y
have nonzero matrix
elements for states with L = 1, L
z
= 1, and n
z
have nonzero matrix elements for
states with L = 1, L
z
= 0. In general, n
a
n
b
) can have nonzero matrix elements
for states with L = 0, 2, and L
z
= 0, 1, 2. Therefore, even in the absence of
interaction between dierent sites, nematic order parameter n
a
n
b
) is nonzero. It can be
calculated using rotor wave functions, and the result is
n
a
n
b
) =
_
_
S+1
2S+3
0 0
0
S+1
2S+3
0
0 0
1
2S+3
_
_
.
To dene the order parameter, which will signal the transition out of the QF phase,
we rewrite n
x
and n
y
via operators n
= n
x
in
y
which have matrix elements between
states with L
z
= 1, and [L[ = 1. We see, that for any state which is an eigenstate of
L
z
, n
a
n
b
) is diagonal and n
x
n
x
) = n
y
n
y
) holds identically. Therefore, appearance of the
non diagonal elements in n
a
n
b
) will signal the transition out of the QF phase. Instability
towards nonzero n
x
n
y
) = (n
+
)
2
(n
)
2
)/4i) will signal the transition to the phase which
is the mixture of states with L
z
= 2, whereas nonzero n
x
n
z
) or n
y
n
z
) indicates the
transition to the phase with L
z
= 1 at each site . Condensation of n
x
n
y
) breaks initial
U(1) symmetry of the model up to Z
2
, since in this case only states with the same parity of
L
z
are allowed. This can be noted by saying that this state is the eigenstate of the operator
e
i
L
iz
,
which counts the parity of L
z
. On the other hand, condensation of n
x
n
z
) or n
y
n
z
) com-
pletely breaks U(1) symmetry. The former superuid phase is analogous to Spin Singlet
Condensate discussed in Ref. [70], whereas the latter superuid corresponds to Polar
Condensate in the same model.
Our analysis which led to the condition (2.80) tells us only about L
iz
, while the
value of L
i
is constrained by Eq. (2.71). When t = 0, only L
i
= S contributes to the
state with quantized L
iz
. However, as we change t, mean eld QF state may become a
mixture of states with xed L
iz
but dierent L
i
: [L = S, L
z
= S), [L = S + 2, L
z
= S),
[L = S + 4, L
z
= S), etc. An example of this scenario is singlet-nematic transition for
N = 2, H = 0. This is dierent from the case of Bose-Hubbard model, where mean eld
Mott insulating state within each lobe is just a Fock state.
In the vicinity of the point H
s
/U
2
= (2S +3)/2 with S satisfying (2.71), only two
states [L = S + 2, L
z
= S + 2) and [L = S, L
z
= S) are important, all other states have
higher energies. If we denote [L = S+2, L
z
= S+2) as spin up state and [L = S, L
z
= S)
as spin down state, this model can be mapped to the S = 1/2 XXZ ferromagnet with
magnetic eld along z axis. Therefore, for small t rst instability out of QF doesnt change
the parity of L
iz
, and n
x
n
z
) is still zero. If the value of t is increased, the system wants
to take advantage of n
ix
n
iz
n
x
n
z
) couplings, and mixture of states with dierent parities
of L
iz
may become favorable at large enough t.
To summarize, when t is increased from 0, there are three possible scenarios for
the behavior of the system:
1) Mean-eld QF state may become a mixture of states with xed L
z
but dierent
L: [S, S), [S + 2, S), [S + 4, S), before the transition to any canted nematic phase
1 2 3 4 5 6 7
H
U
2
10
20
30
40
2zN
2
t
2
U
0
U
2
S1 S3 S5
Canted
Nematic
Figure 2.15: Mean eld phase diagram for odd N 1. On mean eld level, quantized ferro-
magnet(QF) phases correspond to factorized wave functions
i
[L = S, L
z
= S). Transition
to canted nematic phase occurs via condensation of n
x
n
y
) order parameter.
takes place. This phase transition is of the rst order, analogous to singlet-nematic
transition for N = 2.
2) Mean-eld QF state has the form (2.79), when n
x
n
y
) starts condensing. As was
discussed before, this is the case for small t in the vicinity of H
s
. Transition to the
canted nematic is of the second order, and breaks initial U(1) symmetry up to Z
2
.
3) Mean-eld QF state has the form (2.79), when n
x
n
z
) and n
y
n
z
) condense. Both
these order parameters condense at the same point, due to rotational invariance along
z axis. This transition may be favorable for higher values of t, when H is not close to
H
s
.
To nd out which scenario realizes for each lobe, we have performed numerical
mean-eld calculation, and the result is the following: for all lobes, except for S = 0, the
second possibility takes place, and there is a transition from the mean eld state (2.79)
to nonzero value of n
x
n
y
). For S = 0, H = 0 rst order singlet-nematic transition takes
place before the transition into any canted phase. For nematic phase small magnetic eld
creates some nonzero magnetization, i. e. magnetization changes continuously and S = 0
plateau disappears. The results of the mean eld calculations are summarized in Figs. 2.15
and 2.16. In Fig. 2.17 we present schematic picture of magnetization as magnetic eld is
changed.
Now, lets describe the mean eld calculation. For each lobe S one can calculate
the parameters, for which QF state (2.79) becomes unstable to rst scenario. To obtain
this region we look at the variational ground state
i
(cos [L = S, L
z
= S) +e
i
sin [L = S + 2, L
z
= S)), (2.82)
and determine the region in which = 0 minimizes the energy. Within each lobe the
critical value of 2N
2
t
2
/U
0
for which = 0 becomes unstable doesnt depend on magnetic
2 4 6 8
H
U
2
2
4
6
8
10
12
2zN
2
t
2
U
0
U
2
S0 S2 S4 S6
Canted
Nematic
Figure 2.16: Mean eld phase diagram for even N 1. On mean eld level, quantized
ferromagnet(QF) phases correspond to factorized wave functions
i
[L = S, L
z
= S). For
S ,= 0 transition to canted nematic phase occurs via condensation of n
x
n
y
) order parameter.
For S = 0 phase transition is analogous to N = 2 case.
2 4 6 8
H
U
2
2
4
6
8
L
z
even
odd
Figure 2.17: Schematic of the magnetization, L
z
), as a function of magnetic eld. For
odd N 1,
2zN
2
t
2
U
0
= 5U
2
, for even N 1,
2zN
2
t
2
U
0
= 2U
2
. Magnetization inside canted
nematic phases hasnt been calculated, and represented by dashed lines. Note, that for odd
N magnetization starts from zero at H = 0 and then rapidly becomes 1.
eld. This is an artifact of mean-eld calculation, and in principle one can include quantum
uctuations as it has been done for N = 2.
We will calculate the parameters, for which condensation of n
x
n
y
) occurs, assum-
ing that transition to canted phase takes place while = 0. In Landau expansion of the
ground state energy (2.81), contributions which are quadratic in n
x
n
y
) come from second
order tunneling processes to states with higher energies. From [L = S, L
z
= S) the system
can tunnel to states [L = S +2, L
z
= S +2), [L = S +2, L
z
= S 2), [L = S, L
z
= S 2),
[L = S2, L
z
= S2). All matrix elements and energy gaps can be calculated analytically,
and we obtain the phase boundary by requiring the coecient of n
x
n
y
)
2
to become zero.
Analogous considerations give the phase boundaries for n
x
n
z
) and n
y
n
z
) condensation.
In this case, second order states, which contribute to the energy are [L = S+2, L
z
= S+1),
[L = S + 1, L
z
= S 1), and [L = S 1, L
z
= S 1). By comparing critical values for
these two types of transition to canted nematic, we see that for all S condensation of n
x
n
y
)
occurs before the condensation of n
x
n
z
).
Finally, we need to compare 2N
2
t
2
/U
0
at which = 0 becomes unstable in Eq.
(2.82) with that of transition to canted nematic. For all S ,= 0 the latter transition occurs
rst, while for singlet = 0 becomes unstable rst, and the system undergoes rst order
transition, analogous to singlet-nematic transition in the absence of the magnetic eld.
2.6 Global phase diagram
In earlier sections we have established spin structure of insulating phases of spin-
one bosons in the optical lattice in various limits. Here we summarize our arguments and
discuss implications of our results for the global phase diagram.
2.6.1 Two and three dimensional lattices
In two and three dimensional lattices, insulating states with one atom per site are
nematic as long as the perturbation theory approach in t/U
0
remains valid, as was shown in
section 2.3. For an arbitrary odd number of particles per site, N, and in the limit of small
tunneling (Nt)
2
/U
0
<< U
2
, the nematic order in the ground state was also established in
section 2.3. For large, odd N, the nematic order in the ground state can be proven when
(Nt)
2
/U
0
becomes larger than U
2
(but still smaller than U
0
), as was demonstrated in section
2.5. It is also natural to expect that a superuid polar phase develops from the nematic
insulator (both states break spin rotational symmetry without breaking the time reversal
symmetry), so we expect the nematic order even when Nt/U
0
is not small and the system
is close to the superuid-insulator transition. In all cases we nd that insulating phases
with an odd number of particles per site are nematic.
In the case of two particles per site, the results of section 2.4 establish that for
small enough U
2
/U
0
there is a rst order transition between the spin singlet phase (for
small t) and the spin nematic phase (for larger t) at zt
2
c
/U
0
U
2
= 0.5 (z is the coordination
number of the lattice). Analogously, for large, even N, the results of section 2.5 show that
the singlet insulating ground state goes into spin nematic at zN
2
t
2
c
/U
0
U
2
= 5/2. Since for
small enough U
2
/U
0
we expect nematic spin order close to the SI transition into the polar
superuid phase, we propose that in this case insulating phases with an even number of
particles per site are either singlet or nematic with the rst order transition at some critical
value of tunneling t
c
.
In all of our earlier discussions, we assumed that Mott insulating lobes for even
llings are large enough to have the transition into the nematic phase before superuidity
sets in. This assumption is controlled by the smallness of the ratio U
2
/U
0
. Here we will
discuss the superuid - insulator phase boundaries and estimate how small U
2
/U
0
should
be for the singlet-nematic transition to lie inside the Mott phase.
Assuming transition from the spin-singlet insulating phase, the mean-eld cal-
culation of the superuid-insulator phase boundary was given in Ref. [89]. The analysis
presented there shows that the critical value of tunneling, after which the Mott phase doesnt
exist, is given by
U
0
+ 2U
2
zt
SI
=
1
3
(2N + 3 + 2
_
N
2
+ 3N).
We will use this critical value t
SI
as an estimate of the superuid-insulator transition. For
N = 2, singlet-nematic phase transition takes place at zt
2
c
/U
0
U
2
= 0.5. The condition
t
c
< t
SI
for N = 2 is satised, if
zU
2
U
0
< 0.1.
For the case N >> 1 the requirement of t
c
< t
SI
is satised
zU
2
U
0
< 0.2.
One can see that depending on the exact value of zU
2
/U
0
, there are dierent possibilities
for Mott lobes with an even number of particles. When zU
2
/U
0
< 0.1, all insulating phases
with even lling factors are spin singlet for small tunneling and spin nematic for larger
tunneling. For 0.1 < zU
2
/U
0
< 1, insulating phases with large even lling factors have both
singlet and nematic regimes, but insulating states with small even llings have only the
singlet phase. Finally, for 0.2 < zU
2
/U
0
, all insulating phases with even lling factors are in
the spin singlet state. In Figs. 2.18 and 2.19 we combine these results with the schematic
representation of the SI transitions to obtain the global phase diagram.
2.6.2 One dimensional lattices
For one dimensional lattices we established that when N
2
t
2
/U
0
<< U
2
the system
will be in a uniform singlet phase for even llings and in a dimerized singlet phase for odd
llings (when there is only one atom per site the dimerized phase has been veried in the
regime t << U
0
). The nature of magnetic order close to the tips of the insulating lobes
(when the perturbation theory in t is not applicable) is less clear. However, we expect that
the phase diagram for the one dimensional lattice is qualitatively similar to two and three
dimensional cases with one important dierence: instead of the nematic phase, we have a
dimerized singlet.
c +
-0.5 0.5 1 1.5 2 2.5 3
t
1
2
3
4
5
6
N=1
N=2
N=3
N=4
N=5
Nematic
Nematic
Nematic
Nematic
Nematic
Singlet
Singlet
POLAR SUPERFLUID
First order
Second order
transitions
transitions
t t t
Figure 2.18: Global phase diagram for spin-one bosons in 2D and 3D optical lattice, for
zU
2
/U
0
< 0.1. Mean-eld analysis of the superuid to insulator transition was done in Ref.
[89]. In this work we concentrated on discussing the spin structure of the insulating lobes.
Singlet-nematic rst order phase transition for N = 2 takes place for zt
2
/U
0
U
2
= 0.5 (z is
the coordination number of the lattice). For large, even N singlet-nematic phase transition
occurs at zN
2
t
2
/U
0
U
2
= 5/2. t
c
marks the actual rst order phase transition and t
and t
+
are the limits of metastability. Note that the system may also have fragmented superuid
phases for small t that are not shown here [70].
-0.5 0.5 1 1.5 2 2.5 3
t
1
2
3
4
5
6
N=1
N=2
N=3
N=4
N=5
Nematic
Nematic
Nematic
Singlet
Singlet
POLAR SUPERFLUID
Figure 2.19: Global phase diagram for spin-one bosons in 2D and 3D optical lattice, for
zU
2
/U
0
> 0.2. Superuid-Insulator transition for even lling factors takes place before
singlet-nematic transition.
2.7 Detection of spin order in insulating phases
Now we discuss several approaches to detection of the spin-ordering for spin-one
bosons in optical lattices. One way of detecting singlet-nematic transition has already been
noted in section 2.4.6, where we proposed to introduce an easy plane for nematic order
by applying a small magnetic eld, then releasing the trap and measuring the number of
particles of dierent spin components. Spatial separation of dierent spin components can
be achieved by applying magnetic eld gradients during the free fall of the atoms. For the
case of N = 2, with a small magnetic eld applied in the z direction, expectation values
of n(S
z
= 0) and n(S
z
= 1) = n(S
z
= 1) have been calculated and are shown in Fig.
2.13. Since the phase transition is of the rst order, there is a sharp change which can
be measured experimentally. We note that N = 2 case also has particular experimental
advantage over other lling factors due to the absence of three particle losses.
The second approach to experimental detection of singlet and nematic insulating
phases relies on the measurement of excitation spectra. As discussed in sections 2.4.4 and
2.4.5, the singlet phase has a nonzero gap to all excitations, whereas the nematic phase
has gapless spin wave excitations. To measure the excitation spectra, we propose using
Bragg spectroscopy, which was applied successfully to identify sound-like Bogoliubov ex-
citations in condensates of spinless particles [90]. In such experiments the optical lattice
should be illuminated by two laser beams with wave vectors k
1
and k
2
and a frequency
dierence , which is much smaller than their detuning from an atomic resonance. The
intersecting beams create a periodic, traveling intensity modulation that creates external
potential due to ac Stark eect of the form V
cos (qr t). Here we introduce spin in-

dices for the external potential since the ac Stark eect may introduce mixing between
0.5 1 1.5 2
2
4
6
8
10
Momentum
E
n
e
r
g
y
Excited states
Linear spin wave
Gapped excitations
Figure 2.20: Probing dispersion relation using Bragg scattering.
dierent S
z
components. A response of the system to such potential may be calculated
using Fermis golden rule. Interaction is expressed in second-quantized notations such as
V
/2(
(q)e
it
+
(q)e
it
), where
k
a
k+q
a
k
. The scattering rate is given
by
2
h
f
[f[V
[g)[
2
( h E
f
+E
g
).
If the resonance state is far detuned from the excited states, then V
has the form V
.
Such interaction couples only to the total number of particles at each site and doesnt feel
internal spin structure. Low lying excitations in insulating phases dont change the number
of particles on individual sites, so V
interaction wont produce any Bragg peaks for

low lying excitations. Therefore, it is necessary for detection that V
deviate from V
,
which can be achieved by making detuning comparable to level spacing of ne and hyperne
components. From section 2.4.5 we know that for N = 2, nematic spin wave excitations
correspond to S
z
= 1, longitudinal excitation corresponds to S
z
= 0, and there are also
gapped excitations with S
z
= 2. Since the nematic state has S
z
= 0, it is necessary to
have nonzero V
0,1
, V
0,0
, and V
1,1
to observe each kind of these excitations. In Fig. 2.21
we show dependence of N = 2 peak positions on t for xed q.
Let us now comment on the eect of inhomogeneous trapping potential. When
the local trapping potential
i
varies smoothly from site to site, it is not the chemical
potential
i
which is xed across the trap, but the sum
i
+
i
. Therefore, if
i
varies
considerably, we will have insulating regions with dierent occupation numbers as well as
regions with superuid order, all in the same trap, as was discussed in Ref. [3] for the case of
0.72 0.74 0.76 0.78 0.8
t
1
2
3
4
5
Energy
M
e
t
a
s
t
a
b
l
e
longitudinal
S
z
=2 excitation
Spin wave
excitation
Figure 2.21: Positions of Bragg peaks for [q[ = 0.02, z = 6 in the nematic phase with N = 2
atoms per site. Energy is measured in units of U
2
, and t is measured in units of
_
U
0
U
2
z
spinless atoms. Therefore, Bragg scattering experiments for xed q will exhibit resonances
coming from the regions of the lattice at dierent lling factors. Relative intensity of
these resonances will be determined by the relative number of particles in each region.
Interpretation of Bragg experiments will be easier if the trapping potential is not harmonic,
but has sharp borders, so the whole system has essentially the same density.
We will now discuss the experimental approaches to detect the magnetization
plateaus, studied in section 2.5. An important experimental constraint is that the total
magnetization is xed by the initial state of the system. In the discussion in section 2.5
we showed, that when we change magnetic eld, it is energetically favorable to adjust the
magnetization in order to utilize some of the Zeeman energy. We observe, however, that
the spin component parallel to the applied eld is conserved. For example, if the magnetic
eld is along the z axis, S
tot,z
=
i
S
iz
is a good quantum number of the system (spin
non-conserving interactions, such as the dipolar relaxation, are typically small) and should
not change even when H is changing. For a single large trap this feature allowed J. Stenger
et al. to study magnetic properties of spinor condensate [17] at magnetic elds for which
the true ground state should be fully polarized.
A way around spin conservation has been demonstrated in Ref. [17] and relies on
applying spatially varying magnetic elds and performing Stern-Gerlach imaging. Here we
extend these ideas and suggest an approach to experimental observation of the magnetiza-
tion plateaus discussed above. The idea of our rst experiment is shown in Fig. 2.22. We
consider a strongly anisotropic trap in which a magnetic eld gradient is applied parallel
to the long axis of the system. When there is a magnetic eld gradient parallel to the
long axis, there should also be gradients in the transverse directions (both

H and

H
should be zero). We assume that the size of the condensate in the transverse directions is
small enough so that we can neglect the eects of the magnetic eld in transverse directions.
However, the size of the condensate should be larger than the optical lattice period in any
direction, so that we can consider it as a three dimensional system. As we discussed before,
the uniform part of the magnetic eld has no eect on the state of the system, so in our
discussion we set it to zero. To be concrete, we assume that the largest insulating domain
at the center of the trap has a lling factor N = 6 and that our system has been prepared
to have S
tot,z
= 0. In a non uniform magnetic eld, dierent parts of the N = 6 domain
minimize their energy for dierent values of magnetization. When the eld gradient is
suciently large and tunneling is small enough, the locally favored magnetization changes
in a step like fashion from S
site,z
= 6 (per site) on the left to S
site,z
= 6 on the right.
Such state is also consistent with the spin conservation, since S
tot,z
remains zero. So, the
conguration that minimizes the energy has plateaus in the spatial prole of magnetization,
with spin polarizations (per atom) S
atom,z
= 1, 2/3, 1/3, 0, 1/3, 2/3, 1, as shown in Fig.
2.22. Each plateau has a length of the order of 2U
2
/[H[. The appearance of spin plateaus
in a traveling magnetic eld is analogous to the domain structure of condensates in optical
lattices in the presence of a non-uniform global conning potential that was discussed before
[3, 91]. In the latter case the total density is xed by the number of atoms in the trap, but
insulating phases with dierent integer lling factors exist due to the conning potential.
To detect the magnetization plateaus shown in Fig. 2.22, one needs to image dierent parts
of the trap separately, measuring spin polarization per atom as a function of the position:
S
atom,z
) = (n
+
n
)/(n
+
+ n
0
+ n
), where n
, n
0
are the local densities of atoms with
= 1 and 0 respectively. This quantity can be most easily measured in Stern-Gerlach
time-of-ight experiments. If a small gradient of magnetic eld is applied during expansion,
clouds with dierent spin components spatially separate, and one can measure the number
of atoms with dierent spin components using light scattering.
To illustrate system parameters needed to realize this experiment we consider a
cigar shaped condensate of sizes 400 10 10 m. For an optical lattice created with
= 985 nm lasers and 4 10
5
atoms in a condensate we get the maximum density of
six atoms per site in the center of the trap. To observe ve plateaus in this setup, one
would need magnetic eld gradients 100mG/cm, and spatial resolution of 30m. These
parameters have already been achieved in experiments of Ref. [17]. Inside the insulating
phase, the characteristic timescale for spin relaxation between dierent sites is set by the
exchange interactions
ex
= hU
0
/(Nt)
2
. For t = 0.1khz and U
0
= 2khz we nd times of
the order of hundreds of milliseconds. So, if a magnetic eld gradient is applied in the
insulating regime, one needs to wait at least that long for magnetic plateaus to develop.
Experimentally it may be more ecient to apply magnetic eld gradient when the system is
in the superuid regime and then take the system to the insulating state by slowly reducing
t.
We now discuss our second spin decoration approach to experimental detection
of spin gap eects in the insulating regime of spin-one bosons. The idea of this method is
that when a system is prepared with a non-zero magnetization in the absence of magnetic
eld gradients, magnetization gets distributed non-uniformly among insulating domains
with dierent lling factors. Spin polarization appears predominantly in domains with odd
numbers of atoms per site and is pushed out of the regions with even llings. To justify
this conjecture we propose the following argument. In the regime, when tunneling is small
(zN
2
t
2
/U
0
U
2
), there is a crucial dierence in spin susceptibility between odd and even
phases. For even domains, one needs to pay an energy 3U
2
to break a singlet state and
have S
site,z
2. For odd sites the lowest energy state already has S = 1, and energy cost to
atom, z
<S >
N6 N5 N5 N4 N4
-8 -6 -4 -2 2 4 6
zH
2U
2
-1
-0.5
0.5
1
Figure 2.22: A system of spin-one bosons in an optical lattice conned by a parabolic
potential. In the insulating regime the cloud breaks into insulating domains with dierent
integer lling factors. When magnetic eld gradient is applied parallel to the long axis of
the trap, magnetization plateaus develop inside individual insulating domains.
N6 N5 N5 N4 N4
z
1
S
site,z
Figure 2.23: A system of spin-one bosons with a nonzero total magnetization in an optical
lattice (no external magnetic eld). Magnetization gets distributed among regions with odd
lling factors and is pushed out of the regions with even llings.
polarize existing spins is of the order of zN
2
t
2
/U
0
. Therefore, externally imposed nonzero
magnetization will be redistributed in odd insulating domains. For small magnetization per
site, energy goes as
E
odd
(N, S
site,z
) =
1
2
odd
(N)
S
2
site,z
. (2.83)
If
odd
(N) was the same for all N, then magnetization would be distributed uniformly
among all odd domains. In reality,
odd
(N) decreases with increasing N, so we expect
larger magnetization for insulating domains with smaller number of atoms. Quadratic
dependence in (2.83) ensures, however, that all domains with odd ling factors acquire
nite magnetization. So, in experiments we expect to nd a picture of alternating even and
odd domains, in which odd domains have nite magnetization and even domains have none
(see Fig. 2.23). This picture is valid until all odd regions have magnetization S
site,z
= 1. For
the experimental setup discussed earlier this corresponds to S
atom,z
) < 0.1. By performing
spatially resolved measurements of spin polarization one should be able to observe such
modulated structure of magnetization.
2.8 Conclusions
In summary, in this chapter we have considered Mott insulating phases of spin-
one atoms with antiferromagnetic interactions in optical lattices. In the experimentally
interesting limit U
2
U
0
, and deep inside the Mott phases Nt U
0
(N is the lling factor),
we performed detailed calculations for the following cases: i) odd number of particles per
site and (Nt)
2
/U
0
U
2
; ii) two particles per site and an arbitrary ratio of t
2
/U
0
and U
2
; iii)
large number of particles per site N 1 with an arbitrary ratio of (Nt)
2
/U
0
and U
2
. Based
on this analysis we argued that in two and three dimensional lattices insulating phases with
an odd number of particles per site are always nematic. For an even number of particles
per site, there is either a spin singlet phase or a rst order phase transition between spin
singlet and nematic phases controlled by the depth of the optical lattice. The resulting
global phase diagrams are shown in Figs. 2.18 and 2.19. We have considered excitations
for singlet and nematic phases and have analyzed the eects of magnetic eld. For one
dimensional lattices we have found dimerized singlet phases for insulating states with odd
llings. We also discussed dierent experimental techniques to identify the proposed phases.
Chapter 3
Exactly solvable case of a
one-dimensional Bose-Fermi
mixture
In this chapter we consider a one dimensional interacting Bose-Fermi mixture with
equal masses of bosons and fermions, and with equal and repulsive interactions between
Bose-Fermi and Bose-Bose particles. Such a system can be realized in current experiments
with ultracold Bose-Fermi mixtures. We apply the Bethe-ansatz technique to nd the exact
ground state energy at zero temperature for any value of interaction strength and density
ratio between bosons and fermions. We use it to prove the absence of demixing, contrary to
prediction of a mean eld approximation. Combining the exact solution with local density
approximation (LDA) in a harmonic trap, we calculate the density proles and frequencies
of collective modes in various limits. In the strongly interacting regime, we predict the
appearance of low-lying collective oscillations which correspond to the counterow of the
two species. In the strongly interacting regime we use exact wave function to calculate
the single particle correlation functions for bosons and fermions at low temperatures under
periodic boundary conditions. Fourier transform of the correlation function is a momentum
distribution, which can be measured in time-of-ight experiments or using Bragg scattering.
We derive an analytical formula, which allows us to calculate correlation functions at all dis-
tances numerically for a polynomial time in the system size. We investigate numerically two
strong singularities of the momentum distribution for fermions at k
f
and k
f
+2k
b
. We show
that in a strongly interacting regime correlation functions change dramatically as temper-
ature changes from 0 to a small temperature E
f
/ E
f
, where E
f
= (hn)
2
/(2m), n
is the total density and = mg/( h
2
n) 1 is the Lieb-Liniger parameter. A strong change
of the momentum distribution in a small range of temperatures can be used to perform a
thermometry at very small temperatures.
3.1 Introduction
Recent developments in cooling and trapping of cold atoms open exciting oppor-
tunities for experimental studies of interacting systems under well controlled conditions.
52
Chapter 3: Exactly solvable case of a one-dimensional Bose-Fermi mixture 53
Current experiments [18, 93, 19] can deal not only with single component gases, but with
various atomic mixtures. Using Feshbach[94, 95] resonances and/or optical lattices[3, 4] one
can tune dierent parameters, and drive the systems towards strongly correlated regime.
The eect of correlations is most prominent for low dimensional systems, and recent ex-
perimental realization[42, 43] of a strongly interacting Tonks-Girardeau (TG) gas of bosons
opens new perspectives in experimental studies of strongly interacting systems in 1D[96]. In
this chapter we investigate Bose-Fermi mixtures in 1D, using exact techniques of the Bethe
ansatz. Some of the results presented here have been reported earlier [97].
Most of the theoretical research on Bose-Fermi mixtures[98, 99, 100] so far has
been concentrated on higher dimensional systems, and only recently 1D systems started
attracting attention. Several properties of such systems have been investigated so far,
including phase separation[101, 102, 103], fermion pairing[104], possibility of charge density
wave (CDW) formation[105] and long distance behavior of correlation functions[106].
A 1D interacting Bose-Fermi mixture is described by the Hamiltonian
H =
_
L
0
dx(
h
2
2m
b
b
+
h
2
2m
f
f
) +
_
L
0
dx(
1
2
g
bb
b
+g
bf
b
). (3.1)
Here,
b
,
f
are boson and fermion operators, m
b
, m
f
are the masses, and g
bb
, g
bf
are Bose-
Bose and Bose-Fermi interaction strengths. The model (3.1) is exactly solvable, when[107]
m
f
= m
b
= m, g
bb
= g
bf
= g > 0. (3.2)
This corresponds to the situation when masses are the same, and Bose-Bose and Bose-Fermi
interaction strengths are the same and positive. Although conditions (3.2) are somewhat
restrictive, the exactly solvable case is relevant to current experiments (the experimental
situation will be analyzed in detail in section 3.7 ) and can be used to check the validity of
dierent approximate approaches. Model (3.1) under conditions (3.2) has been considered in
the literature before[107], but its properties have not been investigated in detail. After the
appearance of our initial report [97], two additional articles [106, 108] used Bethe ansatz to
investigate the same model. We use the exact solution to calculate the ground state energy
and investigate phase separation and collective modes at zero temperature. For strongly
interacting regime, we calculate single particle correlation functions, and consider the eects
of small temperature on correlation functions and density proles.
This chapter is organized as follows. In section 3.2 we review the Bethe ansatz
solution for Bose-Fermi mixture and compare it to the solution for two-species fermion
mixture. In section 3.3 we obtain the energy numerically in the thermodynamic limit.
We use it to prove the absence of demixing under conditions (3.2), contrary to prediction
of a mean eld [101] approximation. In section 3.4 we combine exact solution with local
density approximation (LDA) in a harmonic trap, and calculate the density proles and
frequencies of collective modes in various limits. In the strongly interacting regime, we pre-
dict the appearance of low-lying collective oscillations which correspond to the counterow
of the two species. In section 3.5 we use exact wave function in the strongly interacting
regime to calculate the single particle correlation functions for bosons and fermions at zero
54 Chapter 3: Exactly solvable case of a one-dimensional Bose-Fermi mixture
temperature under periodic boundary conditions. We derive an analytical formula, which
allows us to calculate correlation functions at all distances numerically for a polynomial
time in system size. In section 3.6 we extend the results of section 3.5 for low temperatures.
We also calculate the evolution of the zero temperature density prole at small nonzero
temperatures. We show, that in strongly interacting regime correlation functions change
dramatically as temperature is raised from 0 to a small value. Finally in section 3.7 we
analyze the experimental situation and make concluding remarks.
3.2 Bethe ansatz solution
In this section we will briey review the solution [107] of the model (3.1) under
periodic boundary conditions and compare it to the solution of Yang of the spin-
1
2
interacting
fermions [109, 110], for the sake of completeness. More details on Yangs solution can be
found in [111, 112, 114, 115, 113].
In rst quantization, Hamiltonian (3.1) can be written as
H =
N
i=1
2
x
2
i
+ 2c
i<j
(x
i
x
j
), c > 0. (3.3)
Here we have assumed m = 1/2 and h = 1, to stay in line with the literature on the subject.
Later in the discussion of the collective modes we will introduce the mass of atoms, but it
should be clear from the context whether we have assumed m = 1/2 or not. Parameter c
in (3.3) is related to parameters of (3.1) as follows:
c =
mg
h
2
. (3.4)
The wave function is supposed to be symmetric with respect to indices i =
1, ..., M (bosons) and antisymmetric with respect to i = M + 1, ..., N (fermions). In
the rst stage, Yangs solution doesnt impose any symmetry constraint on the wave func-
tion. In the second stage, periodic boundary conditions are resolved with the help of extra
Bethe-ansatz. This idea has been generalized by Sutherland [116] for the case of N-fermion
species. The results presented here can be simply derived from Sutherlands work.
In Yangs solution, one assumes the generalized coordinate Bethe wave function of
the following form: for 0 < x
Q
1
< x
Q
2
< ... < x
Q
N
< L
=
P
[Q, P]e
i
P
k
P
i
x
Q
i
, E =
i
k
2
i
. (3.5)
where k
1
, ..., k
N
is a set of unequal numbers, P is an arbitrary permutation from S
N
and
[Q, P] is N! N! matrix. Lets denote each column of this matrix as N! dimensional vector
P
. Delta function potential in (3.3) is equivalent to the following boundary condition for
the derivatives of the wave function:
(

x
j

x
k
)
x
j
=x
k
+0
(

x
j

x
k
)
x
j
=x
k
0
= 2c
x
j
=x
k
, (3.6)
and the continuity condition reads
x
j
=x
k
+0
=
x
j
=x
k
0
. (3.7)
Suppose Q and Q
are two permutations, such that Q

k
= Q
k
, for k ,= i, i +
1, and Q
i
= Q
i+1
, Q
i
= Q
i+1
. Similarly, P and P
are two permutations, such that

P
k
= P
k
, for k ,= i, i + 1, and P
i
= P
i+1
, P
i
= P
i+1
. To satisfy (3.6) and (3.7) for
x
Q
i
= x
Q
i+1
independently of other x, one has to impose two conditions for four coecients
[Q, P], [Q
, P], [Q, P
], [Q
, P
]. Using these two conditions, we can express [Q, P
], [Q
, P
]
via [Q, P], [Q
, P]. These requirements can be simply written as a condition between

P
and
P
:
P
= Y
i,i+1
P
i
,P
i+1
P
. (3.8)
Operators Y are dened as
Y
l,m
i,j
=

ij
1 +
ij
+
1
1 +
ij
P
lm
, (3.9)
where
ij
=
ic
k
i
k
j
,
and

P
lm
is an operator acting on a vector
P
which interchanges the elements with indices
Q
l
and Q
m
. Using Y operators one can express any
P
via
0
, where
0
is a column for
P = identity. However, arbitrary permutation P can be represented as a combination
of neighboring transpositions by dierent means. Independence of the nal result from
a particular choice of neighboring transpositions can be checked based on the following
Yang-Baxter Relations:
Y
a,b
i,j
Y
a,b
j,i
= 1, (3.10)
Y
a,b
j,k
Y
b,c
i,k
Y
a,b
i,j
= Y
b,c
i,j
Y
a,b
i,k
Y
b,c
j,k
. (3.11)
Operators Y
i,i+1
P
i
,P
i+1
exchange the momentum labels P
i
and P
i+1
, while

P
i,i+1
interchange
relative position labels Q
i
and Q
i+1
. It is convenient to dene combined operator, which
exchanges both labels:
X
ij
=

P
ij
Y
ij
ij
=
1
ij

P
ij
1 +
ij
. (3.12)
Using this denition, periodic boundary conditions can be written as N matrix eigenvalue
equations:
X
j+1,j
X
j+2,j
...X
N,j
X
1,j
...X
j1,j
0
= e
ik
j
L
0
. (3.13)
The procedure outlined above reduces equations for N! N! coecients to N eigenvalue
equations for N! dimensional vector. Imposing some symmetry on
0
simplies the system
further. If
0
is antisymmetric with respect to particle permutations (fermions) , then
P
ij
= 1 and e
ik
j
L
= 1. The system of equations is the same as for non interacting fermions,
as expected. If
0
is symmetric (bosons),

P
ij
= 1 and the system is equivalent to periodic
boundary conditions of Lieb-Liniger model [40].
If one needs to consider two-species system,
0
has the symmetry of the corre-
sponding permutation group representation (Young tableau). Instead of solving Eq. (3.13),
it is convenient to consider the similar problem in the conjugate representation. If
0
is
antisymmetric with respect to both permutations of the rst M indices and the rest N M
(two-species fermions), eigenstate in conjugate representation is symmetric with respect
to rst M indices and is also symmetric with respect to permutations of the rest N M
indices. Similarly, in conjugate representation for Bose-Fermi mixture with M bosons and
N M fermions should be chosen to be antisymmetric for permutations of M boson in-
dices and symmetric with respect to permutations of N M fermion indices. The periodic
boundary conditions are (note the change of the sign in the denition of X
ij
compared to
X
ij
):
X
j+1,j
X
j+2,j
...X
N,j
X
1,j
...X
j1,j
= e
ik
j
L
, (3.14)
X
ij
=
1 +
ij

P
ij
1 +
ij
. (3.15)
Since N!-dimensional vector has symmetry Hamiltonian, it has C
M
N
inequivalent compo-
nents, characterized by the positions y
i
of M spin-down fermions (or M bosons respectively).
One can think of the components of the vector as of the values of the spin wave function,
dened on an auxiliary one-dimensional lattice of size N. C
M
N
independent values of cor-
respond to C
M
N
values of the wave function of M particles with coordinates y
i
, dened
on this auxiliary lattice (since is symmetric for N M fermion indices, these are consid-
ered to be vacancies). Wave function should be symmetric with respect to exchange of two
particles for two-species fermions, and antisymmetric for the case of Bose-Fermi mixture.
To preserve the terminology of the two-species fermion solution for the case of Bose-Fermi
mixture, later in the text we will always refer to the wave function on an auxiliary lattice
as to spin wave function, although it has a direct meaning only for two-species fermion
case.
First, one can solve the problem for M = 1[117]. In this case there is no dierence
between two-species fermions or Bose-Fermi mixture. It can be shown (detailed deriva-
tions are available in the appendix of [115]), that in this case wave function in conjugate
representation is
(M = 1) = F(, y) =
y1
j=1
k
j
+ic/2
k
j+1
ic/2
, (3.16)
where new spectral parameter satises the following equation:
N
i=1
k
i
+ic/2
k
i
ic/2
= 1. (3.17)
Periodic boundary conditions simplify to
e
ik
j
L
=
k
j
+ic/2
k
j
ic/2
, j = 1, ..., N (3.18)
In an auxiliary lattice the wave function of one spin-deviate (or boson) F(, y) plays the role
similar to one-particle basis function e
ikx
of the original coordinate Bethe ansatz, spectral
parameter being the analog of the momentum k.
In the case when M > 1, Yang suggested that the solution of Eqs. (3.14)-(3.15)
again has the form of Bethe ansatz in the spin subspace: for 1 y
1
< y
2
< ... < y
M
N
=
R
A(R)
M
i=1
F(
R
i
, y
i
), (3.19)
where
1
, ...,
M
is a set of unequal numbers, R is an arbitrary permutation from S
M
. It
can be shown [109, 115], that this ansatz solves (3.14)-(3.15) for two-species fermion system,
if
A(R
)
A(R)
=

R
i+1

R
i
ic
R
i+1

R
i
+ic
, (3.20)
similarly to bosonic relations of Lieb-Liniger model[40]. Here R and R
are two permutations

from S
M
such that R
k
= R
k
, for k ,= i, i + 1, and R
i
= R
i+1
, R
i
= R
i+1
. The set of , k
has to satisfy the following set of equations:
i=1
k
i
+ic/2
k
i
ic/2
=
M
=1
+ic
ic
, = 1, ..., M, (3.21)
e
ik
j
L
=
M
=1
k
j

+ic/2
k
j

ic/2
, j = 1, ..., N. (3.22)
For the Bose-Fermi mixture, has to be antisymmetric for permutations of y
i
variables. This problem has actually been solved by Sutherland[116], although he was
interested not in Bose-Fermi mixture, but fermion model with several species. He has shown,
that if one doesnt specify the symmetry of for y
i
variables and applies the generalized
ansatz
=
R
[G, R]
M
i=1
F(
R
i
, y
G
i
) (3.23)
for 1 y
G
1
< y
G
2
< ... < y
G
M
N, then columns of M! M! dimensional matrix [G, R]
are related similarly to (3.8):
R
= Y
i,i+1
R
i
,R
i+1
R
. (3.24)
Y
operators are dened as

Y
l,m
i,j
=

ij
+

P
lm
1
ij
,
ij
=
ic
j
. (3.25)
For two-species fermions in conjugate representation we have

P
lm
= 1, and (3.25) is equiva-
lent to (3.20). For Bose-Fermi mixture in conjugate representation we have

P
lm
= 1, and
(3.25) is much more simple:
Y
l,m
i,j
= 1. (3.26)
Therefore, spin part of wave function is constructed by total antisymmetrization of single
spin wave functions, similar to Slater determinant for fermionic particles:
= det(F(
i
, y
j
)). (3.27)
Periodic boundary conditions for Bose-Fermi mixture are:
N
i=1
k
i
+ic/2
k
i
ic/2
= 1, = 1, ..., M, (3.28)
e
ik
j
L
=
M
=1
k
j

+ic/2
k
j

ic/2
, j = 1, ..., N. (3.29)
One can prove that all solutions of (3.28)-(3.29) are always real, which is a major simpli-
cation for the analysis of both ground and excited states (see Appendix B).
If one introduces function
(k) = 2 tan
1
(k/c), (3.30)
the system (3.28)-(3.29) can be rewritten as
k
j
L = 2I
j
+
M
=1
(2k
j
2
), (3.31)
2I
=
N
j=1
(2
2k
j
). (3.32)
I
j
and I
are integer or half integer quantum numbers (depending on the parity of M and
N), which characterize the state. The ground state corresponds to
I
= (M 1)/2, (M 3)/2, ...., (M 1)/2, (3.33)

I
j
= (N 1)/2, (N 3)/2, ...., (N 1)/2. (3.34)
In the thermodynamic limit, M, N, L have to approach innity proportionally. If one in-
troduces density of k roots (k) and density of roots (), (3.28)-(3.29) simplies to two
coupled integral equations
2(k) = 1 +
_
B
B
4c()d
c
2
+ 4( k)
2
, (3.35)
2() =
_
Q
Q
4c()d
c
2
+ 4( )
2
. (3.36)
Normalization conditions and energy are given by
N/L =
_
Q
Q
(k)dk, (3.37)
M/L =
_
B
B
()d, (3.38)
E/L =
_
Q
Q
k
2
(k)dk. (3.39)
These equations can be solved numerically and the results will be presented in the
next section. The numerical solution of these equations allows us to investigate the possi-
bility of phase separation predicted in [101]. Combined with local density approximation, it
can be used to investigate density proles and collective oscillation modes in external elds.
3.3 Numerical solution and analysis of instabilities
In this section we will solve the system of equations (3.35)-(3.39) numerically, and
obtain the ground state energy as a function of interaction strength and densities. This
solution will be used to analyze the instability to demixing[102, 101, 103].
Substituting (3.36) into (3.35), and performing analytically integration over , one
obtains an integral equation for function (k). Similarly to [40], it is convenient to redene
the variables before solving this equation numerically. Lets introduce the variables , x, y, b
and a function g(x) dened as
c = Q, = xQ, k = yQ, B = bQ, (Qx) = g(x). (3.40)
In new variables, integral equation depends on two parameters b and :
2g(y) = 1 +
_
1
1
2g(x)dx
2(
2
+ (x y)
2
)
(tan
1
2(b y)
+ tan
1
2(b x)
+
tan
1
2(b +y)
+ tan
1
2(b +x)
+

2(x y)
log

2
+ 4(b x)
2
2
+ 4(b y)
2
2
+ 4(b +y)
2
2
+ 4(b +x)
2
). (3.41)
Substituting for the new variables, (3.37)-(3.39) become
=
cL
N
=

_
1
1
g(x)dx
, (3.42)
M
N
=
_
1
1
(tan
1
2(bx)
+ tan
1
2(b+x)
)g(x)dx
_
1
1
g(x)dx
, (3.43)
E =
N
3
L
2
e(, b) =
N
3
L
2
_
1
1
x
2
g(x)dx
_
_
1
1
g(x)dx
_
3
. (3.44)
Integral equation (3.41) can be solved numerically as a function of two parameters b and
, applying Simpson rule for an integral approximation on a grid x
i
= 1 + (i 1)/n, i =
1, ..., 2n+1. This gives a system of 2n+1 linear equations for discrete values g(x
i
), which
can be solved by standard methods. Using (3.42)-(3.44), one can obtain parametrically three
functions (, b), M/N = (, b), e(, b). Then one can numerically invert two of them to
get (, ) and b(, ), and obtain function e(, ). The resulting function is shown in Fig.
3.1. When = 0, the system is purely fermionic, and non interacting. When = 1, the
system is purely bosonic, and numerically obtained energy coincides with the result of [40].
If = 0, bosons and fermions dont interact, and e(, ) = (
2
/3)(1 )
3
.
Tonks-Girardeau (TG) regime of strong interactions, 1, is an interesting case,
in which one can analytically nd the dependence of energies on relative densities. In (3.41)
0
0.2
0.4
0.6
0.8
1
0
5
10
15
0
1
2
3
0
0.2
0.4
0.6
0.8
model
LiebLiniger
e(,)
weak interactions
TG regime
f
r
e
e

f
e
r
m
i
o
n
s
Figure 3.1: Energy of the ground state is given by E = e(, ) h

2
N
3
/(2mL
2
), where =
mg/( h
2
n), and = M/N is the boson fraction. When = 0, the system is purely fermionic,
and the energy doesnt depend on interactions. When = 1, the system is purely bosonic,
and numerically obtained energy coincides with the result of [40]. If = 0, bosons and
fermions dont interact, and e(, ) = (
2
/3)(1 )
3
.
one can neglect the dependence of the kernel on x and y, and g(x) becomes a constant g,
which satises the equation
2g = 1 +
8g
_
tan
1
2b
+
2b
2
+ 4b
2
_
, (3.45)
while (3.43) reads
=
2
tan
1
2b
. (3.46)
After some algebra energy is rewritten as
e(, ) =

2
3
_
1
4
( +
sin
) +
12
2
( +
sin
)
2
_
+O(
1
3
). (3.47)
Using exact solutions, one can analyze demixing instabilities[102, 101, 103] for
repulsive Bose-Fermi mixtures. In the absence of external potential the Bose-Fermi mixture
is stable, if the compressibility matrix
_

b
n
b
b
n
f
f
n
b
f
n
f
_
(3.48)
is positively dened. Here, n
b
is the boson density, and n
f
is the fermion density while
b
and
f
are the bosonic and fermionic chemical potentials, given by
b
=
N
2
L
2
_
3e(, )
e
+ (1 )
e
_
, (3.49)
f
=
N
2
L
2
_
3e(, )
e
_
. (3.50)
The fact that the matrix (3.48) is positively dened can be checked numerically
for any value of and , and proves that Bose-Fermi mixture with the same Bose-Fermi
and Bose-Bose interactions is stable with respect to demixing for any values of boson and
fermion densities. We note, that the absence of demixing for one particular value of the
density has been checked in the original article by Lai and Yang[107]. Although an exact
solution is available only under conditions (3.2), small deviations from these should not
dramatically change the energy e(, ). Therefore, we expect the 1D mixtures to remain
stable to demixing in the vicinity of the integrable line (3.2) for any interaction strength.
Recently this has been checked numerically in Quantum Monte Carlo studies for a systems
of up to 14 atoms[103].
Note, that prediction of [101] about demixing at suciently strong interactions
in this case is incorrect, since it is based on the mean eld approximation. Indeed, the
demixing condition there reads
n
f

m
f
g
2
bf
g
bb
h
2
2
. (3.51)
For g
bf
= g
bb
and n
b
= n
f
this is equivalent to
=
mg
h
2
n

2
2
4.9. (3.52)
Clearly, this condition is incompatible with mean-eld approximation, which is valid for
< 1.
For weakly interacting case one can use mean eld approximation to calculate
energy and chemical potentials[101, 108] :
E = L
_
g
2
n
2
b
+gn
b
n
f
+
h
2
2
2m
n
3
f
3
_
,
b
= g(n
b
+n
f
),
f
= gn
b
+
h
2
2
2m
n
2
f
. (3.53)
For the strong interactions, up to corrections of order 1/
3
,
E = L
_
h
2
2
2m
(n
f
+n
b
)
3
3
_ _
1
4
( +
sin
) +
12
2
( +
sin
)
2
_
, (3.54)
f
=
h
2
2
2m
(n
f
+n
b
)
2
(1 +
1
3
_
16( +
sin
) + 4(1 + cos )
_
+
+
1
2
( +
sin
)
_
20( +
sin
) 8(1 + cos )
_
), (3.55)
b
=
h
2
2
2m
(n
f
+n
b
)
2
(1 +
1
3
_
16( +
sin
) + 4( 1)(1 + cos )
_
+
+
1
2
( +
sin
)
_
20( +
sin
) + 8(1 )(1 + cos )

_
). (3.56)
3.4 Local density approximation and collective modes
So far our arguments have been limited to the case of periodic boundary conditions
without external connement. This is the case in which the many-body interacting model
(3.1) is exactly solvable in the mathematical sense. If one adds an external harmonic po-
tential, strictly speaking the model is not solvable any more. However, if external potential
varies slowly enough (precise conditions for the case of Bose gas have been formulated in
[118]), one can safely use local density approximation (LDA) to analyze the density proles
and collective modes in a harmonic trap. In the local density approximation, one assumes
that in a slowly varying external harmonic trap the chemical potential changes according
to
0
b
(x) +
m
2
b
x
2
2
=
0
b
(0),
0
f
(x) +
m
2
f
x
2
2
=
0
f
(0). (3.57)
Let us consider the case when external harmonic conning potential oscillator
frequencies are the same for bosons and fermions. We note, however, that one can also
analyze the case when
b
,=
f
in a similar way. We consider
b
=
f
=
0
, (3.58)
since in this case the distribution of the relative boson and fermion densities is controlled
only by interactions, and not by external potential, which couples only to total density.
Eqs. (3.57) with
b
=
f
=
0
imply that densities of bosons and fermions in the region
where bosons and fermions coexist are governed by
0
f
(x) +
m
2
0
x
2
2
=
0
f
(0),
0
b
(x)
0
f
(x) = (n
f
+n
b
)
2
e
=
0
b
(0)
0
f
(0). (3.59)
One can show, that these equations cannot be simultaneously satised for the whole cloud,
and the mixture phase separates in an external potential given by (3.58). For both strong
and weak interactions bosons and fermions coexist in the central part, but the outer sections
consist of Fermi gas only. In the weakly interacting limit, this can be interpreted as an eect
of the Fermi pressure[19]: while bosons can condense to the center of the trap, Pauli principle
bosons
fermions
x
x
f
-1 -0.5 0.5 1
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Figure 3.2: Densities of Bose and Fermi gases in weakly interacting regime at zero temper-
ature. Lieb-Liniger parameter in the center of a trap is
0
= 0.18, overall number of bosons
equals number of fermions. Total density in the center of a trap is taken to be 1.
pushes fermions apart. As interactions get stronger, the relative distribution of bosons and
fermions changes. Figs. 3.2 and 3.3 contrast the limits of strong and weak interactions. For
strong interactions, the fermion density shows strong non-monotonous behavior.
When interactions are small, Eqs. (3.53) and (3.59) imply that in the region of
coexistence the densities are given by
n
0
b
(x) = n
0
b
(0)(1
x
2
x
2
b
), n
0
f
(x) = n
0
f
(0), for x
2
< x
2
b
. (3.60)
Outside of the region of coexistence, density of fermions decays as the square root of inverse
parabola:
n
0
b
(x) = 0, n
0
f
(x) =
n
0
f
(0)
_
1
x
2
b
x
2
f
1
x
2
x
2
f
, for x
2
b
< x
2
< x
2
f
. (3.61)
Parameters x
f
and x
b
are given by
x
2
b
=
2gn
0
b
(0)
m
2
0
, x
2
f
= x
2
b
+
( hn
0
f
(0))
2
(m
0
)
2
. (3.62)
A typical graph of density distribution for weakly interacting case is shown is
shown in Fig. 3.2.
If eective
0
is much bigger than 1 in the center of a harmonic trap, the total
density n
0
(x) follows Tonks-Girardeau density prole
n
0
(x) = n
0
(0)
1
x
2
x
2
f
. (3.63)
From Eqs. (3.47) and (3.59) the distribution of (x) is controlled by the following equation:
n
0
(x)
3
(1 + cos ((x))) = n
0
(0)
3
(1 + cos ((0))). (3.64)
x
x
f
bosons fermions
-1 -0.5 0.5 1
0.1
0.2
0.3
0.4
0.5
0.6
Figure 3.3: Densities of Bose and Fermi gases in strongly interacting regime at zero temper-
ature. Lieb-Liniger parameter in the center of a trap is
0
1, overall number of bosons
equals number of fermions. Total density in the center of the trap is taken to be 1.
Since 1 + cos ((x)) is bounded and n
0
(x) goes to 0 near the edges of the cloud, this
equation cant be satised for all x
2
< x
2
f
, which means that only fermions will be present
at the edges of the cloud, similarly to weakly interacting regime. Density distribution for
equal number of bosons and fermions is shown in Fig. 3.3. The form of the prole is
universal, as long as
0
1 and the temperature is zero. The evolution of this prole for
nonzero temperatures is shown in Fig. 3.14.
Recent experiments[119] demonstrated that collective oscillations of 1D gas provide
useful information about interactions in the system. Here we will numerically investigate
collective modes of the system, by solving hydrodynamic equations of motion. These equa-
tions have to be solved with proper boundary conditions at the edge of the bosonic and
fermionic clouds. Within the region of coexistence of bosons and fermions, such oscillations
can be described by four hydrodynamic equations[123]
t
n
b
+

x
(n
b
v
b
) = 0, (3.65)
m

t
v
b
+

x
(
b
+V
ext,b
+
1
2
mv
2
b
) = 0, (3.66)
t
n
f
+

x
(n
f
v
f
) = 0, (3.67)
m

t
v
f
+

x
(
f
+V
ext,f
+
1
2
mv
2
f
) = 0. (3.68)
In certain cases, analytical solutions of hydrodynamic equations are available[122,
123] and provide the frequencies of collective modes. When an analytic solution is not
available, the sum rule approach has been used[122, 123, 124, 125] to obtain an upper
bound for the frequencies of collective excitations. The disadvantage of the latter approach
is an ambiguity in the choice of multipole operator which excites a particular mode, espe-
cially for multicomponent systems[125]. Here we develop an ecient numerical procedure
for solving hydrodynamical equations in 1D, which doesnt involve additional sum rule
approximation.
While looking at low amplitude oscillations, it is sucient to substitute
n
b
(x, t) = n
0
b
(x) +n
b
(x)e
it
, (3.69)
n
b
(x, t)v
b
(x, t) = n
0
b
(x)v
b
(x)e
it
, (3.70)
n
f
(x, t) = n
0
f
(x) +n
f
(x)e
it
, (3.71)
n
f
(x, t)v
f
(x, t) = n
0
f
(x, t)v
f
(x)e
it
, (3.72)
b
+V
ext,b
+
1
2
mv
2
b
= const
1
+
b
(x)e
it
=
const
1
+ (n
b
(x)
b
n
b
+n
f
(x)
b
n
b
)e
it
, (3.73)
f
+V
ext,f
+
1
2
mv
2
f
= const
2
+
f
(x)e
it
=
const
2
+ (n
b
(x)
f
n
b
+n
f
(x)
f
n
f
)e
it
. (3.74)
Here, n
0
b
(x) and n
0
f
(x) are densities obtained within local density approximation. The
linearized system of hydrodynamic equations can be written as:
m
2
_
n
b
(x)
n
f
(x)
_
=
_
_
n
0
b
(x) 0
0 n
0
f
(x)
_
__

b
n
b
b
n
f
f
n
b
f
n
f
_
_
n
b
(x)
n
f
(x)
_
__
. (3.75)
For numerical solutions and boundary conditions it is more convenient to work with inde-
pendent functions
b
(x),
f
(x). The system of equations becomes
m
2
_

b
(x)
f
(x)
_
=
_

b
n
b
b
n
f
f
n
b
f
n
f
_
__
n
0
b
(x) 0
0 n
0
f
(x)
_
_

b
(x)
f
(x)
__
. (3.76)
Outside of the region of coexistence of bosons and fermions,
out
f
satises the following
equation:
m
2
out
f
=
out
f
n
f
_
n
out
f
(x)
out
f
. (3.77)
All modes can be classied by their parity with respect to x x substitution, and
will be investigated by parity-dependent numerical procedure. We will consider equations
only in the positive half of the cloud. For even modes, one may require two additional
conditions:
f
(x = 0) = 0,
b
(x = 0) = 0. (3.78)
For odd modes, analogous conditions are
f
(x = 0) = 0,
b
(x = 0) = 0. (3.79)
Boundary conditions for fermions at the edge of the bosonic cloud, x
b
, correspond
to the continuity of v
f
and
f
. Continuity of the velocity can be obtained by integrating
continuity equation (3.67) in the vicinity of x
b
. From Eq. (3.68) it is equivalent to
out
f
(x = x
b
+ 0) =
f
(x = x
b
0). (3.80)
The second condition can be obtained by integrating (3.68) in the vicinity of x
b
:
out
f
(x = x
b
+ 0) =
f
(x = x
b
0). (3.81)
One may see, that these conditions do not imply that n
out
f
(x = x
b
+0) = n
f
(x =
x
b
0). This can be easily illustrated by the dipole mode, where v
f
(x) = v
b
(x) =
const, n
f
= n
0
f
(x), which is clearly discontinuous for proles shown in Figs. 3.2 and
3.3.
Two additional conditions come from the absence of the bosonic(fermionic) ow
at x
b
(x
f
) :
n
0
b
(x)v
b
(x)[
xx
b
0
= 0, (3.82)
n
0
f
(x)v
f
(x)[
xx
f
0
= 0. (3.83)
Outside of the region of coexistence, the chemical potential and density of fermions
are given by
out
f
(n
out
f
)
2
, n
out
f

_
1 (x/x
f
)
2
, where x
f
is the fermionic cloud size. In
dimensionless variables u = x/x
f
, Eq. (3.77) can be written as
2
0
out
f
= (1 u
2
)
out
f
u
2
u
out
f
u
. (3.84)
For this equation, there exists a general nonzero solution which satises (3.83):
out
f
= cos(

0
arccos
x
x
f
). (3.85)
Substituting this into (3.80)-(3.81), one has to solve eigen mode equations numer-
ically for x < x
b
, with ve boundary conditions (3.80),(3.81),(3.82) and (3.78) or (3.79)
depending on the parity. These boundary conditions are compatible, only if is an eigen
frequency. Using four of these boundary conditions, the system of two second order dif-
ferential equations can be solved numerically for any . To nd a numerical solution we
choose to leave out condition (3.82), and check later if it is satised to identify the eigen
frequencies.
The most precise way to check (3.82) numerically is based on equations of motion.
For even modes, v
b
(0) = 0, and integrating (3.65) from 0 till x
b
, one obtains
_
x
b
0
n
b
(x)dx =
1
i
(n
b
(x = x
b
)v
b
(x = x
b
) n
b
(x = 0)v
b
(x = 0)]) = 0. (3.86)
For odd modes, from Eq. (3.66) v
b
(0) = i
b
(x = 0)/(m), and integrating (3.65) from 0
till x
b
, one obtains
_
x
b
0
n
b
(x)dx =
1
i
(n
b
(x = x
b
)v
b
(x = x
b
) n
b
(x = 0)v
b
(x = 0)]) =
n
b
(x = 0)
b
(x = 0)
m
2
. (3.87)
0
"out of phase" dipole mode

"in phase" breathing mode
"out of phase" breathing mode
0.2 0.4 0.6 0.8 1 1.2
0.75
1
1.25
1.5
1.75
2
2.25
"in phase" dipole mode
Figure 3.4: Frequencies of collective excitations in mean eld regime versus Lieb-Liniger
parameter in the center of a trap
0
. Total number of bosons equals the number of fermions.
Even in mean eld regime frequency of out of phase oscillations gets smaller as interactions
get stronger.
When a numerical solution for
b
(x),
f
(x) is available, conditions (3.86) or (3.87) can
be checked numerically using
n
b
(x) =
f
n
f
b
(x)

f
n
b
f
(x)
f
n
f
b
n
b

f
n
b
b
n
f
. (3.88)
First we apply this numerical procedure for weakly-interacting regime, and the
frequencies of collective modes are shown in Fig. 3.4. When
0
0, boson and fermion
clouds do not interact, and collective modes coincide with purely bosonic or fermionic
modes, with frequencies[123]
f
= n
0
and
b
=
0
_
n(n + 1)/2. Modes which corre-
spond to /
0
= 1,
3, 2,
6 are shown in Fig. 3.4. As interactions get stronger, bo-

son and fermion clouds get coupled, and all the modes except for Kohn dipole mode
change their frequency. For Kohn dipole mode, boson and fermion density uctuations
are given by n
f
= n
0
f
(x), n
b
= n
0
b
(x). In Figs 3.5-3.7 we show density uctuations
for three other modes in the region of coexistence for a particular choice of parameters
0
= 0.394, x
b
/x
f
= 0.6 and equal total number of bosons and fermions. Modes for which
the frequency goes down due to coupling between boson and fermion clouds correspond to
the collective excitations with opposite signs in density uctuations of boson and fermion
clouds. In TG regime these modes continuously transform into out of phase low-lying
modes which do not change the total density. For weak interactions lowest mode is an out
of phase dipole excitation, after that comes in phase Kohn dipole mode (center of mass
oscillation), out of phase even mode, in phase even mode, second out of phase odd
mode.
Lets consider the Tonks-Girardeau regime, when energy is well approximated by
(3.47). Since dependence of the energy on relative boson fraction (x) is 1/ times smaller
n
f
n
b
0.1 0.2 0.3 0.4 0.5 0.6

-2.5
-2
-1.5
-1
-0.5
0.5
f
arb. units
x
x
Figure 3.5: Fermion and boson density uctuations of out of phase dipole mode
(/
0
)
2
= 0.757 for
0
= 0.394, x
b
/x
f
= 0.6. Total number of bosons equals the num-
ber of fermions. Outside of the region of coexistence of boson and fermion clouds,
n
f
(x)
1
1(x/x
f
)
2
cos(

0
arccos
x
x
f
)
f
x
x
n
f
n
b
0.1 0.2 0.3 0.4 0.5 0.6

-10
-5
5
arb. units
Figure 3.6: Fermion and boson density uctuations of out of phase breathing mode
(/
0
)
2
= 2.51 for
0
= 0.394, x
b
/x
f
= 0.6. Total number of bosons equals the number
of fermions. Outside of the region of coexistence of boson and fermion clouds, n
f
(x)
1
1(x/x
f
)
2
cos(

0
arccos
x
x
f
)
f
x
x
n
f
n
b
0.1 0.2 0.3 0.4 0.5 0.6

-0.2
0.2
0.4
0.6
arb. units
Figure 3.7: Fermion and boson density uctuations of in phase breathing mode
(/
0
)
2
= 3.585 for
0
= 0.394, x
b
/x
f
= 0.6. Total number of bosons equals the num-
ber of fermions. Outside of the region of coexistence of boson and fermion clouds,
n
f
(x)
1
1(x/x
f
)
2
cos(

0
arccos
x
x
f
)
0.2 0.4 0.6 0.8 1

2
4
6
8
10
Figure 3.8: Dependence of the frequency of lowest lying out of phase modes for
0
1
on overall boson fraction , where
0
is the Lieb-Liniger parameter in the center of a
trap. Characteristic scale of out of phase oscillations in strongly interacting regime is
0
/
0

0
. Total density in phase modes
n
= n
0
have much higher frequency for
0
1 and are not shown here.
than dependence on the total density, the energetic penalty for changing relative density
of bosons and fermions is small. Thus there should be low-lying modes, which correspond
to an oscillation of the relative density between bosons and fermions, while total density is
kept xed up to 1/ corrections. In addition to these low-lying out of phase oscillations
of boson and fermion clouds, there will be in phase density modes, which correspond to
oscillations of the total density. Since up to 1/ corrections dependence of the energy on
total density in TG regime is the same as for free non interacting fermions, energy of these
excitations is given by [123] = n
0
, up to small corrections of the order of 1/.
When , relative compressibility goes to zero as 1/, so from Eq. (3.76)
the energy of low-lying modes goes to zero as 1/
0
, where
0
is a Lieb-Liniger parameter
in the center of a trap. Performing a numerical procedure outlined above, one can obtain
the dependence of the frequencies of low-lying out of phase modes on relative density
of bosons and fermions. The results of these calculations are shown in Fig. 3.8, and are
parameterized by the overall boson fraction and
0
. It turns out that the lowest lying mode
is odd, and after that the parity of collective excitations alternates signs. For out of
phase modes the signs of density uctuations and velocities of boson and fermion clouds
are opposite. One can easily understand, why the energy grows as the boson fraction is
decreased: the size of the boson cloud shrinks, and the wave vector of the corresponding
excitation increases, leading to an increase of frequency. One should note that for very small
overall boson fraction
0
1 is not enough to separate energy scales for out of phase and
in phase oscillations, and also conditions for applicability of LDA become more stringent.
3.5 Zero-temperature correlation functions in Tonks-Girardeau
regime
The calculation of the collective modes in the previous section relies only on the
dependence of the energy e(, ) on densities of bosons and fermions. Collective modes can
be used in experiments [119, 120] to check to some extent quantitatively the equation of the
state of the system[121]. However, only some part of the information about the ground state
properties is encoded in the energy: indeed, the energy and collective modes of the strongly
interacting Lieb-Liniger gas are the same as for free fermions[41, 40], while the correla-
tion functions are dramatically dierent[126]. Single particle correlation functions can be
measured experimentally using Bragg spectroscopy[130] or time of ight measurements[43].
Generally, it is much harder to calculate the correlation functions than the energy from
Bethe ansatz solution. Most of the progress in this direction has been achieved for the case
of strong interactions[138]. Recently, there have been some reports [127], in which pseud-
ofermionization method has been used to calculate correlation functions for spin-
1
2
fermion
Hubbard model for intermediate interaction strengths. In this section, we will analyze the
correlation functions in the regime of strong interactions, using the factorization of orbital
and spin degrees of freedom similarly to the case of spin-
1
2
fermions[128, 129]. Our calcu-
lations in this section are performed for the periodic boundary conditions, when the many
body problem is strictly solvable in the mathematical sense. We will obtain a representation
of correlation functions through the determinants of some matrices, with the size of these
matrices scaling linearly with the number of the particles. These determinants can be eas-
ily evaluated numerically, and provide a straightforward way to study correlation functions
quantitatively at all distance scales. This determinant representation can be generalized to
nonzero temperatures, and the results of this generalization will be presented in the next
section.
3.5.1 Factorization of spin and orbital degrees of freedom
The regime of strong interactions can be investigated by neglecting k
i
compared
to
, c in (3.28)-(3.29). A simplied system of equations for spectral parameters is

_
+ic/2
ic/2
_
N
= 1, = 1, ..., M, (3.89)
e
ik
j
L
=
M
=1
+ic/2
ic/2
, j = 1, ..., N. (3.90)
We see that the spin part is decoupled from orbital degrees in the Bethe equations.
Equation (3.89) for ground state spin rapidities can be resolved as
+ic/2
ic/2
= e
i2/N
, = 1, ..., M, (3.91)
where
is a set of integer spin wave vectors. Since the details of calculations depend on
the parity of M and N, from now on we will assume that N is even, and M is odd. Ground
state corresponds to
occupying Fermi sea (, ), so from (3.91) ground state spin

wave vectors are
i
= (M 1)/2 +N/2, ..., N/2, ..., (M 1)/2 +N/2. (3.92)
This choice of spin wave vectors will be justied later, in section 3.6. From equation
(3.90) it follows that ground state orbital wave vectors are
k
i
= (N 1)/L, ..., /L, /L, ..., (N 1)/L. (3.93)
Eq. (3.16) for F(, y) simplies to
F(
, y
i
) =
_
+ic/2
+ic/2
_
y
i
1
= e
i
2
N
(y
i
1)
, (3.94)
and spin wave function (3.27) can be represented as a Slater determinant of M single
particle plane waves in spin space:
= e
i
2
N
P
det[e
i
2
N

i
y
j
]. (3.95)
Orbital part of the wave function also simplies into a Slater determinant, since all Yang
matrices Y
a,b
i,j
in (3.9) are equal to 1.
The ground state is written as a product of two Slater determinants, describing
orbital and spin degrees of freedom:
(x
1
, ..., x
N
) det[e
ik
i
x
j
] det[e
i
2
N

i
y
j
]. (3.96)
Here x
1
, ..., x
M
are coordinates of bosons, x
M+1
, ..., x
N
are coordinates of fermions, and y
i
is the order in which the particle x
i
appears, if the set x
1
, ..., x
N
is ordered. In other words,
if
0 x
Q
1
x
Q
2
.... x
Q
N
L, (3.97)
then y
1
, ..., y
N
= Q
1
(1), ..., Q
1
(N). (3.98)
The rst determinant depends on positions of both bosons and fermions, while the second
determinant depends only on relative positions of bosons y
1
, ..., y
M
. Normalization prefactor
will be determined later to give a correct value of the density. One can conrm that
symmetry properties of wave function are as required: transposition of two fermions aects
only rst determinant, therefore wave function acquires 1 sign. Transposition of two
bosons changes signs of both rst and second determinants, so the wave function doesnt
change.
Similar factorization of wave function into spin and orbital degrees of freedom has
been observed in [129] for one dimensional spin-
1
2
Hubbard model. In that case, spin wave
function is a ground state of spin-
1
2
antiferromagnetic Heisenberg model, and is much more
complicated than (3.95).
It might seem that spin degrees are now independent of orbital degrees, but
this is not true, since it is the relative position of orbital degrees which determines spin
coordinates. If one wants to calculate, for example, Bose-Bose correlation function, one has
to x position of x
1
and x
1
and integrate (x
1
, x
2
, .., x
N
)
(x
1
, x
2
..., x
N
) over x
2
, ..., x
N
.
However, there are C
M
N
inequivalent spin distributions, and integration in each subspace
(3.97) has to be performed separately. For spin-
1
2
fermions on a lattice in [129] this inte-
gration becomes a summation, and it has been done numerically for up to 32 cites. This
summation requires computational resources which scale as an exponential of the number
of particles. Here we will report a method to perform integrations for a polynomial time,
which will allow to go for larger system sizes (easily up to 100 on a desktop PC) and study
correlation functions much more accurately.
3.5.2 Bose-Bose correlation function
Lets describe a procedure to calculate Bose-Bose correlation functions of the
model. First, we will use translational symmetry of the model to x the positions of the
rst particle at points x
1
= 0, x
1
= . Instead of writing wave function as a function of
positions of M bosons and N M fermions, lets introduce a set of N ordered variables
Z = 0 z
1
z
2
... z
N
L, (3.99)
which describe positions of the atoms, without specication of bosonic or fermionic nature
of the particle. If any two particles exchange their positions, they are described by the
same set (3.99). In addition to (3.99) one has to introduce a permutation y which species
positions of bosons: y
1
, ..., y
M
are boson positions, and y
M+1
, ..., y
N
are fermion positions
in an auxiliary lattice: z
y
i
= x
i
. In this new parameterization the normalized wave function
is (normalization will be derived later in this subsection)
(z
1
, z
2
, .., z
N
; y) =
1
_
(N M)!M!L
N
N
M
det[e
ik
i
z
j
] det[e
i
2
N

i
y
j
](1)
y
. (3.100)
Here and later we denote a sign factor
(1)
y
=
Ni>j1
Sign(y
i
y
j
). (3.101)
One should note, that second determinant has a size MM, and depends only on y
1
, ..., y
M
.
Dependence of wave function on y
M+1
, ..., y
N
comes only through sign prefactor. For each
particular set of y
1
, ..., y
M
there are (N M)! dierent congurations of y
M+1
, ..., y
N
, for
which wave function only changes its sign depending on relative positions of y
M+1
, ..., y
N
.
To calculate correlation function we should be able to calculate a product of wave
functions at the points
x
1
= 0, x
1
= , x
2
= x
2
, ...., x
N
= x
N
. (3.102)
Let Z be an ordered set for x
i
variables:
Z = z
1
= 0 z
2
... z
N
L. (3.103)
If we denote an ordered set for x
i
variables as Z
, then using (3.102) one can conclude that

Z
is obtained from Z by removing z

1
= 0, inserting an extra coordinate z
d
= , and shifting
variables which are to the left of it:
Z
= 0 z
1
= z
2
... z
d1
= z
d
z
d
= z
d+1
= z
d+1
... z
N
= z
N
L.
(3.104)
Spin states y and y
are connected by
y
1
= 1, y
1
= d,
y
i
= y
i
1, for 1 < y
i
d,
y
i
= y
i
, for d < y
i
. (3.105)
The correlation function can be written as
b
(0, ) = M
_
(0, x
2
, .., x
N
)
(, x
2
..., x
N
)dx
2
...dx
N
=
N
d=1
y
_
dz
2
...dz
N
(1)
y
(1)
y
(N M)!(M 1)!L
N
N
M
det[e
ik
i
z
j
]
det[e
i
2
N

i
y
j
] det[e
ik
i
z
j
] det[e
i
2
N

i
y
j
], (3.106)
where integration over dz
i
and summation over y are done subject to non uniform (3.104)-
(3.105). One can observe now, that limits of integration in (3.104) depend only on
and d. These limits are independent of y, and the function under integral factorizes into
zdependent and ydependent parts. Similarly, summation over y doesnt depend on
precise values of or z
i
, but the dependence comes through d. Therefore, the density matrix
can be written as
b
(0, ) =
1
(N M)!(M 1)!L
N
N
M
N
d=1
I(d, )S
b
(d), (3.107)
where I(d, ) is a an integral
I(d, ) =
_
det[e
ik
i
z
j
] det[e
ik
i
z
j
]dz
2
...dz
N
(3.108)
subject to non uniform (3.104), and S
b
(d) is an expectation value of a translation operator
over a symmetrized Slater determinant wave function:
S
b
(d) =
y
det[e
i
2
N

i
y
j
] det[e
i
2
N

i
y
j
](1)
y
(1)
y
. (3.109)
Normalization can be determined using the following argument: if = 0, then only con-
tribution from d = 1 does not vanish. One can calculate I(1, 0) = NL
N1
and S
b
(1) =
(N M)!(M)!N
M1
, since these follow from normalizations of orbital and spin wave
functions. Since we want
b
(0, 0) = M/L, we can x the normalization prefactor in (3.100).
Calculation of a many-body integral I(d, )
Lets describe the calculation of an integral I(d, ). From now on we will assume
that L = 1. First, since k
i
are equidistant wave vectors (3.93), one can use Vandermonde
formula to simplify the determinants:
det[e
ik
l
z
j
] = e
i(N1)(z
1
+...+z
N
)
det[e
i2(l1)z
j
] = e
i(N1)(z
1
+...+z
N
)
j1<j2
(e
i2z
j2
e
i2z
j1
), l = 1, ..., N,
det[e
ik
l
z
j
] = e
i(N1)(z
1
+...+z
N
)
det[e
i2(l1)z
j
] = e
i(N1)(z
1
+...+z
N
)
j1<j2
(e
i2z
j2
e
i2z
j1
), l = 1, ..., N. (3.110)
Using this representation, the fact that z
1
= 0 and (3.104), one can rewrite these N N
determinants as a product of (N 1) (N 1) determinant and a prefactor:
det[e
ik
l
z
j
] = e
i(N1)(t
1
+...+t
N1
)
det[e
i2(l1)t
l
]
N1
i=1
(e
i2t
i
1), l = 1, ..., N 1
det[e
ik
l
z
j
] = (1)
d1
e
i(N1)(+t
1
+...+t
N1
)
det[e
i2(l1)t
l
]
N1
i=1
(e
i2t
i
e
i2
), l = 1, ..., N 1, (3.111)
where we introduced N 1 variables of integration t
i
, so that
t
i
= z
i+1
. (3.112)
The factor (1)
d1
arises since z
d
= , and to write (3.111) we changed signs of d1 terms
in (3.110). Integration subspace is dened as
0 t
1
... t
d1
t
d
... t
N1
1 (3.113)
One can extend this subspace as follows:
T = 0 t
1
, ..., t
d1
t
d
, ..., t
N1
1. (3.114)
Indeed, the expression under integral doesnt change, when t
i
< and t
j
< change
their positions (similarly for t
i
> and t
j
> ), so this extension just adds prefactor
1/((d 1)!(N d)!). Finally, we have
I(d, ) =
(1)
d1
e
i(N1)
(d 1)!(N d)!
_
t
i
T
det[e
i2(l1)t
l
] det[e
i2(l1)t
l
]
N1
i=1
(e
i2t
i
1)(e
i2t
i
e
i2
)dt
1
...dt
N1
. (3.115)
At this point we use a trick from Ref. [126], in which Toeplitz determinant representation
for strongly interacting Bose gas was derived. Lets expand the determinants under integrals
using a permutation formula for determinants:
I(d, ) =
(1)
d1
e
i(N1)
(d 1)!(N d)!
_
t
i
T
PS
N1
S
N1
(1)
P
(1)
P
N1
i=1
e
i2((P
i
1)(P
i
1))t
i
(e
i2t
i
1)(e
i2t
i
e
i2
)dt
1
...dt
N1
(3.116)
From summation over P, P
we can go to summation over P, Q, where P
= QP. Also, one

can remove non uniform (3.114) by introducing two functions
f
1
(, t) = (e
i2t
1)(e
i2t
e
i2
) for t < , 0 otherwise,
f
2
(, t) = (e
i2t
1)(e
i2t
e
i2
) for t > , 0 otherwise. (3.117)
I(d, ) becomes
(1)
d1
e
i(N1)
(d 1)!(N d)!
PS
N1
QS
N1
(1)
Q
(
d1
i=1
_
1
0
e
i2(P
i
Q
P
i
)t
i
f
1
(, t
i
)dt
i
)
(
N1
i=d
_
1
0
e
i2(P
i
Q
P
i
)t
i
f
2
(, t
i
)dt
i
) (3.118)
If f
1
(, t) and f
2
(, t) were the same, as in Ref. [126], the expression being summed wouldnt
depend on P, and the summation over Q would give a determinant, with the same elements
along diagonals (Toeplitz determinant). In our case, for each given P the expression is P
dependent, and the result doesnt have the Toeplitz form. However, introducing additional
phase variable, one can recast the expression as an integral of some Toeplitz determinant.
The desired expression has the form:
I(d, ) = (1)
d1
e
i(N1)
_
2
0
d
2
e
i(d1)
Q
(1)
Q
N1
i=1
_
1
0
e
i2(iQ
i
)c
i
(e
i
f
1
(, c
i
) +f
2
(, c
i
))dc
i
), (3.119)
where c
i
is a dummy variable of integration. Integration over is analogous to projection of
the BCS wave function on a state with a xed number of particles. Nonzero terms appear
after integration over , if in the expansion of the product of brackets for some d1 brackets
f
1
is chosen instead of f
2
. If this choice is made at brackets with numbers P
1
, ..., P
d1
then
the contribution from such a choice exactly corresponds to a term in (3.118). However, each
choice of brackets corresponds to (N d)!(d 1)! dierent permutations, and this cancels
the same combinatoric factor in the denominator of (3.118).
Summation over Q is nothing but a determinant, and nally we have
I(d, ) = (1)
d1
e
i(N1)
_
2
0
d
2
e
i(d1)
det
_
_
c
0
() c
1
() ... c
N2
()
c
1
() c
0
() ... c
N3
()
... ... ... ...
c
(N2)
() c
(N3)
() ... c
0
()
_
_
, (3.120)
where
c
j
() =
_
1
0
e
ijx
(e
i
f
1
(, x) +f
2
(, x))dx (3.121)
The expression in (3.120) without an integral over is a generating function of I(d, )
with the weights e
i()(d1)
, and integration over extracts a particular term out of this
generating function.
What we achieved in this section is to represent a complicated N 1 fold integral
as an integral over one phase variable, which can be done numerically in a polynomial time
over N.
Calculation of S
b
(d)
Calculation of S
b
(d) is very similar in spirit to calculation of the previous subsec-
tion. Integration over x
i
corresponds to summation over y
i
, and corresponds to d. Final
result is a determinant of some matrix. Due to the shift operator (3.105) this determinant
does not have a Toeplitz form, but this is not important for a numerical evaluation.
We need to calculate
S
b
(d) =
y
det[e
i
2
N

i
y
j
] det[e
i
2
N

i
y
j
](1)
y
(1)
y
, (3.122)
where
i
is a set (3.92). Denition of y
according to (3.105) can be rewritten as

y
1
= 1, y
1
= d,
y
i
= y
i
+
Sign(y
i
d) 1
2
, i = 2, ..., N, (3.123)
where
Sign(x) = 1, x > 0,
Sign(x) = 1, x 0. (3.124)
Sign prefactor in (3.122) can be rewritten as
(1)
y
(1)
y
i>j
Sign(y
i
y
j
)
i>j
Sign(y
i
y
j
) =
N
i=2
Sign(y
i
d) = (1)
d1
. (3.125)
We see, that (3.122) depends only on y
1
, ..., y
M
, so from now on we will consider a summation
in y
1
, ..., y
M
variables. Summation over y
M+1
, ..., y
N
gives a trivial combinatorial prefactor
(N M)!. Furthermore, we can extend possible values of y
1
, ..., y
M
to y
i
= y
j
, i ,= j, since
for such congurations rst determinant in (3.122) is 0, and they dont change the value of
S
b
(d) :
y = 1 y
2
, y
3
, ..., y
M
N. (3.126)
Lets use the fact that
i
is a set of equidistant numbers (3.92), and rewrite de-
terminants using Vandermonde formula, similar to (3.110):
det[e
i
2
N

i
y
j
] = e
i
2
N
((M1)/2+N/2)(1+...+y
M
)
det[e
i
2
N
(l1)y
j
] =
e
i
2
N
((M1)/2+N/2)(1+...+y
M
)
j1<j2
(e
i
2
N
y
j2
e
i
2
N
y
j1
), l = 1, ..., M,
det[e
i
2
N

i
y
j
] = e
i
2
N
((M1)/2+N/2)(d+...+y
M
)
det[e
i
2
N
(l1)y
j
] =
e
i
2
N
((M1)/2+N/2)(d+...+y
M
)
j1<j2
(e
i
2
N
y
j2
e
i
2
N
y
j1
), l = 1, ..., M. (3.127)
For simplicity of notations later, lets introduce t
i
= y
i+1
, t
i
= y
i+1
, i = 1, ..., M 1.
Analogously to (3.111), we extract a determinant of (M 1) (M 1) matrix out of
Vandermonde product:
det[e
i
2
N

i
y
j
] = e
i
2
N
((M1)/2+N/2)(1+t
1
...+t
M1
)
det[e
i
2
N
(l1)t
j
]
M1
i=1
(e
i
2
N
t
i
e
i
2
N
1
), l = 1, ..., M 1,
det[e
i
2
N

i
y
j
] = e
i
2
N
((M1)/2+N/2)(d+t
1
...+t
M1
)
det[e
i
2
N
(l1)t
j
]
M1
i=1
(e
i
2
N
t
i
e
i
2
N
d
), l = 1, ..., M 1. (3.128)
At this point we need to represent the subspace of summation (3.126) as a sum over M
inequivalent partitions, similar to representation (3.107):
S
b
(d) =
M
r=1
(N M)!(M 1)!
(r 1)!(M r)!
S
b
(d, r), (3.129)
where S
b
(d, r) is a result of summation in the T
r
subspace:
T
r
= 1 t
1
, ..., t
r1
d < t
r
, ..., t
M1
N. (3.130)
Note, that S
b
(d, r) = 0 for r > d, since in this case two of t
1
, ..., t
r1
should coincide, and
the wave function becomes 0. Calculation of S
b
(d, r) is very similar to calculation of I(, d).
Lets expand the determinants (3.128) using permutations:
S
b
(d, r) = (1)
d1
e
i
2
N
((M1)/2+N/2)(rd)

PS
M1
S
M1
(1)
P
(1)
P
r1
i=1
(
d
t
i
=1
e
i
2
N
((P
i
1)t
i
(P
i
1)(t
i
1))
(e
i
2
N
t
i
e
i
2
N
)(e
i
2
N
(t
i
1)
e
i
2
N
d
))
M1
i=r
(
N
t
i
=d+1
e
i
2
N
((P
i
1)(P
i
1))t
i
(e
i
2
N
t
i
e
i
2
N
)(e
i
2
N
t
i
e
i
2
N
d
)) (3.131)
= QP. Also, one

can analytically perform summation over t
i
in each of the brackets, since it is a combination
of geometrical progressions (this is analogous to integration over t
i
variables in previous
subsection):
S
b
(d, r) = (1)
d1
e
i
2
N
((M1)/2+N/2)(rd)
PS
M1
QS
M1
(1)
Q
r1
i=1
c
1
(d, Q
P
i
, P
i
)
M1
i=r
c
2
(d, Q
P
i
, P
i
), (3.132)
where
c
1
(d, j, l) = e
i
2
N
(l1)
t=d
t=1
e
i
2
N
(jl)t
(e
i
2
N
d
e
i
2
N
)(e
i
2
N
(t1)
e
i
2
N
d
),
c
2
(d, j, l) =
t=N
t=d+1
e
i
2
N
(jl)t
(e
i
2
N
d
e
i
2
N
)(e
i
2
N
t
e
i
2
N
d
) (3.133)
are independent of r. At this point, we can use the phase variable integration trick to get
rid of summation over P, and then represent summation over Q as a determinant:
S
b
(d, r) = (r 1)!(M r)!(1)
d1
e
i
2
N
((M1)/2+N/2)(rd)
_
2
0
d
2
e
i(r1)
det
_
_
c(, 1, 1) c(, 2, 1) ... c(, M 1, 1)
c(, 1, 2) c(, 2, 2) ... c(, M 1, 2)
... ... ... ...
c(, 1, M 1) c(, 2, M 1) ... c(, M 1, M 1)
_
_
,
where
c(, j, l) = e
i
c
1
(d, j, l) +c
2
(d, j, l). (3.134)
We can analytically perform summation over r in (3.129), since the determinant and c(, j, l)
are independent of r :
S
b
(d) =
M
r=1
(N M)!(M 1)!
(r 1)!(M r)!
S
b
(d, r) = (N M)!(M 1)!(1)
d1
e
i
2
N
((M1)/2+N/2)(d1)
_
2
0
d
2
(
M
r=1
e
i(
2
N
((M1)/2+N/2))(r1)
)
det
_
_
c(, 1, 1) c(, 2, 1) ... c(, M 1, 1)
c(, 1, 2) c(, 2, 2) ... c(, M 1, 2)
... ... ... ...
c(, 1, M 1) c(, 2, M 1) ... c(, M 1, M 1)
_
_
. (3.135)
Expansion of the determinant (3.135) in a series over e
i
has terms up to e
i(M1)
:
det() =
M1
n=0
f
n
e
in
. (3.136)
Summation over r and integration over lead to
_
2
0
d
2
(
M
r=1
e
i(
2
N
((M1)/2+N/2))(r1)
) det() =
_
2
0
d
2
M1
n=0
M
r=1
f
n
e
i(
2
N
((M1)/2+N/2))(r1)+in
=
M1
n=0
f
n
e
i
2
N
((M1)/2+N/2)n
= det(
2
N
((M 1)/2 +N/2)). (3.137)
Finally, if we introduce a notation
0
= 2((M 1)/2 +N/2)/N,
S
b
(d) = (N M)!(M 1)!(1)
d1
e
i
2
N
((M1)/2+N/2)(d1)
det
_
_
c(
0
, 1, 1) c(
0
, 2, 1) ... c(
0
, M 1, 1)
c(
0
, 1, 2) c(
0
, 2, 2) ... c(
0
, M 1, 2)
... ... ... ...
c(
0
, 1, M 1) c(
0
, 2, M 1) ... c(
0
, M 1, M 1)
_
_
. (3.138)
3.5.3 Fermi-Fermi correlation function
Calculation of fermionic correlation function closely reminds of the calculation of
Bose-Bose correlation function, so we will be relatively schematic in our derivations. First,
one splits integration into integration over orbital coordinates z
i
from the set
Z = 0 z
1
z
2
... z
N
L, (3.139)
and summation over spin variables. Integration over orbital variables is absolutely iden-
tical to the Bose-Bose case, the dierence comes only from spin part S
f
(d) :
f
(0, ) = (N M)
_
(x
1
, x
2
, ..., 0)
(x
1
, x
2
..., )dx
1
...dx
N1
=
N
d=1
y
_
dz
2
...dz
N
(1)
y
(1)
y
(N M 1)!M!L
N
N
M
det[e
ik
i
z
j
] det[e
i
2
N

i
y
j
]
det[e
ik
i
z
j
] det[e
i
2
N

i
y
j
] =
1
(N M 1)!M!L
N
N
M
N
d=1
I(d, )S
f
(d), (3.140)
where I(d, ) is given by (3.120), and
S
f
(d) =
y
det[e
i
2
N

i
y
j
] det[e
i
2
N

i
y
j
](1)
y
(1)
y
. (3.141)
In (3.141) y
and y are related by

y
N
= 1, y
N
= d,
y
i
= y
i
+
Sign(y
i
d) 1
2
, i = 1, ..., N 1. (3.142)
Similarly to (3.125) sign prefactor can be rewritten as
(1)
y
(1)
y
i>j
Sign(y
i
y
j
)
i>j
Sign(y
i
y
j
) = (1)
N1
N1
j=1
Sign(d y
j
) = (1)
d1
.
(3.143)
We see that (3.141) depends only on y
1
, ..., y
M
, so from now on we will consider a
summation in y
1
, ..., y
M
variables. Summation over y
M+1
, ..., y
N1
gives a trivial combina-
torial prefactor (N M 1)!. Furthermore, we can extend possible values of y
1
, ..., y
M
to
y
i
= y
j
, i ,= j, since for such congurations the rst determinant in (3.141) is 0, and they
dont change the value of S
f
(d) :
y = 2 y
1
, y
2
, ..., y
M
N. (3.144)
We can to represent the subspace of summation (3.144) as a sum of M + 1 in-
equivalent partitions, similar to representation (3.129):
S
f
(d) =
M+1
r=1
(N M 1)!(M)!
(r 1)!(M r + 1)!
S
f
(d, r), (3.145)
where S
f
(d, r) is a result of summation in the T
r
subspace:
T
r
= 2 t
1
, ..., t
r1
d < t
r
, ..., t
M
N. (3.146)
The product of two determinants in (3.141) is rewritten as
det[e
i
2
N

i
y
j
] det[e
i
2
N

i
y
j
] = e
i
2
N
((M1)/2+N/2)(r1)
det[e
i
2
N
(l1)y
j
]
det[e
i
2
N
(l1)y
j
], l = 1, ..., M. (3.147)
We can expand the determinants (3.147) using permutations:
S
f
(d, r) = (1)
d1
e
i
2
N
((M1)/2+N/2)(r1)
PS
M
S
M
(1)
P
(1)
P
r1
i=1
(
d
t
i
=2
e
i
2
N
((P
i
1)t
i
(P
i
1)(t
i
1))
)
M
i=r
(
N
t
i
=d+1
e
i
2
N
((P
i
1)(P
i
1))t
i
). (3.148)
= QP. Also, one

i
in each of the brackets, since it is a geometric
series.
S
f
(d, r) = (1)
d1
e
i
2
N
((M1)/2+N/2)(r1)
PS
M
QS
M
(1)
Q
r1
i=1
c
1
f
(d, Q
P
i
, P
i
)
M
i=r
c
2
f
(d, Q
P
i
, P
i
), (3.149)
where
c
1
f
(d, j, l) = e
i
2
N
(l1)
t=d
t=2
e
i
2
N
(jl)t
,
c
2
f
(d, j, l) =
t=N
t=d+1
e
i
2
N
(jl)t
(3.150)
are independent of r. At this point, we can use the phase variable integration trick to get
rid of summation over P, and then represent summation over Q as a determinant:
S
f
(d, r) = (r 1)!(M r + 1)!(1)
d1
e
i
2
N
((M1)/2+N/2)(r1)
_
2
0
d
2
e
i(r1)
det
_
_
c
f
(, 1, 1) c
f
(, 2, 1) ... c
f
(, M, 1)
c
f
(, 1, 2) c
f
(, 2, 2) ... c
f
(, M, 2)
... ... ... ...
c
f
(, 1, M) c
f
(, 2, M) ... c
f
(, M, M)
_
_
, (3.151)
where
c
f
(, j, l) = e
i
c
1
f
(d, j, l) +c
2
f
(d, j, l). (3.152)
We can analytically perform summation over r in (3.145), since the form of the determinant
and c
f
(, j, l) are independent of r, and rdependent combinatorial prefactor cancels:
S
f
(d) =
M+1
r=1
(N M 1)!M!
(r 1)!(M r + 1)!
S
f
(d, r) = (N M 1)!(M)!(1)
d1
_
2
0
d
2
e
i(
2
N
((M1)/2+N/2))(M+1)
1
e
i(
2
N
((M1)/2+N/2))
1
det
_
_
c
f
(, 1, 1) c
f
(, 2, 1) ... c
f
(, M, 1)
c
f
(, 1, 2) c
f
(, 2, 2) ... c
f
(, M, 2)
... ... ... ...
c
f
(, 1, M) c
f
(, 2, M) ... c
f
(, M, M)
_
_
. (3.153)
Analogously to the case of bosons, integration over is equivalent to substitution
0
=
2((M 1)/2 +N/2)/N to the determinant, and the nal expression is
S
f
(d) = (N M 1)!(M)!(1)
d1
det
_
_
c
f
(
0
, 1, 1) c
f
(
0
, 2, 1) ... c
f
(
0
, M, 1)
c
f
(
0
, 1, 2) c
f
(
0
, 2, 2) ... c
f
(
0
, M, 2)
... ... ... ...
c
f
(
0
, 1, M) c
f
(
0
, 2, M) ... c
f
(
0
, M, M)
_
_
. (3.154)
3.5.4 Numerical evaluation of correlation functions and Luttinger param-
eters
Using the results of the previous sections, one can evaluate correlation functions
on a ring numerically and extract both long-range and short range behavior of correla-
tion functions. Calculation of all determinants requires polynomial time in their size, and
systems of up to N = 100 atoms can be easily investigated on a desktop PC. Fourier
transform of the correlation function is an occupation number n(k), which can be mea-
sured directly in time-of-ight experiments[43] or using Bragg spectroscopy[130]. Recently,
long distance correlation functions of the model under consideration have been investigated
based on conformal eld theory (CFT) arguments[106]. Our determinant representations
for strongly interacting mixture can be used to obtain these correlation functions at all
distances, and compare their large distance asymptotic behavior with predictions of CFT.
In Fig. 3.9 we show numerically evaluated Bose-Bose correlation function for
M = 15, N = 30. Since we used periodic boundary conditions, the correlation function is
periodic in . To extract universal long-distance correlation functions from our calculation,
one has to t the numerical results using general Luttinger liquid asymptotic behavior. In
the thermodynamic limit long range behavior is
b
(0, ) [[
1/(2K
b
)
, (3.155)
where K
b
is a bosonic Luttinger Liquid parameter. This formula is valid, if is bigger then
any non-universal short-range scale of the model. In our case, such short-range scale is given
by the interbosonic distance, which is L/M. For a nite size system, general arguments of
L
(0, )
n
b
b
0.2 0.4 0.6 0.8 1
0.2
0.4
0.6
0.8
1
Figure 3.9: Normalized Bose-Bose correlation function on a circle as a function of the
distance (here n
b
is boson density). Due to periodic boundary conditions correlation
function is periodic with a period L, where L is the size of the system. Numerical evaluation
is done for M = 15, N = 30.
conformal invariance[131, 132] imply that the correlation function has the form
b
(0, )
1
[ sin

L
[
1/(2K
b
)
. (3.156)
We tted numerically obtained correlation functions with (3.156), and the results
coincide with the formula
K
b
=
1
( 1)
2
1
, (3.157)
obtained in [106] based on CFT arguments. One can see subleading oscillations in the
numerical evaluation, but their quantitative analysis would require more numerical eort.
Fourier transform of
b
(0, ) is a monotonously decreasing function, which has a singularity
at k = 0, governed by Luttinger liquid parameter K
b
:
n
b
(k) [k[
1+1/(2K
b
)
for k 0. (3.158)
Fermionic correlation functions can also be obtained using the results of the pre-
vious section, and space dependence of a typical correlation function is presented in gure
3.10. The oscillations are reminiscent of Friedel oscillations of the ideal Fermi gas. Their
large distance decay is controlled by Luttinger liquid behavior.
One can investigate Fourier transform of the correlation function, which is an
occupation number. The results for dierent boson fractions are shown in Figs. 3.11-3.13.
In gure 3.11 densities of bosons and fermions are almost equal. Fermi step at k
f
gets
smeared out by interactions, but relative change of occupation number as k
f
is crossed is
signicant. As boson fraction is decreasing, the discontinuity appears at k
f
+ 2k
b
, and it
gets stronger as M/N decreases (see Figs. 3.12,3.13). The presence of this discontinuity
L
(0, )
n
f
f
0.02 0.04 0.06 0.08 0.1
0.2
0.4
0.6
0.8
1
Figure 3.10: Normalized Fermi-Fermi correlation function on a circle as a function of the
distance (here n
f
is fermion density and L is the size of the system). Oscillations are
reminiscent of Friedel oscillations of the ideal Fermi gas, and their large distance decay is
controlled by Luttinger liquid behavior. Numerical evaluation is done for M = 51, N = 100.
has been predicted in [106], based on CFT arguments, and here we quantify the strength
of the eect. One should note, that discontinuity at k
f
+ 2k
b
is a direct signature of the
interactions and its detection can serve as an unambiguous verication of our theory.
3.6 Tonks-Girardeau regime at low temperatures
In the previous sections we considered density proles and developed an algorithm
to calculate the correlation functions of the ground state of the Bose-Fermi wave vector
(3.1) in the strongly interacting regime. An important question, which is very relevant
experimentally, is the eect of nite temperatures. In principle, one can use techniques of
the thermodynamic Bethe ansatz [112] to obtain free energy at nonzero temperatures as the
function of interaction strength and densities. Combined with local density approximation,
it can be used to calculate density proles for any interaction strength. In this section we will
limit our discussion to eects of small nonzero temperatures T E
f
= (hn)
2
/(2m) only
for strongly interacting regime. We will show the evolution of the density prole (see Fig.
3.14) in a harmonic trap and calculate the correlation functions under periodic boundary
conditions. The eect of nonzero temperatures on correlation functions is particularly
interesting for strongly interacting multicomponent systems (as has been emphasized for
the case of Bose-Bose and Fermi-Fermi mixtures in [134]), due to considerable change of the
momentum distribution in the very narrow range of the temperatures of the order of E
f
/.
For the case of Bose-Bose or Fermi-Fermi mixture it was possible[134] to obtain correlation
functions only in the two limiting cases T E
f
/ and E
f
/ T E
f
. For Bose-Fermi
mixture, we are able to calculate correlation functions for any ratio between E
f
/ and
T E
f
(see Fig. 3.15). By adding an imaginary part to T, the procedure presented in this
kL
2
k
f
20 40 60 80 100
0.1
0.2
0.3
0.4
0.5
0.6
0.7
n (k)
f
Figure 3.11: Fourier transform of the Fermi-Fermi correlation function for M = 51, N = 100.
Fermi step at k
f
gets smeared out by interactions, but relative change of occupation number
as k
f
is crossed is signicant.
kL
2
k
f
+2k
b
k
f
20 40 60 80 100
0.2
0.4
0.6
0.8
n (k)
f
Fermi step at k
f
gets smeared out by interactions, and additional discontinuity appears at
k
f
+ 2k
b
.
kL
2
n (k)
f
k
f
+2k
b
k
f
20 40 60 80 100
0.2
0.4
0.6
0.8
Discontinuity at k
f
+ 2k
b
gets stronger as M/N decreases.
section can be also easily generalized for non equal time correlations.
3.6.1 Low energy excitations in Tonks-Girardeau regime.
As has been discussed in section 3.3, for 1 there are two energy scales in the
problem: the rst energy scale is the Fermi energy of orbital motion E
f
= (hn)
2
/(2m),
while the second is the spin wave (relative density oscillation) energy E
f
/. The second
energy scale is present only in strongly interacting multicomponent systems, as has been
emphasized earlier[133, 134]. Density proles and correlation functions we have considered
earlier are valid in the regime, when temperature is smaller than both of these energy scales:
T E
f
/ E
f
. (3.159)
However, interesting phenomena[133, 134, 135, 136] can be analyzed in the spin disor-
deredregime, when
E
f
/ T E
f
. (3.160)
This regime has attracted lots of attention recently in the context of electrons in 1d quantum
wires[133, 135, 136]. In spin disordered regime, spin degrees of freedom are completely
disordered, while orbital degrees are not aected much. From the point of view of orbital
degrees, this is still a low-temperature regime, since T E
f
. The energy of the system
doesnt change too much, while momentum distribution changes dramatically as temper-
ature changes from 0 to the order of several E
f
/. Spin disordered regime exists only
for multicomponent systems and a crossover from true ground state to spin disordered
regime provides a unique opportunity to study the eects of low temperatures on a highly
correlated strongly interacting system. Spin disordered limit is likely to be reached rst
in the experiments, and a signicant change of the density prole and of the momentum
distribution as regime (3.159) is reached can be used as a way to calibrate the temperatures
much smaller than E
f
.
Only two limiting cases (3.159)-(3.160) have been investigated for spin-
1
2
fermion
and boson mixtures, since in these cases spin wave functions are related to eigenstates of
spin-
1
2
Heisenberg wave vector, and have a complicated structure. In the case of a Bose-
Fermi mixture, spin wave functions correspond to non interacting fermionized single-spin
excitations, and one can calculate correlation functions in the whole low-temperature limit,
investigating crossover from true ground state to spin disordered limit:
E
f
/, T E
f
. (3.161)
In the following calculations, we will neglect the inuence of nonzero temperature
on orbital degrees, and will always assume that orbital degrees are not excited. This as-
sumption will aect the results only at distances, at which the correlation functions are
already very small due to eects of spin excitations.
In the zeroth order in 1/ expansion, energies of all spin states are degenerate,
and solutions of Bethe equations are given by
_
+ic/2
ic/2
_
N
= 1, = 1, ..., M, (3.162)
e
ik
j
L
=
M
=1
+ic/2
ic/2
, j = 1, ..., N. (3.163)
In the next order in 1/ expansion, both k
j
and
i
acquire corrections of the order of 1/.
Since energy depends only on (k), we need to calculate corrections to (k) in the leading
order. According to (3.35), to calculate 1/ correction to (k), one can use
i
in the zeroth
order, given by (3.162):
2(k) = 1 +
1
L
M
i=1
4c
4
2
i
+c
2
(3.164)
is independent of k in the rst order of 1/ expansion. If we dene spin wave vectors
according to
+ic/2
ic/2
= e
i2/N
, = 1, ..., M, (3.165)
energy of the state with spin wave vectors
i
in 1/ order is given by
E(,
i
) =

2
3
N
2
L
2
(N
4
i=1
(1 cos
2
i
N
)). (3.166)
Allowed values for spin wave vectors are
K =
i
1, ..., N,
i
<
j
for i < j. (3.167)
The number of spin excitations (we will call them magnons from now on) is xed to be
the number of bosons, and dierent spin wave vectors cannot coincide. Hence, magnons
have a fermionic statistics. The eect of nonzero temperatures is to average the correlations
over the dierent sets of possible
i
from (3.167).
According to (3.166) in the rst order in 1/ expansion magnons do not interact
with each other, and the total energy is the sum of separate magnon energies. Magnon
energy spectrum is
() =
4
2
3
N
2
L
2
(cos
2
N
1) =
4E
f
3
(cos
2
N
1). (3.168)
Lowest state corresponds to = N/2, and as the number of magnons increases, spin wave
vectors near N/2 start being occupied. Eq. (3.168) proves the choice (3.92) for the true
ground state at zero temperature.
3.6.2 Density proles
In this subsection we will analyze the behavior of the strongly interacting mixture
in a harmonic trap at low temperatures. Similarly to section 3.4 we consider the case
b
=
f
=
0
. (3.169)
According to (3.59), within the region of coexistence densities are governed by equations
0
f
(x) +
m
2
0
x
2
2
=
0
f
(0),
0
b
(x)
0
f
(x) =
0
b
(0)
0
f
(0). (3.170)
Similarly to the case of T = 0, total density is given by (3.63):
n
0
(x) = n
0
(0)
1
x
2
x
2
f
, (3.171)
and has a weak temperature dependence. On the other hand, relative density is controlled
by solutions of the second equation (3.170), and its dependence on temperature is quite
strong. It turns out that in strongly interacting regime
b

f
can be easily calculated
using formulas from the previous subsection.
b
f
is the change of the free energy, when
one boson is added and one fermion is removed from the mixture. In the language of the
magnons this corresponds to an addition of one magnon. Therefore, one obtains
f
=
m
, (3.172)
where
m
is the chemical potential of the magnons with energy spectrum (3.168). As has
been noted earlier, magnons obey fermionic statistics (only one magnon can occupy each
state) and do not interact, so one can use Fermi distribution for their occupation number.
The chemical potential for magnons
m
as a function of and T can be obtained numerically
from the normalization condition for the total number of magnons, which reads
=
_
2
0
1
e
1
T
(
4E
f
3
(cos k1)m)
+ 1
dk
2
. (3.173)
Then one can use LDA to obtain the density proles. In Fig. 3.14 we show the density of
fermions for the case in which total number of bosons equals total number of fermions. One
x
x
f
n
n
f
0
E
f
0
0
T=10
E
f
0
0
T=0.01
0.2 0.4 0.6 0.8 1
0.1
0.2
0.3
0.4
0.5
0.6
Figure 3.14: Evolution of the fermionic density prole in the strongly interacting
regime as the function of a temperature. Four graphs correspond to temperatures
0.01E
0
f
/
0
, 0.2E
0
f
/
0
, E
0
f
/
0
and 10E
0
f
/
0
. Here
0
is the Lieb-Liniger parameter in the
center of the trap, x
f
is the size of the fermionic trap, and E
0
f
= (hn
0
)
2
/(2m), where n
0
is
the total density in the center of the trap. Overall number of bosons equals the number of
fermions. Non monotonous behavior of the fermion density prole persists up to T E
0
f
/
0
.
The total density prole doesnt change considerably in this range of the temperatures, and
is given by n(x) = n
0
_
1 x
2
/x
2
f
.
sees that density prole changes considerably at the temperatures of the order of E
0
f
/
0
,
where E
0
f
and
0
are the Fermi energy and Lieb-Liniger parameter in the center of the
trap. For E
0
f
/
0
T E
0
f
boson fraction is uniform along the trap. As temperature
is lowered, more bosons condense towards the center of the trap, and fermionic density
behaves non-monotonously as a function of the distance form the center of the trap.
3.6.3 Fermi-Fermi correlations
From now on we will consider the periodic boundary conditions, when the many
body problem is strictly solvable in the mathematical sense. We will rst describe the
calculation of fermionic correlations, since it is simpler than calculation of Bose correlations.
To calculate temperature averaged correlation functions, we should be able to calculate
f
(0, ; T) =
1

K
e
P
i
(
i
)/T

K
e
P
i
(
i
)/T
(N M)
_
(
1
, ...,
M
; x
1
, x
2
, ..., 0)
(
1
, ...,
M
; x
1
, x
2
..., )dx
1
...dx
N1
. (3.174)
Denominator in (3.174) is a partition function of non interacting fermions in a micro canon-
ical ensemble. It can be written as
Z =

K
e
P
i
(
i
)/T
=
_
2
0
e
iM
d
2
N
=1
(1 +e
i
e
()/T
) (3.175)
The numerator can be simplied using the factorization of spin and orbital parts, similarly
to (3.140):
f
(0, ; T) =
1
Z
1
(N M 1)!M!L
N
N
M

N
d=1
K
e
P
i
(
i
)/T
S
f
(
1
, ...,
M
; d)I(d, ;
1
, ...,
M
). (3.176)
Here S
f
(
1
, ...,
M
; d) is an expression (3.141) for an arbitrary choice of
i
belonging to
(3.167):
S
f
(
1
, ...,
M
; d) =
y
det[e
i
2
N

i
y
j
] det[e
i
2
N

i
y
j
](1)
y
(1)
y
. (3.177)
I(d, ;
1
, ...,
M
) is an integral (3.108), whose dependence on
1
, ...,
M
comes only through
boundary conditions (3.163). If
M
i=1
i
mod N = D, where D = 1, ..., N 1, then the
set of k
i
which minimizes kinetic energy is uniquely dened:
2
L
(N/2 +D/N),
2
L
(N/2 + 1 +D/N), ...,
2
L
(N/2 1 +D/N). (3.178)
If D = 0, then there are two degenerate sets of k
i
, and each of them should be taken with
a weight 1/2. Taking this into account, I(d, ;
1
, ...,
M
) can be expressed as
I(d, ;
1
, ...,
M
) = I(d, )
N
D=0
(1

N
(D)
2
)
N
(D
M
i=1
i
)e
(D
N
2
)
2i
N

, (3.179)
where
N
(x) =
_
1 if x mod N = 0,
0 otherwise
(3.180)
N
(x) can be represented as a Fourier sum,
N
(x) =
1
N
N1
p=0
e
2i
N
px
. (3.181)
Taking this into account, the correlation function (3.176) is rewritten as
f
(0, ; T) =
1
Z
1
(N M 1)!M!L
N
N
M+1
N
d=1
I(d, )
N
D=0
(1

N
(D)
2
)e
(D
N
2
)
2i
N

N1
p=0
e
2i
N
pD
S
f
(d; p; T), (3.182)
where
S
f
(d; p; T) =
K
e
P
i
(
2i
N
p
i
+(
i
)/T)
S
f
(
1
, ...,
M
; d). (3.183)
Calculation of S
f
(
1
, ...,
M
; d) closely reminds a calculation of S
f
(d) in section
3.5.3, so we will present only a brief derivation.
S
f
(
1
, ...,
M
; d) =
M+1
r=1
(N M 1)!(M)!
(r 1)!(M r + 1)!
S
f
(
1
, ...,
M
; d, r), (3.184)
where S
f
(
1
, ...,
M
; d, r) is a product of two determinants:
S
f
(
1
, ...,
M
; d, r) = (1)
d1

PS
M
S
M
(1)
P
(1)
P
r1
i=1
(
d
t
i
=2
e
i
2
N
(
P
i
t
i
P
i
(t
i
1))
)
M
i=r
(
N
t
i
=d+1
e
i
2
N
(
P
P
i
)t
i
). (3.185)
= QP. Also, one

i
in each of the brackets, since it is a geometric
series.
S
f
(
1
, ...,
M
; d, r) = (1)
d1
PS
M
QS
M
(1)
Q
r1
i=1
g
1
f
(d,
Q
P
i
,
P
i
)
M
i=r
g
2
f
(d,
Q
P
i
,
P
i
),
(3.186)
where
g
1
f
(d, j, l) = e
i
2
N
l
t=d
t=2
e
i
2
N
(jl)t
,
g
2
f
(d, j, l) =
t=N
t=d+1
e
i
2
N
(jl)t
(3.187)
are independent of r. We can use the phase variable integration trick to get rid of sum-
mation over P, and then represent summation over Q as a determinant:
S
f
(
1
, ...,
M
; d, r) = (r 1)!(M r + 1)!(1)
d1
_
2
0
d
2
e
i(r1)
det
_
_
g
f
(,
1
,
1
) g
f
(,
1
,
2
) ... g
f
(,
1
,
M
)
g
f
(,
2
,
1
) g
f
(,
2
,
2
) ... g
f
(,
2
,
M
)
... ... ... ...
g
f
(,
M
,
1
) g
f
(,
M
,
2
) ... g
f
(,
M
,
M
)
_
_
, (3.188)
where
g
f
(, j, l) = e
i
g
1
f
(d, j, l) +g
2
f
(d, j, l). (3.189)
We can analytically perform summation over r, since the form of the determinant and
g
f
(, j, l) are independent of r, and combinatorial prefactor cancels in (3.186). Similarly
to (3.138) we represent summation over r and integration over as a substitution
0
= 0,
and obtain the following result:
S
f
(
1
, ...,
M
; d) = (N M 1)!M!(1)
d1
det
_
_
g
f
(0,
1
,
1
) g
f
(0,
1
,
2
) ... g
f
(0,
1
,
M
)
g
f
(0,
2
,
1
) g
f
(0,
2
,
2
) ... g
f
(0,
2
,
M
)
... ... ... ...
g
f
(0,
M
,
1
) g
f
(0,
M
,
2
) ... g
f
(0,
M
,
M
)
_
_
. (3.190)
To calculate S
f
(d; p; T) we have to sum (3.190) for dierent choices of
i
with
i
dependent
prefactor. One can take these prefactors into account by multiplying each row in (3.190) by
f(
i
) = e
(
2i
N
p
i
+(
i
)/T)
, (3.191)
since only one term from each row appears in the expansion of the determinant:
S
f
(d; p; T) = (N M 1)!M!(1)
d1
N
1
=1
...
N
M
=1
det
_
_
f(
1
)g
f
(0,
1
,
1
) f(
1
)g
f
(0,
1
,
2
) ... f(
1
)g
f
(0,
1
,
M
)
f(
2
)g
f
(0,
2
,
1
) f(
2
)g
f
(0,
2
,
2
) ... f(
2
)g
f
(0,
2
,
M
)
... ... ... ...
f(
M
)g
f
(0,
M
,
1
) f(
M
)g
f
(0,
M
,
2
) ... f(
M
)g
f
(0,
M
,
M
)
_
_
. (3.192)
Summations over
i
in (3.192) can be performed analytically, since each choice of
i
is a
term in the expansion of the Fredholm determinant[137]. The desired expression has the
form:
S
f
(d; p; T) = (N M 1)!M!(1)
d1
_
2
0
d
2
e
i(NM)
det
_
_
e
i
+f(1)g
f
(0, 1, 1) f(1)g
f
(0, 1, 2) ... f(1)g
f
(0, 1, N)
f(2)g
f
(0, 2, 1) e
i
+f(2)g
f
(0, 2, 2) ... f(2)g
f
(0, 2, N)
... ... ... ...
f(N)g
f
(0, N, 1) f(N)g
f
(0, N, 2) ... e
i
+f(N)g
f
(0, N, N)
_
_
. (3.193)
Integration over extracts terms from the determinant which have e
i(NM)
dependence.
Such terms appear, when NM e
i
elements in the expansion of the determinant are taken
along the diagonal. If e
i
are chosen in the rows except for
1
, ...,
M
, then contribution from
such choice of e
i
is a minor which equals f(
1
)...f(
M
)S
f
(
1
, ...,
M
; d). Thus evaluation
of the prefactor in the e
i(NM)
dependence of the determinant corresponds to summation
of f(
1
)...f(
M
)S
f
(
1
, ...,
M
; d) over possible sets of
i
.
Finally, substituting (3.193) into (3.182), one can evaluate numerically Fermi-
Fermi correlation functions for any temperature and ratio between boson and fermion den-
sity in low temperature limit.
In Fig. 3.15 we show numerically evaluated Fermi-Fermi correlation function for
M = 15, N = 30 and several temperatures, ranging from T = 0 to T = 5
4E
f
3
. At this low
temperature region Fermi-Fermi correlation function changes considerably due to transition
from true ground state to spin disordered regime. In spin disordered regime Fermi
singularity at k
f
gets completely smeared out by thermal spin excitations.
k
f
4E
f
3
T=5
kL
2 5 10 15 20 25 30
0.2
0.4
0.6
T=0
n (k)
f
Figure 3.15: Fourier transform of the Fermi-Fermi correlation function for M = 15, N =
30. Four graphs correspond to temperatures 0, 0.1
4E
f
3
, 0.5
4E
f
3
, 5
4E
f
3
. In the range of the
temperatures E
f
/ E
f
fermionic correlation function changes considerably due to
transition from true ground state to spin disordered regime. In spin disordered regime
Fermi singularity at k
f
gets completely smeared out by thermal spin excitations.
3.6.4 Bose-Bose correlation function
Bose-Bose correlation functions also change as T goes up. However, since for T = 0
n
b
(k) doesnt have any interesting structure except for singularity at k = 0, the eects of
nonzero temperatures will not be as dramatic as for fermionic correlations. We present the
results here mainly for the sake of completeness. Calculations in this subsection are similar
to what has been done in the previous subsection. The correlation function can be written
as
b
(0, ; T) =
1
K
e
P
i
(
i
)/T
K
e
P
i
(
i
)/T
M
_
(
1
, ...,
M
; 0, x
2
, ..., x
N
)
(
1
, ...,
M
; , x
2
..., x
N
)dx
2
...dx
N
. (3.194)
Similarly to (3.182), this can be written as
b
(0, ; T) =
1
Z
1
(N M)!(M 1)!L
N
N
M+1
N
d=1
I(d, )
N
D=0
(1

N
(D)
2
)e
(D
N
2
)
2i
N

N1
p=0
e
2i
N
pD
S
b
(d; p; T), (3.195)
where
S
b
(d; p; T) =
K
e
P
i
(
2i
N
p
i
+(
i
)/T)
S
b
(
1
, ...,
M
; d). (3.196)
Here S
b
(
1
, ...,
M
; d) is an expression (3.122) for an arbitrary choice of
i
belong-
ing to (3.167):
S
b
(
1
, ...,
M
; d) =
y
det[e
i
2
N

i
y
j
] det[e
i
2
N

i
y
j
](1)
y
(1)
y
. (3.197)
Similarly to (3.129), it can be written as
S
b
(
1
, ...,
M
; d) =
M
r=1
(N M)!(M 1)!
(r 1)!(M r)!
S
b
(
1
, ...,
M
; d, r), (3.198)
where S
b
(
1
, ...,
M
; d, r) is a result of the summation of (3.197) in the following subspace:
1 y
2
, ..., y
r
d < y
r+1
, ..., y
M
N. (3.199)
We can expand determinants of (3.197) using permutations:
S
b
(
1
, ...,
M
; d, r) = (1)
d1

PS
M
S
M
(1)
P
(1)
P
e
i
2
N
(
P
1
1
P
1
d)
r
i=2
(
d
y
i
=2
e
i
2
N
(
P
i
y
i
P
i
(y
i
1))
)
M
i=r+1
(
N
y
i
=d+1
e
i
2
N
(
P
P
i
)y
i
). (3.200)
= QP. Also , one

can analytically perform summation over y
i
in each of the brackets, since it is a geometrical
progression. Compared to the case of fermions, there are 3 types of the brackets:
S
f
(
1
, ...,
M
; d, r) = (1)
d1

PS
M
QS
M
(1)
Q
g
0
b
(d,
Q
P
1
,
P
1
)
r
i=2
g
1
b
(d,
Q
P
i
,
P
i
)
M
i=r
g
2
b
(d,
Q
P
i
,
P
i
), (3.201)
where
g
0
b
(d, j, l) = e
i
2
N
(j1ld)
,
g
1
b
(d, j, l) = e
i
2
N
l
t=d
t=2
e
i
2
N
(jl)t
,
g
2
b
(d, j, l) =
t=N
t=d+1
e
i
2
N
(jl)t
(3.202)
We can use phase integration trick to represent (3.201) as an integral of some determinant,
but there will be two phase variables, since there are 3 types of inequivalent brackets:
S
b
(
1
, ...,
M
; d, r) = (r 1)!(M r)!(1)
d1
_
2
0
e
i(r1)
d
2
_
2
0
d
2
e
i
det
_
_
g
b
(, ,
1
,
1
) g
b
(, ,
1
,
2
) ... g
b
(, ,
1
,
M
)
g
b
(, ,
2
,
1
) g
b
(, ,
2
,
2
) ... g
b
(, ,
2
,
M
)
... ... ... ...
g
b
(, ,
M
,
1
) g
b
(, ,
M
,
2
) ... g
b
(, ,
M
,
M
)
_
_
, (3.203)
where
g
b
(, , j, l) = e
i
g
0
b
(d, j, l) +e
i
g
1
b
(d, j, l) +g
2
b
(d, j, l). (3.204)
After integration over , determinant in (3.203) has terms up to e
i(M1)
, therefore inte-
gration over and summation according to (3.198) are equivalent to substitution
0
= 0 :
S
b
(
1
, ...,
M
; d) = (N M)!(M 1)!(1)
d1
_
2
0
d
2
e
i
det
_
_
g
b
(0, ,
1
,
1
) g
b
(0, ,
1
,
2
) ... g
b
(0, ,
1
,
M
)
g
b
(0, ,
2
,
1
) g
b
(0, ,
2
,
2
) ... g
b
(0, ,
2
,
M
)
... ... ... ...
g
b
(0, ,
M
,
1
) g
b
(0, ,
M
,
2
) ... g
b
(0, ,
M
,
M
)
_
_
. (3.205)
Integral over can be simplied further, since the determinant in (3.205) has a form A
0
+
A
1
e
i
. The form above follows from the formula for the determinant of the sum of the
matrices (see page 221 of [138]) and the fact that a part of the matrix which depends on e
i
has a rank 1. Lets for a moment introduce a notation z = e
i
. Integration over with a
weigh e
i
extracts the term A
1
, which can be alternatively written as a dierence between
two determinants, one with z = 1 and the other with z = 0 (g
f
(, j, l) is given by (3.189)):
S
b
(
1
, ...,
M
; d) = (N M)!(M 1)!(1)
d1
det
_
_
g
b
(0, 0,
1
,
1
) g
b
(0, 0,
1
,
2
) ... g
b
(0, 0,
1
,
M
)
g
b
(0, 0,
2
,
1
) g
b
(0, 0,
2
,
2
) ... g
b
(0, 0,
2
,
M
)
... ... ... ...
g
b
(0, 0,
M
,
1
) g
b
(0, 0,
M
,
2
) ... g
b
(0, 0,
M
,
M
)
_
_
(N M)!(M 1)!(1)
d1
det
_
_
g
f
(0,
1
,
1
) g
f
(0,
1
,
2
) ... g
f
(0,
1
,
M
)
g
f
(0,
2
,
1
) g
f
(0,
2
,
2
) ... g
f
(0,
2
,
M
)
... ... ... ...
g
f
(0,
M
,
1
) g
f
(0,
M
,
2
) ... g
f
(0,
M
,
M
)
_
_
. (3.206)
We note, that a similar trick is explained on the page 609 of [139]. Next, summation
over dierent
i
can be performed similarly to the case of fermions:
S
b
(d; p; T) = (N M)!(M 1)!(1)
d1
_
2
0
d
2
e
i(NM)
det
_
_
e
i
+f(1)g
b
(0, 0, 1, 1) f(1)g
b
(0, 0, 1, 2) ... f(1)g
b
(0, 0, 1, N)
f(2)g
b
(0, 0, 2, 1) e
i
+f(2)g
b
(0, 0, 2, 2) ... f(2)g
b
(0, 0, 2, N)
... ... ... ...
f(N)g
b
(0, 0, N, 1) f(N)g
b
(0, 0, N, 2) ... e
i
+f(N)g
b
(0, 0, N, N)
_
N M
M
S
f
(d; p; T), (3.207)
where S
f
(d; p; T) is dened in (3.193).
3.7 Experimental considerations and conclusions
In this section we will consider in detail possible ways to realize the system under
investigation in experiments with cold atoms.
An array of one dimensional tubes of cold atoms along x direction has been realized
experimentally using strong optical lattices in two dimensions[42, 43, 96, 119, 140] y and
z. The large number of tubes provides a good imaging quality, but the number of atoms
and the ratio between Bose and Fermi particle numbers varies from tube to tube, and may
complicate the interpretation of the experiments (one of the ways to x the ratio between
Bose and Fermi numbers for all tubes will be discussed later). In addition, due to harmonic
connement along the axis of the tube, boson and fermion densities vary within each tube,
which causes non-homogeneous broadening of the momentum distribution. Alternatively,
single copies of one dimensional mixtures with constant densities along the axis can be
realized in micro traps on a chip[141], or using cold atoms in a 1d box potential[142]. Here
we will mostly concentrate on a realization of 1d system using strong 2D optical lattice in
y and z directions.
First of the conditions (3.2), m
b
= m
f
, is approximately satised for isotopes of
the atoms, and one can expect our theory to be valid with high accuracy for isotopes. Some
of the promising candidates are
39(41)
K
40
K[143],
171
Y b +
172
Y b[144], and
86(84)
Rb
87(85)
Rb[145]. Dierent isotopes of potassium have already been cooled to quantum degeneracy
[93, 146] by sympathetic cooling with Rb. There is another way to satisfy the rst condition
of (3.2) using already available degenerate mixtures[93]. If one uses an additional optical
lattice along the x direction with lling factors much smaller than one, then (3.1) is an
eective Hamiltonian describing this system with the eective masses determined by the
tunneling, similarly to a recent realization of Tonks-Girardeau gas for bosons[43]. Finally, we
note that one can realize experimentally the model, which has the same energy eigenvalues
as (3.1), using a mixture of two bosonic atoms (see next paragraph). If one chooses two
magnetic sublevels of the same atom, equality of masses will be satised automatically.
Second of the conditions (3.2), g
bb
= g
bf
> 0, can also be satised in current
experiments, using a combination of several approaches. First, one can use Feshbach reso-
nances to control the interactions: this is particularly straightforward for LiNa of KRb
mixtures, where resonances have already been observed experimentally[95, 94]. Second, we
point out that it is sucient to have equal (positive) signs for the two scattering lengths, but
not necessarily their magnitudes. Well away from connement induced resonances[39], 1D
interactions are given by g
bb
= 2 h
b
a
bb
, g
bf
= 2 ha
bf
f
(m
b
+m
f
)/(
b
m
b
+
f
m
f
),
where
b
,
f
are radial connement frequencies, and a
bb
, a
bf
are 3D scattering lengths.
For a xed value of a
bb
/a
bf
, one can always choose the detuning of the optical lattice laser
frequencies in such a way that g
bb
= g
bf
. Next, one can vary the intensity of the y, z optical
lattice beams and change g, while always being on the integrable line of the phase diagram.
Combination of these two approaches to control 1D interactions gives a lot of freedom for
experimental realization of equal one dimensional interactions. Finally, lets describe how to
realize the bosonic model, which has the same eigenvalues as the model (3.1). Bosonic sys-
tem is characterized by 3 interaction parameters, g
11
, g
22
, g
12
. If one tunes g
11
to +, then
bosons of type 1 get fermionized within the same type, and the model will be equivalent
in terms of energy spectrum, density proles and collective modes to (3.1). Note, however,
that single-particle correlation functions will be dierent, and the results of sections 3.5 and
3.6 (except for 3.6.2) are not applicable. This general equivalence between Bose-Bose and
Bose-Fermi models is valid for any ratio between g
22
and g
12
. One can push this result even
further, by tuning g
22
to +. In this case eigenstates of (3.1) are equivalent to spin1/2
fermionic system[109, 110, 147], and some predictions for those systems can be applied for
bosons.
Detection of the properties of the system may be hindered by the fact that both the
number of atoms and relative fraction of bosons vary from tube to tube. However, one can
use Feshbach resonances to x the boson fraction to be = 1/2 in each tube
1
. To do this,
one can use Feshbach resonance for Bose-Fermi scattering to adiabatically create molecules
before loading the mixture in strong y, z optical lattice. If one gets rid of unpaired atoms at
this stage, switches on y, z optical lattice, and adiabatically dissociates the molecules, boson
fraction will be xed in each tube to be = 1/2. Most of our gures have been calculated
for this particular boson fraction. Our results in harmonic traps are presented as functions
1
We thank G. Modugno for pointing out this possibility.
of
0
= mg/( h
2
n
0
), where n
0
is a total density in the center of a one dimensional trap,
and
0
is the Lieb-Liniger [40] parameter in the center of the trap.
0
1 corresponds
to a strongly interacting regime. n
0
varies from tube to tube, and to be able to compare
theoretical predictions precisely with experiments, one should be able to have an optical
access to regions where variation of n
0
is small.
Most of our experimental predictions, except for those in section 3.6, deal with
zero temperature case. Experimentally, one needs to verify the quantum degeneracy of the
gases in 1D regime. A possible way to identify the onset of quantum degeneracy is based on
density proles[19]. In Figs. 3.2 and 3.3 we show the density proles at zero temperature for
weak and strong interactions, when the harmonic connement frequency
0
is the same for
bosons and fermions. In both cases, only central part is occupied by bosons, and outer shells
consist of fermions only. In addition, for the strong interactions fermion density develops a
strong peak at the edge of bosonic cloud. When the interactions are not strong (
0
< 1),
one can estimate the temperature at which quantum eects become important for ground
state density prole to be of the order of Nh
0
, where N is the total number of atoms in a
tube. In the strongly interacting regime (
0
1), however, the situation is very dierent.
There are two temperature scales in the problem: E
0
f
= (hn
0
)
2
/(2m), and E
0
f
/
0
E
0
f
.
As the temperature goes up from 0 to E
0
f
/
0
, the density prole changes as shown in
gure 3.14, and the peak in the fermion density disappears. However, total density prole
doesnt change much as long as T E
0
f
. This eect can be qualitatively understood as the
demonstration of the fermionization of the Bose-Fermi cloud, as will be explained in the
next paragraph.
First, lets consider the case without a harmonic potential. When interactions are
strong, bosons tend to avoid fermions and other bosons. Whenever coordinates of any two
particles coincide, wave function is close to 0. Eectively, the gas is mutually fermionized,
and the ground state energy of the system is close to the ground state energy of the pure
non interacting Fermi gas with a density equal to the total density of bosons and fermions.
Dependence of the energy on the relative density (or boson fraction ) appears only in the
next order in 1/ expansion, and two rst terms in this expansion are given by (3.54). Since
dependence of the energy on boson fraction is 1 times smaller than dependence on
total density, the quantum degeneracy temperature for relative density excitations is also
times smaller than quantum degeneracy temperature for fermions with density n, hence
it is E
f
/. When harmonic trap is present at T = 0, the relative density distributes itself
to minimize the total energy. As temperature becomes of the order of several E
0
f
/
0
, almost
all relative density modes get excited, and boson fraction becomes uniform along the trap.
Total density modes are still not excited, since their quantum degeneracy temperature is
E
0
f
, and therefore the total density prole doesnt change much. Temperature E
0
f
/
0
, is
important not only for density distribution, but also for correlation functions, as will be
discussed later.
Knowing the exact dependence of the energy as the function of densities and inter-
actions allows us to investigate not only the static properties, but also dynamic behavior.
In section 3.4 we developed a two-uid hydrodynamic approach to calculate the frequencies
of collective oscillations. In the strongly interacting limit we predict the appearance of
low-lying modes, with a frequency scaling
0
/
0
. These modes correspond to out of
phase oscillations of boson and fermion clouds that keep the total density approximately
constant. These modes can be understood as follows: due to fermionization eects discussed
in previous paragraph, for
0
1 the energetic penalty for changing the relative density of
bosons and fermions is small, and hence it doesnt cost too much energy to create out of
phase oscillations that dont change the total density. The dependence of the frequencies
of low-lying oscillations with small quantum numbers on overall boson fraction in a tube
is shown in gure 3.8. In addition to low lying out of phase oscillations, the cloud has
in phase oscillations, with the frequencies
n
= n
0
, similarly to Tonks-Girardeau gas of
bosons[123]. These modes have frequencies considerably higher than out of phase modes,
and are not shown in gure 3.8. One can excite any of these modes by adding a pertur-
bation of the matching frequency, similarly to what has been done to bosons in [119]. A
dierent manifestation of the slow out of phase dynamics can be observed looking at the
evolution of density perturbations: initial perturbation will split into fast in phase part,
moving at Fermi velocity, and slow out of phase part. This is similar to spin-charge
separation, proposed for Fermi[147] or Bose[148] spin 1/2 mixtures. When interactions
are not strong (
0
< 1), one can obtain frequencies of all modes using mean-eld energy.
Figure 3.4 shows the dependence of frequencies for equal number of bosons and fermions
( = 1/2) on
0
. Even in the mean eld regime frequency of out of phase oscillations gets
smaller as interactions get stronger.
Finally, lets discuss theoretically the most interesting and sensitive measure of the
correlations, single particle correlation function, considered in sections 3.5 and 3.6. Fourier
transform of the single particle correlation function is an occupation number, and it can be
measured experimentally using Bragg spectroscopy[130] or time of ight measurements[43].
We can calculate these correlation functions in strongly interacting regime under periodic
boundary conditions for any temperatures. At zero temperature Bose momentum distribu-
tion has a singularity (3.158) at k = 0 reminiscent of BEC in higher dimensions, and its
strength is controlled by Luttinger liquid parameter K
b
, which depends only on boson frac-
tion for strong interactions. For fermions, momentum distribution has a lot of interesting
features. At zero temperature, several momentum distributions are presented in Figs. 3.11,
3.12, 3.13. One sees, that due to strong interactions, Fermi step at k
f
gets smeared out
even at T = 0, and n
f
(k) is considerably dierent from 0 at wave vectors far away from k
f
.
However, total change in n
f
(k) as one crosses k
f
is quite large. In addition, n
f
(k) develops
an extra singularity[106] at k
f
+2k
b
, and the strength of this singularity is higher for small
boson fractions. As the temperature rises, momentum distribution changes considerably
in the region of low temperatures of the order of E
f
/, and its evolution as a function
of temperature is shown in gure 3.15. For E
f
/ T E
f
, one enters so called spin
disordered regime[133, 134], where singularity at k
f
gets completely washed out, and for
equal densities of bosons and fermions momentum distribution gets almost twice as wide
compared to T E
0
f
/. A strong change of the momentum distribution in a small range
of temperatures can be used to perform a thermometry at very small temperatures. To
verify experimentally exact numerical correlation functions one needs to work with systems
at constant densities along x direction. Such constant density can be achieved in experi-
ments with micro traps[141], or in 2D arrays of tubes, if one makes a very shallow harmonic
connement, and creates strong box-like impenetrable potential at the sides of the tubes
with the help of additional lasers. If the system is in harmonic trap, lots of the features of
correlations themselves (i.e. singularity at k
f
+ 2k
b
) get washed out due to averaging over
inhomogeneous density prole[149]. However, the averaged correlation function still shows
signicant change in the region of temperatures of the order of E
0
f
/
0
, and the results for
T E
0
f
/
0
and E
0
f
/
0
T E
0
f
are shown in Fig. 3.16. The point where N
f
(k) has a
discontinuous derivative for T = 0 corresponds to the Fermi Hamiltonian for the maximal
density of fermions (at the edge of the bosonic cloud). For comparison, we also show N
f
(k)
for the same number of fermions in the same trap for non interacting case.
In conclusion, we presented a model for interacting Bose-Fermi mixture in 1D,
which is exactly solvable by Bethe ansatz technique. We obtained the energy numerically in
the thermodynamic limit, and used it to prove the absence of the demixing under conditions
(3.2), contrary to prediction of a mean eld approximation. Combining exact solution with
local density approximation (LDA) in a harmonic trap, we calculated the density proles
and frequencies of collective modes in various limits. In the strongly interacting regime,
we predicted the appearance of low-lying collective oscillations which correspond to the
counterow of the two species. In the strongly interacting regime we used exact wave
function to calculate the single particle correlation functions for bosons and fermions at
zero temperature under periodic boundary conditions. We derived an analytical formula,
which allows to calculate correlation functions at all distances numerically for a polynomial
time in system size. We investigated numerically two strong singularities of the momentum
distribution for fermions at k
f
and k
f
+2k
b
. We extended the results for correlation functions
for low temperatures, and calculated correlation functions in the crossover regime from
T = 0 to spin disordered regime. We also calculated the evolution of the density prole
in a harmonic trap at small nonzero temperatures. We showed, that in strongly interacting
regime correlation functions change dramatically as temperature changes from 0 to a small
temperature E
f
/ E
f
, where E
f
= (hn)
2
/(2m), n is the total density and is the
Lieb-Liniger parameter. Finally, we analyzed the experimental situation, proposed several
ways to implement the exactly solvable wave vector and combined the results for correlation
functions with LDA.
0
f
E
T>>
0
x
f
N (k)
f
0
n 2
0.2 0.4 0.6 0.8 1
0.25
0.5
0.75
1
1.25
1.5
T=0
Free fermions
k
Figure 3.16: Momentum distribution for fermions after averaging by inhomogeneous density
prole for harmonic connement. The results are shown for T = 0, (thick line), E
0
f
/
0

T E
0
f
(normal line) and for the same number of non interacting fermions (dashed).
Overall number of bosons in a trap equals the total number of fermions, n
0
is the total
density in the center of the trap, E
0
f
= (hn
0
)
2
/(2m),
0
1 is the Lieb-Liniger parameter
in the center of the trap, and 2x
f
is the total size of the cloud. In the range of the
temperatures E
0
f
/
0
fermionic correlation function changes considerably due to transition
from true ground state to spin disordered regime.
Chapter 4
Breakdown of the local density
approximation in interacting
systems of cold fermions in
strongly anisotropic traps
During the past few years, considerable experimental progress has been achieved
in creating systems of strongly interacting ultracold fermions [46]. One of the primary
motivations for this research eort is simulating strongly correlated electron systems such
as high-temperature superconductors and quantum magnets. The number of particles used
in experiments with cold atoms is large but not macroscopic, so they may also exhibit
mesoscopic phenomena. In this chapter, we discuss several important manifestations of such
mesoscopic eects in spin-polarized mixtures of ultracold fermionic atoms in the vicinity of
a Feshbach resonance [47, 48, 150, 151, 152]. Focusing on the BEC side of the resonance,
we demonstrate that some of the unusual features observed in experiments by Partridge et
al. [48], including a double-peak structure in the axial distribution of the density dierence
and a polarization-dependent aspect ratio of the minority species [153], appear as a result of
strong anisotropy of a conning potential. Our starting point is a rigorous proof that neither
of the two phenomena can take place when the conning potential is parabolic and the
local density approximation (LDA) holds (this has also been noted in Ref. [152]). A recent
preprint by Zwierlein and Ketterle [154] argued that the unharmonicity of the conning
potential should have contributed to the double-peak structure observed in Ref. [48]. In
this chapter, we show that even for a harmonic potential with a strongly anisotropic trap,
such as the one used in experiments of Partridge et al., beyond-LDA corrections give rise to
both the double-peak structure in the axial density dierence and a polarization-dependent
aspect ratio.
In this chapter, we consider spin-polarized fermion mixtures on the BEC side of
the Feshbach resonance at T = 0. At low temperatures and deep into the BEC limit, all
minority-component fermions are paired, forming stable bosonic molecules. So the system
can be thought of as a Bose-Fermi mixture, where bosons are tightly bound molecules
and fermions are excess or unpaired fermions of the majority species. Boson-boson (i.e.,
102
Chapter 4: Breakdown of the local density approximation in interacting systems of cold
fermions in strongly anisotropic traps 103
molecule-molecule) and boson-fermion (i.e., molecule-fermion) scattering lengths can be
related to the scattering length between two species of fermions as [155, 156, 157]
a
bb
= 0.6a, a
bf
= 1.18a. (4.1)
Before presenting a formal analysis, we provide a simple qualitative picture of
our results. The leading correction to the LDA comes from including gradient terms in
expressions for the chemical potentials of bosons and fermions. The gradient term for
bosons corresponds to the kinetic energy term of the Gross-Pitaevskii equation and has
been thoroughly analyzed in literature [27]. Gradient terms for fermions have also been
considered in the context of nuclear physics (see chapter 4 of Ref. [159]). Interestingly, it
turns out that gradient terms for fermions have a small numerical prefactor, and have a
negligible eect in the considered region of the parameters. For strongly anisotropic traps,
gradient terms smooth the density distribution of bosons in the tightly conned radial
direction at the boundary of the boson cloud (see the inset to Fig. 4.1). This means that
there are extra bosons in the radial direction compared to the LDA model. These extra
bosons interact repulsively with unpaired fermions and push them out in the axial direction
toward the long tips of the trap, leading to a double-peak structure in the axial density
of excess fermions (see Fig. 4.2). We point out that we do not nd a critical value of
polarization beyond which the double peak structure appears, although the peak strength
depends on polarization, scattering length, and the total number of particles in the system
(see Fig. 4.3).
The evolution of the aspect ratio of the minority species is also easy to understand
in the BEC limit that we consider. The aspect ratio of minority species is the aspect ratio
of bosons (molecules). When the LDA applies, the aspect ratio of bosons is equal to the
ratio of conning frequencies (in experiments reported in Ref. [48], this ratio is around
50). In the extreme limit of full polarization, one can consider a single boson (molecule)
in an eective potential created by the conning potential and excess fermions. Solving
a single-particle Schrodinger equation for the boson gives a wave function with the aspect
ratio equal to the square root of the ratio of conning frequencies (this corresponds to the
extreme breakdown of the LDA, see also [158]). When the number of bosons is small but
nite, one does not nd such a dramatic decrease of the aspect ratio, since even a small
number of bosons leads to a change in the fermion distribution (see the inset to Fig. 4.2).
However, this argument explains a general trend of decreasing aspect ratio of minority
species with increasing polarization.
The microscopic model that we consider is the interacting Bose-Fermi Hamiltonian
H =
_
d
3
r
_
h
2
2(2m)
b
+
h
2
2m
f
+
1
2
g
bb
b
+g
bf
b
_
. (4.2)
Here
f
(
b
) is a fermion (boson) operator and m is the original fermion mass. The
interaction parameters are g
bb
=
2h
2
a
bb
m
and g
bf
=
3h
2
a
bf
m
. The Hamiltonian (4.2) with the
parameters (4.1) can be rigorously derived in the dilute limit n
1/3
b
a 1, n
1/3
f
a 1, where
104
fermions in strongly anisotropic traps
n
b
and n
f
are local boson and fermion densities, from a single-channel model of a wide
Feshbach resonance [157]. It has been found out numerically [160] in the absence of density
imbalance that fermion mixture is well described by the model of interacting bosons up to
k
f
a 1.
Within the LDA and assuming a harmonic connement, the density proles are
given as solutions of the following equations:
f
(n
b
, n
f
) +V (x, y, z) = const
1
, (4.3)
b
(n
b
, n
f
) + 2V (x, y, z) = const
2
, (4.4)
where V (x, y, z) =
m
2
z
z
2
2
+
m
2
(x
2
+y
2
)
2
. Notice that in the experiment of Ref. [48], the
trap is highly anisotropic =
/
z
= 48.6 1. Boson and fermion densities depend on
coordinates (x, y, z) through the potential V (x, y, z) only. Hence we should have identical
densities of each species for points that have the same value of the conning potential.
Thus the aspect ratio of both clouds should always be equal to the trap anisotropy . Then
3D densities along the z axis, n
b
(z) n
b
(0, 0, z) and n
f
(z) n
f
(0, 0, z), provide complete
information about densities everywhere in the trap. From 3D densities n
b,f
(z) one can
calculate axial densities as
n
a
f
(z) =
_
dxdy n
f
_
_
(x
2
+y
2
)/
2
+z
2
_
=
2
2
_

0
r dr n
f
(
_
r
2
+z
2
) =

2
_

z
2
dt n
f
(
t). (4.5)
For z > 0, the derivative of the axial density is
dn
a
f
(z)
dz
=
2
2
zn
f
(z) 0. (4.6)
This provides a proof of the absence of peaks in the axial density away from z = 0. We note
that a similar statement can be made in the unitary and BCS regimes: if the density of
the majority component is larger than that of the minority component at any point in the
system, then the axial density cannot have a peak except at z = 0. Within the LDA and
harmonic trapping, phase separation is signaled not by the appearance of a double-peak
structure in the unpaired fermion axial density, but by the existence of an extended region
where the axial density dierence has a zero derivative.
We now consider beyond-LDA corrections, which arise due to the spatial derivatives
of the densities. At the mean-eld level [99], correlations between the Bose and Fermi clouds
are neglected and one treats the fermion (boson) density as providing an external potential
for the bosons (fermions). Including the gradient corrections discussed earlier, up to second
order in gradients of the densities, we nd [27, 159]
E =
_
d
3
r
_
1
2
g
bb
n
2
b
+g
bf
n
b
n
f
+V (r)(n
f
+ 2n
b
)
+
h
2
2m
_
[
n
b
[
2
2
+
(6
2
n
f
)
5/3
10
2
+
(n
f
)
2
36n
f
+

2
n
f
3
_
_
. (4.7)
m
(10 m )
18 3
n
f
(10 m )
18 3
b
n
m
50 100 150 200 250 300 350
0.5
1
1.5
2
300 400 500 600
0.02
0.04
0.06
0.08
0.1
0.12
Figure 4.1: Rescaled 3D densities of fermions in the Bose-Fermi model (4.2). These corre-
spond to density dierences of the majority and minority components n
f
= n
. LDA
result (dotted), n
f
(r = 0, z = x) (dashed), and n
f
(r = x/, z = 0) (solid) for harmonic
connement
z
/2 = 7.2 Hz and
/2 = 350 Hz, scattering length corresponding to

B = 754 G, N
+ N
= 9.46 10
4
, and P = 9.5%. Inset: Bose densities for the same
conditions.
In principle, for strong variations of the fermion density, higher-order terms in gradients have
to be included. The gradient expansion above is not suitable for studies of shell structure
[161], but since these eects are small in the case we are studying, this level of approximation
is sucient. Even though p-wave superuidity of fermions due to boson-induced interactions
has been predicted (see, e.g., Ref. [151]), the value of the gap is exponentially small in
1/n
1/3
f
a and cannot aect signicantly the density proles obtained from the expression
above.
Taking variations of E with the density, one obtains corrections to the local chem-
ical potential (valid in the region of nonzero fermion and boson densities),
f
=
h
2
2m
_
(6
2
n
f
)
2/3
+
1
36
_
(n
f
)
2
n
2
f

2
n
f
n
f
/2
_
_
+g
bf
n
b
,
b
=
h
2
4m
2
b
n
b
n
b
+g
bb
n
b
+g
bf
n
f
.
Notice that beyond-LDA corrections to the fermionic chemical potential carry a small pref-
actor of 1/36 compared to the analogous term for bosons.
To investigate numerically density proles in the presence of kinetic terms, we
use the steepest descent method of functional minimization (used previously for bosons in
anisotropic traps [162]). We investigate the eect of beyond-LDA corrections for typical
parameters used in experiments (cf. Ref. [48]). In Fig. 4.1, we take harmonic connement
with
z
/2 = 7.2 Hz and
/2 = 350 Hz, a scattering length that corresponds to B =

754 G, number of particles N = N
+ N
= 2N
b
+ N
f
= 9.46 10
4
, and polarization
106
0.2 0.4 0.6 0.8
0.65
0.75
0.8
0.85
0.9
0.95
P
1 a
R
r
R
200 400 600 800 1000
2
4
6
8
10
12
6 1
n
f
a
m
(10 m )
Figure 4.2: Axial densities of fermions (n
a
f
= n
a
n
a
) within LDA (dotted) and in the

presence of beyond-LDA corrections (solid), for the same parameters as in Fig. 4.1. The
inset shows the dependence of the anisotropy of the Bose cloud on polarization P, with
other parameters being the same as in Fig. 4.1. R
a
and R
r
are obtained from the ts to
the columnar density as discussed in the text.
P = (N
)/(N
+ N
) = N
f
/(N
f
+ 2N
b
) 10%. The LDA density prole is almost
completely phase separated due to the strong repulsion between bosons and fermions that
spatially overlap only in a small region. In Fig. 4.1, we compare rescaled 3D density proles
in radial and axial directions with LDA results, and in Fig. 4.2 we do the same comparison for
axial densities and identical values of the system parameters. Modication of 3D densities
can be of the order of 30% for fermions, and is much stronger in the radial direction. This
is expected, since the gradient corrections in the radial direction are
2
2.4 10
3
times
larger. We observe a peak in the axial density, with the density at the maximum about 10%
higher than at z = 0. The modication of the axial density is less prominent than changes
in the 3D density prole due to the integration entering the denition of the axial density.
Figure 4.3 shows the dependence of the relative strength of the double-peak struc-
ture (e) on polarization (P), where e is dened as e = [n
a
f
(z)[
max
n
a
f
(0)]/n
a
f
(0). It decreases
with increasing P, since for a larger polarization a smaller fraction of the fermion cloud gets
aected by the boundary of the boson cloud. Being essentially a nite-size eect, e also
decreases with increasing total number of particles. However, this dependence is pretty
weak, since the size of the cloud depends weakly on the total number of atoms (only as
N
1/6
for the unitary regime within the LDA). The inset to Fig. 4.3 shows the dependence
of e on scattering length (a). As a increases, the region of spatial overlap of the Bose and
Fermi clouds shrinks due to stronger repulsion, and this leads to larger gradients and larger
beyond-LDA corrections for the boson cloud. In addition, due to the enhanced repulsion,
the fermions are more sensitive to the corrections to the boson cloud. Both of these eects
lead to an increase of e for larger a. We present results as a function of K
f
a, where K
f
is dened as in Ref. [48]. Notice, however, that there is no universality in this curve:
indeed, increasing the total number for xed a would increase K
f
a, while e would decrease,
K
f
a
0.2 0.4 0.6 0.8 1.2
0.05
0.1
0.15
0.2
e
0.1 0.2 0.3 0.4 0.5
0.1
0.2
0.3
0.4
0.5
P
e
Figure 4.3: Dependence of the strength of the double-peak structure e on polarization P,
with other parameters being the same as in Fig. 4.1. The inset shows how e changes with
a when we move away from the point considered in Fig. 4.1 (marked with an arrow). K
f
is
dened as in Ref. [48].
as discussed earlier. Although K
f
a in Fig. 4.3 is of the order of 1, the main physics we are
interested in takes place at the boundary of the boson cloud, where local k
f
a is considerably
smaller, so the treatment of the system as a Bose-Fermi mixture is well justied.
Beyond-LDA corrections increase the size of the boson cloud in the radial direction
compared to the LDA result, thus reducing the aspect ratio of the bosonic cloud compared
to . As the polarization (P) increases, for a xed total number of atoms, the size of
the bosonic cloud decreases and the importance of beyond-LDA eects grows. The inset
to Fig. 4.2 shows the evolution of anisotropy as a function of polarization for harmonic
connement. R
a
and R
r
are dened as follows: (i) the columnar density n
c
b
(r, z) is obtained
from n
b
(r, z), assuming radial symmetry in the x-y plane; (ii) one then ts non interacting-
fermion density distributions to the columnar density along axial and radial directions,
n
c
b
(r, 0) = A(1
r
2
R
2
r
)
2
and n
c
b
(0, z) = B(1
z
2
R
2
a
)
2
. Notice these functional forms are not
always suitable and the ts provide just an estimate of the radii. The aspect ratio of the
fermionic cloud does not get signicantly modied in a harmonic connement.
So far we discussed only the BEC regime, where controllable analytic theory is
available. We developed a consistent qualitative picture of the appearance of peaks in the
axial density dierence, n
a
f
= n
a
n
a
, and a decrease of the aspect ratio of minority species

with increasing polarization. Quantitatively, the numbers we obtain in the BEC regime are
smaller than what is observed experimentally in Ref. [48] for the unitary regime, but of the
same order of magnitude as the eects of unharmonicity discussed in Ref. [154]. Now we
will comment on the relevance of our ndings for the unitary limit, where most experiments
are performed. Results from the BEC regime should not be directly extrapolated toward
the unitary regime, since the treatment of the system as a Bose-Fermi mixture starts to
break down, and the form of beyond-LDA corrections in the unitary regime is not known.
Hence the statements made in this paragraph are only speculative. One can estimate,
within the LDA, the chemical potential in the center of the trap based on Fig. 2C of
108
Ref. [48]: for a typical conguration it is 10 h
, where h
is the radial level spacing.

The chemical potential at the boundary of the inner cloud is about three to four times h
,
whereas the gradient contribution in the radial direction, which is neglected in the LDA, is
h
2
2
/[m(R)
2
] h
(where R is the dierence in axial radii). It is thus clear that

beyond-LDA corrections might also be relevant for the unitary regime. The appearance of
double peaks has been interpreted in Ref. [48] as evidence for phase separation of excess
fermions from a paired central core. Although peaks should not appear in the LDA, the
picture we have in the BEC limit provides some support for this interpretation. When
excess fermions spatially overlap with the paired core, they are less sensitive to beyond-
LDA corrections, which are important only at the boundary of the paired cloud. As phase
separation takes place, excess fermions are expelled to the boundary, where beyond-LDA
eects become more important. These corrections are stronger in the radial direction, hence
fermions are pushed in the axial direction to larger [z[. This may give rise to a pronounced
double-peak structure in the axial density. The high anisotropy of the trap is important for
this scenario.
To summarize, we considered unbalanced fermions in the BEC limit for T = 0. We
proved a theorem that prohibits peaks in the axial density for z ,= 0 within the LDA and
for the harmonic trapping potential. We showed that for strongly anisotropic connement,
beyond-LDA corrections produce a double-peak structure in the axial density, and change
the aspect ratio of the inner cloud. We discussed the implications of our ndings for the
unitary regime.
Chapter 5
Fundamental noise in matter
interferometers
This chapter discusses two eects which contribute to the reduction of the inter-
ference fringe contrast in matter interferometers. The rst eect is the shot noise arising
from a nite number of atoms used in experiments. Focusing on a single shot measurement
we provide explicit calculations of the full distribution functions of the fringe contrast for
the interference of either the coherent or the number states of atoms. Another mechanism
of the suppression of the amplitude of interference fringes discussed in this chapter is the
quantum and thermal uctuations of the order parameter in low dimensional condensates.
We summarize recent theoretical and experimental studies demonstrating that suppression
of the interference fringe contrast and its shot to shot variations can be used to study corre-
lation functions within individual condensates. We also discuss full distribution functions of
the fringe amplitudes for one and two dimensional condensates and review their connection
to high order correlation functions. We point out intriguing mathematical connections be-
tween the distribution functions of interference fringe amplitudes and several other problems
in eld theory, systems of correlated electrons, and statistical physics.
5.1 Introduction
5.1.1 Interference experiments with cold atoms
From the earliest days of quantum mechanics its probabilistic nature was the cause
of many surprises and controversies [163]. Perhaps the most unusual manifestation of the
quantum uncertainty is a quantum noise: measurements performed on identical quantum
mechanical systems can produce results which are dierent from one experimental run to
another. At the level of a single particle quantum mechanics, the quantum noise is no longer
a research topic but is discussed in undergraduate physics textbooks [164]. However, the
situation is dierent when we talk about quantum mechanics of many-body systems. One
can ask seemingly simple questions to which there is no obvious answer: does it still make
sense to talk about quantum noise when discussing measurements on many-body quantum
states? How does the quantum noise manifest itself? Can one use this noise to extract
109
110 Chapter 5: Fundamental noise in matter interferometers
nontrivial information about the system?
The idea of the quantum noise analysis of many-particle systems is common to
many areas of condensed matter physics [165, 166, 167, 168] and quantum optics [169, 174,
173, 170, 171, 172]. In the eld of ultracold atoms it has been successfully employed in a
variety of recent experiments [175, 176, 177, 178, 179, 180, 181] with many more theoretical
proposals awaiting their turn [182, 183, 184, 185, 186, 187, 188]. This chapter addresses a
very specic experimental probe of the cold atomic ensembles interference experiments.
Our discussion focuses on a variety of interesting and important phenomena which originate
from the fundamental quantum and/or thermal noise of cold atoms condensates and can be
studied in interference experiments. Although focused on the specic type of experiments,
the general methodology discussed in this chapter can be extended to a variety of other
measurements on systems of cold atoms.
Interference experiments constitute an important part of the modern toolbox for
studying ultracold atoms. Original experiments used large three dimensional Bose-Einstein
condensates (BEC) to demonstrate macroscopic coherence [15]. More recently interference
experiments have been done with one and two dimensional condensates [189, 190, 191,
192, 193, 194] and demonstrated the important role of uctuations in low dimensional
systems. Matter interferometers using cold atoms [195, 196, 197, 198, 199, 200, 190, 191,
204, 201, 202, 203, 208, 205, 206, 207, 209, 210, 211] have been considered for applications
in accelerometry, gravitometry, search for quantum gravity, and many other areas (for a
review see Ref. [212]). Interference experiments have been used to measure the condensate
formation [20, 213] as well as the critical properties of the BEC transition [214]. What is
common to most interference experiments is the focus on the phase of interference patterns.
Suppression of the fringe contrast is considered to be a spurious eect caused by noise
and uctuations. However this chapter focuses on understanding physical phenomena that
underlie the imperfect visibility of interference fringes. As we discuss below, suppression of
the fringe visibility comes from fundamental physical phenomena, such as the noise intrinsic
to performing a classical measurement on a quantum mechanical wave function (shot noise)
or classical and quantum uctuations of the order parameter.
In this chapter we discuss how one can use analysis of the contrast of interference
fringes to learn about uctuations of the order parameter [215, 216, 217, 218, 193, 194]. We
will demonstrate that important information is contained not only in the average contrast
but also in its shot to shot variations. For example, when we discuss uctuating condensates,
we will show that high moments of the interference fringe amplitudes contain information
about high order correlation functions and thus provide valuable information about the
system.
The basic scheme of interference experiments is shown in Fig. 5.1. Originally
two condensates are located at distance d away from each other. At some point they
are allowed to expand ballistically
1
until sizes of the clouds become much larger than the
1
Readers may be concerned that the initial part of expansion may not be purely ballistic and atomic
collisions may take place in expanding clouds. In the case of tight connement such collisions are rare and
have minimal eect. One can use such approximation in a variety of experimental situations, because the
density of the condensate falls of rapidly during the rst stages of the expansion. Eects of the interactions
on the interference signal have been recently discussed in Refs. [219, 220, 221]. It is worth noting that the
eect of collisions can be minimized by tuning the scattering length to zero at the moment of the release
Chapter 5: Fundamental noise in matter interferometers 111
Time of
flight
d
Figure 5.1: Schematic view of the interference experiments with ultracold atoms. Two
condensates, originally separated by the distance d, are released from the traps and expand
until the clouds overlap. The imaging beam measures the density of the atoms after the
expansion. Quantum interference leads to the periodically modulated density projected
on the screen. Projected density image is taken from the actual experimental data of
Hadzibabic et al. [193]. Adapted by permission from Macmillan Publishers Ltd : Nature
([193]), copyright (2006).
original separation between the clouds d. After the expansion the density is measured by
shining a laser beam through the cloud. Interference leads to the appearance of the density
modulation at a wave vector Q = md/ht (see Fig. 5.1 and discussion in section 5.2).
When the two condensates are coherent, the position of interference fringes is determined
by the relative phase between the two clouds. Surprisingly the interference pattern will be
observed even for two independent condensates which do not have a well dened relative
phase (see e.g. Fig. 5.2). To beginning readers it may seem confusing that we can observe
interference in the absence of coherence between the two clouds. Or even more confusing, we
discuss interference patterns when both clouds are number states and phases of individual
condensates are not well dened. Several theoretical frameworks have been introduced
to understand the origin of interference patterns in the absence of phase coherence [222,
223, 224, 225, 26]. In this chapter we explain the origin of interference fringes using the
language of correlation functions and point out connections to Hanbury Brown and Twiss
(HBT) experiments [169] in optics. This section provides a simple heuristic picture and a
more formal discussion is left to section 5.2.
What one measures in experiments is the density prole after the expansion, (r).
Interference pattern appears as the density modulation (r) =
Q
e
iQr
+c.c. + const. The
absolute value of
Q
determines the amplitude of interference fringes and its phase denes
the position of the fringe maxima and minima. Schematically we can write
Q
e
i
1
i
2
,
where
1,2
are phases of the two condensates before the expansion (see discussion in section
5.2). In the absence of coherence e
i
1
i
2
) = 0, which implies that
Q
) = 0. The fact
that the average vanishes, however, does not mean that interference fringes are absent in
using magnetic Feshbach resonances [28]. See also discussion after Eq. (5.43).
Figure 5.2: Interference pattern from the rst set of experiments with sodium atoms. Taken
from Ref. [15]. Reprinted with permission from AAAS.
each individual shot. In the present case this only shows that the phase of interference
fringes is random in each shot. We remind the readers that taking an expectation value in
quantum mechanics implies averaging over many measurements. On the other hand, we can
focus on the amplitude of interference fringes and accept the fact that we can not predict
their phase. Then we need to consider the quantity which does not vanish after averaging
over the unpredictable phase dierence. One such quantity is given by the density-density
correlation function (r)(r
) = [
Q
[
2
(e
iQ(rr
)
+c.c.) +other terms. The right hand side of
the last equation does not vanish when we average over random phases
1,2
, and we nd
nite expectation value
(r)(r
)) = 2[
Q
[
2
) cos(Q(r r
)) + const. (5.1)
What this correlation function tells us is that in a single shot we can not predict whether
at a given point r we will have a minimum or a maximum of the density modulation.
However what we can say is that if there is a maximum at point r, it will be followed by
another maximum a distance 2/Q away. While this simple argument explains the origin of
interference patterns from independent condensates, it leaves many questions unanswered.
For example, it is not obvious how accurately one can represent two independent condensates
using states with a well dened but unknown phase dierence. Also it is not clear how to
generalize this analysis to elongated condensates, when we need to go beyond the single
mode approximation and include phase uctuations within individual systems. This chapter
will present a uniform approach for addressing these and many other questions.
When the focus of interference experiments is on measuring the phase, one usually
averages interference patterns obtained in several shots. The result is easy to interpret:
an average of many experimental runs is precisely what we dene as a quantum mechani-
cal average. However in experiments with independent condensates, summing interference
patterns is not appropriate. The phase of interference patterns is random from shot to
1
k
2
k
r
' r
Figure 5.3: Schematic view of Hanbury Brown and Twiss noise correlation experiment as
an example of intensity interferometry. Detectors at positions r, r
measure the intensity of

light coming from two distant incoherent sources. The correlator (denoted by the box)
measures the coincidence events and thus the intensity-intensity correlation function.
shot and adding individual images washes out interference fringes completely (for a nice
experimental demonstration see Ref. [226]). Hence, in this case one needs to focus on in-
terference patterns obtained in individual shots. In the absence of averaging, a single shot
measurement contains noise. Thus to characterize such experiments we need to provide
both the average value and the shot to shot uctuations of the fringe contrast. The most
comprehensive description of the uctuating variable comes from providing its full distri-
bution function. Theoretical calculations of the distribution functions of the fringe contrast
will be the central part of this chapter.
It is useful to point out the analogy between the approach discussed in this chapter
and the famous Hanbury Brown and Twiss experiments in optics. The original motivation
for HBT experiments came from astronomy: the goal was to measure the angle between
two incoherent stellar sources such as two dierent points on the surface of the star. Since
the two sources are incoherent, this can not be done using a single detector: rst order
interference is absent and the measured signal is simply the sum of the two intensities [174].
The insight of HBT was to use two detectors and measure the correlation function of the
two intensities as a function of the relative distance between the two detectors. One nds
that this correlation function is given by
I(r)I(r
)) cos
_
(k
1
k
2
)(r r
)
_
+ const, (5.2)
where k
1,2
are wave vectors of photons arriving from the two points on the surface of the
star, r, r
are positions of the detectors, and I(r), I(r
) are the intensities measured in the

two detectors (see Fig. 5.3). Hence, the main idea of HBT experiments is that information
is contained not only in the average signal I(r)), but also in the noise. Such noise can be
characterized by looking at higher order correlations. In astronomy, HBT experiments were
used to measure several important properties of distant stars, including their angular sizes
and the surface temperature [227].
There is an obvious analogy between equations (5.1) and (5.2), but there is also
an important dierence. HBT stellar interferometers operate in real time, and averaging
over time is built into the measuring procedure. In these experiments, uctuations of
I(r)I(r
) I(r)I(r
)), which would correspond to higher order correlations in I(r), are

not easy to measure. On the other hand interference experiments with cold atoms are
of a single-shot type: each measurement is destructive and gives a certain density prole
(r). A single image contains information not only about the two-point correlation function,
(r)(r
)), but about higher order correlations as well. The most important information
for our purposes is contained in the interference pattern at a wave vector Q. Essentially
each interference pattern constitutes a classical measurement of the quantum mechanical
operator
Q
. We remind the readers that we expect the phase of
Q
to be random, so the
quantity of interest will be [
Q
[
2
. By performing measurements several times we will nd
not only the average value of this operator, but also its higher moments. Ultimately we
should be able to reconstruct the entire distribution function for [
Q
[
2
. So the simplied and
idealized procedure that we analyze is the following: one performs interference experiments
many times. Each experiment is analyzed by doing a Fourier transform of the density to
extract
Q
. The histogram of the measured values of [
Q
[
2
will be the main subject of this
chapter. We will demonstrate the wealth of information that can be extracted from analysis
of such histograms.
As a passing note, we mention that the setup considered in Fig. 5.1 is not the only
possible conguration for interference experiments. In another common setup one makes
several copies of the same cloud using Bragg pulses, and observes an interference between
them [228, 229, 230, 231, 232, 233, 234].
It is useful to put our work in the general perspective of noise analysis in physical
systems. Understanding photon uctuations is at the heart of modern quantum optics and
provides a basis for creation, detection, and manipulation of non-classical states of light
[173]. The eld of quantum optics has a long and fruitful tradition of using the higher-
order correlation functions as well as the shot noise to characterize the quantum states of
light. The notion of higher-order degree of coherence was rst introduced by R. Glauber
in 1963 [170], also by Klauder and Sudarshan [171] and by Mandel and Wolf [172]. The
photoelectron counting distribution function reveals such non-classical features of light as
antibunching [235], sub-Poissonian statistics and probe of violation of Bell inequalities. In
particular, third-order correlations provide a test for distinguishing between quantum and
hidden variable theories in a way analogous to that provided by the Greenberger-Horne-
Zeilinger test of local hidden variable theories [236]. Interference of independent laser beams
was rst observed in Ref. [237] and stimulated a number of theoretical studies (for reviews
see Refs. [172, 238, 239]).
In condensed matter physics, noise analysis was also suggested as a powerful ap-
proach for analyzing electron systems [165]. It was demonstrated theoretically that in
certain mesoscopic systems current uctuations should contain more information than the
average current itself. In particular, the third and higher moments contain important
quantum information on interaction eects, entanglement and relaxation processes (see e.g.
Refs. [240, 241]). Specic proposals exist for detecting statistics of quasi-particles [242], un-
derstanding transmission properties of small conductors [166] and observing entanglement
between electrons [167]. Perhaps the most spectacular experimental success of the noise
analysis in electronic systems has been the demonstration of the fractional charge of quasi-
particles in the fractional quantum Hall regime [168]. However generally the noise analysis
in condensed matter systems did not become the detection tool of the same prominence
as in quantum optics. The main reason for this is the excruciating diculty of the noise
measurements in solid state experiments. One often needs to measure a signal which is only
a part in a million of the unwanted technical noise.
In the eld of ultracold atoms, experiments analyzing quantum noise are only be-
ginning. However we have already seen spectacular success in several recent experiments.
The analysis of noise correlations in the time of ight experiments [182] was used to demon-
strate fermionic pairing [175] as well as HBT type correlations for atoms in optical lattices
[176, 177, 178]. Single atom detectors have been used to demonstrate HBT noise correlations
for cold atoms [179, 180, 181]. Strongly interacting systems of cold atoms are expected to
realize analogues of important models of condensed matter systems [1, 38, 49]. Being able
to study noise in such systems should provide an important new perspective on strongly
correlated states of matter and have a profound eect on many areas of physics. The rst
success in this direction was the recent observation of the Berezinskii-Kosterlitz-Thouless
(BKT) transition [243, 244] by Hadzibabic et al. in Ref. [193].
5.1.2 Fundamental sources of noise in interference experiments with mat-
ter
Two fundamental sources of uctuations in the amplitude of interference fringes
are the shot noise and the order parameter uctuations within individual condensates.
Shot noise comes from the nite number of atoms used in the experiments. Let
us discuss limiting cases rst. Consider an interference experiment with one atom. Before
the expansion the atom is in a perfect superposition between the two wells. After the
expansion we get a perfectly periodic wave function (r) = 2C cos(
Qr+
2
) (for a more
detailed discussion see section 5.2). The expectation value of the density operator is (r)) =
[(r)[
2
) = 2[C[
2
(cos(Qr + ) + 1). However this average value will not be measured in a
single shot. A single measurement nds the atom at a single point. The expectation value
of the density determines probabilities with which we can nd atom at any given point,
but in a single measurement we collapse a quantum mechanical wave function and observe
the atom at one point only. Can one reconstruct the entire amplitude of the interference
pattern
Q
= [C[
2
e
i
from a single measurement? Obviously, the answer is no. In the
opposite case of a very large number of atoms in the same single particle state one should
be able to reconstruct a complete interference pattern from a single measurement, since
the measurement of positions of many atoms performs a statistical averaging implicit in
quantum mechanics. In the general case of experiments with a nite number of atoms,
the question that arises is how well one can determine the amplitude of interference fringes
from doing a single shot measurement. Formulated more accurately the problem is to
determine probabilities of nding a certain amplitude of interference fringes, [
Q
[
2
, in a
single measurement.
Fluctuations of the order parameter are particularly important for low dimensional
systems. If the condensates are conned in one [245, 42, 43, 158] or two [192, 193, 194, 246,
158] dimensions, then the true long-range order may not exist. Rigorous theorems forbid
2 (L)
Q
Z
X
IMAGING BEAM
L
Y
Figure 5.4: Simplied setup of interference experiments with 1D Bose liquids (see e.g.
Ref. [190]). Two parallel condensates are extended in the x direction. After atoms are
released from the trap, clouds are imaged by the laser beam propagating along the z axis.
Meandering structure of the interference pattern arises from phase uctuations along the
condensates. The net interference amplitude
Q
(L) is dened from the density integrated
along the section of length L.
true long range order in two dimensional systems at nite temperature and in one dimen-
sion even at zero temperature [247, 248, 75]. What this means is that low dimensional
condensates can not be characterized by a single phase and we need to take into account
spatial uctuations of the order parameter. The eects of such uctuations on interference
experiments are illustrated schematically in Fig. 5.4. Two one dimensional clouds expand
in the transverse direction. Each point along the condensates has a local interference pat-
tern, but in the presence of phase uctuations (either thermal or quantum), these patterns
are not in phase with each other. It is natural to dene the net interference amplitude
from the density integrated over the axis of the system (the so called columnar density).
In many experiments such integration is done by the measurement procedure itself. For
example, systems such as shown in Fig. 5.4 originally had imaging done along the axis of
the interferometer [190]. Then the laser beam integrates the atomic densities within the
imaging length. Integrating over local interference patterns which are not in phase with
each other leads to a reduced contrast of the net interference fringes.
In earlier literature smearing of interference patterns by uctuations was consid-
ered an unwanted eect [15]. However, the point of view presented in these notes is quite
the opposite. Suppression of the fringe contrast is an interesting eect which tells us about
important phenomena in Bose condensates, such as thermal and quantum uctuations of
the order parameter. By analyzing such suppression we can extract non-trivial informa-
tion about the system. In particular, it has been shown in Ref. [215] that the scaling of
the average interference signal with the observation area contains information about the
two-point correlation functions within each cloud. Recent experiments [193] by Hadzibabic
et al. used this approach to observe the BKT transition in two dimensional condensates
(see discussion in section 5.4). We also note that such experiments can be used to extract
information which is dicult to obtain by other means. For example, in section 5.4 we show
that the analysis of the high moments of the contrast tells us about high order correlations
within individual clouds.
This chapter is organized as follows. In section 5.2 we discuss how interference
fringes appear for ideal non interacting 3D BECs at zero temperature. In section 5.3 we
analyze the shot noise for ideal condensates. Ideal condensates are understood as clouds
of non-interacting atoms which occupy a single mode within each of the traps before the
expansion. The problem of interference of independent condensates of ideal bosons has been
analyzed extensively in literature before [222, 223, 224, 225, 26, 254, 249, 252, 253, 250, 251].
In particular, in a recent important paper [255], Polkovnikov showed that the variance of
the fringe amplitude decreases as an inverse power of N, with a non-universal coecient
which contains information about the state of each cloud (e.g. coherent states vs Fock
states). In this chapter we develop a general formalism for calculating the full distribution
functions of the fringe amplitudes in interference experiments with ideal condensates with a
nite number of atoms. We apply this formalism to obtain distribution functions for several
experimentally relevant cases such as states with a well dened phase dierence between
the two clouds and Fock states of atoms. Eects of the order parameter uctuations are
discussed in section 5.4. We obtain distribution functions for both one and two dimensional
condensates in the limit when the number of particles is large and the shot noise can be
neglected. We also discuss intriguing mathematical connections between these distribution
functions and a number of important problems in physics, such as the quantum impurity
problem in a low dimensional interacting electron system [256] or the distribution of rough-
ness in systems with 1/f noise [257]. In this chapter we do not address the issue of technical
noise which is obviously important for understanding real experiments. In the concluding
section 5.5, however, we comment on the experimental requirements for observing some of
the phenomena discussed in this chapter.
5.2 Interference of ideal condensates
In this section we discuss why interference fringes appear for ideal non interacting
BECs at zero temperature. We follow Ref. [26], and introduce notations for subsequent
sections. First we consider the case of two clouds with a well dened relative phase, where
the appearance of interference fringes can be understood at a single particle level. Then
we show that almost ideal interference fringes appear even when two expanding clouds are
uncorrelated, provided that the number of particles in each cloud is large.
5.2.1 Interference of condensates with a well dened relative phase
Basics of interference experiments. First quantized representation
To illustrate how interference fringes arise, let us start by considering a simple case
of two BEC clouds with a well dened relative phase. Here we neglect interactions between
atoms, so initially all atoms are assumed to be in the same single-particle state (single mode
approximation). After the conning potential is removed, the single-particle state evolves
with time, but many body wave function remains in the product state. Interference appears
as a result of single particle wave function evolution, which can be studied in detail.
Normalized single particle wave functions for two clouds will be denoted as
1
(r, t)
and
2
(r, t), and the initial relative phase is . If the total number of particles equals N, then
the complete wave function of the system in the rst quantized notations at any moment
of time is given by
(r
1
, ..., r
N
, t) =
N
n=1
1
2
(
1
(r
n
, t)e
i/2
+
2
(r
n
, t)e
i/2
). (5.3)
This wave function satises the proper symmetry requirements for permutations of r
i
and
r
j
, and evolution of
1
(r, t) and
2
(r, t) is controlled by the single particle Schrodinger
equation. Initial overlap of the states
1
(r)
1
(r, 0) and
2
(r)
2
(r, 0) is assumed to
be negligible:
_
dr
1
(r)
2
(r) 0. (5.4)
The expectation value of the total density corresponding to the wave function (5.3)
is
(r, t)) =
N
2
_
[
1
(r, t)[
2
+[
2
(r, t)[
2
+ 2Re
_
e
i
1
(r, t)
2
(r, t)
_
. (5.5)
The expectation value of the total density displays an interference pattern due to the last
term of Eq. (5.5). As a simple example, let us assume that
1
(r, t) and
2
(r, t) are initially
in the Gaussian states centered at points d/2, and their widths are R
0
d. Then the
evolution of the single particle wave functions can be simply calculated, and the result is
1
(r, t) =
1
(R
2
t
)
3/4
e
(rd/2)
2
(1+i ht/mR
2
0
)
2R
2
t
, (5.6)
2
(r, t) =
1
(R
2
t
)
3/4
e
(r+d/2)
2
(1+i ht/mR
2
0
)
2R
2
t
, (5.7)
where the widths of the wave packets, R
t
, at time t are given by
R
2
t
= R
2
0
+
_
ht
mR
0
_
2
. (5.8)
We will be interested in the regime, when the sizes of the clouds R
t
are much larger than
the original distance between the clouds, that is
R
t
d R
0
. (5.9)
In this regime, the clouds overlap strongly, and the real parts of the exponents in Eqs. (5.6)
and (5.7) are responsible for the broad overall density prole. Imaginary parts in the same
2 1
d
r
R
0
Figure 5.5: Schematic view of the interference experiment with 3D condensates.
exponents give rise to interference eects in the last term of Eq. (5.5). Thus the interference
part of the density is equal to
N
(R
2
t
)
3/2
e
r
2
+d
2
/4
R
2
t
cos
_
h
m
rd
R
2
0
R
2
t
t +
_
. (5.10)
For suciently large t, one can substitute R
t
ht/mR
0
, and obtain oscillations of the
density at wave vector Q = md/ht, with positions of the minima and the maxima controlled
by the relative phase . The Fourier transform of the density at wave vector Q is
Q
) =
_
dr(r, t)e
iQr
)
N
2
e
i
. (5.11)
Physically Q can be understood as the momentum dierence of the two particles
which have been released from the two traps and arrive simultaneously at the detection
point r. This can be seen from the following quasi-classical argument. A particle released
from the condensate one and detected at time t at point r has momentum
Q
1
= m(r d/2)/( ht). (5.12)
During the expansion this particle picks up a phase Q
1
(r d/2). Analogously a particle
originating from the condensate two has momentum
Q
2
= m(r +d/2)/( ht) (5.13)
and picks up a phase Q
2
(r +d/2). The interference pattern arises from the oscillating
structure of the phase dierence with the wave vector of oscillations
Q = Q
1
Q
2
= md/ht. (5.14)
This simple argument shows that as long as the original sizes of the clouds are much smaller
than the distance between them, after sucient expansion one should observe oscillations
of the density at the wave vector Q determined by the distance between the clouds.
The interference patterns which we introduced up to this point appear as a result
of the time evolution of single particle states. The many-body nature of the state comes
into Eq. (5.5) only as a prefactor N. In principle, one could have done the same experiment
with only one particle. The same result as Eq. (5.5) can be obtained in this case by doing
experiments many times and averaging over individual experiments: in each particular
realization the particle is observed at some random point r with probability (r, t)). In
experiments with a large number of atoms, N, used in each shot, each absorption image is
a result of N measurements of single particle wave functions. This performs the statistical
averaging implicit in quantum mechanics, and leads to the density prole close to Eq. (5.5)
in each shot.
Second quantized representation
To set up the stage for later we will now present the discussion leading to Eq. (5.11)
using the second-quantized formalism. Wave function (5.3) in the second quantization at
t = 0 can be written as
[, N) =
1
(2
N
N!)
1/2
(a
1
e
i/2
+a
2
e
i/2
)
N
[0). (5.15)
Here a
1
and a
2
are creation operators for clouds one and two:
a
i
=
_
dr
i
(r)
(r). (5.16)
(r) is the second quantized operator for the boson eld, which satises the usual commu-
tation relations
_
(r
),

(r)
_
= (r
r),
_
(r
),

(r)
_
=
_
(r
),

(r)
_
= 0. (5.17)
Operators a
1
, a
2
and their conjugates satisfy the canonical boson commutation relations:
_
a
i
, a
j
_
=
ij
, [a
i
, a
j
] =
_
a
i
, a
j
_
= 0. (5.18)
Dierent initial states can be simply written using the Fock basis of operators a
1
and a
2
.
For example, the initial state for two independent condensates with N
1
and N
2
particles in
clouds 1 and 2 can be conveniently written as
[N
1
, N
2
) =
1
N
1
!N
2
!
(a
1
)
N
1
(a
2
)
N
2
[0). (5.19)
The initial state written in the Fock basis of operators a
1
and a
2
contains all information
about properties of the interference amplitudes. During the free expansion the occupation
numbers of states one and two do not change, and only the single particle wave functions
1
(r, t) and
2
(r, t) evolve. After the expansion, the many-body wave function at time t
can be obtained from the initial state written in the Fock basis of operators a
1
and a
2
using
substitutions
a
1

_
dr
1
(r, t)

(r), (5.20)
a
2

_
dr
2
(r, t)
(r). (5.21)
For example, substituting Eqs. (5.20)-(5.21) into Eq. (5.15), the wave function (5.3) con-
sidered earlier is written as
[, N, t) =
1
(2
N
N!)
1/2
__
dr(
1
(r, t)e
i/2
+
2
(r, t)e
i/2
)
(r)
_
N
[0). (5.22)
In the long time limit (5.9) considered earlier, single-particle wave functions
1
(r, t),
2
(r, t) can be written as
1
(r, t) = u
1
(r, t)e
iQ
1
r
, (5.23)
2
(r, t) = u
2
(r, t)e
iQ
2
r
, (5.24)
where Q
1
, Q
2
are dened by Eqs. (5.12)-(5.13) and u
1
(r, t), u
2
(r, t) are slowly varying real
functions, which determine the overall density proles. Since clouds overlap strongly after
the expansion, these functions are normalized according to
_
u
1
(r, t)
2
dr = 1,
_
u
2
(r, t)
2
dr = 1, (5.25)
_
u
1
(r, t)u
2
(r, t)dr 1. (5.26)
The operator, which corresponds to the amplitude of density oscillation at wave
vector Q is written in the second quantized notations as

Q
=
_
dr (r)e
iQr
=
_
dr
(r)
(r)e
iQr
. (5.27)
To nd out the statistical average of the amplitude of density oscillations for state
(5.22), we need to evaluate the following matrix element:
Q
) = , N, t[
Q
[, N, t) = , N, t[
_
dr
(r)
(r)e
iQr
[, N, t). (5.28)
To evaluate such matrix elements, rst we need to know how annihilation operator

(r) acts
on a state [, N, t). Since [, N, t) is obtained from [, N) by substitutions (5.20)-(5.21), it
is easy to see that
, N, t[
(r)
(r)[, N, t) = , N[
_
a
1
u
1
(r, t)e
iQ
1
r
+a
2
u
2
(r, t)e
iQ
2
r
_
_
a
1
u
1
(r, t)e
iQ
1
r
+a
2
u
2
(r, t)e
iQ
2
r
_
[, N),
where [, N) is dened in Eq. (5.15) and is written only in terms of a
1
and a
2
. Integration
over dr in (5.28) can be done using normalization (5.26), and assuming that u
i
(r, t) vary
at scales much larger than 1/Q. Since Q = Q
1
Q
2
, evaluation of Fourier transform picks
only the product of the rst term in the rst parentheses and of the second term in the
second parentheses of the equation above. We obtain
Q
) = , N[a
1
a
2
_
u
1
(r, t)u
2
(r, t)dr[, N) =
, N[a
1
a
2
[, N) = , N 1[
_
_
N
2
e
i/2
_
2
[, N 1) =
N
2
e
i
, (5.29)
which is the same as obtained from a single particle discussion in Eq. (5.11).
The example above illustrates how the matrix elements of many-particle operators
at time t can be evaluated using initial states written in the Fock basis of operators a
1
and
a
2
. In general, when one needs to evaluate an expectation value of some normal ordered
combination of operators

(r), and

(r) over the nal state at time t, one needs to make

substitutions
(r) a
1
u
1
(r, t)e
iQ
1
r
+a
2
u
2
(r, t)e
iQ
2
r
, (5.30)
(r) a
1
u
1
(r, t)e
iQ
1
r
+a
2
u
2
(r, t)e
iQ
2
r
, (5.31)
and evaluate matrix elements over the t = 0 state, written in the Fock basis of operators a
1
and a
2
. It is important that the expression needs to be normal ordered using commutation
relations (5.17) before making substitutions (5.30)-(5.31), since after substitutions (5.30)-
(5.31) elds

(r),

(r) do not satisfy the exact commutation relations (5.17).

Another way of seeing this is to realize that in Heisenberg representation (see e.g.
chapter 6 of Ref. [258]) substitutions (5.30)-(5.31) perform the time evolution of a product
of boson annihilation operators
(r
1
, t)...
(r
n
, t) = e
i

Ht

(r
1
)...
(r
n
)e
i

Ht
(5.32)
only when this product acts on states of the form
[
0
) =
1
N
1
!N
2
!
(a
1
)
N
1
(a
2
)
N
2
[0). (5.33)
Thus for calculating the expectation values of the form
0
[e
i

Ht
(r
1
)... (r
n
)e
i

Ht
[
0
) (5.34)
one rst needs to normal order (r
1
)... (r
n
) using commutation relations (5.17) as
(r
1
)... (r
n
) =
mn
f
m
(r
1
, ..., r
n
)
(r
1
)...
(r
m
)(r
m
)...
(r
1
), (5.35)
and only then use substitutions (5.30)-(5.31) to evaluate matrix elements:
0
[e
i

Ht
(r
1
)... (r
n
)e
i

Ht
[
0
) =
mn
f
m
(r
1
, ..., r
n
)
0
[e
i

Ht

(r
1
)...
(r
m
)e
i

Ht
e
i

Ht

(r
m
)...
(r
1
)e
i

Ht
[
0
) =
mn
f
m
(r
1
, ..., r
n
)
0
[
_
a
1
u
1
(r
1
, t)e
iQ
1
r
1
+a
2
u
2
(r
1
, t)e
iQ
2
r
1
_
...
_
a
1
u
1
(r
1
, t)e
iQ
1
r
1
+a
2
u
2
(r
1
, t)e
iQ
2
r
1
_
[
0
). (5.36)
5.2.2 Interference of independent clouds
The surprising phenomenon which was observed in Ref. [15] is the appearance
of interference fringes when condensates are completely independent. To illustrate how
interference fringes appear for independent clouds, let us now discuss the case in which the
numbers of particles in each of the clouds, N
1
and N
2
, are xed, hence the phase dierence
between the two clouds is not well dened. Initial state in the Fock basis in this case is
given by (5.19):
[N
1
, N
2
) =
1
N
1
!N
2
!
(a
1
)
N
1
(a
2
)
N
2
[0). (5.37)
Using the formalism of second quantization explained earlier, one can evaluate
Q
) by
analogy to Eq. (5.29):
Q
) = N
1
, N
2
[a
1
a
2
[N
1
, N
2
) = 0. (5.38)
However,
Q
) = 0 doesnt imply that there are no interference eects for independent
condensates. Indeed,
Q
) gives only the statistical average over many experiments, accord-
ing to the usual interpretation of expectation values of operators in quantum mechanics.
Being a quantum operator,
Q
has non vanishing quantum uctuations. In each particular
realization of experiment, complex number
Q
can have a nonzero value. To show this, let
us evaluate [
Q
[
2
), which is the density-density correlation function at wave vector Q :
[
Q
[
2
) =
Q

Q
) =
_
drdr
(r)

(r)
(r
(r
)e
iQ(rr
)
) =
_
drdr
(r)
(r
(r)
(r
)e
iQ(rr
)
+
_
dr
(r)
(r)). (5.39)
These matrix elements can be evaluated using the second quantization prescription of the
previous section, and the result is
[
Q
[
2
) = N
1
, N
2
[a
1
a
2
a
1
a
2
+a
1
a
1
+a
2
a
2
[N
1
, N
2
) = N
1
N
2
+N
1
+N
2
. (5.40)
In the limit of large N
1
= N
2
= N/2, the leading contribution to [
Q
[
2
) is the same as
for the state with the xed phase. Information about the full distribution of the quantum
operator
Q
is contained in higher moments of the distribution. If one considers higher
moments of the type [
Q
[
2n
) =
n
Q
n
Q
), the leading contribution in the limit of large N
1
and N
2
will again have the form
[
Q
[
2n
) = (N
1
N
2
)
n
_
1 +O(
1
N
1
) +O(
1
N
2
)
_
. (5.41)
Corrections which appear because of the normal ordering result in subleading terms which
are denoted by O(1/N
1
) +O(1/N
2
). The leading term implies that in the limit of large N
1
and N
2
, the distribution function of [
Q
[
2
is highly peaked near the value N
1
N
2
, with the
relative width which is proportional to the inverse square root of number of particles. Any
operator of the form
n
Q
m
Q
will have zero expectation value for m ,= n similar to
Q
, which
means that the phase of the complex number
Q
is uniformly distributed from 0 to 2. The
expectation value of any operator which depends on the phase of
Q
becomes zero due to
the averaging over the phase.
The physical picture which emerges from the calculations is the following [26,
223, 224]: for two independent ideal clouds in the limit of large N the absolute value of
interference fringe amplitude is the same as for the state with a xed relative phase, but
the position of the intensity minima uctuates from shot to shot. The state with a xed
number of particles is a superposition of states with xed relative phases. For example,
[N/2, N/2) =
_
N
2
_
1/4
_
2
0
d
2
[, N). (5.42)
In the limit of large N the phase states are almost orthogonal, and the measurement picks
some value of the relative phase. Since the relative phase is not well dened for independent
clouds, in each particular experiment the positions of the minima will uctuate from shot to
shot. To distinguish independent clouds from states which have correlated relative phases,
one needs to do a series of experiments and measure not only the absolute magnitude of
interference fringes, but also the positions of the minima. Experiments which distinguish
states with a xed relative phase from some other many body states are already being done,
and can be used, for example, to measure the temperature [259] or to study the dynamical
evolution [206, 190, 191, 209, 210] of the relative phase.
5.3 Full counting statistics of shot noise
As has been explained in the previous section, for experiments with independent
clouds the average interference amplitude depends only on the number of particles per cloud.
In this section we consider not only the average interference amplitude, but also its shot to
shot uctuations due to a nite number of atoms in the clouds. We will demonstrate that
while the average value of [
Q
[
2
depends only on the number of particles per cloud, the full
distribution function of the variable
R = [
Q
[
2
(5.43)
contains information about the states of individual clouds. Our analysis is motivated by
the earlier work of Polkovnikov [255], who showed that the variance of the fringe amplitude
decreases as the inverse power of the number of particles per cloud, N, with a non-universal
coecient which contains information about the state of the clouds. Experimental observa-
tion of the eects discussed in this section requires systems with a small number of atoms.
This may be realized with micro-BECs on chips [206, 207, 190, 191, 209, 208, 210, 211].
The shot noise for nite N has a fundamental nature, which stems from the prob-
abilistic nature of quantum mechanics. Distributions of R obtained below correspond to
the following idealized experimental procedure: release the conning potential and take
an absorption image of the columnar density on an ideal CCD camera with 100% eciency
(photon shot noise is ignored). To obtain the amplitude of interference fringes,
Q
, extract a
Fourier component of the density at wave vector Q from each image separately. The results
of many experiments give the histogram W(R) of the values of R = [
Q
[
2
. We note that
the quantum observable

R dened in such way is a many-body operator, and calculation of
its full distribution is a non-trivial task, even when all atoms are in the same state, such as
for the case of a well dened relative phase between atoms in the two wells. In this section
we develop a general method to nd distribution functions of R analytically.
We note that in our idealized setting we nd interference patterns at a well dened
wave vector Q. We expect that the nite size of the systems in transverse direction after
expansion and collisions during the initial stage of expansion broaden the peak in the
Fourier space to a nite, but small range of wave vectors around Q [260]. Hence one should
consider R = [
Q
[
2
as an integral over the peak in the Fourier image of the density. Also in
this chapter we will discuss the amplitude of the interference fringes whereas experimental
papers typically discuss the visibility of the interference patterns. The two quantities dier
only by a trivial rescaling.
Results for independent clouds in coherent (solid) and number (dashed) states for
N
1
= N
2
= 100 are presented in Fig. 5.6. One can see, that even for a relatively large
number of atoms, N = 100, uctuations due to shot noise are appreciable . In Fig. 5.7 we
compare the full counting statistics of R = [
Q
[
2
for independent clouds in coherent (dashed)
and number (dotted) states with N
1
= N
2
= 20, and for clouds with a xed relative phase
(solid) with total number of atoms N = N
1
+N
2
= 40. Distribution functions for the cases
of (i) well dened relative phase between the clouds and (ii) xed number of atoms in each
cloud are very close. They become indistinguishable when R is rescaled by its average
value R), although each of them diers considerably from the Gaussian distribution. The
distribution function of R is wider for coherent states compared to number states, as was
suggested in Ref. [255] based on the study of the variance of the two distributions. Hence
the conclusion is that coherent and number states can be easily distinguished based on the
statistics of uctuations of R relative to its average value.
In principle, when the two clouds are prepared with the same relative phase over
many experiments, it is possible to distinguish independent condensates in number states
from states with a xed relative phase using a set of several interference experiments: min-
ima positions are uniformly distributed for independent clouds, while for states with a xed
relative phase, positions of the interference minima are always at the same points in space.
However one can imagine the situation when the clouds are prepared in state with a xed
relative phase but the relative phase itself is random from realization to realization. Our
results show that it is practically impossible to distinguish such states from number states
0.5 1 1.5 2
RR
0.5
1
1.5
2
2.5
WRR
Figure 5.6: Rescaled distribution functions of R = [
Q
[
2
for independent clouds in the
coherent states (solid) or in the states with a well dened numbers of atoms (dashed). Here
N
1
= N
2
= 100.
by looking at the distribution of the amplitude of interference fringes. Our calculations be-
low provide additional support to the physical interpretation [26, 223, 225, 224] presented
in the previous section. Fluctuations of the absolute value of the interference amplitude are
the same when clouds have random relative phase and when clouds are prepared in number
states so the random relative phase is measurement induced.
The method developed in this section can be generalized to a variety of experi-
mental situations, i.e. several independent condensates. Dierent squeezed states within
individual condensates can be considered, and measurement of full counting statistics of
shot noise can be used as an experimental probe to distinguish between dierent correlated
states.
As noted earlier, we will be interested in the full distribution function of the
positive denite quantum observable
R = [
Q
[
2
=
Q

Q
, (5.44)
dened by Eq. (5.39). To calculate its full distribution function, W(R), one needs to
know expressions for higher moments
R
n
). After that, one has to solve the problem of
moments, i.e. to recover the distribution function on the (0, ) interval using all moments.
In general, this procedure is numerically hard and unstable, unless higher moments have a
certain analytical form. If the expression for higher moments is known analytically, then one
can sometimes avoid the problem of moments by calculating the so-called characteristic
function, (), which is the Laplace transform of W(R) :
() =
_

0
e
R
W(R)dR =
_

0
i=0
(R)
n
n!
W(R)dR =
i=0
()
n
R
n
)
n!
(5.45)
If () can be calculated analytically, then W(R) can be recovered by the inverse Laplace
transform. In our case it is more practical to calculate not the characteristic function, but
200 400 600 800 1000 1200
R
0.0005
0.001
0.0015
0.002
0.0025
0.003
WR
Figure 5.7: Distribution functions of R = [
Q
[
2
for independent clouds in the coherent
(dashed) and in the number (dotted) states with N
1
= N
2
= 20. The solid line is a distribu-
tion function of R for clouds with a xed relative phase with total number N = N
1
+N
2
= 40.
Distribution functions for states with a xed relative phase and with xed numbers are very
close, and become indistinguishable when R is normalized by its average value R).
the analog of the Hankel transformation [261, 262] of W(R), given by
Z(i) =
n=
n=0
(i)
2n
(n!)
2

R
n
). (5.46)
Using the expansion of the zeroth order Bessel function, one can write
Z(i) =
_

0
W(R)J
0
(2
R)dR. (5.47)
(5.48)
The inversion of the transformation can be found using the orthogonality condition for the
zeroth order Bessel functions
_
0
J
0
(x)J
0
(y)[x[d = ([x[ [y[), which gives
W(R) = 2
_

0
Z(i)J
0
(2
R)d. (5.49)
By the end of this section we will provide analytical expressions for Z(i) for certain cases
(see Eqs. (5.61), (5.68), (5.71)), from which W(R) can be obtained by simple numerical
integration according to Eq. (5.49).
To proceed we note that
Q
and
Q
commute with each other:
[
Q
,
Q
] = 0. (5.50)
Operators
Q
are understood as in Eq. (5.27), without the projection on single particle
states
1
(r, t),
2
(r, t) as in Eqs. (5.30)-(5.31). The latter substitutions can be only done
after normal ordering (see discussion in section 5.2.1). Hence we nd
Z(i) =
_
2
0
d
2
e
i(
Q
e
i
+
Q
e
i
)
). (5.51)
Indeed, after expanding the exponent and integration, only even degrees of i survive, and
non vanishing terms are exactly what is needed for Eq. (5.46).
The normal ordering of Eq. (5.51) can be done using the following identity:
e
R
f(r)

(r)

(r)dr
=: e
R
(e
f(r)
1)

(r)

(r)dr
: . (5.52)
Here we have assumed that operators
(r), (r), have the canonical commutation relations

given by Eq.(5.17). Eq. (5.52) is a generalization of the simpler identity [263]:
e
a
a
=: e
(e
1)a
a
: (5.53)
for operators which obey the commutation relations
[a, a
] = 1. (5.54)
The normal ordering signs : : mean that all creation operators should be put to the left of
annihilation operators in Taylor expansion of expressions being ordered. To illustrate the
meaning of Eq. (5.53), let us consider the expansions of left and right side up to
2
. The
left hand side is
1 +a
a +

2
2
a
aa
a +O(
3
) = 1 + ( +

2
2
)a
a +

2
2
a
aa +O(
3
). (5.55)
The right hand side is
1 + (e
1) : a
a : +
(e
1)
2
2
: a
aa
a : +O(
3
) =
1 + ( +

2
2
+O(
3
)) : a
a : +
2
+O(
3
)
2
: a
aa
a : +O(
3
) =
1 + ( +

2
2
)a
a +

2
2
a
aa +O(
3
).
Eq. (5.53) holds not only up to
2
, but to all orders in and plays an important role in
quantum optics.
Using the denition of
Q
given by Eq. (5.27), one can apply Eq. (5.52) with
f(r) = 2icos (Qr +) and rewrite Z(i) in Eq. (5.51) as
Z(i) = :
_
2
0
d
2
e
R
(e
2i cos (Qr+)
1)

(r)

(r)dr
:). (5.56)
5.3.1 Interference of two independent coherent condensates
Let us rst explain how to evaluate Z(i) for independent clouds in coherent states
[170, 92] of operators a
1
and a
2
with eigenvalues
N
1
e
i
1
and
N
2
e
i
2
. Since coherent
states form a complete basis, any initial state can be expanded in this basis, and thus
the problem of calculating W(R) is essentially solved for arbitrary initial states. Coherent
states are convenient, since they are the eigen states of the annihilation operator, and the
annihilation operator acts on them as a cnumber. Hence after making substitutions (5.30)-
(5.31) into the normal ordered expression, one can substitute operators a
i
, a
i
by numbers
N
i
e
i
i
,
N
i
e
i
i
. Since the normal ordered expression is obtained by the normal ordering
of the Taylor expansion of Eq. (5.56), one needs to collect the Taylor series back. For
coherent states, the whole procedure is equivalent to removing the normal ordering signs
in Eq. (5.56), making substitutions (5.30)-(5.31) and treating operators a
i
, a
i
as numbers
N
i
e
i
i
,
N
i
e
i
i
. Thus we obtain
Z(i;
N
1
e
i
1
,
N
2
e
i
2
) =
_
2
0
d
2
e
R
(e
2icos (Qr+)
1)(N
1
u
1
(r,t)
2
+N
2
u
2
(r,t)
2
+2
N
1
N
2
cos (
1
2
+Qr)u
1
(r,t)u
2
(r,t))dr
. (5.57)
Similar to section 5.2.1 we assume the that normalized functions u
1
(r, t) and u
2
(r, t) strongly
overlap and vary at scales much larger than 1/Q, which is equivalent to
_
e
inQr
u
(r, t)u
(r, t)dr =
n0
. (5.58)
Then integration over dr in the exponent of Eq. (5.57) can be done using the following
equations:
_
_
e
2icos (Qr+)
1
_
(N
1
u
1
(r, t)
2
+N
2
u
1
(r, t)
2
)dr =
m=
m=1
(i)
2m
(2m)!
_
(N
1
u
1
(r, t)
2
+N
2
u
1
(r, t)
2
)(e
i(Qr+)
+e
i(Qr+)
)
2m
dr =
m=
m=1
(N
1
+N
2
)
(i)
2m
(2m)!
(2m)!
m!m!
= (J
0
(2) 1)(N
1
+N
2
); (5.59)
2
_
N
1
N
2
_
_
e
2icos (Qr+)
1
_
cos (
1
2
+Qr)u
1
(r, t)u
2
(r, t)dr =
_
N
1
N
2
m=
m=0
(i)
2m+1
(2m + 1)!
(2m+ 1)!
m!(m+ 1)!
_
e
i(
1
2
)
+e
i(
1
2
)
_
=
2i
_
N
1
N
2
J
1
(2) cos (
1
2
). (5.60)
Substituting Eq. (5.59) and Eq. (5.60) into Eq. (5.57), and integrating over , we
nally obtain the central result of this section:
Z(i;
N
1
e
i
1
,
N
2
e
i
2
) =
_
2
0
d
2
e
(J
0
(2)1)(N
1
+N
2
)+2i
N
1
N
2
J
1
(2) cos (
1
2
)
=
e
(J
0
(2)1)(N
1
+N
2
)
J
0
_
2
_
N
1
N
2
J
1
(2)
_
.(5.61)
5.3.2 Interference of independent clouds in number states
Let us now explain how to calculate Z
f
(i, N
1
, N
2
) for the Fock states with the
number of particles equal to N
1
and N
2
. First, we need to expand the Fock states [N
1
, N
2
)
using the coherent states basis. Since the basis of coherent states is overcomplete [92], there
are many ways to do a decomposition. For our purposes it is convenient to use
[N
1
, N
2
) =
(a
1
)
N
1
N
1
!
(a
2
)
N
2
N
2
!
[0) =
_
N
1
!N
2
!
N
1
N
2
e
_
2
0
_
2
0
d
1R
d
2R
(2)
2
e
iN
1
1R
iN
2
2R
[e
i
1R
, e
i
2R
), (5.62)
where is an arbitrary real positive number. Coherent states are given by
[e
i
1R
, e
i
2R
) = e
2
+e
i
1Ra
1
+e
i
2Ra
2
[0), (5.63)
and the overlap between them equals
e
i
1L
, e
i
2L
[e
i
1R
, e
i
2R
) = e
2
(e
i
1R
i
1L+e
i
2R
i
2L2)
. (5.64)
One can also expand the bra state N
1
, N
2
[ similarly to (5.62) by introducing integration
variables
1L
,
2L
. For any given values of ,
1L
,
2L
,
1R
, and
2R
, matrix elements be-
tween coherent states can be evaluated as in previous section, and by simple modication
of Eq. (5.61) we obtain
Z
f
(i, N
1
, N
2
) =
_ _ _ _ _
dd
1L
d
2L
d
1R
d
2R
(2)
5
N
1
!N
2
!
2N
1
2N
2
e
2
2
e
iN
1
(
1L
1R
)+iN
2
(
2L
2R
)
e
i
1L
, e
i
2L
[e
i
1R
, e
i
2R
)
e
(J
0
(2)1)
2
(e
i
1R
i
1L+e
i
2R
i
2L)
e
i
2
J
1
(2)(e
i(
1L
+
2R
)
+e
i(
1R
+
2L
)
)
. (5.65)
One can now substitute Eq.(5.64) into the equation above, introduce variables
1
=
1L
1R
and
2
=
2L
2R
, (5.66)
and integrate over ,
1R
,
2R
. Multiple cancellations occur, and eventually we obtain
Z
f
(i, N
1
, N
2
) =
_ _
d
1
d
2
(2)
2
N
1
!N
2
!
2N
1
2N
2
e
iN
1
1
+iN
2
e
J
0
(2)
2
(e
i
1 +e
i
2)
J
0
_
2
2
J
1
(2)e
i(
1
+
2
)/2
_
. (5.67)
Both integrations in the equation above can be done in a closed form for arbitrary positive
integer N
1
and N
2
using hypergeometric functions. Here we will present the results only
for N
1
= N
2
= N. One needs to expand the last exponent and J
0
(2
2
J
1
(2)e
i(
1
+
2
)/2
)
using Taylor series. After integration over d
1
and d
2
dependence on disappears, and
we obtain the nal result for Fock states:
Z
f
(i, N, N) =
N
k=0
(N!)
2
J
0
(2)
2k
k!
2
(iJ
1
(2))
2(Nk)
(N k)!
2
=
2
F
1
_
n, n; 1;
J
0
(2)
2
J
1
(2)
2
_
(iJ
1
(2))
2N
, (5.68)
where
2
F
1
(a, b; c; x) in a hypergeometric function dened by
2
F
1
(a, b; c; x) = 1 +
ab
c
x
1!
+
a(a + 1)b(b + 1)
c(c + 1)
x
2
2!
+... (5.69)
5.3.3 Clouds with a well dened relative phase
Let us now consider the case of clouds with a xed relative phase, when the initial
state [
0
, N) at t = 0 is given by Eq. (5.15). This state can be expanded using the coherent
states basis as
[
0
, N) =
1
(2
N
N!)
1/2
(a
1
e
i
0
/2
+a
2
e
i
0
/2
)
N
[0) =
N!(
2)
N
e
2
_
2
0
d
R
2
e
iN
R
[e
i
R
+i
0
/2
, e
i
R
i
0
/2
).
A similar expansion can be written for bra- vector
0
, N[ using the phase variable
L
.
Coherent states and their overlaps are given by Eqs. (5.63)-(5.64), and one obtains an
expression for the generating function Z(i, N) similar to Eq. (5.65):
Z(i, N) =
_ _ _
dd
L
d
R
(2)
3
N!(
2)
2N
e
2
2
e
iN(
L
R
)
e
i
L
+i
0
2
, e
i
L
i
0
2
[e
i
R
+i
0
2
, e
i
R
i
0
2
)
e
2(J
0
(2)1)
2
e
i
R
i
L
e
i
2
J
1
(2)(e
i(
0
+
R
L
)
+e
i(
0
+
L
R
)
)
.
The integrand depends only on the dierence =
L

R
, and the integration over
can be performed analytically. Dependence of Z(i, N) on
0
drops out as expected:
Z(i, N) =
_ _
d
2
N!
2N
2
N
e
iN
e
2J
0
(2)
2
e
i
J
0
_
2
2
J
1
(2)e
i
_
. (5.70)
Expanding the last exponent and J
0
_
2
2
J
1
(2)e
i
_
in the expression above and
integrating over , we obtain the nal expression for even N :
Z(i, N) = N!2
N
N/2
k=0
(iJ
1
(2))
2k
(k!)
2
(2J
0
(2))
N2k
(N 2k)!
=
2
N
(iJ
1
(2))
N
C
N
N
_
J
0
(2)
iJ
1
(2)
_
, (5.71)
where C
n
(x) is a Gegenbauer polynomial [262].
5.4 Interference of low-dimensional gases
As we discussed in previous sections, for macroscopic three dimensional Bose-
Einstein Condensates the long-range phase coherence manifests itself in the nearly perfect
interference fringes between two independent condensates [15]. For low dimensional Bose
gases, the situation is dierent, since phase uctuations are very eective in destroying
the long-range order. In one dimension, long-range coherence is prohibited even at zero
temperature [75], while in two dimensions any nonzero temperature destroys long-range
order [248]. In addition, the Berezinskii-Kosterlitz-Thouless (BKT) phase transition occurs
Figure 5.8: Experimental setup for interference of 2D gases. Note that the interference
patterns are straight at low temperatures indicating suppressed phase uctuations. Mean-
dering patterns at high temperatures come from strong phase uctuations. Reprinted with
permission from Macmillan Publishers Ltd: Nature ([193]), copyright (2006).
[243, 244]. It separates the low temperature phase with power-law correlations from the
high temperature phase with short-range correlations. Phase uctuations reduce the aver-
age visibility of the interference fringes, and result in the shot to shot uctuations of the
visibility.
In this section, we discuss how measurements of interference fringes can reveal
information about spatial correlations within individual condensates. The typical exper-
imental setups are shown in Figs. 5.4 and 5.8. They correspond to the so-called open
boundary conditions (OBC). Essentially the OBC mean that the imaged area is cut out of
a larger system. As a theoretical model one can also consider a one-dimensional condensate
with periodic boundary conditions (PBC), which corresponds to interference experiments
with two coaxial rings lying in two parallel xy planes. While this model is somewhat arti-
cial from the experimental point of view (see, however, Ref. [264]), it allows a very elegant
theoretical analysis, hence we will discuss it in this chapter as well.
The conning potential is highly anisotropic, and after it is switched o, the clouds
predominantly expand in the transverse direction, while no signicant expansion occurs in
the axial (for 1D gases) or in-plane (for 2D gases) directions. For low-dimensional gases
the phase of the condensate doesnt have a long-range order due to quantum or thermal
uctuations. Locally, the phase determines the positions of the minima of the absorption
intensity, and uctuations of the phase lead to uctuations of the interference fringe po-
sitions along the condensates, as shown in Fig. 5.4. Fluctuations of the fringe positions
contain information about the original phase uctuations present in the system, which are
preserved during expansion.
To extract information about fringe position uctuations for the 1D case, we will
integrate the intensity along the axes of the clouds. Fluctuations of the relative phase
result in uctuations of the minima positions for dierent x. For each y, the image can
be integrated along the x direction to give the integrated fringe amplitude
Q
(L) (see Fig.
5.4). Note that the integrated fringe amplitude depends on the integration length L. One
experimental image can be used to extract information for dierent values of L. Many
images are still required to obtain distribution functions for each L. For 2D gases, the setup
is analogous and is shown in Fig 5.8. Here, a part of the integration is performed by the
imaging beam itself. The size of the integration area along the direction of the imaging beam
can be controlled by applying magnetic eld gradients, so that only a specied section of
the cloud is resonant with the probe light.
The operator which corresponds to the fringe amplitude
Q
(L), illustrated in Fig.
5.4, is the same as
Q
dened by Eq. (5.27), where the integration along x dimension
is limited to the section of length L. Let us rst consider the expectation value [
Q
(L)[
2
)
(expectation values of operators which depend on the phase of
Q
(L) vanish, similar to 3D
case, since two clouds are assumed to be independent). One has to use modied formulas
(5.30)-(5.31), where operators a
i
, a
i
are now allowed to have xdependence. In the limit
when the number of particles in the section of size L is large, the average value of [
Q
(L)[
2
)
is given by [215]
[
Q
(L)[
2
) =
_
L
0
_
L
0
dx
1
dx
1
a
1
(x
1
)a
2
(x
1
)a
1
(x
1
)a
2
(x
1
)). (5.72)
Note that in Eq. (5.72) we used the normal ordered form of the operators which means that
we neglect the shot noise considered in section 5.3. This is justied for long condensates
as we discuss below. From now on, we will concentrate on the case in which independent
clouds are identical and have the same density of particles with equal interaction strengths.
Then
[
Q
(L)[
2
) =
_
L
0
_
L
0
dx
1
dx
1
a
(x
1
)a(x
1
))
2
. (5.73)
To gain intuition into the physical meaning of the average amplitude of interference
fringes, we address two limiting cases. First, consider the situation in which a
(x)a(0))
decays exponentially with distance and the correlation length is given by << L. Then
Eq. (5.73) implies that [
Q
(L)[
L, which has a simple physical interpretation. Since

the phase is only coherent over a length , the system is eectively equivalent to a series
of L/ pairs of independent condensates. Each pair contributes interference fringes with a
constant amplitude proportional to and a random phase. The total amplitude
Q
(L) is
the result of adding L/ independent vectors of constant length and random direction.
Adding random uncorrelated vectors gives a zero average except for a typical square root
uctuation. Thus scaling of the absolute value of the net interference amplitude is
L.
This observation is similar in spirit to that made of interference between 30 independent
condensates in a chain in Ref. [226]. Fringes can be seen, though their average amplitude is
suppressed by a factor of
30 compared to the interference between two condensates. Now

consider the opposite limit of perfect condensates, for which a
(x)a(0)) is constant. In this

case Eq. (5.73) implies that [
Q
(L)[ L. Pictorially this is the result of adding vectors
which are all aligned, resulting in a fringe amplitude, the absolute value of which scales as
the total size of the system.
The methods developed in this section for analyzing [
Q
(L)[
2
can be applied to
condensates with either uniform or non uniform densities. For simplicity, we concentrate
on the case in which L is much smaller than the size of the clouds, so the change in the
atomic density along the clouds can be ignored. In this case correlation functions for 1D
gases are described by the Luttinger liquid theory [265, 132]. For OBC at zero temperature
the two-point correlation functions are given by
a
(x)a(y)) (
h
/[x y[)
1/2K
. (5.74)
Here is the particle density,
h
is the healing length, which also serves as the short range
cuto, and K is the so-called Luttinger parameter, which characterizes the strength of
interactions. For bosons with a repulsive short-range potential, K ranges between 1 and ,
with K = 1 corresponding to strong interactions, or impenetrable bosons, while K
for weakly interacting bosons. Substituting Eq. (5.74) into Eq. (5.73) and assuming that
L
h
, we obtain [215]
[
Q
(L)[
2
) =

C
2
L
2
_
h
L
_
1/K
, (5.75)
where

C is a constant of order unity. We see that the amplitude of the interference fringes
[
Q
(L)[) scales as a non trivial power of the imaging length. In the non interacting limit
(K ) the scaling is linear [
Q
(L)[) L, as expected for a fully coherent system.
Interestingly, [
Q
(L)[)
L appears in the hard core limit (K = 1), as in systems with

short range correlations which were discussed above.
One may be concerned that Eq. (5.75) gives only the long distance asymptotic
behavior of the correlation functions, and does not describe the short distance behavior.
From Eq. (5.73) one nds that the contribution of the short distance part of the correlation
functions to [
Q
(L)[
2
) scales as L. In the physically relevant case of K > 1 and in the
limit of large L this contribution is smaller than Eq. (5.75). We note that in principle one
can use the exact Bethe ansatz solution of the Lieb-Liniger model [40] to obtain correlation
functions valid at all distances [266]. Another contribution which has been neglected is that
of shot noise. The shot noise contribution to [
Q
(L)[
2
) comes from the normal ordering of
operators, and equals
_
L
0
_
L
0
dx
1
dx
1
_
a
1
(x
1
)a
2
(x
1
)a
2
(x
1
)a
1
(x
1
)) a
1
(x
1
)a
2
(x
1
)a
1
(x
1
)a
2
(x
1
))
_
=
_
L
0
_
L
0
dx
1
dx
1
(x
1
x
1
)a
1
(x
1
)a
1
(x
1
)) = n
1D
L. (5.76)
In the limit of large L and K > 1 this is again a subleading contribution and can be
neglected.
For 2D, one can use similar approach to describe the contrast distribution at nite
temperature below the BKT transition. We note that we assume that the temperature is
small enough such that 2D gas is in a quasi-condensate regime [267, 268], when only phase
uctuations are present. In this case, correlation functions are given by [243, 244, 131]
a
(r)a(0))
_
h
r
_
(T)
, (5.77)
where (T) = mT/(2h
2
s
(T)) depends on the temperature and the superuid density
s
(T). The BKT transition takes place at the universal value
c
(T
c
) = 1/4. To keep con-
nection to the 1D case, we introduce
K = 1/(2(T)), (5.78)
and restrict our attention to K > K
c
= 2. For temperatures above the BKT transition, Eq.
(5.77) doesnt hold and correlations decay exponentially. This means that the integrated
interference amplitude will only increase as the square root of the integration area [215].
Fig. 5.8 illustrates experiments performed with 2D gases to identify the BKT
transition [193]. Two independent 2D condensates are conned in transverse directions
using an optical lattice potential. After the optical potential is switched o and clouds
expand, the density is imaged on a CCD camera. When temperature is small, interference
fringes are straight lines. As the temperature is increased, the fringes start to meander
due to spatial uctuations of the phase. Integrating the image along the section of variable
length L in x direction gives L dependent fringe amplitude [
Q
(L)[). Scaling of this
amplitude with L contains information about (T), which is expected to have a universal
value
c
(T
c
) = 1/4 at the BKT transition.
Fig. 5.9 illustrates the procedure used to extract scaling exponents in experiments.
[
Q
(L)[
2
) (denoted as
C
2
) in Fig. 5.9A) is plotted as a function of L, and its scaling with
L in a certain range (see Ref. [193] for more details) is used to extract the exponent (T)
(denoted as in Fig. 5.9B). The average central contrast c
0
serves as a thermometer, such
that smaller values of c
0
correspond to higher temperatures. Above the BKT temperature
the value of (T) extracted from interference experiments is expected [215] to be equal to
0.5, while at the transition point it is equal to
c
(T
c
) = 1/4. Fig. 5.9B shows a sudden
change of the exponent in a relatively narrow range of temperatures. This change is remi-
niscent of the universal jump in the superuid density for 2D helium lms [269]. Of course,
since experiments are done with nite systems and imaging done along the ydirection per-
forms averaging over the inhomogeneous density prole, one shouldnt expect a universal
jump, but rather a crossover. The presence of the trap also aects parameters of the BKT
transition [270, 271].
The BKT transition is an example of a topological transition which is driven by
the unbinding of vortices, and the remarkable feature of experiments in Ref. [193] is the
ability to independently resolve the vortices. When only one of the condensates has a
vortex, the interference pattern will have a disclination-like structure [192]. It was shown
experimentally in Ref. [193] that proliferation of vortices occurs at the same point in the
parameter space as the jump in the scaling exponent. In the following sections we will
show that in uniform systems not only the scaling exponent, but also the full distribution
function of the fringe contrast has a universal form at the BKT transition.
5.4.1 Interference amplitudes: from high moments to full distribution
functions
Measuring atom density to obtain an interference pattern is a classical measure-
ment on a quantum mechanical wave function. The process of the measurement itself
introduces an intrinsic quantum mechanical noise. In other words, from shot to shot we will
not get precisely the same value of the amplitude of interference fringes. Expressions for
[
Q
(L)[
2
) which we derived in Eqs. (5.72)-(5.75) correspond to averaging over many shots.
For example, data points from Fig. 5.9A correspond to averaging over approximately one
hundred measurements [194]. However each individual shot will give the value of [
Q
(L)[
2
which may be dierent from its average value.
An interesting question to consider is how this amplitude uctuates from one
experimental run to another. To address this question we need to consider higher moments
of the operator [
Q
(L)[
2
. Generalizing the argument which lead to Eq. (5.73), we obtain
[
Q
(L)[
2n
) =
_
L
0
...
_
L
0
dx
1
...dx
n
dx
1
...dx
n
[a
(x
1
)...a
(x
n
)a(x
1
)...a(x
n
))[
2
. (5.79)
In Eq. (5.79) we used a normal ordered correlation function similar to Eq. (5.73).
One can calculate [255] corrections due to normal ordering for higher moments of [
Q
(L)[
2
,
and show that in the limit of large L and K > 1 they can be neglected.
From Eq. (5.73) we observe that [
Q
(L)[
2
) contains information about two-point
correlation functions of individual clouds. Eq. (5.79) shows that higher moments of [
Q
(L)[
2
Figure 5.9: Emergence of quasi-long-range order in a 2D gas. a, Examples of average
integrated interference contrasts
C
2
) are shown for low (blue circles) and high (red squares)
temperatures; L
x
is the integration length along x direction. The lines are ts to the data
using the power-law functions 1/(L
x
)
2
. b, Exponent as a function of the central contrast
c
0
. Central contrast c
0
serves as a thermometer, such that smaller values of c
0
correspond
to higher temperatures. Dashed lines indicate theoretically expected values of above and
below the BKT transition in a uniform system. Reprinted with permission from Macmillan
Publishers Ltd: Nature ([193]), copyright (2006).
contain information about higher order correlation functions. The full distribution of the
uctuating variable [
Q
(L)[
2
contains information about all high order correlation functions.
In Luttinger liquid theory, uctuations of the phase are described by the Gaussian
action. For Gaussian actions, higher order correlation functions are simply related to two-
point correlation functions (see e.g. Ref. [131]):
a
(x
1
) . . . a
(x
n
)a(x
1
) . . . a(x
n
)) =
ij
a
(x
i
)a(x
j
))
i<j
a
(x
i
)a(x
j
))
i<j
a
(x
i
)a(x
j
))
. (5.80)
Using this formula together with Eq. (5.74), the higher moments of fringe amplitudes can
be written as
[
Q
(L)[
2n
) = A
2n
0
Z
2n
, where A
0
=
_
C
2
1/K
h
L
21/K
, (5.81)
C is a constant of the order of unity and for OBC in 1D
Z
2n
=
_
1
0
...
_
1
0
du
1
...dv
n
i<j
[u
i
u
j
[
i<j
[v
i
v
j
[
ij
[u
i
v
j
[
1
K
=
_
1
0
...
_
1
0
du
1
...dv
n
e
1
K
P
i<j
G(u
i
,u
j
)+
P
i<j
G(v
i
,v
j
)
P
ij
G(u
i
,v
j
)
!
. (5.82)
Here for OBC G(x, y) is given by
G(x, y) = log [x y[. (5.83)
Integrals similar to Eq. (5.82) appeared in literature before [272, 273], but they are not
easy to compute.
The Gaussian model possesses a powerful conformal symmetry [132, 131], which
dictates the form of the correlation functions for periodic boundary conditions or nonzero
temperatures. For PBC with circumference of the condensates equal to the imaging length,
the change in the correlation functions leads to [132]
G
per
(x, y) = log
1
sin [x y[. (5.84)

For nonzero temperature, Z
2n
depends on K and the thermal length
T
= hv
s
/(k
B
T), where
v
s
is the sound velocity:
G(x, y,

T
L
) = log
_
T
L
sinh
[x y[L
T
_
. (5.85)
The analysis in this section doesnt depend on the particular form of G(x, y). The only
general restriction is
G(x, y) = G(y, x). (5.86)
Eq. (5.82) is also valid for 2D case below the BKT transition temperature, with
u
i
and v
i
being 2D variables on a rectangle with G(x, y) = log [x y[, and with a properly
redened A
0
.
In what follows, we will be interested in distribution functions W() of the variable
= [
Q
(L)[
2
/A
2
0
and of its normalized version = [
Q
(L)[
2
/
Q
(L)
2
). By the denition of
higher moments
Z
2n
=
_

0
W()
n
d, and
Z
2n
Z
n
2
=
_

0
W( )
n
d . (5.87)
The general problem we consider now is how to construct distribution functions
W() given by Eq. (5.87) with Z
2n
given by Eq. (5.82). G(x, y) can be an arbitrary
symmetric function of x and y, and not necessarily the function of their dierence. This
allows us to study systems with non-uniform density in external traps, where one needs
to use modied correlation functions [268]. In section 5.4.2 we will show that W() is
connected to the partition functions of various Sine-Gordon models and Coulomb gases,
and methods developed in this section provide a new non perturbative tool for calculating
partition functions of such models.
One can think of Z
2n
in Eq. (5.82) as of a partition function of a classical two-
component gas of ctitious charged particles in a microcanonical ensemble, with K being
the temperature. We rst briey comment on the limiting cases K 1, and K 1.
For K 1, the expansion of Z
2n
gives Z
2n
1 in the zeroth order. This means that
W( ) ( 1), i.e. we have a very narrow function peaked near its average value.
Higher order terms in the expansion give

W( ) a small width of the order of 1/K, and
are studied in detail in Appendix C. For 1D gases at zero temperature with K 1
or nonzero temperatures with
T
K/L 1 the distribution function

W( ) is Poissonian
irrespective of OBC or PBC:

W( ) = e

. To demonstrate this we need to verify that
Z
2n
/Z
n
2
=
_
0

n
e

d = n!. This can be shown using the classical gas analogy: as
K 1, Z
2
=
_
1/[x y[
1/K
dxdy starts to diverge for x y, and the main contribution
to Z
2n
comes from molecular states of the two-component gas, i.e. from the parts of the
conguration space in which each particle has a particle of the opposite charge in its
neighborhood; n! is just the number of ways to form such pairs. In this language K 1
corresponds to the plasma phase of the classical charged gas, and the evolution of the
distribution function can be understood as a formation of molecules out of the plasma
phase as the temperature (i.e. the Luttinger parameter K) is lowered. For nite
T
, the
main contribution to Z
2n
comes from distances
T
K/L, and if this parameter is much
smaller than 1, then again molecular contributions dominate.
There is a simple physical interpretation of W() for both K 1 and K 1.
When bosons do not interact and K = , there should be no phase uctuations within
individual condensates. Hence in each experiment we should nd a perfect interference pat-
tern, although positions of the density minima are unpredictable (see discussion in section
5.2). Alternatively, in the regime of strong interactions when K 1 we can think of the
net interference as a result of adding many random uncorrelated two dimensional vectors
(see discussion after Eq. (5.73)). Earlier we used the fact that for a random walk, the net
displacement is proportional to the square root of the number of steps. But we also know
that for 2D random walks, the distribution function of the square of the net displacement
is Poissonian, which is what we nd for

W( ).
As K varies from 1 to , the distribution function

W( ) should evolve from
being a very broad Poissonian function to a narrow delta function. The evolution of the
distribution in the intermediate regimes will be studied in detail below. In Appendix C
we develop a systematic expansion of Z
2n
in powers of 1/K for large K, which works for
dierent dimensions and boundary conditions, and investigate the distribution functions
in this limit. The rest of this section is organized as follows. In section 5.4.2 we discuss
the connection of our problem to the Sine-Gordon models, which describe various physical
problems, ranging from ux line in superconductors [274] to string theory [275]. In section
5.4.3 we present a novel non perturbative solution which is applicable for any value of K
and for various dimensions and boundary conditions. We discuss the connection of the
distribution functions of fringe visibilities to the statistics of random surfaces [218], and
prove that for 1D case with periodic boundary conditions in the limit of large K, the
distribution of fringe visibilities is given by the Gumbel distribution, one of the extreme
value statistical distributions [276].
5.4.2 Connection of the fringe visibility distribution functions to the par-
tition functions of Sine-Gordon models
To illustrate the connection of the fringe visibility distribution functions to the
partition functions of Sine-Gordon models, we start from a 1D system with periodic bound-
ary conditions. As shown earlier, for a one-dimensional ring condensate with PBC and
circumference equal to the imaging length, the correlation functions lead to
G
per
(x, y) = log [
1
sin((x y))[. (5.88)

This function describes a two-component gas of particles interacting via a 1D Coulomb
potential, which is logarithmic in the interparticle separation. For PBC the distance is a
chord function sin([x y[)/. It is common to dene the grand partition function of such
Coulomb gas as
Z
per
(g) =
n=0
g
2n
(n!)
2
Z
per
2n
, (5.89)
where Z
per
2n
is given by Eq. (5.82) with G(x, y) substituted by G
per
(x, y).
Several dierent physical problems can be related to the partition function given
by Eq. (5.89). Here we mention only a few examples: a) the anisotropic Kondo model [277];
b) the quantum impurity model in one dimensional interacting electron systems introduced
in Ref. [256], which has been extensively studied in the context of edge states in Quantum
Hall systems [278]; c) the background-independent string theory and the model of strings
attached to a D-brane originally introduced in Ref. [275] (see Ref. [279] for a recent
review); d) Calogero-Sutherland model [280], which has numerous application as an eective
model; e) ux line pinning in superconductors [274]; f) quantum tunneling in the presence
of dissipation within Caldeira-Leggett model [281]; g) interference of two one-dimensional
condensates [215, 216, 218].
We now briey describe how the expression given by Eq. (5.89) appears as a
partition function in physical systems. We consider the imaginary time action
S
per
(g) =
1
2
_

dx
_
1
0
d[(
)
2
+ (
x
)
2
] + 2g
_
1
0
d cos[(x = 0, )] (5.90)
known as the boundary Sine-Gordon model [282]. The quantum eld (x, ) in Eq. (5.90)
is dened on an innite line in x direction and is assumed to be periodic along direction:
(x, ) (x, + 1). The interaction term is present only at x = 0. A typical physical
system which is described by action (5.90) is an interacting 1D electron liquid scattered by
an impurity [256]. In this case the cos-term describes backscattering of electrons within the
Luttinger liquid formalism.
The partition function of Eq. (5.90) is dened as [272]
Z
per
(g) =
_
Te
Sper(g)
_
Te
Sper(0)
. (5.91)
To show that Eqs. (5.89) and (5.91) dene the same Z
per
(g), one can expand Z
per
(g) in
Eq. (5.91) in Taylor series of the coupling g. Non vanishing contributions come from com-
binations which have equal number of exp(+i) and exp(i) terms. This is essentially
a charge neutrality condition for the Coulomb gas. If we identify
2
2
=
1
K
, (5.92)
and use the expression for the multi-point correlation functions of the Gaussian model (see
e.g. Ref. [131]), we obtain the expression given by Eq. (5.89).
One can relate Z
per
(g) to the solution of the problem of moments given by Eq.
(5.87):
Z
per
(g) =
n=0
g
2n
(n!)
2
Z
per
2n
=
n=0
g
2n
(n!)
2
n
) =
n=0
g
2n
(n!)
2
_

0
W()
n
d =
_

0
W() I
0
(2g
) d. (5.93)
Z
per
(g) is essentially a Hankel transformation [261] of W(), and inverting Eq. (5.93) we can
express the probability W() through the partition function Z
per
(g). Noting that I
0
(ix) =
J
0
(x) and using the completeness relation for Bessel functions,
_
0
J
0
(x)J
0
(y)[x[d =
([x[ [y[), we obtain
W() = 2
_

0
Z
per
(ig)J
0
(2g
)gdg. (5.94)
It is important that the last equation has the partition function at imaginary value of the
coupling constant. This should be understood as an analytic continuation of Z
per
(g). For
PBC in 1D, partition functions Z
per
(g) can be evaluated using the exact solution of the
boundary Sine-Gordon model with periodic boundary conditions, and this approach has
been used in Ref. [216] to calculate distribution functions.
For open boundary conditions one can also write the grand canonical partition
function as
Z(g) =
n=0
g
2n
(n!)
2
Z
2n
, (5.95)
where Z
2n
are given by Eqs. (5.82),(5.83). We can express Eq. (5.95) as a partition function
of a certain Sine-Gordon model:
Z(g) =
_
Te
S(g)
_
Te
S(0)
, (5.96)
where
S(g) = K
_

dx
_

d[(
)
2
+ (
x
)
2
] + 2g
_
1
0
d cos[2(x = 0, )]. (5.97)
Note dierent limits of the integration in the rst and the second terms. One can see that
Eqs. (5.95) and (5.96) dene the same Z(g) by observing that log ([x y[) /(2) is a free
propagator of the Gaussian action on a plane.
So far we established the relation between the distribution functions of the ampli-
tude of interference fringes of 1D condensates and the boundary Sine-Gordon models (5.90),
(5.97). Generalization of this argument to the case of 2D condensates is straightforward.
Higher moments of the interference amplitude are given by integrals of type (5.82), but each
u and v is now a two-dimensional coordinate. The distribution function of fringe amplitudes
is then related to the partition function of the bulk Sine-Gordon model, given by
S
2D
(g) = K
_

dx
_

d[(
)
2
+ (
x
)
2
] + 2g
_
1
0
_
1
0
ddxcos[2(x, )]. (5.98)
Expanding the partition function corresponding to action (5.98) in powers of g, one nds
the expression identical to Eq. (5.93), but with W() corresponding to interference of 2D
condensates.
When we describe interference of systems with OBC, we use Eqs. (5.97) and (5.98),
in which the eld (x, ) is dened on a whole plane, and is not periodic in . In both these
cases the interaction is present only in some part of the system, so translational invariance
is lost in both x and dimensions. Thus to calculate W() using Eq. (5.94) one needs to
calculate partition functions of inhomogeneous Sine-Gordon models. Exact solution of Sine-
Gordon model is available only for periodic boundary conditions, so to treat open boundary
conditions one needs to develop alternative methods. Conversely, if one has a solution for
W(), this provides a tool for calculating partition functions of inhomogeneous Sine-Gordon
models and Coulomb gases using Eq. (5.93). In section 5.4.3 we present a novel mapping of
W() for arbitrary G(x, y) to the statistical properties of random surfaces, which provides
a new tool for calculating partition functions of a wide class of Sine-Gordon models and
Coulomb gases. In general, partition functions with inhomogeneous g(x, ) can be evaluated
using this mapping as well. In this case the function g(x, ) will appear in the integrand of
Eq. (5.98). We point out that the mapping which we use is not related to the existence of
the exact solution of Sine-Gordon models, but relies only on the structure of the correlation
functions in the absence of interactions. We also note that a suitable extension of our method
can be used to compute correlation functions of Coulomb-gas models in equilibrium and
non-equilibrium situations.
5.4.3 Non perturbative solution for the general case
In this section we present a novel approach for calculating W(), which is based on
a mapping to the statistics of random surfaces [218]. We use this method to evaluate W()
numerically for a variety of situations. We point out that our method is not related to the
existence of the exact solution of Sine-Gordon models, but relies only on the structure of the
multi-point correlation functions in the absence of interactions. As has been discussed in
section 5.4.2, W() is connected to the partition functions of Sine-Gordon models and the
partition functions of Coulomb gases. Our mapping provides a new non perturbative tool
for calculating partition functions of such models. We expect that there should be numerous
applications of our new method to other physical problems which can be related to Sine-
Gordon and Coulomb gas models. As a concrete application of our approach we prove
analytically that for periodic boundary conditions in the limit of large K the distribution
of fringe visibilities is given by the Gumbel distribution, one of the extreme value statistical
distributions [276]. In this case the distribution function of the normalized fringe amplitudes
equals
W( ) = Ke
K( 1)e
K( 1)
, (5.99)
where 0.577 is the Euler gamma-constant. The Gumbel distribution frequently appears
in various problems (see e.g. books [276] for historic introduction), including number theory
[283], 1/f noise [257], Kardar-Parisi-Zhang growth [284], free Bose gas [285], etc. Probably
the whole distribution function in our interference context can in general be characterized
as a certain extreme-value statistics.
Mapping to the statistics of random surfaces
We start by observing that G(x, y) is real and symmetric. Hence it can be diago-
nalized on (0, 1) by solving the eigenvalue equations
_
1
0
G(x, y)
f
(y)dy = G(f)
f
(x). (5.100)
Here f is an integer index, which goes from 1 to .
f
(x) can be chosen to be real and
normalized according to
_
1
0
f
(x)
k
(x)dx = (f, k). (5.101)
Then, G(x, y) is given by
G(x, y) =
f=
f=1
G(f)
f
(x)
f
(y). (5.102)
Decomposition given by Eqs. (5.100)-(5.102) is similar to diagonalization of a symmetric
matrix by nding its eigenvectors and eigenvalues. Now we can write Z
2n
from Eq. (5.82)
as
Z
2n
=
_
1
0
...
_
1
0
du
1
...du
n
dv
1
...dv
n
e
1
K
(
P
i<j
G(u
i
,u
j
)+
P
i<j
G(v
i
,v
j
)
P
ij
G(u
i
,v
j
))
=
_
1
0
...
_
1
0
du
1
...du
n
dv
1
...dv
n
e
P
f
G(f)
K
"
(
P
i

f
(u
i
)
)
2
+
(
P
i

f
(v
i
)
)
2
P
i

f
(u
i
)
2
P
i

f
(v
i
)
2
2
(
P
i

f
(u
i
))(
P
i

f
(v
i
))
#
=
_
1
0
...
_
1
0
du
1
...dv
n
e
P
f
G(f)
2K
h
(
P
i=n
i=1

f
(u
i
)
f
(v
i
))
2
P
i=n
i=1
(
f
(u
i
)
2
+
f
(v
i
)
2
)
i
. (5.103)
Square of the rst isum in the last line above can be decoupled by introducing
Hubbard-Stratonovich integrations over auxiliary variables t
f
:
Z
2n
=
_
...
_
du
1
...dv
n
f=
f=1
_
dt
f
e
t
2
f
2
e
P
i
t
f
q
G(f)
K
(
f
(u
i
)
f
(v
i
))
G(f)
2K
(
f
(u
i
)
2
+
f
(v
i
)
2
)
2
.
Now we can simply integrate over du
1
, ..., dv
n
, since all integrals over uvariables
are the same (integrals over v variables are also identical):
Z
2n
=
_
_
_
f=
f=1
_
t
2
f
2
dt
f
2
_
_
_g(t
f
)
n
g(t
f
)
n
, (5.104)
where
g(t
f
) =
_
1
0
dx e
P
f
t
f
q
G(f)
K

f
(x)
G(f)
2K

f
(x)
2
. (5.105)
If all eigenvalues G(f) are negative, then
g(t
f
) = g(t
f
)
, g(t
f
)g(t
f
) = [g(t
f
)[
2
.
From comparison of Eqs. (5.87) and (5.104) we obtain the central result of this section
W() =
f=
f=1
_
t
2
f
2
dt
f
_
[g(t
f
)[
2
. (5.106)
Eq. (5.106) can be used to simulate distributions W() using Monte-Carlo ap-
proach. First, one needs to solve integral Eqs. (5.100)-(5.102) numerically to obtain eigen
functions and eigen vectors. Then one needs to choose random numbers t
f
from the
Gaussian ensemble, and plot the histogram of the results for [g(t
f
)[
2
. In what follows we
perform simulations of W() with up to N = 10
6
10
7
realizations of t
f
and smooth the
W( )
W( )
1 2 3 4
0.2
0.4
0.6
0.8
1
1.2
1.4
0.5 1 1.5 2
0.25
0.5
0.75
1
1.25
1.5
Figure 5.10: Distribution functions of the normalized interference amplitude

W( ) at
T = 0 for 1D gases with open boundary conditions, shown for Luttinger parameters
K = 2 (dashed), K = 3 (dotted), and K = 5 (solid). The inset shows a comparison between
open (solid) and periodic (dashed) boundary conditions for K = 5. The gure is taken from
Ref. [218].
data. We use a nite value of f
max
and check for convergence with f
max
, typically 30.
The average ) is always kept within 1% from its expected value. For most of the presented
results, all eigenvalues G(f) are negative, and Eq. (5.106) can be directly applied. Special
care should be taken of 1D case with nonzero temperature, where one of the eigenvalues can
be positive. This situation can be handled by subtracting suciently large positive constant
C from G(x, y,

T
L
), which makes all eigenvalues negative. According to Eqs. (5.82) and
(5.87), this leads to rescaling of by a factor e
C/K
, which can be easily taken into account.
In Fig. 5.10 we show scaled distribution of contrast

W( ) at T = 0 for 1D gases
with OBC for various K. The inset shows a comparison between OBC and PBC for K = 5.
In Fig. 5.11 we show the scaled distribution of contrast for 1D gas with OBC at nonzero
temperature and K = 5. As has been discussed earlier, for
T
K/L 1 the distribution is
Poissonian and wide, while for K 1 and
T
K/L 1 it is very narrow. The evolution
of the full distribution function of the visibilities as L is varied can be used to precisely
measure the thermal length,
T
= hv
s
/(k
B
T), and to extract the temperature. As seen in
Fig. 5.11, at T ,= 0 the distribution function has characteristic features, i.e. it is generally
non-symmetric and can have a minimum. These features can be used to distinguish the
noise due to uctuations of the phase from technical noise. Finally, in Fig. 5.12 we show the
scaled distribution of contrast for 2D gas with unity aspect ratio of imaging area and OBC
below the BKT temperature. Above the BKT temperature distribution functions become
Poissonian for L , where is the correlation length. In 2D one cannot describe the
crossover at L similar to 1D, since the action which describes the uctuations of the
phase is not quadratic in this region, and Eq. (5.82) doesnt hold.
The result (5.106) can be interpreted as the mapping of our problem to the statis-
tics of random surfaces (strings in 1D), subject to classical noise. In this mapping,
f
(x)
are the eigen modes of the surface vibrations, t
f
are the uctuating mode amplitudes, and
W( )
0.5 1 1.5 2 2.5 3
0.2
0.4
0.6
0.8
1
1.2
1.4

W( ) for a 1D
Bose gas with open boundary conditions at nonzero temperature and K = 5. Dierent curves
correspond to ratios K
T
/L = (solid), K
T
/L = 1 (dotted), and K
T
/L = 0.25 (dashed).
T
is the thermal correlation length, K is the Luttinger parameter and L is the imaging
length. The gure is taken from Ref. [218].
W( )
0.5 1 1.5 2
0.5
1
1.5
2
2.5

W( ) for a
two dimensional Bose gas with the aspect ratio of the imaging area equal to unity and open
boundary conditions. Temperature is below the Berezinskii-Kosterlitz-Thouless (BKT)
transition temperature. Dierent curves correspond to (T) =
c
(T
c
) = 1/4 (the BKT
transition point, solid), (T) = 1/6 (dashed line) and (T) = 1/10 (dotted line). Above the
BKT transition temperature the distribution function is Poissonian.
[G(f)[ is the noise power. This mapping holds as long as all eigenvalues G(f) are negative,
as discussed earlier. Innite dimensional integral over a t
f
variables can be understood
as an averaging over uctuations of the surface. For particular realization of noise variables
t
f
, a complex valued surface coordinate at point x is given by
h(x; t
f
) =
f
t
f
_
G(f)
K

f
(x)
G(f)
2K

f
(x)
2
. (5.107)
g(t
f
) is a number dened for each realization of a random surface t
f
, given by Eq.
(5.105). Fringe amplitude for each realization of t
f
variables is given by
= [g(t
f
)[
2
=
_
1
0
dxe
h(x;{t
f
})
2
. (5.108)
From interference of 1D Bose liquids of weakly interacting atoms to extreme
value statistics
To illustrate the power of interpretation of the interference fringe amplitudes in
terms of random surfaces, we now prove analytically that for periodic boundary conditions
in 1D and in the limit of large K, the normalized distribution

W( ) is given by the Gumbel
distribution (5.99), one of the extreme value statistical distributions. We note that to prove
such a result using 1/K expansion described in Appendix C, one needs to go to the innite
order of the perturbation theory, so our result is essentially non-perturbative in K.
The Gumbel function
P
G
(x) = e
(x+)e
(x+)
(5.109)
plays the same role in the extreme value statistics [276] as the usual Gaussian distribution
does in the statistics of the average value. According to the central limit theorem, the aver-
age value of N numbers taken from the same ensemble in the limit of large N is distributed
according to Gaussian function. One can prove similar theorems for the distribution of the
extreme, i.e. largest value of N 1 numbers taken from the same ensemble. The Gumbel
function (5.109) is one of the universal functions describing ensembles which decay faster
than any algebraic function at innity. The Gumbel distribution often appears in applied
mathematics, including problems in nance or climate studies, where it is used to describe
rare events such as stock market crashes or earthquakes.
For the periodic case corresponding to Eq. (5.84), the eigenvalue problem (5.100)-
(5.102) can be solved analytically, and the noise has 1/f power spectrum:
G
per
(x, y) = log
1
sin[x y[ = log 2
f=
f=1
1
2f
_
(
2 cos 2fx)(
2 cos 2fy) + (
2 sin 2fx)(
2 sin 2fy)
_
. (5.110)
Eigen modes are given by simple harmonic functions, and all of them, except one, are doubly
degenerate: we will denote corresponding noise variables as t
1,f
, t
2,f
and eigen values as
G(f) = 1/(2f). Simplication in the limit K 1 stems from the fact that exponent in
Eq. (5.108) can be expanded in Taylor series, since h(x; t
f
) is small for large K according
to its denition (5.107). In this case it can be shown that the distribution of is linearly
related to roughness, or mean square uctuation of the surface, as dened in Ref. [257].
It has been shown in Ref. [257] that 1/f noise results in the Gumbel distribution of the
roughness, which has been interpreted in terms of extreme value statistics in Ref. [286].
Terms of the order of 1/
K vanish in the Taylor expansion of Eq. (5.108), since

the average values of cos (2fx) and sin (2fx) on interval (0, 1) are equal to 0. To order
1/K, we obtain
1 +
1
K
log 2
f=1
1
2fK
(t
2
1,f
+t
2
2,f
2). (5.111)
The constant term log 2 in G
per
(x, y) gives a constant rescaling of the distribution func-
tion of , and doesnt show up in the normalized fringe amplitude :
1
f=1
1
2fK
(t
2
1,f
+t
2
2,f
2). (5.112)
Thus to the leading order in 1/K expansion, the distribution Y (x) of the rescaled variable
x = K( 1) =
f=1
1
2f
(t
2
1,f
+t
2
2,f
2) (5.113)
equals
Y (x) =
f=
f=1
_

dt
1,f
dt
2,f
2
e
f=
P
f=1
t
2
1,f
+t
2
2,f
2
_
_
x
f=
f=1
1
2f
(t
2
1,f
+t
2
2,f
2)
_
_
. (5.114)
To prove Eq. (5.99), we need to show Y (x) = P
G
(x), where the Gumbel function P
G
(x) is
given by Eq. (5.109). Let us introduce new positive variables
u
f
=
(t
2
1,f
+t
2
2,f
)
2f
> 0, (5.115)
then
Y (x) =
f=
f=1
_

0
fdu
f
e
P
f=
f=1
fu
f
_
_
x
f=
f=1
(u
f
1/f)
_
_
. (5.116)
To prove that Y (x) = P
G
(x) we will calculate their Fourier transforms:
Y (is) =
_

dxY (x)e
isx
=
f=
f=1
fdu
f
e
fu
f
e
is(u
f
1/f)
=
f=
f=1
fe
is/f
f is
=
e
is
[1 is], (5.117)
P
G
(is) =
_

dxP
G
(x)e
isx
=
_

dxe
(x+)e
(x+)
e
isx
= e
is
[1 is]. (5.118)
The proof above doesnt illustrate the meaning of Y (x) as of a distribution of extreme
value. Here we will follow the method of Ref. [286] and explicitly construct the variable,
the extreme value of which generates the Gumbel distribution. Let us impose a nite cuto
f
max
= N, and at the end of calculation we will send N to innity. If one identies
z
1
= u
N
(5.119)
z
2
= u
N1
+u
N
, (5.120)
... (5.121)
z
N
= u
1
+u
2
+... +u
N
, (5.122)
then z
1
, ..., z
N
is an ordered set (since u
i
> 0) of outcomes taken from Poissonian distri-
bution, since
e
P
f=N
f=1
fu
f
= e
P
f=N
f=1
z
f
, (5.123)
and Jacobian of transformation from variables u
f
to z
f
variables is unity. Then
Y
N
(x) = N!
_

0
e
z
1
dz
1
_

z
1
e
z
2
dz
2
...
_

z
N1
e
z
N
dz
N
_
_
x
_
_
z
N

N
f=1
1
f
_
_
_
_
is nothing other than the shifted distribution of the largest of N numbers taken from the
Poissonian distribution, and in the limit of large N this distribution converges to the Gumbel
function.
One can understand the appearance of the Gumbel distribution by noting that
for K 1 the distribution function of the interference amplitude is dominated by rare
uctuations of the random periodic 1D strings, which are spatially well localized. The
Gumbel distribution was introduced precisely to describe similar rare events such as stock
market crashes or earthquakes. For open boundary conditions the universal distribution for
large K is slightly dierent from the Gumbel function, similar to 1/f noise in other systems
[257]. But the main properties like the presence of asymmetry or the asymptotic form of
the tails are preserved.
5.5 Conclusions
5.5.1 Summary
When we discuss interference experiments with ultracold atoms, the conventional
idea of the particle-wave duality takes a new meaning. On the one hand, these experiments
probe phase coherence which is typically associated with coherent non-interacting waves.
On the other hand, one can use powerful tools of atomic physics to control interactions
between atoms in a wide range and to reach the regime of strong correlations. One can
also prepare atomic systems in states which would be dicult if not impossible to obtain
in optics, e.g. low dimensional condensates with strong thermal or quantum uctuations.
This remarkable combination places interference experiments with ultracold atoms in a
unique position: they can address a problem of how the interactions, correlations, and
uctuations aect the coherent properties of matter. This question appears in many areas
of physics, including high energy and condensed matter physics, nonlinear quantum optics,
and quantum information. While the naive answer that interactions suppress interference
turns out to be correct in most cases, the goal of this chapter was to demonstrate that the
quantitative analysis of this suppression can provide a lot of nontrivial information about
the original systems.
We discussed two eects which contribute to the reduction of the interference
fringe contrast in matter interferometers. The rst eect is the shot noise arising from
a nite number of atoms used in a single measurement. This analysis is particularly
important for interference experiments with independent condensates in which the po-
sition of interference fringes is random and averaging over many shots can not be per-
formed. In this case one needs to rely on single shot measurements to observe interfer-
ence patterns. While interference of independent condensates has been discussed before
[222, 223, 224, 225, 26, 254, 249, 252, 253, 250, 251], to our knowledge, we provide the
rst derivation of the full distribution function of the amplitude of interference fringes.
Another mechanism of the suppression of the amplitude of interference fringes discussed
in this chapter is the quantum and thermal uctuations of the order parameter in low di-
mensional condensates. The motivation for this discussion comes from the observation that
interference experiments between independent uctuating condensates can be used to study
correlation functions in such systems [215]. For example, one can use the scaling of the in-
tegrated amplitude of interference patterns to analyze two point correlation functions. This
method has been successfully applied by Hadzibabic et al. [193] to observe the Berezinskii-
Kosterlitz-Thouless transition in two dimensional condensates. One conceptual approach to
understanding interference experiments with independent condensates is to consider them
as analogues of the Hanbury Brown and Twiss experiments in optics [169]. In the latter
experiments interference between incoherent light sources appears not in the average sig-
nal but in the higher order correlation function. One important dierence however is that
matter interference experiments are of a single shot type and information is contained not
only in the average fringe contrast but also in the variation of the signal between individ-
ual shots. In particular, higher moments of the amplitude of interference fringes contain
information about higher order correlation functions [215]. A complete theoretical descrip-
tion of the fringe contrast variations is contained in the full distribution functions of the
fringe amplitudes, which we calculate for one and two dimensional condensates [216, 218]
in the limit when the number of atoms is large and the shot noise can be neglected [255].
An important aspect of this chapter was identifying intriguing mathematical connections
which exist between the problem of calculating distribution functions of interference fringe
amplitudes and several other problems in eld theory and statistical physics, such as the
quantum impurity problem [256], tunneling in the presence of the dissipation [281], Sine-
Gordon models, and so on. We developed a novel mapping of a wide class of such problems
to the statistics of random surfaces, which provided a complete non perturbative solution.
In certain cases we have analytically proven [218] the relationship between the distribution
function of fringe amplitudes and the universal extreme value statistical distribution [276].
5.5.2 Some experimental issues
We now comment on a few issues relevant for experimental analysis of noise in
interference experiments. The amount of information contained in the experimentally mea-
sured distribution function is directly related to the number of cumulants which can be
accurately extracted. This includes the second cumulant k
2
, which corresponds to the
width of the distribution; the third cumulant k
3
, which is related to skewness, g
1
= k
3
/k
3/2
2
,
and describes the asymmetry of the distribution function, and so on. In general, the statis-
tical error in determining the n
th
order cumulant after N measurements scales as
_
A
n
/N,
where A
n
is a constant which grows with n and depends on the higher moments of the
distribution. For example, to experimentally distinguish between the normal and the Gum-
bel distributions, it is necessary that the statistical error in skewness is at least a factor
of two smaller than the mean skewness, which is g
1
1.14 for the Gumbel distribution
. Thus the minimal number of measurements required is N
min
24/g
2
1
20, where we
used A
3
6, appropriate for the normal distribution [287]. In practice the required number
of measurements may be higher because of the inuence of other possible sources of noise.
However, it is certainly experimentally feasible.
Another experimentally relevant issue is the eect of the inhomogeneous density
due to the parabolic conning potential. While the approach discussed in this chapter can be
extended to include the inhomogeneous density prole, interpretation of the experimental
results is more straightforward when density variations can be neglected. We note that
when the condensate density varies gradually in space, the power-law decay of the correlation
functions is not strongly aected [288], except that the exponent may be dierent in dierent
parts of the trap (correlation function exponents typically depend on the density). We
expect that qualitatively our results will not change provided that the power law decay of
the correlation functions is much stronger than the change of the condensate density in the
measured part of the cloud. To be more precise, the best comparison with theory can be
achieved when the observation region L is much smaller than the size of the condensate,
determined by the eective Thomas-Fermi length [289], R
TF
= (3Nh
2
/(m
2
2
a
1D
))
1/3
(here
N is the number of atoms of mass m and a
1D
is the one-dimensional scattering length).
As long as L remains much larger the healing length, our analysis is valid. In the regime
of weakly interacting atoms, one can show that the ratio between the eective Luttinger
parameter K at the center of the harmonic trap and at the boundary of the observation
region is given by 1 L
2
/8R
2
TF
, thus giving only a small correction to the distribution
function computed in the central region. One can also reach similar conclusions in the
strongly interacting regime. It is also worth pointing out that we expect the limiting case
of the Poissonian distribution to be particularly robust to the inhomogeneous density of
atoms. Indeed, the Poissonian distribution is related to the fast 1/
x decay of the one-

particle correlation functions in the strongly interacting limit. This scaling is a universal
feature of the Tonks-Girardeau limit of bosons and is not aected by the weak harmonic
trap [290].
5.5.3 Outlook
Before concluding this part of the thesis, we would like to discuss questions which
still need to be understood in the context of interference experiments with ultracold atoms.
We also suggest an outlook for future theoretical work.
Combining shot noise with the order parameter uctuations.
A careful reader has undoubtedly noticed that we discussed either the shot noise or the
order parameter uctuations. At this point we still lack theoretical tools which would
allow to include both eects simultaneously. One of the diculties is that such analysis
requires the knowledge of the correlation functions for all distances rather than the long
distance asymptotic form. Indeed, in section 5.4 we showed that the short distance part
of the correlation functions gives contribution of the same order as the shot noise. In the
particular case of the interference of 2D condensates, the knowledge of the short distance
behavior of the correlation functions is needed to include the eect of the vortex excitations
below the BKT transition.
Stacks of independent condensates
In this chapter we focused on interference patterns from a single pair of condensates. How-
ever in experiments one often has a stack of several condensates (see e.g. Ref. [192]). In
this case interference arises from all possible pairs, and the system provides intrinsic aver-
aging and suppression of the noise. For a nite number of condensates, self-averaging is not
complete and one expects nite uctuations of the fringe contrast. It would be useful to
generalize the analysis of the shot noise and order parameter uctuations to such systems.
Dynamics of interacting atoms.
One of the advantages of the cold atoms systems is the possibility to study non equilibrium
coherent dynamics of interacting systems. In particular, dynamical splitting of a single
condensate into a pair of condensates has been performed in experiments on microchips [191,
206, 209, 210] and stimulated theoretical work on the subject [293, 294, 291, 292, 295, 296].
Similar experiments can also be done using superlattice potentials in optical lattices which
are now available in experiments [297, 298]. While analysis of fringe amplitude distribution
functions presented in this chapter dealt exclusively with systems in the thermodynamic
equilibrium, it would be interesting to generalize it to systems undergoing non equilibrium
dynamical evolution.
Interference experiments with fermions
The discussion presented in these notes was limited to the case of interference of bosons.
Such experiments can also be done with fermions [299], which are available experimentally
in dierent dimensions [18, 300, 96, 301]. For fermions, modulation of the density can be
related to fermion antibunching [302, 303, 182]. Analysis of the noise of the fringe contrast
visibility for fermions would be an interesting problem too.
Generalization to other systems
We note that mapping of the Coulomb gas into the statistics of random surfaces introduced
in section 5.4.3 should have applications beyond calculating the distribution functions of the
interference fringe amplitudes. This is a new non perturbative tool to calculate partition
functions of a variety of other systems that can be represented as Coulomb gas models.
Examples include quantum impurity-related problems [256], Sine-Gordon models where
interaction strength can depend on position, and many others. Our mapping is not related
to the existence of the exact solution of Sine-Gordon models, but relies only on the factorable
structure of the many-point correlation functions in the absence of interactions, which is a
general property of a Gaussian action.
Appendix A
Appendix to Chapter 2
A.1 Derivation of the eective magnetic Hamiltonian for in-
sulating states with odd number of atoms
To be able to derive J
0
, J
1
, J
2
dependence on n, we should know how to write
down explicitly, in terms of creation and annihilation operators, all of the states that we
are interested in. To do this we introduce singlet pair creation operator (summation over
repeated indices is presumed over x, y, z):
= a
p
a
p
= (a
0
)
2
2a
+
,
which has the following commutation relations:
[a
p
,
] = 2a
p
, [a
0
,
] = 2a
0
, [a
] = 2a
,
[a
] = 0, [S
a
,
] = 0, [
S
2
,
] = 0. (A.1)
Since
commutes with spin operators, it doesnt change total spin and spin components but
just adds two bosons. Therefore, we can construct unnormalized spin states for arbitrary
n in the following way: rst, write down a state with necessary spin for a small number of
particles; second, apply
as many times as needed to get the desired number of particles.

Using this procedure we obtain that states
a
+
(
)
n
[0 >, a
)
n
[0 >, a
0
(
)
n
[0 >
belong to S = 1 low energy subspace and are orthogonal since they are dierent eigenvectors
of S
z
. Making orthogonal transformation that leads to a
x
, a
y
, a
z
basis, we can write three
orthonormal states with S = 1 as
[S = 1, p, 2n + 1 >=
1
_
f(n; 1)
a
p
(
)
n
[0 >,
where normalization factor
f(n; s) = s!n!2
n
(2n + 2s + 1)!!
(2s + 1)!!
154
Appendix A: Appendix to Chapter 2 155
was calculated in Ref. [65]. In our calculation later we will need more general normalization
factors, so rst we will derive the way to normalize our spin states.
A.1.1 Normalization of the states
We will be interested in normalization of the states
[a, b, n >= a
a
a
b
(
)
n
[0 >,
and calculation of
f(a, b, p, q, n) =< a, b, n[p, q, n >=< 0[()
n
a
a
a
b
a
p
a
q
(
)
n
[0 >,
where a, b, p, q x, y, z. Lets consider a coherent state
e
a
x
x
1
+a
y
x
2
+a
z
x
3
[0 > .
We observe that this state is a linear combination of Fock states, with the coecients being
polynomials in x
1
, x
2
, x
3
. To extract the weight of the state [a, b, n >, we need to calculate
the quantity
[a, b, n >= [T
n
a,b
(x)e
a
x
x
1
+a
y
x
2
+a
z
x
3
[0 >]
x=0
,
where
T
n
a,b
(x) =
xa
x
b
(
x
)
n
.
We use the normalization condition for coherent states (see e.g. Ref. [92])
< 0[e
axy
1
+ayy
2
+azy
3
e
a
x
x
1
+a
y
x
2
+a
z
x
3
[0 >= e
x
1
y
1
+x
2
y
2
+x
3
y
3
to calculate
[T
n
a,b
(x)T
n
p,q
(y)e
x
1
y
1
+x
2
y
2
+x
3
y
3
][
x=0,y=0
=
T
n
a,b
(x)T
n
p,q
(y)
(x
1
y
1
+x
2
y
2
+x
3
y
3
)
2n+2
(2n + 2)!
.
We can expand T
n
a,b
(x)T
n
p,q
(y) using the extended Newton binomial formula :
T
n
a,b
(x)T
n
p,q
(y) =
n
1
+n
2
+n
3
=n
n!
n
1
!n
2
!n
3
!
2n
1
x
1

2n
2
x
2

2n
3
x
3

xa
x
b
m
1
+m
2
+m
3
=n
n!
m
1
!m
2
!m
3
!
2m
1
y
1

2m
2
y
2

2m
3
y
3

yp
yq
.
Lets rst consider the case when sets a, b and p, q coincide. We have two essentially
dierent cases: a = b and a ,= b. Without loss of generality, suppose for the rst case
a = b = p = q = x. Then,
< a, a, n[a, a, n >=
n
1
+n
2
+n
3
=n,l
1
+l
2
+l
3
=2n+2
(
n!
n
1
!n
2
!n
3
!
)
2
1
l
1
!l
2
!l
3
!
156 Appendix A: Appendix to Chapter 2

(
x
1
y
1
)
2n
1
+2
(
x
2
y
2
)
2n
2
(
x
3
y
3
)
2n
3
(x
1
y
1
)
l
1
(x
2
y
2
)
l
2
(x
3
y
3
)
l
3
=
n
1
+n
2
+n
3
=n
(
n!
n
1
!n
2
!n
3
!
)
2
(2n
1
+ 2)!(2n
2
)!(2n
3
)!.
This double sum can be calculated; the answer is
f(x, x, x, x, n) =
2
15
(3 + 2n)(5 + 3n)(2n + 1)!.
For the case of dierent indices, the normalization is
f(x, y, x, y, n) =
n
1
+n
2
+n
3
=n,l
1
+l
2
+l
3
=2n+2
(
n!
n
1
!n
2
!n
3
!
)
2
1
l
1
!l
2
!l
3
!
(
x
1
y
1
)
2n
1
+1
(
x
2
y
2
)
2n
2
+1
(
x
3
y
3
)
2n
3
(x
1
y
1
)
l
1
(x
2
y
2
)
l
2
(x
3
y
3
)
l
3
=
n
1
+n
2
+n
3
=n
(
n!
n
1
!n
2
!n
3
!
)
2
(2n
1
+ 1)!(2n
2
+ 1)!(2n
3
)! =
1
15
(3 + 2n)(5 + 2n)(2n + 1)!.
Now lets consider the case when a, b and p, q dont coincide. There are 4 essentially
dierent cases:
(x, y, x, z), (x, x, y, z), (x, x, x, y), (x, x, y, y).
All other normalizations can be obtained from these by proper permutation of indices. In
the rst three cases, overlap is zero since (say, for the rst case) a nonzero value comes from
the term that satises the conditions
2n
1
+ 1 = 2m
1
+ 1, 2n
2
+ 1 = 2m
2
, 2n
3
= 2m
3
+ 1,
which doesnt have integer solutions. However, in the fourth case, overlap of the states is
not zero: it comes from the terms obeying
n
1
+ 1 = m
1
, n
2
= m
2
+ 1, n
3
= m
3
.
We can calculate this quantity analogously to the previous calculation, but we can use
another trick:
< 0[()
n+1
(
)
n+1
[0 >= 3 < 0[()
n
a
x
a
x
a
x
a
x
(
)
n
[0 > +6 < 0[()
n
a
x
a
x
a
y
a
y
(
)
n
[0 > .
Therefore,
f(x, x, y, y, n) =
1
6
((2n + 3)! 3
2
15
(3 + 2n)(5 + 3n)(2n + 1)!) =
2
15
n(3 + 2n)(2n + 1)!.
Since sometimes it will be more convenient for us to work in +, , 0 basis, lets also write
down all nonzero overlaps in this basis (up to trivial permutations):
f(, , , , n) =< 0[()
n
a
)
n
[0 >=
2
15
(5 + 2n)(3 + 2n)(2n + 1)!,
f(0, 0, 0, 0, n) =< 0[()
n
a
0
a
0
a
0
a
0
(
)
n
[0 >=
2
15
(5 + 3n)(3 + 2n)(2n + 1)!,
f(0, , 0, , n) =< 0[()
n
a
a
0
a
0
(
)
n
[0 >=
1
15
(5 + 2n)(3 + 2n)(2n + 1)!,
f(+, , +, , n) =< 0[()
n
a
+
a
+
a
)
n
[0 >=
1
15
(5 + 4n)(3 + 2n)(2n + 1)!,
f(+, , 0, 0, n) =< 0[()
n
a
+
a
0
a
0
(
)
n
[0 >=
2
15
n(3 + 2n)(2n + 1)!.
0
So, rst lets calculate the energy for total spin 0 state. From known Clebsch-
Gordon coecients, this state is
[S = 0, S
z
= 0 >=
1
3
([1, 1 >
i
[1, 1 >
j
[1, 0 >
i
[1, 0 >
j
+[1, 1 >
i
[1, 1 >
j
).
Rewriting it via creation operators, we get the normalized state
[S
i
+S
j
= 0 >=
1
3f(n, 1)
a
i,p
a
j,p
(
i
)
n
(
j
)
n
[0 > .
In the second order of perturbation theory, the energy expectation value is
|m
t
2
E
m
E
0
[ < m[(a
ip
a
jp
+a
jp
a
ip
)[S
i
+S
j
> [
2
.
The intermediate state [m > cannot correspond to S
i
= 2, S
j
= 0 since in this case two
spins cant add to form a singlet. There are always at least two possible states:
S
i
= 0, n
i
= 2n + 2, S
j
= 0, n
j
= 2n, S
i
+S
j
= 0
and i j. The matrix element for each of these states is
1
3f(n, 1)
< 0[
n+1
a
p
a
q
(
)
n
[0 >
i
< 0[
n
a
p
a
q
(
)
n
[0 >
j
1
_
(f(n + 1, 0)f(n, 0))
=
1
3f(n, 1)
< 0[
n+1
a
p
a
q
(
)
n
[0 >
i

pq
f(n, 1)
1
_
(f(n + 1, 0)f(n, 0))
=
_
f(n + 1, 0)
3f(n, 0)
_
.
Finally, this term gives
2t
2
3(U
0
2U
2
)
f(n + 1, 0)
f(n, 0)
=
4t
2
(n + 1)(2n + 3)
3(U
0
2U
2
)
.
In general case, n ,= 0, there can also be intermediate states
S
i
= 2, n
i
= 2n + 2, S
j
= 2, n
j
= 2n, S
i
+S
j
= 0
and i j because S
i
= 2 and S
j
= 2 can form a singlet. However, in the case of n = 0,
such a term doesnt exist but can appear if a
jp
, acting on a
jq
(
j
)
n
[0 >, breaks a singlet
part and the result has a mixture of S = 2. But for n = 0 there is no singlet part, so these
intermediate states are absent. One should also note that even for large n we wont have
terms with higher spins, i.e. S
i
= 4, because our perturbation in the Hilbert space of i th
site is a vector and can have matrix elements only for the states with spins diering by 1.
From general considerations we know that for S
i
+S
j
= 0 energy has the form
2t
2
(
f
1
(n)
U
0
2U
2
+
f
2
(n)
U
0
+ 4U
2
),
where the term with U
0
2U
2
in the denominator comes from the processes of the rst kind,
and the term with U
0
+ 4U
2
in the denominator comes from the processes of the second
kind. We calculated f
1
(n) earlier to be 2(n +1)(2n +3)/3. Now, to nd f
2
(n), we take the
limit U
2
0. In this case we dont need to know the exact form of states with S
i
= S
j
= 0
and S
i
= S
j
= 2, S
i
+ S
j
= 0 since their energy is the same. We can take the intermediate
state [m > to be
[m >=
1
M
(a
i,p
a
j,p
+a
j,p
a
i,p
)a
i,q
a
j,q
(
i
)
n
(
j
)
n
[0 >,
where M is normalization of the intermediate state. Then, the second order energy is
t
2
[ < m[m >
3f(n,1)
[
2
U
0
=
t
2
U
0
M
3f(n, 1)
2
. (A.2)
Using formulas from the previous subsection, we can calculate M. Then, we can write an
equation
2(f
1
(n) +f
2
(n)) =
M
3f(n, 1)
2
= 2 +
6
5
n(5 + 2n),
and nally obtain
0
=
4t
2
(n + 1)(2n + 3)
3(U
0
2U
2
)

16t
2
n(5 + 2n)
15(U
0
+ 4U
2
)
.
1
Lets consider the case S
i
+S
j
= 1. There are no intermediate states S
i
= 0, S
j
= 2
or S
i
= 0, S
j
= 0 since these states can only form S
i
+S
j
= 2 and S
i
+S
j
= 0, respectively.
S
i
= 2, S
j
= 2 can add up to form S
j
+S
j
= 1, and this is the only contribution. S
i
+S
j
= 1
subspace is three dimensional, and from rotational invariance we can choose any state we
want to calculate the energy. Lets choose our initial state to be S
i
= 1, S
j
= 1, S
i
+ S
j
=
1, S
iz
+ S
jz
= 0. From known Clebsch-Gordon coecients, we can write this state using
creation and annihilation operators as :
[S
i
+S
j
= 1 >=
a
i,+
a
j,
a
i,
a
j,+
2f(n, 1)
(
i
)
n
(
j
)
n
[0 > .
Normalization of
(a
i,+
a
j,+
+a
i,0
a
j,0
+a
i,
a
j,
)(a
i,+
a
j,
a
i,
a
j,+
)[S
i
+S
j
= 1 >
equals
2(f(+, 0, n)(2n)
2
f(+, 0, n 1) +f(+, +, n)(2n)
2
f(+, +, n 1)) =
4
45
n(1 + 2n)(3 + 2n)
2
(5 + 2n)(2n + 1)!
2
.
Hence, the energy for this case is
1
=
4t
2
n(5 + 2n)
5(U
0
+ 4U
2
)
.
2
Lets choose the normalized state, for which we will calculate the second order
energy, to be the state with total S
iz
+S
jz
= 2 :
[S
i
+S
j
= 2 >=
1
f(n, 1)
a
i,+
a
j,+
(
i
)
n
(
j
)
n
[0 > .
The intermediate state should also have S
i
+S
j
= 2, S
iz
+S
jz
= 2. Such a state cant belong
to S
i
= 0, S
j
= 0 subspace since this pair of spins cant add up to form S
i
+S
j
= 2. There
are four possible intermediate states with
S
i
= 2, n
i
= 2n + 2, S
j
= 0, n
j
= 2n,
S
i
= 0, n
i
= 2n + 2, S
j
= 2, n
j
= 2n,
and i j. In the rst case, matrix element equals
1
f(n, 1)
< 0[
n
a
+
a
+
a
+
a
+
(
)
n
[0 >
i
< 0[
n
a
+
a
+
(
)
n
[0 >
j
1
_
f(n, 2)f(n, 0)
=
1
f(n, 1)
f(n, 2)f(n, 1)
1
_
(f(n, 2)f(n, 0))
=
_
f(n, 2)
f(n, 0)
_
.
Finally, this term contributes to the energy
2t
2
(U
0
+U
2
)
f(n, 2)
f(n, 0)
=
4t
2
(2n + 3)(2n + 5)
15(U
0
+U
2
)
.
The second process contributes with the same energy denominator but with a dierent
dependence on n:
16t
2
n(n + 1)
15(U
0
+U
2
)
.
Now, lets consider the contribution from the S
i
= 2, S
j
= 2, S
i
+ S
j
= 2 inter-
mediate state. Using the same trick as at the end of the calculation of
0
, we only need to
calculate the norm of
(a
i,+
a
j,+
+a
i,0
a
j,0
+a
i,
a
j,
)a
i,+
a
j,+
(
i
)
n
(
j
)
n
[0 > .
This quantity equals
f(+, +, n)(f(0, 0, n 1) 2(2n + 2)f(+, , 0, n 1) + (2n + 2)
2
f(+, , n 1))+
f(+, 0, n)(2n)
2
f(+, , n 1) +f(+, , n)(2n)
2
f(+, +, n 1) =
2
225
(3 + 2n)
2
(5 + 3n)(5 + 6n)(2n + 1)!
2
.
Then, taking a limit U
2
0, we obtain the contribution from this process to be
28t
2
n(5 + 2n)
75(U
0
+ 4U
2
)
.
Finally,
2
=
28t
2
n(5 + 2n)
75(U
0
+ 4U
2
)

4(15 + 20n + 8n
2
)
15(U
0
+U
2
)
.
A.2 Derivation of the eective magnetic Hamiltonian for the
insulating state with two atoms
To derive the eective Hamiltonian, we should be able to calculate matrix elements
in Eq. (2.31). Since energy and matrix elements in each subspace dont depend on z
projection of total spin, we can choose S
z
components at our convenience. We can express
any state [E
1
>, ..., [E
8
> using known Clebsch-Gordon coecients. For the state from
[E
8
> with S
z
= 0, we have
[E
8
, S
z
>=
m
C
4,0
2,m,2,m
[N
1
= 2, S
1
= 2, S
1z
= m > [N
2
= 2, S
2
= 2, S
2z
= m > .
We can write any of the states [N
i
, S
i
, S
iz
> via creation and annihilation operators since
we know how to express spin operators via creation and annihilation operators (2.11).
Evaluation of e
6
e
8
is quite simple since total spin conservation of the tunneling term
doesnt allow mixing of these subspaces with any other. Therefore, as in Eq. (A.2), we just
need to calculate the normalization of the states into which particles can tunnel. Using this
procedure, we obtain energies (2.30).
Now lets consider energy in the [E
1
>, [E
2
> subspace. From [E
1
> bosons can
hop only into high energy states
[N
1
= 3, N
2
= 1, S
1
= S
2
= 1, S
1
+S
2
= 0 >,
[N
1
= 1, N
2
= 3, S
1
= S
2
= 1, S
1
+S
2
= 0 >, (A.3)
since in the Hilbert space of each spin can change only by 1. For N = 2 from [E
2
>,
bosons can also tunnel only to these states since spins 3 and 1 cannot add to form total
spin 0. Therefore, our exact Hamiltonian in the basis of [E
1
>, [E
2
> and high energy states
(A.3) has the form
_
_
0 0 V
1
V
1
0 6U
2
V
2
V
2
V
1
V
2
U
0
0
V
1
V
2
0 U
0
_
_
.
We can diagonalize this wave vector in the low energy [E
1
>, [E
2
> subspace in the limit
V
1
, V
2
U
0
, U
2
U
0
.
First, we integrate out high energy levels this is done as described in Ref. [51]. We use
the following matrix identity:
_
_
A B
C D
_
1
_
ij
= [(A BD
1
C)
1
]
ij
,
where 1 i, j 2.
In our case, D has the form U
0
I
2
, so it is easy to calculate inverse matrix. Finally,
our eective Hamiltonian has the form
_
0 0
0 6U
2
_
2
U
0
_
V
2
1
V
1
V
2
V
1
V
2
V
2
2
_
.
Now we can diagonalize this 2 2 matrix; its energy levels are
3U
2
V
2
1
+V
2
2
U
0
(3U
2
V
2
1
+V
2
2
U
0
)
2
+ 12
U
2
V
2
1
U
0
.
Using expressions for all states of interest in Fock basis, we can calculate
V
1
= t
_
10
3
, V
2
= t
_
8
3
,
which leads to Eqs. (2.20)-(2.21).
Now, lets calculate energies for the S
i
+S
j
= 2 subspace. In this case bosons can
hop to 4 states:
[N
1
= 3, N
2
= 1, S
1
= S
2
= 1, S
1
+S
2
= 2 >,
[N
1
= 1, N
2
= 3, S
1
= S
2
= 1, S
1
+S
2
= 2 >,
[N
1
= 1, N
2
= 3, S
1
= 1, S
2
= 3, S
1
+S
2
= 2 >,
[N
1
= 3, N
2
= 1, S
1
= 3, S
2
= 1, S
1
+S
2
= 2 > .
Matrix A has the form (in E
3
, E
4
, E
5
basis)
_
_
3U
2
0 0
0 3U
2
0
0 0 6U
2
_
_
,
and matrix B has the form
_
_
V
1
V
2
V
3
0
V
2
V
1
0 V
3
V
4
V
4
V
5
V
5
_
_
,
and C = B
T
. The eective Hamiltonian is
_
_
3U
2
0 0
0 3U
2
0
0 0 6U
2
_
_
1
U
0
_
_
V
2
1
+V
2
2
+V
2
3
2V
1
V
2
(V
1
+V
2
)V
4
+V
3
V
5
2V
1
V
2
V
2
1
+V
2
2
+V
2
3
(V
1
+V
2
)V
4
+V
3
V
5
(V
1
+V
2
)V
4
+V
3
V
5
(V
1
+V
2
)V
4
+V
3
V
5
2(V
2
4
+V
2
5
)
_
_
. (A.4)
From the symmetry of this matrix, one eigenvalue is easy to determine it corresponds to
eigenvector (1, 1, 0)
T
and equals
3U
2
V
2
1
+V
2
2
+V
2
3
2V
1
V
2
U
0
.
Since one of the eigenvectors is known, two other eigenvalues are also easy to obtain they
are solutions of a quadratic equation. Calculating V
i
gives
V
1
= 2
_
2
15
t, V
2
=
_
10
3
t, V
3
=
_
14
5
t, V
4
=
_
14
15
t, V
5
=
_
2
5
t.
Energy levels are
3U
2
4
t
2
U
0
,
1
2
_
9U
2
12t
2
144
t
4
U
2
0
+ 40
t
2
U
0
U
2
+ 9U
2
2
_
.
A.3 Mean eld solution for the case of two bosons per site
Our mean eld state depends on 12 variables 6 complex numbers, subject to nor-
malization (2.40). However, energy is clearly the same for all states that can be transformed
into each other by global SU(2) rotations, so that gives us 3 conditions we can choose. We
also have an overall U(1) phase freedom, so the number of independent parameters reduces
to 12 3 1 1 = 7. We can parameterize
a
m
= b
m
e
im
.
It is convenient to choose 3 non uniform to be:
b
0
= 0,
1
=
1
.
We can always nd an axis of quantization for which the rst condition is satised and then
we can satisfy the last condition on phases by rotating along chosen axis of quantization.
We have 7 degrees of freedom: eight variables b
2
, b
1
,
2
,
1
, with the normalization
constraint
b
2
2
+b
2
2
+b
2
1
+b
2
1
= 1.
We can express mean eld energy via these degrees of freedom, and it will be some polyno-
mial up to the fourth order in b
m
(we wont explicitly write down this polynomial). First,
lets make a transformation that diagonalizes the part quadratic in b
2
, b
1
. That orthogonal
transformation is
b
2
=
u
2
+v
2
2
, b
1
=
u
1
v
1
2
,
b
1
=
u
1
+v
1
2
, b
2
=
u
2
+v
2
2
.
Now we solve the normalization constraint by
v
1
= sin sin
1
, v
2
= cos sin
2
,
u
1
= sin cos
1
, u
2
= cos cos
2
.
The numerical procedure now consists of xing the values of t and and minimizing the
expression for energy over six angles using steepest descents method. Using this procedure
we can numerically nd energy as a function of for xed t. If the minimum is attained at
= 0, then mean eld wave function is a trivial singlet. Result of this minimization leads
to the state
b
m
= (
1
6
,
1
3
, 0,
1
3
,
1
6
)
T
that can be rewritten as Eq. (2.42) after rotation. Though this result was obtained numer-
ically, we can check analytically that at this point all rst derivatives of the energy over
angles ,
1
and
2
vanish.
A.4 Large N expansion
In this Appendix we prove some properties of wave functions [)
N
, given by Eq.
(2.59). The normalization factor F in Eq. (2.59) may be found by considering the overlap
of two states. It is sucient to consider wave functions constructed of only two single
particle states, since rotation of the spinor basis can always bring our pure condensate
wave functions to this form:
[N, n
1
) =
1
F
(cos
1
a
x
+ sin
1
a
y
)
N
[0),
[N, n
2
) =
1
F
(cos
2
a
x
+ sin
2
a
y
)
N
[0).
(A.5)
We have
N, n
1
[N, n
2
) =
1
F
2
0[(cos
1
a
x
+ sin
1
a
y
)
N
(cos
2
a
x
+ sin
2
a
y
)
N
[0) =
1
F
2
N
k=0
(C
k
N
)
2
(cos
1
cos
2
)
k
(sin
1
sin
2
)
Nk
0[a
k
x
a
Nk
y
(a
x
)
k
(a
y
)
Nk
[0) =
N!
F
2
N
k=0
C
k
N
(cos
1
cos
2
)
k
(sin
1
sin
2
)
Nk
=
N!
F
2
cos
N
(
1
2
) =
N!
F
2
(n
1
n
2
)
N
.
Orthogonality and normalization for large N now become obvious after noting that for
(n
1
n
2
) = cos and /2 we have cos
N
e
N
2
/2
.
To prove Eq. (2.65) we consider wave functions
[N, n) =
1
F
N
(n
x
a
x
+n
y
a
y
)
N
[0),
[N + 1, n
) =
1
F
N+1
(n
x
a
x
+n
y
a
y
)
N
[0). (A.6)
Simple calculation gives
N + 1, n
[a
x
[N, n) =
(N + 1)!
F
N
F
N+1
n
x
(nn
)
N
= (N + 1)
1/2
n
x
N
(n n
). (A.7)
Appendix B
In this appendix we will prove that all solutions of equations (3.28)-(3.29)
N
i=1
k
i
+ic/2
k
i
ic/2
= 1, = 1, ..., M, (B.1)
e
ik
j
L
=
M
=1
k
j

+ic/2
k
j

ic/2
, j = 1, ..., N (B.2)
are always real. This is a major simplication for the analysis of the excited states compared
to spin-
1
2
fermion systems, where one has to consider complex solutions[111].
Suppose that solutions of (B.1)-(B.2) are complex numbers, such that
inf Imk
j
= k
supImk
j
= k
+
, (B.3)
inf Im
supIm
=
+
. (B.4)
We need to prove that k
= k
+
=
=
+
= 0.
First, lets prove that
k
, (B.5)
+
k
+
. (B.6)
Suppose that (B.5) is not valid, i. e.
: Imk
j
Im
> 0 j. (B.7)
Then
k
j

+ic/2
k
j

+ic/2
> 1 j, (B.8)
and absolute value of the left hand side of Eq. (B.1) is bigger than 1, which contradicts the
equation. Equation (B.6) can be proven similarly.
Now, lets prove that
k
+
0, (B.9)
k
0. (B.10)
165
166 Appendix B: Appendix to Chapter 3
These equations together with (B.5)-(B.6) would imply k
= k
+
=
=
+
= 0.
Suppose that (B.9) is not valid, i. e. j : Imk
j
= k
+
> 0. From (B.6) it follows
that
Imk
j
Im
0 , (B.11)
therefore

k
j

+ic/2
k
j

+ic/2
1 , (B.12)
and absolute value of the right hand side of equation (B.2) is not smaller than 1. On the
other hand, by assumption, left hand side of this equation is smaller than 1 :
[e
ik
j
L
[ = e
k
+
L
< 1. (B.13)
Contradiction proves the validity of (B.9), and (B.10) can be proven similarly.
Appendix C
In this Appendix, we will describe a systematic diagrammatic technique to cal-
culate Z
2n
or Z
per
2n
as an expansion in small parameter 1/K. It corresponds to the high
temperature limit of the classical gas analogy discussed in section 5.4.1. This expansion
can be applied both in 1D or 2D, and can be used to study the limiting distribution at large
K, which for PBC in 1D has been conjectured [216] and proven [218] to be the Gumbel
distribution [276].
C.1 Expansion to order (1/K)
2
.
We will start from the 1D case by expanding the exponent in Eq. (5.82):
Z
2n
=
_
1
0
...
_
1
0
du
1
...dv
n
e
1
K
(
P
i<j
G(u
i
,u
j
)+
P
i<j
G(v
i
,v
j
)
P
ij
G(u
i
,v
j
))
=
_
1
0
...
_
1
0
du
1
...dv
n
m=0
1
m!K
m
_
_
i<j
G(u
i
, u
j
) +
i<j
G(v
i
, v
j
)
ij
G(u
i
, v
j
)
_
_
m
. (C.1)
In the rst order of expansion in powers of 1/K, G(x, y) dependence comes only
through one integral
I
0
=
_
1
0
_
1
0
dxdyG(x, y) (C.2)
after the integration in (C.1).
The prefactor depends only on n, and can be calculated analytically: the total
number of uu terms is n(n1)/2, the total number of v v terms is also n(n1)/2, and
the total number of uv terms is n
2
. The latter come with 1 sign, so the total expression
for Z
2n
up to O(1/K
2
) is
Z
2n
= 1
1
K
nI
0
+O(
1
K
2
). (C.3)
167
168 Appendix C: Appendix to Chapter 5
x
z t
y y
y
x
z x
x y
y
2,2 1,2 3,2
I I
I
G(x,y)
G(z,t)
G(x,y)
G(y,z)
G(x,y)
G(x,y)
Figure C.1: Topologically inequivalent diagrams, corresponding to 1/K
2
terms in expansion
of Z
2n
.
In second order, dependence of Z
2n
on G(x, y) comes through three dierent integrals:
I
1,2
=
_
1
0
_
1
0
_
1
0
_
1
0
dxdydzdtG(x, y)G(z, t) =
__
1
0
_
1
0
dxdyG(x, y)
_
2
= I
2
0
, (C.4)
I
2,2
=
_
1
0
_
1
0
_
1
0
dxdydzG(x, y)G(y, z), (C.5)
I
3,2
=
_
1
0
_
1
0
dxdyG(x, y)G(x, y). (C.6)
Pictorially, these expressions can be represented by the corresponding diagrams,
shown in Fig. C.1. There, horizontal solid line corresponds to G(x
i
, x
j
). If two ends are
connected by a dashed line, then in the integral these two ends should correspond to the
same variable. To write the expression for Z
2n
in 1/K
2
order, one has to calculate the total
number of terms corresponding to I
1,2
, I
2,2
, I
3,2
. Clearly, it will be some universal polynomial
depending on n, which is determined by combinatorics. Here we describe how to do this
combinatorics in detail.
When parenthesis are multiplied in (C.1), there are several types of expressions
which appear, and come with dierent signs to the integrand:
_
_
i<j
G(u
i
, u
j
) +
i<j
G(v
i
, v
j
)
ij
G(u
i
, v
j
)
_
_
2
=
G(u
i1
, v
j1
)G(u
i2
, v
j2
) +
i1<j1,i2<j2
G(u
i1
, u
j1
)G(u
i2
, u
j2
) +G(v
i1
, v
j1
)G(v
i2
, v
j2
) + 2G(u
i1
, u
j1
)G(v
i2
, v
j2
)
2
i1<j1,i2,j2
(G(u
i1
, u
j1
)G(u
i2
, v
j2
) +G(v
i1
, v
j1
)G(u
i2
, v
j2
)). (C.7)
Appendix C: Appendix to Chapter 5 169
A)
B)
Figure C.2: Topologically equivalent diagrams, corresponding to a)I
2,2
and b)I
3,2
.
These are (signs are indicated)
+G(u
i1
, u
j1
)G(u
i2
, u
j2
), and equivalent G(v
i1
, v
j1
)G(v
i2
, v
j2
); (C.8)
+G(u
i1
, v
j1
)G(u
i2
, v
j2
); (C.9)
G(u
i1
, u
j1
)G(u
i2
, v
j2
), and equivalent G(v
i1
, v
j1
)G(u
i2
, v
j2
); (C.10)
+G(u
i1
, u
j1
)G(v
i2
, v
j2
); (C.11)
For terms which have G(u
ik
, u
jk
) or G(v
ik
, v
jk
), k = 1, 2 there is a restriction
that ik < jk, but there is no such restriction for G(u
ik
, v
jk
).
Lets calculate the total number of I
1,2
, I
2,2
and I
3,2
terms for uuuu expressions
of type (C.8). Total number of I
1,2
terms compatible with restrictions is
1
2!
2
n(n 1)(n 2)(n 3). (C.12)
Prefactor 1/2!
2
appears since pairs i1, j1 and i2, j2 are ordered , while n(n 1)(n
2)(n 3) is the total number of ways to choose a sequence of four dierent not ordered
numbers out of the set of size n.
Total number of I
2,2
terms compatible with restrictions is
1
2!
2
n(n 1)(n 2) 4. (C.13)
Here the situation is similar to I
1,2
, but there is a topological prefactor of 4, which
corresponds to four topologically equivalent diagrams for I
2,2
, as shown in Fig. C.2.a.
Finally, total number of I
3,2
terms is
1
2!
2
n(n 1) 2. (C.14)
Here, 2 is again a topological prefactor, corresponding to two topologically equivalent
diagrams shown in Fig. C.2.b.
Overall, u u u u term of (C.8) gives
1
4
n(n 1)(n 2)(n 3)I
1,2
+n(n 1)(n 2)I
2,2
+
1
2
n(n 1)I
3,2
(C.15)
As a simple check of combinatorics one can calculate the total numbers of u u u u
terms of type (C.8). It is
1
2!
2
n(n 1)(n 2)(n 3) +
1
2!
2
n(n 1)(n 2) 4 +
1
2!
2
n(n 1) 2 =
_
n(n 1)
2!
_
2
, (C.16)
as it should be from the calculation which neglects the dashed lines, and doesnt impose
any restrictions on the terms at dierent horizontal rows.
Analogously, one can calculate polynomials for expressions (C.9)-(C.11), and the
results are as follows, respectively:
n
2
(n 1)
2
I
1,2
+ 2n
2
(n 1)I
2,2
+n
2
I
3,2
, (C.17)
1
2
n
2
(n 1)(n 2)I
1,2
n
2
(n 1)I
2,2
, (C.18)
1
2
2
n
2
(n 1)
2
I
1,2
. (C.19)
Summing all the terms with corresponding prefactors from (C.7), we obtain
Z
2n
= 1
1
K
nI
0
+
1
2!K
2
(P
1,2
I
1,2
+P
2,2
I
2,2
+P
3,2
I
3,2
) +O(
1
K
3
) =
1
1
K
nI
0
+
1
2!K
2
(3n(n 1)I
1,2
4n(n 1)I
2,2
+ (2n 1)nI
3,2
) +O(
1
K
3
), (C.20)
where P
1,2
= 3n(n 1), P
2,2
= 4n(n 1), and P
3,2
= (2n 1)n are universal polynomials
of n. Notice, that due to sign cancellations of terms in (C.8) (C.11), the overall degree of
polynomials in 1/K
2
order is m = 2, compared to naively expected 2m = 4.
Using (C.20), one can calculate Z
2n
/Z
n
2
up to O(1/K
3
) :
Z
2n
Z
n
2
= 1 +
n(n 1)
K
2
(I
1,2
2I
2,2
+I
3,2
) +O(
1
K
3
) (C.21)
Calculated values of I
0
, I
1,2
, I
2,2
, I
3,2
and I
per
0
, I
per
1,2
, I
per
2,2
, I
per
3,2
are
I
0
=
3
2
= 1.5, I
per
0
= log 2 1.83788; (C.22)
I
1,2
=
9
4
= 2.25, I
per
1,2
= (log 2)
2
3.37779; (C.23)
I
2,2
=
51
2
18
2.28502, I
per
2,2
= (log 2)
2
3.37779; (C.24)
I
3,2
=
7
2
= 3.5, I
per
3,2
=

2
+ 12(log 2)
2
12
4.20026. (C.25)
Thus, for K , the limiting ratio of the widths of the distributions for OBC and PBC
are
Z
4
/(Z
2
)
2
1
Z
per
4
/(Z
per
2
)
2
1

I
1,2
2I
2,2
+I
3,2
(I
per
1,2
2I
per
2,2
+I
per
3,2
)
1.43465. (C.26)
C.2 General properties of (1/K)
m
terms, and expansion to
order (1/K)
5
.
From the expansion of the previous subsection, one can formulate the general
properties of the diagram technique to calculate terms up to (1/K)
m
:
a) First, one has to draw all possible topologically inequivalent diagrams, which consist
of m horizontal solid lines, with some of the ends connected by dashed lines. Each
end can have at most two dashed lines coming out of it.
b) Ends which are connected by a dashed line correspond to the same variable. Diagrams
for which two opposite ends of the horizontal line correspond to the same variable are
excluded.
c) The expression which corresponds to a diagram is constructed the following way: if
variables at the end of a given solid horizontal line are x and y, then G(x, y) should
be put as one of the terms in the product under the integrand. Thus the integrand
consists of the product of function G of some variables m times. Diagrams for which
the integrands are the same up to relabeling of the variables are considered to be
identical.
d) All free variables should be integrated from 0 to 1.
Diagrams can be connected or disconnected. For example, in Fig. C.1 diagram
corresponding to I
1,2
is disconnected, and diagrams corresponding to I
2,2
and I
3,2
are con-
nected. The integral which corresponds to a disconnected diagram is a product of expres-
sions, corresponding to its parts: for example, I
1,2
= I
0
I
0
.
If the number of topologically inequivalent diagrams of order m is g(m), then the
term of the order 1/K
m
in Z
2n
has the following form:
_
1
0
...
_
1
0
du
1
...dv
n
1
m!K
m
_
_
i<j
G(u
i
, u
j
) +
i<j
G(v
i
, v
j
)
ij
G(u
i
, v
j
)
_
_
m
=
1
m!K
m
(
g(m)
r=1
P
r,m
(n)I
r,m
), (C.27)
where P
r,m
(n) are universal polynomials of n, which can be calculated combinatorially, as
described in section C.1.
Polynomials P
r,m
(n) should satisfy the following requirements:
a) For positive integer n, the values of P
r,m
(n) are integer, and
P
r,m
(0) = 0. (C.28)
b) If one sets G(x, y) = 1, then I
r,m
= 1, and lhs of (C.27) can be trivially calculated.
This implies
g(m)
r=1
P
r,m
(n) = (n)
m
. (C.29)
c) The degree of each polynomial P
r,m
(n) is not larger than m. This has been shown
above for m = 1 and m = 2, and has been checked up to m = 5, although we
didnt succeed in proving it directly. This conjecture is supported by the fact, that it
guarantees that for any G(x, y) for K distributions t on the top of each other
after proper rescaling (see next section).
One can do the combinatorial calculations similar to previous section for m = 3,
and it takes about a day by hand. We will only show the results here. For m = 3 there
are g(3) = 8 topologically inequivalent diagrams, which are shown in Fig. C.3, out of which
5 diagrams (I
1,3
to I
5,3
) are connected. The expressions corresponding to each diagram and
universal polynomials are, respectively:
I
1,3
=
_ _ _ _
dxdydzdtG(x, y)G(x, z)G(z, t), P
1,3
(n) = 12n(n 1)(2n 3),
I
2,3
=
_
1
0
_
1
0
_
1
0
_
1
0
dxdydzdtG(x, y)G(x, z)G(x, t), P
2,3
(n) = 12n(n 1),
I
3,3
=
_
1
0
_
1
0
_
1
0
dxdydzG(x, y)
2
G(x, z), P
3,3
(n) = 12n(n 1),
I
4,3
=
_
1
0
_
1
0
_
1
0
dxdydzG(x, y)G(x, z)G(y, z), P
4,3
(n) = 4n(n 1)(2n 1),
I
5,3
=
_
1
0
_
1
0
dxdyG(x, y)
3
, P
5,3
(n) = n,
I
6,3
= I
2,2
I
0
, P
6,3
(n) = 36n(n 1)(n 2),
I
7,3
= I
3,2
I
0
, P
7,3
(n) = 3n(n 1)(2n 3),
I
8,3
= I
1,2
I
0
= I
3
0
, P
8,3
(n) = 15n(n 1)(n 2).
Numerically evaluated integrals for OBC and PBC are
I
1,3
= 3.49399, I
per
1,3
= 6.20797; (C.30)
I
2,3
= 3.5268, I
per
2,3
= 6.20797; (C.31)
I
3,3
= 5.32704, I
per
3,3
= 7.71956; (C.32)
I
4,3
= 4.21255, I
per
4,3
= 6.50848; (C.33)
I
5,3
= 11.25, I
per
5,3
= 12.5458; (C.34)
I
6,3
= 3.42753, I
per
6,3
= 6.20797; (C.35)
I
7,3
= 5.25, I
per
7,3
= 7.71956; (C.36)
I
8,3
= 3.375, I
per
8,3
= 6.20797. (C.37)
For m > 3 it becomes too cumbersome to manually calculate universal polynomials P
r,m
.
We wrote a program in Mathematica, which expands m th term of (C.1) directly, and
calculates the values P
r,m
(0), ..., P
r,m
(n) using powerful pattern recognition tools. Next,
P
r,m
(n) is recovered using Newtons formula. Results for m = 3 can be recovered that way.
One can also check in each order that the degree of P
r,m
(n) is not larger than m. For each
m the program needs as an input all topologically inequivalent diagrams, and currently the
I
1,3
I I I
I I I
2,3 3,3 4,3
I
5,3
6,3 7,3 8,3
Figure C.3: Topologically inequivalent diagrams for m = 3. Diagrams I
1,3
I
5,3
are con-
nected and I
6,3
I
8,3
are disconnected.
results have been extended to m = 5. For m = 4, the overall number of diagrams is 23,
out of which 12 are irreducible, and shown in Fig. C.4. For m = 5, the overall number
of diagrams is 66, out of which 33 are irreducible. Numerical prefactors of polynomials
P
r,m
(n) grow with m, while their overall sum has a prefactor 1, as follows from (C.29). For
example, one of the polynomials for m = 5 has a prefactor of 384. This puts stringent
requirements on the errors in calculation of I
r,m
(n), so going beyond m = 5 will require
additional numerical eort. For m = 5 results are reliable, since we can check the numerical
accuracy of the calculation of integrals by comparison with the analytically proven [218]
Gumbel distribution.
While calculation of original Z
2n
requires 2n dimensional numerical integration,
calculation of the I
r,m
requires at most mdimensional numerical integration. This can be
seen from the fact that even for irreducible diagrams, only the parts which contain loops
have to be integrated numerically. Horizontal lines which have a free end can be integrated
out analytically, and the dimension of the numerical integral has at most dimension of the
loop. For example, rst and third horizontal bars in I
1,3
of Fig C.3 can be integrated out
using the following identity:
_
1
0
dy log [x y[ = 1 + log (1 x) xlog (1 x) +xlog x, (C.38)
so one has to do only 2-dimensional integral numerically. Analogously, diagram I
2,3
requires
only one dimensional numerical integration. In each order m there is only one diagram which
requires mdimensional integration, i.e. I
4,3
in Fig.C.3 for m = 3. All the rest require at
most (m1) dimensional integration.
Figure C.4: Irreducible diagrams for m = 4.
C.3 Properties of the K distribution
For K distribution function becomes very narrow, and it is an interesting
question to investigate the limiting behavior of the distribution function. Let us consider the
distribution of normalized contrast , dened by (5.87). Due to normalization the following
relations hold:
_

0
W( )d 1,
_

0

W( )d 1. (C.39)
For large K distribution function is peaked near = 1, and its width is propor-
tional to 1/K. To calculate the properties of the universal function we assume that it decays
relatively fast (exponential decay is enough) away from = 1 for large K, so we can extend
integrations to innity. If we dene the uctuation of the normalized contrast

= 1
with distribution function

W
0
(
) =

W(1 +

), then
0 =
_

0
( 1)

W( )d =
_

1

W
0
(
)d

W
0
(
)d

M
1
, (C.40)
1 +
Z
4
Z
2
2
=
_

0
(
2
1)

W( )d
_

(2
2
)

W
0
(
)d
2

M
1
+

M
2
, (C.41)
1 +
Z
6
Z
3
2
=
_

0
(
3
1)

W( )d 3

M
1
+ 3

M
2
+

M
3
, (C.42)
..., (C.43)
where

M
1
,

M
2
,

M
3
are the moments of

W
0
(
). Inverting (C.40)-(C.42), one can nd

M
2
,

M
3
, ...
as series expansion in powers of 1/K. If for large K the distribution functions collapse on
the top of each other after proper rescaling, as has been conjectured in [216], then the
expansion of

M
m
in powers of 1/K should start from 1/K
m
. Below we will show, that this
is necessarily true for all n, if the degrees of universal polynomials P
r,m
(n) are not larger
than m and P
r,m
(0) = 0.
To show this, let us consider an analog of Eqs. (C.40)-(C.42) for the uctuation
of the unnormalized contrast = 1 with distribution W
0
() = W(1 + ) (normalized
contrast considered in Appendix C.3 is related to by simple rescaling). Then
Z
2n
=
_

0
n
W()d =
_

1
(1 +)
n
W
0
()d
_

i=n
i=0
C
i
n
i
W
0
()d =
i=n
i=0
C
i
n
M
i
, (C.44)
where M
i
is the i th moment of W
0
(). We need to show that expansion of M
i
in powers
of 1/K starts only from the terms of the order of 1/K
i
. This can be seen from expansion
(C.27) together with (C.44). If degrees of P
r,m
(n) are not larger than m, then (C.27) means
that 1/K
m
terms in Z
2n
grow at most as n
m
for large n. On the other hand, C
i
n
=
n!
i!(ni)!
grows as n
i
for large n, and if expansion of M
i
in powers of 1/K starts before 1/K
i
, this
will contradict the previous sentence.
To nd the rst nontrivial contribution to

M
m
, one has to go to the order 1/K
m
in the expansion. Using results for m = 5 calculated above for K , one can deduce
the limiting behavior of K
2

M
2
, ..., K
5

M
5
for periodic and non-periodic cases, and compare
it with the result of the Gumbel distribution
W
G
(
) = Ke
K
e
K
, (C.45)
where 0.577 is the Euler gamma-constant. One obtains:
Non-periodic : K
2

M
2
2.35991, K
3

M
3
5.105577, K
4

M
4
37.5258,
K
5

M
5
242.492,
M
2
3
M
3
2
1.98336; (C.46)
Periodic : K
2

M
per
2
1.64493, K
3

M
per
3
2.40411, K
4

M
per
4
14.6114,
K
5

M
per
5
64.4321; (C.47)
Gumbel : K
2

M
G
2
1.64493, K
3

M
G
3
2.40411, K
4

M
G
4
14.6114,
K
5

M
G
5
64.4321,
(

M
G
3
)
2
(

M
G
2
)
3
1.29857. (C.48)
Clearly, for periodic case the agreement with the Gumbel distribution is excellent,
but for non-periodic case, one can unambiguously conclude that limiting function is NOT the
Gumbel function. For non-periodic case the function is more widely distributed compared
to the Gumbel function, as evident from the higher moments. Here for completeness we
provide numerical results for Z
2n
, Z
per
2n
up to m = 5 :
Z
per
2n
= 1 +
1.8379n
K
+
0.4112n + 2.5114n
2
K
2
+
0.1002n 0.1548n
2
+ 2.1456n
3
K
3
+
0.03382n + 1.0804n
2
0.7399n
3
+ 1.7369n
4
K
4
+
0.01535n + 1.05941n
2
0.161727n
3
+ 0.172302n
4
+ 0.93098n
5
K
5
,
Z
2n
= 1 +
1.5n
K
+
0.554956n + 2.30496n
2
K
2
+
0.049657n + 0.343838n
2
+ 1.481505n
3
K
3
+
0.04741n + 2.0329n
2
1.8735n
3
+ 1.8256n
4
K
4
+
0.0106468n + 3.14868n
2
4.1829n
3
+ 2.8980n
4
+ 0.0943039n
5
K
5
.
C.4 D=2
Here we will briey consider the results of the expansion in powers of 1/K for 2-
dimensional case. The correlation function below the BKT transition is given by Eq. (5.77),
and for square imaging area with unity aspect ratio, all discussions of one dimensional case
carry over, with substitutions
_
1
0
du
i

_
1
0
_
1
0
d
2
u
i
,
_
1
0
dv
i

_
1
0
_
1
0
d
2
v
i
, G(x, y) = log [x y[, (C.49)
and (5.78). In 2D case, there is one extra degree of freedom which can be controlled in
experiments, which is the aspect ratio of the observation region. If the aspect ratio of the
observation region is very large, then the distribution function is essentially the same as
in the one-dimensional case. Below we concentrate on the case with unity aspect ratio.
Similarly to one-dimensional case, the integral
_
1
0
_
1
0
dx
1
dy
1
1
2
log
_
(x
1
x
2
)
2
+ (y
1
y
2
)
2
_
(C.50)
can be evaluated analytically, which somewhat simplies the numerical evaluation. However,
the dimensions of integrals grow fast with m, and here we present results only up to m = 3 :
Z
2n
= 1 +
0.805087n
K
+
0.0740n + 0.3342n
2
K
2
+
0.015n + 0.045n
2
+ 0.241n
3
K
3
.
The error in numerical coecient due to integration is of the order of the last reported digit.
Compared to 1D case, convergence is faster, which is consistent with our general ideology
that the role of uctuations is larger in systems with lower dimensions.
Bibliography
[1] J. R. Anglin and W. Ketterle, Nature 416, 211 (2002).
[2] M.P.A. Fisher et al., Phys. Rev. B 40, 546 (1989).
[3] D. Jaksch et. al., Phys. Rev. Lett. 81, 3108 (1998).
[4] M. Greiner et al., Nature 415, 39 (2002).
[5] M. H. Anderson, J. R. Ensher, M. R. Matthews, C. E. Wieman, and E. A. Cornell,
Science 269, 198 (1995).
[6] C.C. Bradley, C.A. Sackett, J.J. Tollett, and R.G. Hulet, Phys. Rev. Lett. 75, 1687
(1995).
[7] K.B. Davis, M.O. Mewes, M.R. Andrews, N.J. van Druten, D.S. Durfee, D.M. Kurn,
and W. Ketterle, Phys. Rev. Lett. 75, 3969 (1995).
[8] K. Burnett, P. S. Julienne, P. D. Lett, E. Tiesinga and C. J. Williams, Nature 416,
225 (2002).
[9] H. J. Metcalf and P. van der Straten, Laser Cooling and Trapping (Springer, 1999).
[10] L.D. Landau and E.M. Lifshitz, Statistical physics (Pergamon, London, 1958).
[11] L.D. Landau and E.M. Lifshitz, Quantum Mechanics (Pergamon, London, 1959).
[12] J. Weiner, V. S. Bagnato, S. Zilio, P. S. Julienne, Rev. Mod. Phys. 71, 1 (1999).
[13] E.P. Gross, Nuovo Cimento 20, 454 (1961).
[14] L.P. Pitaevskii, Sov. Phys. JETP 13, 451 (1961).
[15] M. R. Andrews et al., Science 275, 637 (1997).
[16] D.M. Stamper-Kurn et al., Phys. Rev. Lett. 80, 2027 (1998).
[17] J. Stenger et al., Nature 396, 345 (1998).
[18] B. DeMarco and D.S. Jin, Science 285, 1703 (1999).
[19] A.G. Truscott et al., Science 291, 2570 (2001).
177
178 Bibliography
[20] I. Bloch, T.W. Hansch and T. Esslinger, Nature 403, 166 (2000).
[21] J. R. Abo-Shaeer, C. Raman, J. M. Vogels, and W. Ketterle, Science 292, 476 (2001).
[22] K. W. Madison et al., Phys. Rev. Lett. 84, 806 (2000).
[23] W. Ketterle, D.S. Durfee, D.M. Stamper-Kurn, In Bose-Einstein condensation in
atomic gases, Proceedings of the International School of Physics Enrico Fermi,
Course CXL, edited by M. Inguscio, S. Stringari and C.E. Wieman (IOS Press, Ams-
terdam, 1999), pp. 67, cond-mat/9904034.
[24] F. Dalfovo, S. Giorgini, L. P. Pitaevskii, and S. Stringari, Rev. Mod. Phys. 71, 463
(1999).
[25] A.J. Leggett, Rev. Mod. Phys. 73, 307 (2001).
[26] C.J. Pethick and H. Smith, Bose-Einstein condensation in Dilute Gases (Cambridge
University Press, Cambridge, 2002).
[27] L. Pitaevskii and S. Stringari, Bose-Einstein Condensation (Clarendon Press, Oxford,
UK, 2003).
[28] S. Inouye et al., Nature 392, 151 (1998); P. Courteille et al., Phys. Rev. Lett. 81, 69
(1998); J.L. Roberts et al., Phys. Rev. Lett. 81, 5109 (1998).
[29] R. Grimm, M. Weidem uller, and Y.B. Ovchinnikov, Adv. At. Mol. Opt. Phys. 42, 95
(2000).
[30] C. Orzel et al., Science 291, 2386 (2001).
[31] J.I. Cirac and P. Zoller, Science 301, 176 (2003).
[32] I. Bloch, Nature Physics 1, 23 (2005).
[33] H. Feshbach, Ann. Phys. 5, 357 (1958).
[34] E. Tiesinga, B. J. Verhaar, and H. T. C. Stoof, Phys. Rev. A 47, 4114 (1993).
[35] E Timmermans, K. Furuya, P. W. Milonni and A. K. Kerman, Phys. Lett. A 285, 228
(2001).
[36] R.A. Duine and H.T.C. Stoof, Phys. Rep. 396, 115 (2004).
[37] T. Kohler, K. Goral, and P. S. Julienne, Rev. Mod. Phys. 78, 1311 (2006).
[38] D. Jaksch and P. Zoller, Annals of Physics 315, 52 (2005).
[39] M.Olshanii, Phys. Rev. Lett. 81, 938(1998).
[40] E.H. Lieb and W. Liniger, Phys. Rev. 130, 1605 (1963); E.H. Lieb, ibid. 130, 1616
(1963).
Bibliography 179
[41] M. Girardeau, J. Math. Phys. 1, 516 (1960).
[42] T. Kinoshita, T. Wenger and D.S. Weiss, Science, 305, 1125 (2004).
[43] B. Paredes et al., Nature 429, 277 (2004).
[44] N.K. Wilkin and J.M.F Gunn, Phys. Rev. Lett. 84, 6 (2000).
[45] V. Schweikhard, I. Coddington, P. Engels, V. P. Mogendor, and E. A. Cornell, Phys.
Rev. Lett. 92, 040404 (2004); V. Bretin, S. Stock, Y. Seurin, and J. Dalibard, Phys.
Rev. Lett. 92, 050403 (2004).
[46] K. M. OHara, S. L. Hemmer, S. R. Granade, M. E. Gehm, and J.E. Thomas, Science
298, 2179 (2002); C. A. Regal, M. Greiner, and D. S. Jin, Phys. Rev. Lett. 92, 040403
(2004); M. Bartenstein et al., ibid. 92, 120401 (2004); M. W. Zwierlein et al., ibid. 92,
120403 (2004); J. Kinast et al., ibid. 92, 150402 (2004); T. Bourdel et al., ibid. 93,
050401 (2004); G. B. Partridge et al., ibid. 95, 020404 (2005); T. Stoferle et al., ibid.
96, 030401 (2006).
[47] M. W. Zwierlein, A. Schirotzek, C. H. Schunck, and W. Ketterle, Science 311, 492
(2006).
[48] G. B. Partridge, W. Li, R. I. Kamar, Y. Liao, and R. G. Hulet,Science 311, 503 (2006).
[49] M. Lewenstein et al., cond-mat/0606771.
[50] I. Bloch, J. Dalibard, and W. Zwerger, arXiv:0704.3011v1.
[51] A. Auerbach, Interacting Electrons and Quantum Magnetism (Springer-Verlag, 1994).
[52] I. Aeck, Nuclear Physics B 265, 409 (1984).
[53] C. Nayak, K. Shtengel, Phys. Rev. B 64, 064422 (2001).
[54] S. A. Kivelson, Synthetic Metals 125, 99 (2002).
[55] D. A. Ivanov, T. Senthil, Phys. Rev. B 66, 115111 (2002).
[56] M. Levin, X.-G. Wen, Phys. Rev. B 67 ,245316 (2003).
[57] M. Hermele, M.P. A. Fisher, L. Balents, Phys. Rev. B 69, 064404 (2004).
[58] A. A. Nersesyan, A. M. Tsvelik, Phys. Rev. B 67, 024422 (2003).
[59] R. Moessner, S.L. Sondhi, E. Fradkin, Phys.Rev. B 65 024504 (2002).
[60] L. Balents, M. P. A. Fisher, S. M. Girvin, Phys. Rev. B 65, 224412 (2002).
[61] S. Rolston and W. Phillips, private communication.
[62] T. L. Ho, Phys. Rev. Lett. 81, 742 (1998).
180 Bibliography
[63] T. Ohmi and K. Machida, J. Phys. Soc. Jpn. 67, 1822 (1998).
[64] Y. Castin and C. Herzog, C.R. Acad. Sci. serie IV, 2, 419-443 (2001).
[65] T.L. Ho and S.K. Yip, Phys. Rev. Lett. 84, 4031 (2000).
[66] C.K.Law, H. Pu, and N.P. Bigelow, Phys.Rev.Lett. 81, 5257 (1998).
[67] W. V. Liu, X.G. Wen, cond-mat/0201187.
[68] A. B. Kuklov and B. V. Svistunov, Phys. Rev. Lett. 90, 100401 (2003).
[69] L.-M. Duan, E. Demler, M. D. Lukin, Phys. Rev. Lett. 91, 090402 (2003).
[70] E. Demler, F. Zhou, Phys.Rev.Lett. 88, 163001(2002).
[71] F. Zhou Int. Journ. Mod. Phys. B 17, 2643 (2003) and Europhys. Lett. 63, 505 (2003).
[72] S. K. Yip, Phys.Rev.Lett. 90, 250402 (2003).
[73] E. Demler et al., Phys. Rev. B 65, 155103 (2002).
[74] J.P.Burke, Jr., C.H. Greene, and J.L. Bohn, Phys. Rev. Lett. 81, 3355 (1998).
[75] S. Coleman, Commun. Math. Phys. 31, 259 (1973).
[76] B. A. Ivanov and A.K. Kolezhuk, Phys. Rev. B 68, 052401 (2003).
[77] A.V. Chubukov, J. Phys. Condens. Matter 2, 1593 (1990); A.V.Chubukov, Phys. Rev.
B 43, 3337(1991).
[78] G. Fath and J.Solyom, Phys.Rev.B 51, 3620 (1995).
[79] H.H. Chen and P.M. Levy, Phys.Rev. B 7, 4267 (1973).
[80] N.Papanicoulaou, Nucl. Phys. B305 [FS23], 367(1988).
[81] K. Harada and N. Kawashima, J. Phys. Soc. Jpn. 70, 13 (2001); K. Harada and
N.Kawashima, Phys.Rev.B 65, 052403 (2002).
[82] K. Tanaka, A.Tanaka and T. Idogaki, J. Phys. A: Math. Gen. 34, 8767 (2001).
[83] P.G. de Gennes and J. Prost, The Physics of Liquid Crystals(Oxford Press,1993).
[84] A.F. Andreev and I.A. Grischuk, Zh. Exp. Teor. Fiz. 87, 467 (1984).
[85] M. Greiner et.al., Nature 419, 51 (2002).
[86] E.Demler, F. Zhou, F.D.M. Haldane , Report No. ITP-UU-01/09 (2001).
[87] S. Sachdev, Quantum Phase Transitions (Cambridge university press, 1999).
[88] S. Sachdev and T. Senthil, Ann. Phys. 251, 76.
Bibliography 181
[89] S.Tsuchiya, S.Kurihara and T.Kimura, Phys. Rev. A bf 70, 043628 (2004).
[90] D.M. Stamper-Kurn et al., Phys.Rev.Lett. 83, 2876 (1999).
[91] C. Kollath et al., Phys. Rev. A 69, 031601 (2004); S.Bergkvist et al., Phys. Rev. A 70,
053601 (2004).
[92] J.W. Negele and H. Orland, Quantum many-particle systems (Addison-Wesley, 1988).
[93] F. Schreck et al., Phys. Rev. Lett. 87, 080403 (2001); G. Modugno et al., Science 297,
2240 (2002); Z. Hadzibabic et al., Phys. Rev. Lett. 88, 160401 (2002); G. Roati et al.,
Phys. Rev. Lett. 89, 150403 (2002); J. Goldwin et al., Phys. Rev. A 70, 021601(R)
(2004).
[94] A. Simoni et al., Phys. Rev. Lett. 90, 163202 (2003); S. Inouye et al., Phys. Rev. Lett.
93, 183201 (2004); F. Ferlaino et al., Phys. Rev. A 73, 040702(R) (2006).
[95] C.A. Stan et al., Phys. Rev. Lett. 93, 143001 (2004).
[96] H. Moritz et al., Phys. Rev. Lett. 94, 210401 (2005).
[97] A. Imambekov and E.Demler, Phys. Rev. A 73, 021602(R) (2006).
[98] K. Molmer, Phys. Rev. Lett. 80, 1804 (1998).
[99] L. Viverit, C. J. Pethick and H. Smith, Phys. Rev.A 61, 053605 (2000).
[100] H. Heiselberg et al., Phys. Rev. Lett. 85, 2418 (2000); M. J. Bijlsma, B. A. Heringa,
and H. T. C. Stoof, Phys. Rev. A 61, 053601 (2000); L. Viverit and S. Giorgini ,
Phys. Rev. A 66, 063604 (2002); A. Albus, F. Illuminati and J. Eisert, Phys. Rev. A
68, 023606 (2003); H. P. Buchler and G. Blatter, Phys. Rev. Lett. 91, 130404 (2003);
M. Lewenstein et al., Phys. Rev. Lett. 92, 050401 (2004); D.-W. Wang, M.Lukin and
E.Demler, Phys. Rev. A 72, 051604(R) (2005); A. Storozhenko et al., Phys. Rev. A
71,063617 (2005).
[101] K.K. Das, Phys. Rev. Lett. 90, 170403 (2003).
[102] M. A. Cazalilla and A. F. Ho, Phys. Rev. Lett. 91, 150403 (2003).
[103] Y. Takeuchi and H. Mori, cond-mat/0508247; cond-mat/0509048;
cond-mat/0509393.
[104] L. Mathey et al., Phys. Rev. Lett. 93, 120404 (2004).
[105] T. Miyakawa, H. Yabu and T. Suzuki, Phys. Rev. A 70, 013612 (2004); E. Nakano
and H.Yabu, Phys. Rev. A 72, 043602 (2005).
[106] H. Frahm and G. Palacios, Phys. Rev. A 72, 061604(R) (2005).
[107] C.K. Lai and C.N.Yang, Phys. Rev A 3, 393 (1971); C.K.Lai Journ. of Math. Phys.,
15, 954 (1974).
182 Bibliography
[108] M.T. Batchelor, M. Bortz, X.W. Guan, N. Oelkers, Phys. Rev. A 72, 061603(R)
(2005).
[109] C. N. Yang, Phys. Rev. Lett. 19, 1312 (1967).
[110] M. Gaudin, Phys. Lett. A 24, 55 (1967).
[111] M. Gaudin, La Fonction dOnde de Bethe (Paris, Masson, 1983); F. H. L. Essler et
al., The One-Dimensional Hubbard Model (Cambridge University Press, Cambridge,
2005).
[112] M. Takahashi, Thermodynamics of one-dimensional solvable models (Cambridge Uni-
versity Press, 1999).
[113] B. Sutherland, Beautiful models (World Scientic Publishing, 2004).
[114] A.M. Tsvelick and P.B. Wiegmann, Adv. Phys., 32, 453 (1983).
[115] N. Andrei, K. Furuya and J. H. Lowenstein, Rev. Mod. Phys. 55, 331 (1983).
[116] B. Sutherland, Phys. Rev. Lett. 20, 98 (1968).
[117] J.B.McGuire, J. Math. Phys. 6, 432 (1965); ibid, 7, 123 (1966).
[118] K. V. Kheruntsyan et al., Phys. Rev. A 71, 053615 (2005).
[119] H. Moritz et al., Phys. Rev. Lett. 91, 250402 (2003).
[120] M. Bartenstein et al., Phys. Rev. Lett. 92, 203201 (2004)
[121] G. E. Astrakharchik, cond-mat/0507711.
[122] S. Stringari, Phys. Rev. Lett.77, 2360 (1996).
[123] S Stringari and C. Menotti, Phys. Rev. A 66, 043610 (2002).
[124] G.E. Astrakharchik et al, Phys. Rev. Lett. 93, 050402 (2004).
[125] T. Miyakawa, T. Suzuki and H. Yabu, Phys. Rev. A 62, 063613 (2000).
[126] A. Lenard, J. Math. Phys. 5, 930 (1964).
[127] J.M.P. Carmelo, cond-mat/0405411.
[128] F. Woynarovich, J. Phys. C 15, 85 (1982).
[129] M. Ogata and H. Shiba, Phys. Rev. B 41, 2326 (1990).
[130] S. Richard et al., Phys. Rev. Lett. 91, 010405 (2003).
[131] A. O. Gogolin, A. A. Nersesyan and A. M. Tsvelik, Bosonization and strongly corre-
lated systems (Cambridge University Press, 1998).
Bibliography 183
[132] M. A. Cazalilla, Journal of Physics B: AMOP 37, S1-S47 (2004).
[133] V.V. Cheianov and M.B. Zvonarev, Phys. Rev. Lett. 93, 176401 (2004); V.V.
Cheianov and M.B. Zvonarev, J. Phys. A: Math. Gen. 37, 2261 (2004);
[134] V.V. Cheianov, H. Smith and M.B. Zvonarev, Phys.Rev. A 71, 033610 (2005).
[135] G.A. Fiete and L. Balents, Phys. Rev. Lett. 93, 226401 (2004); G.A. Fiete, K.L. Hur
and L. Balents, Phys. Rev. B 72, 125416 (2005).
[136] K.A. Matveev, Phys. Rev. Lett. 92, 106801 (2004); K.A. Matveev, Phys. Rev. B 70,
245319 (2004).
[137] V. I. Smirnov, A course of higher mathematics, Vol IV, p. 24 (Pergamon, Oxford,
1964).
[138] V.E. Korepin, N.M. Bogoliubov and A.G. Izergin, Quantum Inverse Scattering Method
and Correlation Functions (Cambridge University Press,Cambridge, England,1993).
[139] Izergin A G and Pronko A G Nucl. Phys. B 520, 594 (1998).
[140] M. Greiner et al., Phys. Rev. Lett. 87, 160405 (2001).
[141] W. Hansel et al., Nature 413, 498 (2001); H. Ott et al., Phys. Rev. Lett. 87 230401
(2001); S. Groth et al., Applied Physics Letters 85, 2980 (2004); J. Esteve et al., Phys.
Rev. A 70, 043629 (2004); S. Aubin et al., Journal of Low Temperature Physics 140,
377 (2005).
[142] T.P. Meyrath et al., Phys. Rev. A 71, 041604(R) (2005).
[143] R. Cote et al., Phys. Rev. A 57, R4118 (1998) .
[144] K. Honda et al., Phys. Rev. A 66, 021401(R) (2002); Y. Takasu et al., Phys. Rev.
Lett. 91, 040404 (2003); C. Y. Park and T. H. Yoon , Phys. Rev. A 68, 055401 (2003).
[145] J.P. Burke and J.L. Bohn, Phys. Rev. A 59, 1303(1999); S. G. Crane et al., Phys.
Rev. A62, 011402(R) (2000).
[146] G. Modugno et al., Science 294, 1320 (2001).
[147] A. Recati et al., Phys. Rev. Lett. 90, 020401 (2003).
[148] J.N. Fuchs et al., Phys. Rev. Lett. 95, 150402 (2005).
[149] F. Gerbier et al., Phys. Rev. A 67, 051602(R) (2003).
[150] P. F. Bedaque, H. Caldas, and G. Rupak, Phys. Rev. Lett. 91, 247002 (2003); A.
Sedrakian, J. Mur-Petit, A. Polls, and H. Muther, Phys. Rev. A 72, 013613 (2005);
C.-H. Pao and S.-T. Wu, S.-K. Yip, Phys. Rev. B 73, 132506 (2006); D.T. Son and
M. A. Stephanov, Phys. Rev. A 74, 013614 (2006); D. E. Sheehy and L. Radzihovsky,
Phys. Rev. Lett. 96, 060401 (2006); J. Dukelsky, G. Ortiz, S. M. A. Rombouts, and K.
184 Bibliography
Van Houcke, ibid. 96, 180404 (2006); Kun Yang, cond-mat/0508484; P. Pieri and G.C.
Strinati, Phys. Rev. Lett. 96, 150404 (2006); J. Kinnunen, L. M. Jensen, and P. Torma,
ibid. 96, 110403 (2006); W. Yi and L.-M. Duan, Phys. Rev. A 73, 031604(R) (2006);
F. Chevy, Phys. Rev. Lett. 96, 130401 (2006); M. Haque and H. T. C. Stoof, Phys.
Rev. A 74, 011602 (2006); H. Caldas, cond-mat/0601148; Z.-C. Gu, G. Warner, and F.
Zhou, cond-mat/0603091; X.-J. Liu and H. Hu, Europhys. Lett. 75, 364; H. Hu, X.-J.
Liu, Phys. Rev. A 73, 051603(R) (2006); M. Iskin and C. A. R. Sa de Melo, cond-
mat/0604184; K. Machida, T. Mizushima, M. Ichioka, Phys. Rev. Lett. 97, 120407
(2006).
[151] A. Bulgac, M.M. Forbes, and A. Schwenk, Phys. Rev. Lett. 97, 020402 (2006); D. V.
Efremov and L. Viverit, Phys. Rev.B 65, 134519 (2002).
[152] T. N. De Silva and E. J. Mueller, Phys. Rev. A 73, 051602(R) (2006).
[153] R. Hulet, Presentation at APS March Meeting, Baltimore, 15 March 2006.
[154] M. W. Zwierlein and W. Ketterle, cond-mat/0603489.
[155] G. V. Skorniakov and K. A. Ter-Martirosian, Sov. Phys. JETP 4, 648 (1957).
[156] D. S. Petrov, C. Salomon, and G. V. Shlyapnikov, Phys. Rev. Lett. 93, 090404 (2004).
[157] I. V. Brodsky et al., JETP Letters 82, 273 (2005); J. Levinsen, V. Gurarie, Phys.
Rev. A 73, 053607 (2006).
[158] A. Gorlitz et al., Phys. Rev. Lett. 87, 130402 (2001).
[159] M. Brack and R.K. Bhaduri, Semiclassical Physics, Frontiers in Physics Vol. 96
(Addison-Wesley, Reading, MA, 1997).
[160] G. E. Astrakharchik, J. Boronat, J. Casulleras, and S. Giorgini, Phys. Rev. Lett. 93,
200404 (2004).
[161] E. J. Mueller, Phys. Rev. Lett. 93, 190404 (2004).
[162] F. Dalfovo and S. Stringari, Phys. Rev. A 53, 2477 (1996); for each point we use
about 1000 iterations for functions dened on a 100 100 grid.
[163] A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47, 777 (1935)
[164] C. Cohen-Tannoudji, B. Diu, F. Laloe, Quantum Mechanics (Wiley-Interscience,
2006).
[165] L. S. Levitov, in Quantum Noise in Mesoscopic Systems, edited by Yu. V. Nazarov
(Kluwer, 2003).
[166] Ya. M. Blanter and M. B uttiker, Phys. Rep. 336, 2 (2000).
[167] C. W. J. Beenakker and C. Schonenberger, Physics Today 56, 37 (2003).
Bibliography 185
[168] L. Saminadayar et al., Phys. Rev. Lett. 79, 2526 (1997); R. De-Picciotto et al., Nature
389, 162 (1997).
[169] R. Hanbury Brown and R. Q. Twiss, Nature 177, 27 (1956).
[170] R. J. Glauber, Phys. Rev. 130, 2529 (1963).
[171] J. R. Klauder, E. C. G. Sudarshan, Fundamentals of Quantum Optics (W. A. Ben-
jamin, N.Y. 1968).
[172] L. Mandel, E. Wolf, Optical Coherence and Quantum Optics (Cambridge University
Press, Cambridge, UK, 1995).
[173] D.F. Walls and G.J. Milburn, Quantum Optics (Springer, 1995).
[174] M. O. Scully and M. S. Zubairy, Quantum Optics (Cambridge University Press, Cam-
bridge, UK, 1997).
[175] M. Greiner, C. A. Regal, J. T. Stewart, and D. S. Jin, Phys. Rev. Lett. 94, 110401
(2005).
[176] S. Folling, F. Gerbier, A. Widera, O. Mandel, T. Gericke, I. Bloch, Nature 434, 481
(2005).
[177] I. B. Spielman, W. D. Phillips and J. V. Porto, Phys. Rev. Lett. 98, 080404 (2007)
[178] T. Rom et al., Nature 444, 733(2006).
[179] A.

Ottl, S. Ritter, M. Kohl, T. Esslinger, Phys. Rev. Lett. 95, 090404 (2005).
[180] M. Schellekens et al., Science 310, 648 (2005).
[181] T. Jeltes et al., Nature 445, 402 (2007).
[182] E. Altman, E. Demler, and M. D. Lukin, Phys. Rev. A 70, 013603 (2004).
[183] R. W. Cherng and E. Demler, New J. Phys. 9, 7 (2007).
[184] P. Nagornykh and V. Galitski, cond-mat/0612376.
[185] W. Belzig, C. Schroll and C. Bruder, cond-mat/0412269.
[186] A. Lamacraft, cond-mat/0512580.
[187] A. Kuklov and H. Moritz, Phys. Rev. A 75, 013616 (2007).
[188] C. Menotti, C. Trefzger and M.Lewenstein, cond-mat/0612498.
[189] M. Hugbart et al., Eur. Phys. J. D 35,155 (2005).
[190] T. Schumm et al., Nature Physics 1, 57 (2005).
[191] S. Hoerberth et al., Nature Physics 2, 710 (2006).
186 Bibliography
[192] S. Stock et al., Phys. Rev. Lett. 95, 190403 (2005).
[193] Z. Hadzibabic et al., Nature 441, 1118 (2006).
[194] Z. Hadzibabic et al., cond-mat/0609761.
[195] T. Gustavson, P. Bouyer, and M. Kasevich, Phys. Rev. Lett. 78, 2046 (1997).
[196] B. P. Anderson and M. A. Kasevich, Science 282, 1686 (1998).
[197] A. Peters, K. Chung, and S. Chu, Nature 400, 849 (1999).
[198] S. Fray, C. A. Diez, T. W. Hansch, and M. Weitz, Phys. Rev. Lett. 93, 240404 (2004).
[199] P. Clade et al., Phys. Rev. Lett. 96, 033001 (2006).
[200] G. Ferrari, N. Poli, F. Sorrentino, and G. M. Tino, Phys. Rev. Lett. 97, 060402 (2006).
[201] E.A. Hinds, C.J. Vale and M.G. Boshier, Phys. Rev. Lett. 86, 1462 (2001).
[202] W. Hansel, J. Reichel, P.Hommelho and T.W. Hansch, Phys. Rev. A 64, 063607
(2001).
[203] E. Andersson et al., Phys. Rev. Lett. 88, 100401 (2002).
[204] R. Dumke et al., Phys. Rev. Lett. 89, 220402 (2002).
[205] H. Kreutzmann et al., Phys. Rev. Lett. 92, 163201 (2004).
[206] Y. Shin et al., Phys. Rev. Lett. 92, 050405 (2004).
[207] Y. Shin et al., Phys. Rev. A 72, 021604 (2005).
[208] Y.-J. Wang et al., Phys. Rev. Lett. 94, 090405 (2005).
[209] G.-B. Jo et al., Phys. Rev. Lett. 98, 030407 (2007).
[210] G.-B. Jo et al., cond-mat/0703006.
[211] J. Fortagh and C. Zimmermann, Rev. Mod. Phys. 79, 235 (2007).
[212] P. R. Berman (editor), Atom Interferometry (Academic Press, 1997).
[213] S. Ritter et al., Phys. Rev. Lett. 98, 090402 (2007).
[214] T. Donner et al., Science 315 (5818), 1556 (2007).
[215] A. Polkovnikov, E. Altman and E. Demler, Proc. Natl. Acad. Sci. USA 103, 6125
(2006).
[216] V. Gritsev, E. Altman, E. Demler and A. Polkovnikov, Nature Physics 2 , 705 (2006).
[217] V. Gritsev, E. Altman, A. Polkovnikov and E. Demler, AIP Conference Proceedings
869, 173 (2006).
Bibliography 187
[218] A. Imambekov, V. Gritsev and E. Demler, cond-mat/0612011.
[219] H. Xiong et al., New J. Phys. 8, 245 (2006).
[220] L. S. Cederbaum, A. I. Streltsov, Y. B. Band, and O. E. Alon, cond-mat/0701277; L.
S. Cederbaum et al., cond-mat/0607556.
[221] D. Masiello and W. P. Reinhardt, cond-mat/0702067.
[222] R. J. Glauber, in Quantum Optics and Electronics, Les Houches Summer School
Lectures, edited by C. DeWitt, A. Blandin, and C. Cohen-Tannoudji (Gordon and
Breach, New York, 1965).
[223] J. Javanainen and S.M. Yoo, Phys. Rev. Lett. 76, 161 (1996).
[224] Y. Castin and J. Dalibard, Phys. Rev. A 55, 4330 (1997).
[225] J. I. Cirac, C. W. Gardiner, M. Naraschewski and P. Zoller, Phys. Rev. A 54, R3714
(1996).
[226] Z. Hadzibabic, S. Stock, B. Battelier, V. Bretin, and J. Dalibard, Phys. Rev. Lett.
93, 180403 (2004).
[227] G. Baym, Acta Phys. Pol. B 29, 1839 (1998).
[228] E. W. Hagley et al., Phys. Rev. Lett. 83, 3112 (1999).
[229] J. E. Simsarian et al., Phys. Rev. Lett. 85, 2040(2000).
[230] D. Hellweg et al., Phys. Rev. Lett. 91, 010406 (2003).
[231] L. Cacciapuoti et al., Phys. Rev. A 68, 053612 (2003).
[232] K. Bongs and K. Sengstock, Rep. Prog. Phys. 67, 907 (2004).
[233] D. E. Miller et al., Phys. Rev. A 71, 043615 (2005).
[234] M. Hugbart et al., Eur. Phys. J. D 35, 155 (2005).
[235] Y. Qu, S. Singh, C. D. Cantrell, Phys. Rev. Lett. 76, 1236 (1996).
[236] P. Kinsler, Phys. Rev. A 53, 2000 (1996).
[237] G. Magyar and L. Mandel, Nature 198, 255 (1963).
[238] L. Mandel, Rev. Mod. Phys. 71, S274 (1999).
[239] Jan Perina, Quantum Statistics of Linear and Nonlinear Optical Phenomena (D. Rei-
del Publishing Company, 1984).
[240] S. Gustavsson, R. Leturcq, B. Simovic, R. Schleser, T. Ihn, P. Studerus, K. Ensslin,
D. C. Driscoll, A. C. Gossard, Phys. Rev. Lett. 96, 076605 (2006); S. Gustavsson, et.al.
Phys. Rev. B, 74, 195305 (2006).
188 Bibliography
[241] H.-S. Sim and E. V. Sukhorukov, Phys. Rev. Lett. 96, 020407 (2006).
[242] E. Grosfeld, S. H. Simon and A. Stern, cond-mat/0602634.
[243] V. L. Berezinskii, Sov. Phys. JETP 32, 493 (1971); 34, 610 (1972).
[244] J. M. Kosterlitz and D. J. Thouless, J. Phys. C 6, 1181 (1973).
[245] F. Schreck et al., Phys. Rev. Lett. 87, 080403 (2001).
[246] S. Burger Europhys. Lett. 57, 1 (2002); D. Rychtarik, B. Engeser, H.-C. Nagerl, and
R. Grimm, Phys. Rev. Lett. 92, 173003 (2004); N. L. Smith et al., J. Phys. B 38, 223
(2005).
[247] N.D. Mermin and H. Wagner, Phys. Rev. Lett. 17, 1133 (1966).
[248] P. C. Hohenberg, Phys. Rev. 158, 383 (1967).
[249] T. Wong, M. J. Collett, and D. F. Walls, Phys. Rev. A 54, R3718 (1996).
[250] M. Naraschewski et al., Phys. Rev. A 54, 2185 (1996).
[251] Rohrl et al., Phys. Rev. Lett. 78, 4143 (1997).
[252] P. Horak and S.M. Barnett, J. Phys. B 32, 3421 (1999).
[253] K. Mlmer, Phys. Rev. A 65, 021607 (2002).
[254] W. J. Mullin, R. Krotkov and F. Laloe, Am. J. Phys. 74, 880 (2006); cond-
mat/0605038.
[255] A. Polkovnikov, Europhys. Lett. 78, 10006 (2007).
[256] C. L. Kane and M. P. A. Fisher, Phys. Rev. B 46, 15233 (1992).
[257] T. Antal, M. Droz, G. Gyorgyi, and Z. Racz, Phys. Rev. Lett. 87, 240601 (2001);
Phys. Rev. E, 65, 046140 (2002).
[258] A. A. Abrikosov, L. P. Gorkov, I. E. Dzyaloshinski, Methods of Quantum Field Theory
in Statistical Physics, edited by R. A. Silverman (Prentice-Hall, 1963).
[259] R. Gati et al., Appl. Phys. B 82, 207 (2006).
[260] F. Gerbier, S. Folling, A. Widera, and I. Bloch, cond-mat/0701420.
[261] A. D. Polyanin and A. V. Manzhirov, Handbook of Integral Equations (CRC Press,
Boca Raton, 1998).
[262] G. A. Korn and T. E. Korn, Mathematical Handbook for Scientists and Engineers:
Denitions, Theorems, and Formulas for Reference and Review (McGraw-Hill, 1961).
[263] P. Blasiak, quant-ph/0507206.
Bibliography 189
[264] S. Gupta, K.W. Murch, K.L. Moore, T.P. Purdy, and D.M. Stamper-Kurn, Phys.
Rev. Lett. 95, 143201 (2005).
[265] F. D. M. Haldane, Phys. Rev. Lett. 47, 1840 (1981).
[266] J.-S. Caux, P. Calabrese, and N. A. Slavnov, J. Stat. Mech. 0701 (2007) P008.
[267] D. S. Petrov, M. Holzmann, and G. V. Shlyapnikov, Phys. Rev. Lett. 84, 2551 (2000).
[268] D. S. Petrov, D. M. Gangardt, and G. V. Shlyapnikov, J. Phys. IV France 116, 3
(2004).
University Press, 1998).
[269] D. J. Bishop and J. D. Reppy, Phys. Rev. Lett. 40, 1727(1978)
[270] M. Holzmann, G. Baym, J.-P. Blaizot, and F. Laloe cond-mat/0508131.
[271] A. Posazhennikova, Rev. Mod. Phys. 78, 1111 (2006).
[272] P. Fendley, F. Lesage and H. Saleur, J. of Stat. Phys. 79, 799 (1995).
[273] R. Konik and A. LeClair, Nucl. Phys. B479 619 (1996).
[274] I. Aeck, W. Hofstetter, D. R. Nelson, U. Schollwock, Europhys. Lett. 66, 178 (2004);
J.Stat.Mech. 0410, P003(2004).
[275] E. Witten, Phys. Rev. D 47, 3405 (1993).
[276] E. J. Gumbel, Statistics of the extremes (Columbia University Press, 1958); J. Galam-
bos, The assymptotic theory of extreme value statistics (R. E. Krieger Publ. Co.,
Malabar, Florida, 1987).
[277] G. Yuval and P. W. Anderson, Phys. Rev. B 1, 1522 (1970).
[278] H. Saleur, cond-mat/9812110 and cond-mat/0007309.
[279] A. Sen, Int. J. Mod. Phys. A 20, 5513 (2005).
[280] L. Vinet and J. F. van Diejen (editors), Calogero-Moser-Sutherland Models (Springer,
1998).
[281] A. O. Caldeira and A. J. Leggett, Phys. Rev. Lett. 46, 211 (1981); Physica A 121,
587 (1983).
[282] P. Fendley, H. Saleur and N.P. Warner, Nucl. Phys. B430, 577 (1994).
[283] P. Erdos and J. Lehner, Duke Math. J. 8, 335 (1941).
[284] M. Kardar, G. Parisi and Y.C. Zhang, Phys. Rev. Lett. 56, 889 (1986).
[285] A. Comtet, P. Leboeuf and S. N. Majumdar, cond-mat/0610411.
190 Bibliography
[286] E. Bertin, Phys. Rev. Lett. 95, 170601 (2005); E. Bertin and M. Clusel, Journal of
Physics A 39, 7607 (2006).
[287] M. Kendall, A. Stuart, J. Keith Ord, Kendalls Advanced Theory of Statistics, Volume
1: Distribution Theory (A Hodder Arnold Publication, 1994).
[288] K. V. Kheruntsyan, D. M. Gangardt, P. D. Drummond, G. V. Shlyapnikov, Phys.
Rev. A 71, 053615 (2005).
[289] V. Dunjko, V. Lorent and M. Olshanii, Phys. Rev. Lett. 86, 5413 (2001).
[290] A. Minguzzi, D.M. Gangardt, Phys. Rev. Lett. 94, 240404 (2005).
[291] R. Bistritzer and E. Altman, cond-mat/0609147.
[292] A. A. Burkov, M. D. Lukin and E. Demler, cond-mat/0701058.
[293] A. Mebrahtu, A. Sanpera, M. Lewenstein, Phys. Rev. A 73, 033601 (2006).
[294] M. A. Cazalilla, Phys. Rev. Lett. 97, 156403 (2006).
[295] V. Gritsev, A. Polkovnikov, E. Demler, cond-mat/0701421.
[296] V. Gritsev, E. Demler, M. Lukin, A. Polkovnikov, cond-mat/0702343.
[297] I. Bloch, private communication.
[298] T. Porto, private communication.
[299] G. Roati et al., Phys. Rev. Lett. 92, 230402 (2004); G. Modugno et al., Fortschr.
Phys. 52, 1173 (2004).
[300] K. G unter et al., Phys. Rev. Lett. 95, 230401 (2005).
[301] G. Modugno et al., Phys. Rev. A 68, 011601(R) (2003).
[302] M. Henny et al., Science 284, 296 (1999).
[303] W. D. Oliver, J. Kim, R. C. Liu, Y. Yamamoto, Science 284, 299 (1999).

Adilet Imambekov - Strongly Correlated Phenomena With Ultracold Atomic Gases

Încărcat de

Informații document

Descriere originală:

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Adilet Imambekov - Strongly Correlated Phenomena With Ultracold Atomic Gases

Încărcat de

Drepturi de autor:

Formate disponibile

Strongly Correlated Phenomena with Ultracold Atomic

are the usual spin-one operators). The rst term in Eq.

d are equivalent) and corresponds to the

and its eigenvalues. Energies of the states

2 sin 4). (2.43)

0.25, see Fig. 2.5. As

S with zero eigenvalue. Vector

Figure 2.5: Dependence of the energy functional (2.43) on when zt

. We note, however, that the dierence between dierent

Figure 2.6: Dependence of the energy functional (2.43) on when zt

Figure 2.7: Dependence of the energy functional (2.43) on when zt

for spin S = 2 states and similar

c = 1 state [80], which is given by

S) = 0, but there is a nematic order (2.27) when ,= 0. For example, when

k[, and dependence of the spin wave velocity

since at this point uctuations of are not suppressed.

d is directing along the z axis and

H it is again zero and

to show that it acts on the wave function . Therefore,

as products of two operators, which act in

cos (qr t). Here we introduce spin in-

has the form V

interaction wont produce any Bragg peaks for

are two permutations, such that Q

are two permutations, such that

]. Using these two conditions, we can express [Q, P

, P]. These requirements can be simply written as a condition between

are two permutations

operators are dened as

= (M 1)/2, (M 3)/2, ...., (M 1)/2, (3.33)

Figure 3.1: Energy of the ground state is given by E = e(, ) h

) + 8(1 )(1 + cos )

"out of phase" dipole mode

6 are shown in Fig. 3.4. As interactions get stronger, bo-

0.1 0.2 0.3 0.4 0.5 0.6

0.1 0.2 0.3 0.4 0.5 0.6

0.1 0.2 0.3 0.4 0.5 0.6

0.2 0.4 0.6 0.8 1

, c in (3.28)-(3.29). A simplied system of equations for spectral parameters is

occupying Fermi sea (, ), so from (3.91) ground state spin

, then using (3.102) one can conclude that

is obtained from Z by removing z

we can go to summation over P, Q, where P

= QP. Also, one

according to (3.105) can be rewritten as

we can go to summation over P, Q, where P

= QP. Also, one

and y are related by

we can go to summation over P, Q, where P

= QP. Also, one

we can go to summation over P, Q, where P

= QP. Also, one

we can go to summation over P, Q, where P

= QP. Also , one

/2 = 350 Hz, scattering length corresponding to

/2 = 350 Hz, a scattering length that corresponds to B =

) within LDA (dotted) and in the

, and a decrease of the aspect ratio of minority species

is the radial level spacing.

(where R is the dierence in axial radii). It is thus clear that

measure the intensity of

are positions of the detectors, and I(r), I(r