Ultracold bosonic atoms
in optical lattices
by
Ana Maria Rey
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland at College Park in partial fulﬁllment
of the requirements for the degree of
Doctor of Philosophy
2004
Advisory Committee:
Prof. Theodore R. Kirkpatrick, Chairman/Advisor
Dr. Charles W. Clark, Advisor
Prof. BeiLok Hu,
Prof. Steve Rolston,
Prof. J. Robert Dorfman
There are men who ﬁght one day and are good.
There are men who ﬁght one year and are better.
There are some who ﬁght many years and they are better still.
But there are some that ﬁght their whole lives,
these are the ones that are indispensable.
Berthold Brecht
Abstract
Title of Dissertation: Ultracold bosonic atoms
in optical lattices
Ana Maria Rey, Doctor of Philosophy, 2004
Dissertation directed by: Prof. Theodore R. Kirkpatrick
Physics
Dr. Charles W. Clark
National Institute of Standards and Technology
This thesis covers most of my work in the ﬁeld of ultracold atoms loaded in
optical lattices. It makes a route through the physics of cold atoms in periodic
potentials starting from the simple noninteracting system and going into the many
body physics that describes the strongly correlated Mott insulator regime. Even
though this thesis is a theoretical work all the chapters are linked either with
experiments already done or with ongoing experimental eﬀorts.
This thesis can be divided into four diﬀerent parts. The ﬁrst part comprises
chapters 1 to 3. In these chapters, after a brief introduction to the ﬁeld of optical
lattices I review the fundamental aspects pertaining to the physics of systems in
periodic potentials.
The second part deals with the superﬂuid weakly interacting regime where
standard mean ﬁeld techniques can be applied. This is covered in chapters 4 and
5. Speciﬁcally, chapter 4 introduces the discrete nonlinear Schr¨odinger equation
(DNLSE) and uses it to model some experiments. In chapter 5 I go one step
further and include the small quantum ﬂuctuations neglected in the DNLSE by
studying quadratic approximations of the BoseHubbard Hamiltonian.
Chapters 6 to 8 can be grouped as the third part of the thesis. In them I
adopt an eﬀective action formalism, the so called two particle irreducible eﬀective
action (2PI) together with the closed time path (CTP) formalism to study far
fromequilibrium dynamics. The manybody techniques discussed in these chap
ters systematically include higher order quantum corrections, not included in the
quadratic approximations of the Hamiltonian, which we show are crucial for a
correct description of the quantum dynamics outside the very weakly interacting
regime.
Finally, chapter 9 to 11 are devoted to study the Mott insulator phase. In
these chapters using perturbation theory I study the Mott insulator ground state
and its excitation spectrum, the response of the system to Bragg spectroscopy, and
propose a mechanism to correct for the residual quantum coherences inherent to
the Mott insulator ground state. Even though small these are not ideal for the use
of neutral atoms in optical lattice as a tool for quantum computation.
Acknowledgements
I want to use this lines to thank and acknowledge all the people that made this
thesis possible.
First and foremost I would like to thank Charles Clark for giving me the in
valuable opportunity to join the NIST group. Through his intelligent ideas, love
for physics and dedication he gave me not only very good advice and guidance
but also all the encouragement I needed to feel the work I was doing was worth
while and important. It has been a pleasure to work with and learn from such an
extraordinary person.
I also want to thank Prof. Ted Kirkpatrick for serving as my University of
Maryland supervisor. He has always made himself available for help and advice.
His ideas and suggestions were really valuable for the development of this thesis.
I am also grateful to Prof. Bei Lok Hu. He was an invaluable source of knowl
edge and assistance, especially with regard to the 2PICTP formalism discussed
in chapters 6 to 8. There has never been an occasion when I’ve knocked on his
door and he hasn’t given me time. I cannot leave out Esteban Calzetta and Albert
Roura. Discussions with Bei Lok, Esteban and Albert regarding the application
of the 2PICTP formalism to the patterned loaded system made possible the pub
lication of the paper.
I would also like to acknowledge help and support from Carl Williams. He was
my other advisor at NIST, willing to help me all the time and providing me with
very useful ideas.
I also owe special gratitude to everybody in the Physics Laboratory at NIST
and colleagues in the Radiation Physics Building, in particular Blair Blakie, Zach
Dutton, Nicolai Nygaard, Marianna Safranova, Jamie Williams and Mark Ed
wards. Besides colleagues they have been wonderful friends and helped me any
time I needed help or advice. Especially I am greatly indebted with Blair who
was not only my oﬃce mate when he was at NIST and helped me to sort out all
the problems but also for his support and great ideas even now that he is in New
Zealand. I also want to thank Nicolai not only for providing me all the LaTex
style ﬁles for writing this thesis but also for all his time reviewing the manuscript.
Guido Pupillo and Gavin Brennen also deserve a special mention. Because of their
great ideas and critical thinking the work described in chapters 9 to 11 was possi
ble. I can not forget to thank Prof. Keith Burnett. I worked with him during the
summer of 2002 when he came to NIST. Besides an incredible person, he is also
a bright physicist. He was the one who gave me the idea of using the superﬂuid
viii
properties to study the approach to the Mott insulator phase.
I owe my deepest thanks to my family  my mother, my father, my sister and
my brother. They have always stood by me in good and bad times. I really owe
all my achievements to them.
Finally, I have no words to express the gratitude I have to my husband Juan
Gabriel. He is the most important person in my life. We came together to pursue
the PhD. program and without his love and support the completion of this work
would not have been possible.
Table of Contents
List of Figures xii
1 Introduction 1
2 Optical lattices 9
2.1 Basic theory of optical lattices . . . . . . . . . . . . . . . . . . . . 9
2.1.1 AC Stark Shift . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.2 Dissipative interaction . . . . . . . . . . . . . . . . . . . . . 11
2.1.3 Lattice geometry . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Single particle physics . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1 Bloch functions . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.2 Wannier orbitals . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.3 Tightbinding approximation . . . . . . . . . . . . . . . . . 17
2.2.4 Semiclassical dynamics . . . . . . . . . . . . . . . . . . . . . 19
3 The BoseHubbard Hamiltonian and the superﬂuid to Mott insu
lator transition 25
3.1 BoseHubbard Hamiltonian . . . . . . . . . . . . . . . . . . . . . . 25
3.2 SuperﬂuidMott insulator transition . . . . . . . . . . . . . . . . . 27
4 Mean ﬁeld theory 31
4.1 Discrete nonlinear Schr¨odinger equation (DNLSE) . . . . . . . . . 31
4.2 Dragging experiment . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.2.1 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.2.2 Linear free particle model . . . . . . . . . . . . . . . . . . . 35
4.2.3 Eﬀect of the magnetic conﬁnement . . . . . . . . . . . . . . 36
4.2.4 Interaction eﬀects . . . . . . . . . . . . . . . . . . . . . . . 38
4.3 Dynamics of a periodthree patternloaded BoseEinstein conden
sate in an optical lattice . . . . . . . . . . . . . . . . . . . . . . . . 40
4.3.1 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3.2 Case of no external potential . . . . . . . . . . . . . . . . . 43
4.3.3 Dynamics with a constant external force . . . . . . . . . . . 51
x TABLE OF CONTENTS
5 Quadratic approximations 61
5.1 The characteristic Hamiltonian . . . . . . . . . . . . . . . . . . . . 62
5.2 Bogoliubovde Gennes (BdG) equations . . . . . . . . . . . . . . . 62
5.3 Higher order terms . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.3.1 HartreeFockBogoliubov equations . . . . . . . . . . . . . . 66
5.3.2 HFBPopov approximation . . . . . . . . . . . . . . . . . . 67
5.4 The BoseHubbard model and superﬂuidity . . . . . . . . . . . . . 67
5.5 Expectation values . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.6 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.6.1 Translationally invariant lattice . . . . . . . . . . . . . . . . 71
5.6.2 Onedimensional harmonic trap plus lattice . . . . . . . . . 78
5.7 Improved HFBPopov approximation . . . . . . . . . . . . . . . . . 91
5.7.1 The twobody and manybody scattering matrices . . . . . 92
5.7.2 The anomalous average and the manybody scattering matrix 93
5.7.3 Improved Popov approximation . . . . . . . . . . . . . . . . 95
5.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6 The two particle irreducible eﬀective action (2PI) and the closed
time path (CTP) formalism 101
6.1 2PI eﬀective action Γ(z, G) . . . . . . . . . . . . . . . . . . . . . . 102
6.2 Perturbative expansion of Γ
2
(z, G) and approximation schemes . . 107
6.2.1 The standard approaches . . . . . . . . . . . . . . . . . . . 107
6.2.2 Higher order expansions . . . . . . . . . . . . . . . . . . . . 108
6.2.3 Zero mode ﬂuctuactions . . . . . . . . . . . . . . . . . . . . 109
6.3 CTP formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.4 HFB approximation . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.4.1 Equations of motion . . . . . . . . . . . . . . . . . . . . . . 113
6.4.2 Mode expansion of the HFB equations . . . . . . . . . . . . 114
6.5 Secondorder expansion . . . . . . . . . . . . . . . . . . . . . . . . 116
6.5.1 Equations of motion . . . . . . . . . . . . . . . . . . . . . . 116
6.5.2 Conservation laws . . . . . . . . . . . . . . . . . . . . . . . 123
6.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7 Nonequilibrium dynamics of a patterned loaded optical lattice:
Beyond the mean ﬁeld approximation 127
7.1 Mean ﬁeld dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 127
7.1.1 Dynamical evolution . . . . . . . . . . . . . . . . . . . . . . 128
7.1.2 Comparisons with the exact solution . . . . . . . . . . . . . 128
7.2 2PICTP approximations . . . . . . . . . . . . . . . . . . . . . . . 131
7.2.1 Initial conditions and parameters . . . . . . . . . . . . . . 131
7.2.2 Numerical algorithm for the approximate solution . . . . . 132
7.2.3 Results and discussions . . . . . . . . . . . . . . . . . . . . 133
7.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
TABLE OF CONTENTS xi
8 From the 2PICTP approximations to kinetic theories and local
equilibrium solutions 143
8.1 Rewriting the 2PICTP secondorder equations . . . . . . . . . . . 144
8.2 Boltzmann equations . . . . . . . . . . . . . . . . . . . . . . . . . . 147
8.3 Equilibrium properties for a homogeneous system . . . . . . . . . . 152
8.3.1 Quasiparticle formalism . . . . . . . . . . . . . . . . . . . . 153
8.3.2 HFB approximation . . . . . . . . . . . . . . . . . . . . . . 155
8.3.3 Secondorder and Beliaev approximations . . . . . . . . . . 156
8.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
9 Characterizing the Mott Insulator Phase 159
9.1 Commensurate translationally invariant case . . . . . . . . . . . . . 159
9.1.1 Perturbation theory . . . . . . . . . . . . . . . . . . . . . . 159
9.2 Harmonic conﬁnement plus lattice . . . . . . . . . . . . . . . . . . 167
9.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
10 Bragg Spectroscopy 173
10.1 Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
10.2 Observables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
10.3 Linear response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
10.4 Zerotemperature regime . . . . . . . . . . . . . . . . . . . . . . . . 179
10.4.1 Bogoliubov approach . . . . . . . . . . . . . . . . . . . . . . 179
10.4.2 Inhomogeneous system . . . . . . . . . . . . . . . . . . . . . 181
10.5 Mott insulator regime . . . . . . . . . . . . . . . . . . . . . . . . . 183
10.5.1 Commensurate homogeneous system . . . . . . . . . . . . . 184
10.5.2 Inhomogeneous system . . . . . . . . . . . . . . . . . . . . . 186
10.6 Finite temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
10.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
11 Scalable register initialization for quantum computing in an op
tical lattice 193
11.1 Homogeneous dynamics . . . . . . . . . . . . . . . . . . . . . . . . 194
11.2 Dynamics in presence of the external trap . . . . . . . . . . . . . . 197
11.3 Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
11.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
12 Conlusions 207
xii TABLE OF CONTENTS
List of Figures
2.1 AC Stark shift induced by atomlight interaction. The laser fre
quency is ω = 2πν which is detuned from the atomic resonance by
Λ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Optical lattice potential.(a)(e) potentials for diﬀerent conﬁgura
tions of 3 beams, (f) potential for the 4 counterpropagating laser
beam conﬁguration (The two pair of light ﬁelds are made indepen
dent by detuning the common frequency of one pair of beams from
the other. E
R
is the atomic recoil energy, E
R
=
2
k
2
/2m. This
ﬁgure is a courtesy of P. Blair Blakie [34] ) . . . . . . . . . . . . . 14
2.3 Band structure of an optical lattice . . . . . . . . . . . . . . . . . . 16
2.4 Probability density of the Bloch wave function φ
(n=1)
q=0
for diﬀerent
lattice depths. It can be observed the localization of the wave func
tion increases with the lattice increases. . . . . . . . . . . . . . . . 17
2.5 Wannier state logdensity distribution corresponding to the lattice
conﬁgurations shown in Fig. 2.2. Dotted white lines indicate the
potential energy contours. The numbers labels the near neighbor
sites: 1 the nearest neighbors, and so on. This ﬁgure is a courtesy
of P. Blair Blakie [34] ) . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Ground state amplitudes f
(0)
i
as a function of the lattice site i for
diﬀerent values of ξ . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.7 Diﬀusion of a square wave packet in the presence of a linear poten
tial.The parameters for the plot are ξ = 0.1. . . . . . . . . . . . . . 21
2.8 Harmonic conﬁnement plus lattice spectrum: The index n labels
the eigenvalues in increasing energy order. . . . . . . . . . . . . . 24
3.1 Zerotemperature phase diagram. The vertical lines indicate critical
values for diﬀerent ﬁlling factors . . . . . . . . . . . . . . . . . . . 30
4.1 Condensate density in the xdirection at the observation time t
f
.
Light colors areas indicate regions of high condensate density. The
dashed line indicates the ﬁnal location of a comoving point with
the optical lattice. . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
xiv LIST OF FIGURES
4.2 Atom’s average velocity in the lab frame vs. lattice velocity. Line:
results from the Bloch dispersion relation, dots: tightbinding ap
proximation results. v
B
= π/am. . . . . . . . . . . . . . . . . . . 36
4.3 Atom’s average velocity in the lab frame vs. lattice velocity in the
presence of a magnetic trap. v
B
= π/am. . . . . . . . . . . . . . . 37
4.4 Velocity of the atoms predicted by the variational model in the
case when interactions are taken into account (dotted line). The
single particle case is shown in red. The parameters used were
γ(0) = 10, δ(0) = 0 and NU = 2E
R
. . . . . . . . . . . . . . . . . . 39
4.5 Phase diagram UN vs v
latt
for V
o
= 3E
r
and γ
o
= 10. . . . . . . . 40
4.6 Localization of the wave packet in the selffocusing regime v
latt
=
−0.7v
B
. Notice the change in the vertical scale in the last row . . 41
4.7 Evolution of the wave packet in the diﬀusive regime v
latt
= −0.15v
B
. 41
4.8 Patterned loading experiment. . . . . . . . . . . . . . . . . . . . . . 42
4.9 Oscillation period (in units of /J) as a function of the interaction
strength γ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.10 Minimum value of f
2
during an oscillation period as a function of
γ. As γ increases the population imbalance between wells increases
(see text). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.11 Maximum contrast of the Fourier components as a function of γ.
The maximum contrast is deﬁned as ∆
max
≡ ([c
1
[
2
−[c
0
[
2
)
max
with
the maximum value occurring when f
2
is at its minima. . . . . . 48
4.12 Comparison between the population evolution of the central three
wells for the inhomogeneous condensate and the homogeneous model.
Inhomogeneous condensate: stars for the initially populated well
and boxes for the initially empty wells. Homogeneous model: dashed
line represents the initially populated wells, and the solid line rep
resented the initially unpopulated wells. We used γ
eﬀ
as the local
mean ﬁeld energy (see text). The parameters used for the simulation
were J = 0.075E
R
and γ
eﬀ
= 2.64. . . . . . . . . . . . . . . . . . . 49
4.13 Momentum distribution of the inhomogeneous condensate evaluated
at various times for the same parameters as used in Fig.4.12. . . . 50
4.14 Evolution of momentum peak populations. Upper curves: popula
tion of the q = ±2π/3 momentum states. Lower curves: population
of the q = 0 momentum state. Inhomogeneous condensate (dotted),
homogeneous result (solid line), where the comparison is made by re
placing γ by an average mean ﬁeld energy γ
ave
≡ Λ
¸
n
[z
n
[
4
= 1.85.
Parameters are the same as in Fig. 4.12. . . . . . . . . . . . . . . 51
LIST OF FIGURES xv
4.15 Evolution of the normalized population for diﬀerent values of ξ. One
Bloch period is shown in the plots except ξ = 0 where the period
is inﬁnite. Solid line: [
˜
Ψ
3n
[
2
, dotted line: [
˜
Ψ
3n+1
[
2
, dashdot line:
[
˜
Ψ
3n+2
[
2
. The ”nonclassical” motion can be seen where the 3n+1
well populations initially increase more rapidly that the populations
of the 3n+2wells. It can also be seen in the plots that ξ = ξ
res
max
/2
and ξ = ξ
res
max
are resonant values. . . . . . . . . . . . . . . . . . . . 55
4.16 The spectrum of values of external force, ordered in decreasing mag
nitude, for which a population resonance occurs i.e. the values of ξ
for which
˜
Ψ
0
periodically disappears. . . . . . . . . . . . . . . . . 56
4.17 Eﬀects of interactions on generalized Bloch oscillations for the pat
tern loaded system. Evolution of [
˜
Ψ
n
[
2
for various interaction strengths.
Upper plot: ξ = 2ξ
res
max
. Lower plot: ξ = 2ξ
res
max
/7. . . . . . . . . . . 57
4.18 Comparison between the evolution of a inhomogeneous condensate
with the homogeneous result. Inhomogeneous condensate: [
˜
Ψ
0
[
2
(boxes), [
˜
Ψ
1
[
2
(stars), and [
˜
Ψ
2
[
2
(diamonds). Homogeneous case:
[
˜
Ψ
3n
[
2
(dashdot line), [
˜
Ψ
3n+1
[
2
(solid line), and [
˜
Ψ
3n+2
[
2
(dotted
line), where we have taken γ as the local mean ﬁeld energy. The
parameters used were J = 0.075E
R
, ξ = 2 and γ
eﬀ
= 1.59 . . . . . 59
4.19 Evolution of the momentum peak populations. q = 0 (dashed line,
squares), q = −2π/3 (dashdot line, stars), q = 2π/3 (solid line, dia
monds). Inhomogeneous condensate (dotted curves), homogeneous
model (lines) using the same parameters as those in Fig. 4.18, but
replacing γ by an average value γ
ave
= 1.11 for the homogeneous
model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.1 Scattering processes included in the quadratic hamiltonian: a) Di
rect and exchange excitations, b) Pair excitations . . . . . . . . . . 63
5.2 Comparisons of the exact solution and BdG (and HFBPopov) so
lutions as a function of V
eff
= U/J, for a system with M = 3
and ﬁlling factors n = 5 and 5.33. Left: number ﬂuctuations (Ex
act: solid line, BdG(and HFBPopov): dotted line), middle: con
densate fraction (Exact: solid line, BdG(and HFBPopov): dotted
line,analytic (5.79:red line ), right: superﬂuid fraction f
s
(Exact:
solid line, BdG(and HFBPopov): dotted line). The exact second
order term (dashed line) of the superﬂuid fraction, f
(2)
s
is also shown
in these plots. The vertical line shown in the plots is an estimation
of V
crit
eff
when the system is commensurate . . . . . . . . . . . . . 75
xvi LIST OF FIGURES
5.3 Comparisons of the exact solution and BdG(and HFBPopov) solu
tions as a function of V
eff
= U/J, for a system with M = 3 and
ﬁlling factors n = 50 and 50.33. Left: number ﬂuctuations (Exact:
solid line, BdG(and HFBPopov): dotted line), middle: condensate
fraction (Exact: solid line, BdG(and HFBPopov): dotted line, ana
lytic Eq.(5.79): red line ), right: superﬂuid fraction f
s
(Exact: solid
line, BdG(and HFBPopov): dotted line). The exact second order
term (dashed line) of the superﬂuid fraction, f
(2)
s
is also shown in
these plots. In this case the agreement is much better. . . . . . . 76
5.4 Comparisons of the condensate fraction given by the BdG solutions
and the HFB equations as a function of V
eff
= U/J, for a system
with M = 3 and ﬁlling factor n = 10. . . . . . . . . . . . . . . . . 78
5.5 Comparisons of the condensate wave function found by numerically
solving the BdG equations with the ThomasFermi solution. The
parameters used were U/J = 0.2, Ω/J = 9.5 10
−4
and N = 100.
The depletion is also shown in the plot. . . . . . . . . . . . . . . . 82
5.6 Comparisons of the quasiparticle spectrum found by numerically
solving the BdG equations(BdG) with the ThomasFermi solution
(TF) and the noninteracting energies (Free). The parameters used
were U/J = 0.2, Ω/J = 9.5 10
−4
and N = 100. . . . . . . . . . 83
5.7 Lowlying quasiparticle amplitudes found by numerically solving the
BdG equations. The parameters used were U/J = 0.2, Ω/J =
9.5 10
−4
and N = 100. . . . . . . . . . . . . . . . . . . . . . . . 84
5.8 Comparisons between the product F
s
n
G
s
n
calculated from the Thomas
Fermi approximation and the exact numerical solution of the BdG
equations. The parameters used were U/J = 0.2, Ω/J = 9.5 10
−4
and N = 100. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.9 Condensate density (triangles), total density (ﬁlled boxes) and
local depletion (empty diamonds) as a function of the lattice site i
for diﬀerent values of V
eff.
.The site indices i are chosen such that i =
0 corresponds to the center of the trap. Although these quantities
are deﬁned only at the discrete lattice sites we join them to help
visualization. The empty boxes represent the exact solution for the
case J=0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.10 Quasiparticle spectrum predicted by the HFBPopov theory for
diﬀerent values of V
eff
: Empty diamonds (V
eff
= 0.01), stars
(V
eff
= 0.09), crosses (V
eff
= 3), ﬁlled diamonds (V
eff
= 11),
empty boxes (V
eff
= 100) and pentagons (V
eff
= 312). The letter
q labels the quasiparticle energies in increasing order. The quasi
particle energies are in recoil units. . . . . . . . . . . . . . . . . . . 87
LIST OF FIGURES xvii
5.11 Number ﬂuctuations in the self consistent HFBPopov approach as a
function of lattice site for V
eff
= 0.01 (boxes), V
eff
= 0.09 (crosses),
V
eff
= 3 (circles),V
eff
= 11 (triangles), V
eff
= 100 (stars) and
V
eff
= 312 (diamonds). The maximum value reached by the pro
ﬁle decreases as V
eff
is increased. The empty boxes shown for each
of the curves correspond to the number ﬂuctuations predicted by
the homogeneous HFBPopov model using a local density approxi
mation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.12 Quasimomentum distribution as a function of qa, a the lattice spac
ing, q the quasimomentum, for diﬀerent values of V
eff
. . . . . . . . 89
5.13 First order onsite superﬂuid fraction as a function of the lattice
site i for diﬀerent values of V
eff.
.The site indices i are chosen such
that i = 0 corresponds to the center of the trap. Filled boxes: V
eff.
=0.01, empty boxes: V
eff.
= 0.09, empty diamonds: V
eff.
= 3,
stars: V
eff.
= 11, crosses: V
eff.
= 100 and triangles: V
eff.
= 312. . 90
5.14 Diagrammatic representation of the twobody T
2b
scattering matrix.
In this ﬁgure [kl` designates the initial states, [ij` the ﬁnal states
and [np` a set of intermediate states . . . . . . . . . . . . . . . . . 92
5.15 Comparisons of the condensate fraction predicted from the exact
diagonalization of the BHH (red), the improved Popov approxima
tion (yellow), the BdG (and HFBPopov) equations (blue) and the
test ﬁeld approximation (green) as a function of V
eff
= U/J, for a
system with M = 3 and ﬁlling factors n = 5 and 5.33. . . . . . . . 97
5.16 Comparisons of the condensate fraction predicted from the exact
diagonalization of the BHH (red), the improved Popov approxima
tion (yellow), the BdG (and HFBPopov) equations (blue) and the
test ﬁeld approximation (green) as a function of V
eff
= U/J, for a
system with M = 3 and ﬁlling factors n = 50 and 50.33. . . . . . . 98
5.17 Comparisons between n
o
/nT
mb
= f
c
T
mb
and U as a function of U
for a system with M = 3 and ﬁlling factors n = 5 and 50. . . . . . 98
6.1 Twoloop (upper row) and threeloop diagrams (lower row) con
tributing to the eﬀective action. Explicitly, the diagram a) is what
we call the doublebubble , b) the settingsun and c) the basketball. 102
xviii LIST OF FIGURES
7.1 Comparisons between the exact and the DNLSE solutions for six
atoms and three wells. The time is given in units of /J. Top
panel, strongly correlated regime (γ = 12); middle panel, inter
mediate regime (γ = 2); bottom panel, weakly interacting regime
(γ = 0.2). The solid line is the DNLSE prediction for the population
per well: [z
0
(t)[
2
and [z
1,2
(t)[
2
, the triangles are used to represent
the exact solution for the population per well calculated using the
Bose Hubbard Hamiltonian: 'ˆ a
†
0
ˆ a
0
`, 'ˆ a
†
1,2
ˆ a
1,2
`. The pentagons show
the condensate population per well calculated from the exact so
lution: ['ˆ a
0
`[
2
and ['ˆ a
1,2
`[
2
. Due to the symmetry of the initial
periodic conditions the curves for the i = 1 and 2 wells are the same
in all depicted curves . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.2 Comparisons between the exact solution(solid line), the HFB ap
proximation (boxes), the secondorder large ^ approximation (pen
tagons) and the full 2PI secondorder approximation(crosses) for
the very weak interacting regime. The parameters used were M =
3, N = 6, J = 1 and U/J = 1/30. The time is given in units of
/J. In the plots where the population, condensate and deple
tion per well are depicted the top curves correspond to the initially
populated well solutions and the lower to the initially empty wells.
Notice the diﬀerent scale used in the depletion plot. In the momen
tum distribution plot the upper curve corresponds to the k = ±2π/3
intensities and the lower one to the k = 0 quasimomentum intensity. 138
7.3 Comparisons for the case M = 3, N = 8, J = 1 and U/J = 1/3.
The time is given in units of /J. In the plots the abbreviation 1st
is used for the initially occupied well and 2nd for the initially empty
wells. In the quasimomentum plots k = 2π/a is the reciprocal lattice
vector with a the lattice spacing. . . . . . . . . . . . . . . . . . . 139
7.4 Comparison between the evolution of the atomic population per
well for M = 2, J = 1/2, NU/J = 4 and N = 20, 40 and 80. Time
is in units of /J. In the plots P1 stands for the fractional atomic
population in the initially populated wells and P2 for the population
in the initially empty wells. The number of atoms N is explicitly
shown in each panel. . . . . . . . . . . . . . . . . . . . . . . . . . 140
7.5 Time evolution of the condensate population per well and the total
condensate population,for the same parameters as Fig. 7.4. Time is
in units of /J. In the plots C1 stands for the fractional condensate
population in the initially populated well, C2 for the fractional con
densate population in the initially empty one and CT for the total
condensate fraction. . . . . . . . . . . . . . . . . . . . . . . . . . . 141
LIST OF FIGURES xix
7.6 Dynamical evolution of the quasimomentum intensities. The pa
rameters used were M = 2, J = 1/2, NU/J = 4 and N = 20, 40 and
80. Time is in units of /J. In the plots ko denotes the k = 0 quasi
momentum component and k1 the k = π/a one (a the lattice spac
ing). The plots are scaled to set the integrated quasimomentum
density to one for all N . . . . . . . . . . . . . . . . . . . . . . . . 142
9.1 Contour plot of the two dimensional band of the 1ph excitations
to ﬁst order in perturbation theory. In the plot the brighter the
color the higher the energy. The labels are ky = 2π/MR and kx =
2π/MR. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
9.2 Comparisons between the ﬁrst order corrections to the 1ph exci
tations calculated by diagonalizing the BoseHubbard Hamiltonian
inside the 1ph subspace and the analytic solution Eq. (9.22). The
number of sites used for the plot was M = 11. Energies are in units
of J. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
9.3 Comparisons between the energy eigenvalues calculated by the exact
diagonalization of the Hamiltonian (red dots), the diagonalization
of the Hamiltonian in the restricted 1ph subspace (green triangles)
and the analytic solution (blue crosses), Eq. (9.22). The index
n labels the eigenvalues in increasing order of energy. The ﬁlling
factor g, the number of wells M and the ratio U/J is indicated in
each plot. Energies are in units of J. . . . . . . . . . . . . . . . . . 165
9.4 Quasimomentum distribution as a function of U/J for a unit ﬁlled
lattice with six wells (M = 6). The exact solutions are displayed
with solid lines and the ﬁrst order perturbative results with dotted
lines. Due to the lattice symmetry n
4π/3
= n
2π/3
and n
5π/3
= n
π/3
. 166
9.5 Comparisons between the exact (red) and the perturbative (blue)
spectra for a trapped system with N = 9, M = 11 and Ω/J = 1.875
deep in the Mott regime. The lines indicate the energies of the
states that are coupled directly to the ground state. . . . . . . . . 170
10.1 Bragg spectroscopic scheme considered in this paper. The lattice
potential with atoms loaded into the ground band is perturbed by
a shallow running wave perturbation (the Bragg potential). . . . . 174
10.2 Imparted energy: Comparisons between the the exact and Bogoli
ubov approximation for N = M = 9. The horizontal axis is in units
of . For the plot we used Jτ/ = 10. See text Eq. (10.23). . . . . 181
10.3 Comparisons of the imparted energy vs. Bragg frequency curves
calculated from the exact diagonalization of the manybody Hamil
tonian (red) and the from the HFBPopov approximation(blue) for
a trapped system with N = 9, M = 11 and Ω/J = 1.875. The
horizontal axis is in units of . The lines depicted in the plot are
located at the diﬀerent quasiparticle energies excited by the Bragg
perturbation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
xx LIST OF FIGURES
10.4 Density proﬁle calculated from the exact diagonalization of the
manybody Hamiltonian (red) and from the HFBPopov approx
imation (blue) for a trapped system with N = 9, M = 11 and
Ω/J = 1.875. n labels the lattice site. . . . . . . . . . . . . . . . . 184
10.5 Dynamical structure factor vs. ω/J in the ThomasFermi (blue)
and HFBPopov (red) approximations. The horizontal axis is in
units of . The system parameters are U/J = 0.2, Ω/J = 9.510
−4
and N = 100. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
10.6 Eigenvalues of the quasiparticle spectrum plotted in increasing en
ergy order. The vertical scale is in units of . . . . . . . . . . . . 186
10.7 Transfer energy vs. Bragg frequency for a zero temperature homo
geneous system with M = N = 9 and diﬀerent transfer momenta
q. The horizontal axis is in units of . The perturbation time was
set to Jτ/ = 10. Note the change of the vertical scale between the
two panels. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
10.8 Transfer energy vs. Bragg frequency for a zero temperature trapped
system with parameters U/J=45, Ω/J = 1.857, M = 11 and N = 9
and diﬀerent transfer momenta q. The horizontal axis is in units of
. The perturbation time was set to Jτ/ = 10. Note the change
of the vertical scale between the two panels. . . . . . . . . . . . . 187
10.9 Transfer energy as a function of the temperature for diﬀerent U/J
parameters. The plots are for a homogeneous system with N =
M = 9. The horizontal axis in in units of . . . . . . . . . . . . . 191
10.10Transfer energy as a function of the temperature for diﬀerent U/J
parameters. The plots are for an trapped system with M = 11,
N = 9 and Ω/J = 1.875. The horizontal axis in in units of . . . . 192
11.1 Comparisons between the time evolution of the percentage ﬁdelity
calculated by diagonalizing the BoseHubbard Hamiltonian (dots),
the perturbative solution inside the one particle hole subspace (red)
and the analytic solution Eq. 11.6 (blue). . . . . . . . . . . . . . . 196
11.2 Percentage ﬁdelity as a function of time calculated using the per
turbative solution 11.4 (red) and the analytic solution Eq. 11.6
(blue).The number of sites used for the plot is M = 31 . . . . . . 197
11.3 Schematic of an inhomogeneous lattice ﬁlled with N qubits with an
onsite interaction energy U. An externally applied trapping poten
tial of strength V
j
= Ωj
2
, e.g. due to a magnetic ﬁeld, acts to ﬁll
gaps in the central region of the trap. The center subspace { of the
lattice deﬁnes the quantum computer register containing K < N
qubits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
LIST OF FIGURES xxi
11.4 Comparisons between the time evolution of the percentage ﬁdelity
F
com
(K = 5, t) calculated by evolving the initial target state using
the BoseHubbard Hamiltonian (dots), the restricted one particle
hole basis (red) and the analytic solution Eq. 11.16 (blue). In the
plot we assumed a commensurate unit ﬁlled lattice with ﬁve sites
and inﬁnitely high boundaries. . . . . . . . . . . . . . . . . . . . . 200
11.5 Percentage ﬁdelity as a function time calculated using the approx
imated solution 11.16 (blue) and the ﬁdelity found numerically by
restricting the dynamics to the one particle hole subspace. We as
sumed inﬁnity high boundaries and K = 31. . . . . . . . . . . . . 201
11.6 Schematic of the relevant couplings in the problem. The unit ﬁlled
state [T` describing a target quantum register and the states [S
±
j
`
having one doubly occupied lattice site and a neighboring hole are
coupled to ﬁrst order in J. A catalysis laser resonantly couples the
ground states [S
±
j
` to the excited states [M
±
j
` describing a bound
molecule at the doubly occupied site. The bound states quickly de
cay and give the possibility of monitoring population in the “faulty”
register states [S
±
j
`. . . . . . . . . . . . . . . . . . . . . . . . . . . 202
11.7 Population in the unit ﬁlled register state [T`
1
during continuous
measurement of the register beginning in the BoseHubbard ground
state [Ψ
g
`. The plots show dynamics appropriate to tunneling in
one dimension with U/J = 500. (a) Quantum trajectories corre
sponding to a null measurement result for three diﬀerent register
sizes K. The time scale to saturate the target state is independent
of the number of qubits: t
sat
≈ κ
−1
. (b) Long time dynamics for
K = 501, N = 551 and ﬁnite detector eﬃciencies η. The popula
tion in [T`
1
for η = 1 is indistinguishable from one. Also shown is
the oscillatory dynamics at fundamental frequency U described by
Eq. (11.16) if the measurement is turned oﬀ after the target state
is reached. The arrow indicates ρ
T,T
(0). . . . . . . . . . . . . . . . 203
Chapter 1
Introduction
Bose Einstein condensation (BEC) in bosonic gases was predicted by Einstein in
1925 [1] based on the quantum statistics ideas developed by Bose for photons [2].
The basic idea of BEC is that below a critical temperature a macroscopic number
of particles occupy the lowest energy state: as the temperature, T, is decreased the
deBroglie wave length, which scales like T
−1/2
, increases and at the critical point it
becomes comparable to the interparticle mean separation. At this point the wave
functions of the particles are suﬃciently smeared out so that there is always some
overlap and a Bose Einstein condensate is formed. Although Einstein’s prediction
applied to a gas of noninteracting atoms, London suggested that, BEC could be
the mechanism underlying the phenomenon of superﬂuidity in
4
He [3], despite the
strong interactions in this system. Further evidence for this point of view came
from neutron scattering experiments [4].
Experimental eﬀorts to create a BEC in dilute gases date back to the 1980’s
[5]. The ﬁrst experiments concentrated on using atomic Hydrogen but mainly the
large rates of inelastic collision prevented these experiments from succeeding. It
was not until 1995, using the advances made in laser cooling techniques[9], that
BEC in dilute alkali atomic gases was achieved. The ﬁrst series of experiments were
done with Rubidium [6], sodium [7] and lithium [8] vapors. In these experiments
atoms are typically collected in a magnetooptical trap (MOT) and compressed
and cooled to microKelvin temperatures using laser cooling techniques. They are
then transferred to a magnetic trap where evaporative cooling allows the system
to be cooled to nanoKelvin temperatures. At a critical phase space density BEC
takes place. In such a condensate a macroscopic number of atoms, generally up to
10
6
, collectively occupy the lowest energy state.
The experimental realization of BEC in alkali gases opened unique opportu
nities for exploring quantum phenomena on a macroscopic scale. In contrast to
experiments with liquid helium, where the strong interactions between particles
wash out the eﬀects due to the BEC, the relatively weak twoparticle interaction
in dilute alkali atoms allows these systems to be used as a theoretical and experi
mental arena to study coherent matter wave properties. Theoretically, the weakly
interacting regime has the advantage that all the atoms can be described by a
2 Chapter 1 Introduction
single macroscopic wave function. This allows a very intuitive understanding of
the system based on the so called GrossPitaevskii equation (GPE) [10]. Experi
mentally, the macroscopic wave packet can be probed by interference experiments,
where by turning oﬀ the trapping ﬁelds the atoms are allowed to expand and the
wave packets to interfere with each other.
The GPE assumes that all the atoms are in the condensate and neglects com
pletely quantum correlations. In the weakly interacting regime, a description be
yond the simplest mean ﬁeld theory can be made by treating the small quantum
ﬂuctuations as perturbations. This treatment was proposed by Bogoliubov in 1947
[11]. The ﬂuctuations lead to a small depletion of the condensate mode since other
excited states diﬀerent from the condensate get populated. Since in most of the
experiments with dilute gases the condensate depletion is at most 3%, the GPE
together with Bogoliubov analysis have been in general very successful describing
these experiments. Much theoretical and experiment work has been done studying
condensate properties such as condensate collective excitations, phonon modes,
sound velocity and superﬂuid ﬂow phenomena [12][19].
However, weakly interacting dilute gases described by a mean ﬁeld picture
are the simplest many body systems one can possible found. In order to be in the
weakly interacting regime, the ratio between the interaction energy of uncorrelated
atoms at a given density, E
int
, and the quantum kinetic energy needed to correlate
particles by localizing them within a distance of order of the mean interparticle
distance, E
kin
, must be small. For three dimensional systems E
int
∼ n
4π
2
a
s
m
(a
s
is
the scattering length, which fully characterizes the low energy scattering processes,
n is the mean particle density and m is the atomic mass) and E
kin
∼
2
2m
n
2/3
.
Thus, the ratio between these two energies is proportional to n
1/3
a
s
. In dilute
alkali vapors this ratio is generally of order 0.02. To enter the strongly correlated
regime, an obvious way to proceed is either to raise the density or to raise the
scattering length. It is indeed possible to tune the scattering length to large values
by using a Feshbach resonance. This has recently been realized for example in
85
Rb
where the scattering length was tuned over several orders of magnitude [20, 21].
The problem of this approach is, however, that the lifetime of the condensate
strongly decreases due to threebody losses [22]. An entirely diﬀerent way to reach
the strongly correlated regime is by using optical lattices. By increasing the depth
of the optical lattice the ratio between kinetic energy and potential energy can be
changed without aﬀecting the density or the scattering length. The beauty of this
approach is that the lattice depth can be used as an experimental knob to change
the kinetic to interaction energy ratio allowing us to reach diﬀerent manybody
regimes.
• Optical lattices
Optical lattices are periodic Stark shift potentials created by the interference
of two or more laser beams. They have been widely used in atomic physics in
the context of atom diﬀraction [23, 24] with applications to atom optics and atom
interferometry [25]. They have also been used as a way to trap and cool atoms. The
3
ﬁrst experiment where atoms were cooled to the microkelvin regime in a multi
dimensional optical lattice was carried out by Hemmerich et al. [26][28] followed
by Grynberg et al. [29]. There have been various attempts to cool atoms directly
in an optical lattice. Among them we can mention Raman cooling techniques, by
which atoms have recently been partially cooled to the ground sate with ﬁlling
factors of order one [30, 31, 32]. However, one of the most successful techniques
to load ultracold atoms in the ground state, with almost no discernible thermal
component, is by ﬁrst forming a Bose Einsten condensate in a weak magnetic trap
and then adiabatically turning on the lattice by slowly ramping up the intensity
of the laser beams.
Atom dynamics in optical lattices is closely related to electron dynamics in
solid state crystals, but optical lattice have favorable attributes such as the absence
of defects and the high degree of experimental control [33, 34]. When ultracold
bosonic atoms are loaded in shallow lattices, the system is in the weakly interacting
regime and most of the atoms are Bose condensed. Combined with BEC, the
ultimate source of coherent atom , optical lattices provide a way of exploring
a quantum system analogous to electrons in crystals but with complete control
over the lattice and the atoms. Beautiful experiments have been done in this
regime and have provided an elegant demonstration of band structure [35, 36,
37], Bragg scattering and Bloch oscillations [38], as well as coherent matter wave
interferometry [39, 40], superﬂuidity [41] and quantum chaos [42].
There has also been spectacular recent experimental progress in the strongly
correlated regime. It was ﬁrst realized by Jaksch et al. [43], that a BEC loaded in
a lattice potential is a nearly perfect experimental realization of the BoseHubbard
Hamiltonian, which describes bosons with local repulsive interactions in a periodic
potential. M.P.A. Fisher et al. [44] predicted that a system modeled by the
BoseHubbard Hamiltonian exhibits a quantum phase transition from a superﬂuid
to an insulator state (superﬂuidMott insulator transition) as the interactions are
increased. In fact, the Mott insulator transition in a 3D lattice starting from a BEC
has experimentally observed by M. Greiner and coworkers [46]. Moreover, in recent
years there have been many impressive experiments which have demonstrated the
loss of quantum coherence as the system approaches the strongly correlated regime,
for example by measuring number squeezing [45] or by studying the collapse and
revival of coherence in a matter wave ﬁeld [47].
One of the most important potential applications of the Mott insulator tran
sition is to use it as a mean to initialize a quantum computer register. Deep in
the Mott insulator regime the kinetic energy is very small with respect to the in
teraction energy and it is energetically favorable for the atoms to remain localized
without tunneling. The negligible number ﬂuctuations makes it possible to prepare
the ﬁducial state with exactly one atom per site needed to initialize a quantum
computer register [43, 48, 50]. Besides the possibility of high ﬁdelity initialization,
the easy scalability, low noise and high experimental control make ultracold neu
tral atoms loaded in an optical lattice one of the most attractive candidates for
implementations of quantum computation.
4 Chapter 1 Introduction
The overall goal of this thesis is to study equilibrium and non equilibrium prop
erties of cold bosons loaded in optical lattices starting from the superﬂuid regime,
where mean ﬁeld techniques can be applied, and going into the rich and complex
strongly correlated regime where the standard GPE and Bogoliubov treatments
fail to describe the system and a more general framework is required. Most of the
work is done in the context of ongoing experimental eﬀorts, especially the ones
trying to achieve lattice based quantum information processing.
• Overview
In chapter 2, I start by reviewing the theory of optical potentials, the single
particle band structure and the tight binding approximation. In chapter 3, I go
one step further and consider the manybody properties of the system by intro
ducing the BoseHubbard Hamiltonian. I give a review of the generic issues that
characterize the superﬂuid to Mott insulator transition.
In chapter 4, I describe the equilibrium properties of lattice systems in the
superﬂuid regime, where a mean ﬁeld treatment is valid. I study the mean ﬁeld
Discrete Nonlinear Schr¨odinger equation (DNLSE), and use it to model two exper
iments done by the lasercooling and trapping group at NIST. In the ﬁrst one, an
optical lattice was moved and the average displacement of the atoms was used as
a means to probe the band structure of the system. In the second experiment,to
which I refer as the patterned loading experiment [33],the atoms were loaded into
every third site of an optical lattice, with the aim of having large enough spatial
separation to address individual atoms. This patterned loading method may be
a useful technique for the implementation of lattice based quantum computing
proposals.
The DNLSE completely neglects quantum ﬂuctuations. However, if the sys
tem is weakly interacting, the small quantum corrections can be included by using
the Bogoliubov approximation. In the Bogoliubov approximation the complicated
manybody quartic Hamiltonian is reduced to a quadratic one, which can be diago
nalized exactly. Using the Bogoliubov approximation, in chapter 5 I study diﬀerent
standard quadratic approximations in two diﬀerent lattice systems, a translation
ally invariant one with periodic boundary conditions which in general allows ana
lytic solutions, and a system closer to real experimental situations where, besides
the lattice potential, there is a superimposed harmonic conﬁnement potential. I
discuss the Bogoliubov de Genes equations (BdG), the HartreeFockBogoliubov
(HFB) approximation and the HFBPopov approximation. To test the validity of
the diﬀerent approximations and their departure from the exact solution as the
interactions are increased I compare them with the exact numerical diagonaliza
tion of the BoseHubbard Hamiltonian. Because of the exponential scaling of the
dimensionality of the Hilbert space with respect to system size, the exact solu
tion (for a system with N atoms and M wells the number of states scales like
(N +M −1!)/(N!M!)), the numerical comparisons are restricted to systems with
a moderate number of atoms and wells.
In chapter 5, I also report on our idea of using the superﬂuid fraction to study
5
the approach to the Mott insulator transition. By deriving an expression for
the superﬂuid density based on the rigidity of the system under phase variations
we were able to explore the connection between the quantum depletion of the
condensate and the quasimomentum distribution on one hand and the superﬂuid
fraction on the other.
At the end of the chapter, I present my attempt to approach the strongly
correlated regime by using the improved Popov approximation [52, 53, 55]. The
idea presented here is to upgrade the bare potential, which is the one that explicitly
appears in the manybody Hamiltonian, to the manybody scattering matrix. By
upgrading the bare potential, we are properly taking into account the eﬀect of the
surrounding atoms in the properties of binary collisions.
In chapter 6 and 7, motivated by the patterned loading experiment, we adopt
a functional eﬀective action approach capable of dealing with non equilibrium
situations that require a treatment beyond mean ﬁeld theory. Even though a de
scription of the dynamics of the patterned loading system using the DNLSE was
derived in chapter 4, it is shown by comparisons with the exact quantal solution
calculated by time propagating the initial conﬁguration with the BoseHubbard
Hamiltonian that a mean ﬁeld solution is valid only for short times and in the very
weakly interacting regime. To deal with the dynamics far from equilibrium, we
adopt a closed time path (CTP) [56] functionalintegral formalism together with
a twoparticle irreducible (2PI) [57] eﬀective action approach and derive equations
of motion. We retain terms of up to secondorder in the interaction strength when
solving these equations. Under the 2PICTP scheme we consider three diﬀerent
approximations : a) the time dependent HartreeFockBogoliubov (HFB) approx
imation, b) the nexttoleading order 1/^ expansion and c) a full secondorder
perturbative expansion in the interaction strength. We derive mathematical ex
pressions for the equations of motion in chapter 6 and apply them to the particular
case of the patterned loaded lattice in chapter 7. We use this system to illustrate
many basic issues in nonequilibrium quantum ﬁeld theory, such as nonlocal and
nonMarkovian eﬀects, pertaining to the dynamics of quantum correlation and
ﬂuctuations. We show that because the secondorder 2PI approximations include
multiparticle scattering in a systematic way, they are able to capture damping
eﬀects exhibited in the exact solution, which a collisionless approach fails to pro
duce. While the secondorder approximations show a clear improvement over the
HFB approximation, they fail at late times, when interaction eﬀects are signiﬁcant.
The 2PI eﬀective action formalism provides a useful framework where the mean
ﬁeld and the correlation functions are treated on the same selfconsistent footing.
However, it yields dynamical equation of motion that are non local in time and
hard to estimate analytically. The idea in chapter 8 is to simplify the 2PI equations
and to obtain near equilibrium solutions where, kinetic theories that describe ex
citations in systems close to thermal equilibrium are valid. In particular, we show
in this chapter how the full secondorder 2PI equations are in agreement with cur
rent kinetic theories [15],[58][62] and reproduce in equilibrium the higher order
perturbative corrections well known in the literature since Beliaev’s work [63].
6 Chapter 1 Introduction
In chapter 9, I jump into the Mott insulator phase and I study the physical
properties of the system deep in the Mott insulator regime, which is the other
regime where an analytic treatment based on perturbation theory is possible. A
perturbative analysis should be applicable to study current experiments [64] that
reach the strong Mott regime. I derive expressions for the excitation spectrum of
the Mott state for both homogeneous and trapped systems and compare them with
the solutions obtained by exact diagonalization of the BoseHubbard Hamiltonian.
The main purpose of this chapter is to understand the manybody properties of
the Mott insulator ground state and the nature of its manybody excitations which
will be crucially important as more elaborate experiments with optical lattices in
the strongly correlated regime are undertaken.
One key piece of evidence for the Mott insulator phase transition is the loss of
global phase coherence of the matter wavefunction when the lattice depth increases
beyond a critical value [46]. However, there are many possible sources of phase
decoherence in these systems. It is known that substantial decoherence can be
induced by quantum or thermal depletion of the condensate during the loading
process, so loss of coherence is not a proof that the system resides in the Mott
insulator ground state. Indeed, for this reason, in the experiments by Greiner et
al. [46] a potential gradient was applied to the lattice to show the presence of
a gap in the excitation spectrum. In chapter 10, we show that another common
experimental technique, Bragg spectroscopy [65, 66], not only can identify the
excitation gap that opens up in the Mott regime, but also can be used to map
out the excitation spectrum and to determine the temperature of the system when
it is deep in the Mott regime. Speciﬁcally, we study the total momentum and
total energy deposited in the system by the Bragg perturbation calculated under
a linear response analysis and obtain analytical solutions in the superﬂuid and
deep Mott insulator regimes. We test the accuracy of the approximations and
their deviation from the full quantal behavior as usual by comparing them with
numerical solutions obtained by diagonalizing the BoseHubbard Hamiltonian for
a moderate number of atoms and wells.
All of the proposals for quantum computation which utilize a latticetype ar
chitecture have the Mott insulator transition as the initialization scheme to load
exactly one atom per lattice site. Such architecture requires a lattice commensu
rately ﬁlled with atoms, which does not correspond exactly to the ground Mott
insulator state. The ground state has a remaining coherence proportional to the
tunneling matrix element. This degrades the initialization of the quantum com
puter register and can introduce errors during error correction. I ﬁnish this thesis
with chapter 11, where I report on our proposal to solve this problem by using the
spatial inhomogeneity created by a quadratic magnetic trapping potential together
with a continuous measurement procedure which projects out the components of
the wave function with more than one atom in any well.
7
Publications in the PhD work:
• Chapter 4
Dynamics of a periodthree pattern loaded BoseEinstein condensate in an
optical lattice, A. M. Rey, P. B. Blakie, Charles W. Clark, Phys. Rev.
A 67, 053610 (2003).
• Chapter 5
Bogoliubov approach to superﬂuidity of atoms in an optical lattice, Rey AM,
Burnett K, Roth R, Edwards M, Williams CJ, Clark CW, J. Phys. B:
At. Mol. Opt. Phys, 36 , 825 (2003).
• Chapter 6 and 7
Nonequilibrium Dynamics of Optical Lattice  Loaded BEC Atoms: Beyond
HFB Approximation, A. M. Rey, B. L. Hu, E. Calzetta, A. Roura,
Charles W. Clark, Phys. Rev A. 69, 033610 (2004).
BEC with ﬂuctuations: beyond the HFB approximation, A. M. Rey, B. L.
Hu, E. Calzetta, A. Roura, Charles W. Clark (Proceedings of the Laser
Physics Workshop 2003), Las. Phys., 14, 1, (2004).
• Chapter 10
Bragg spectroscopy of ultracold atoms loaded in an optical lattice, Ana Maria
Rey, P. Blair Blakie, Guido Pupillo, Carl J. Williams, Charles W. Clark,
submitted to Phys. Rev. Lett. condmat/0406552.
• Chapter 11
Scalable register initialization for quantum computing in an optical lattice,
G. K. Brennen, G. Pupillo, A. M. Rey, C. W. Clark, C. J. Williams,
submitted to Phys. Rev. Lett., condmat/0312069.
Scalable quantum computation in systems with BoseHubbard dynamics,
Guido Pupillo, Ana M. Rey, Gavin Brennen, Carl J. Williams, Charles
W. Clark, to appear in J. Mod. Opt, quantph/0403052.
8 Chapter 1 Introduction
Chapter 2
Optical lattices
Optical lattices are periodic potentials created by lightmatter interactions. When
an atom interacts with an electromagnetic ﬁeld, the energy of its internal states
depends on the light intensity. Therefore, a spatially dependent intensity induces
a spatially dependent potential energy. If such a modulation is obtained by the
interference of several laser beams, the resultant optical potential felt by the atoms
will have diﬀerent potential wells separated by a distance of the order of the laser
wavelength. The depths of the optical potential wells that can be obtained in an
experiment are in the microKelvin range. Nevertheless, atoms can be trapped in
this potentials when cooled at low temperatures, by laser and evaporative cooling
techniques.
Cold atoms interacting with a spatially modulated optical potential resemble
in many respects electrons in ionlattice potential of a solid crystals . However,
optical lattices have several advantages with respect to solid state systems. They
can be made to be largely free from defects, such defects for example prevented the
observation of Bloch oscillations in crystalline solids. Optical lattices also can be
controlled very easily by changing the laser ﬁeld properties. For example the lattice
depth can be changed by modifying the laser intensity, the lattice can be moved
by changing the polarization of the light or chirping the laser frequency and the
lattice geometry can be modiﬁed by changing the laser conﬁguration. Moreover,
in contrast to solids, where the lattice spacings are generally of order of Angstrom
units, the lattice constants in optical lattices are typically three order of magnitude
larger.
The idea of this chapter is to introduce the basic theory of optical lattices and
to review the single particle properties of atoms loaded in such periodic potentials.
2.1 Basic theory of optical lattices
Neutral atoms interact with light in both dissipative and conservative ways.
The conservative interaction comes from the interaction of the light ﬁeld with
the induced dipole moment of the atom which causes a shift in the potential
10 Chapter 2 Optical lattices
energy called acStark shift. On the other hand, the dissipation comes due to the
absorption of photons followed by spontaneous emission. Because of conservation
of momentum, the net eﬀect is a dissipative force on the atoms caused by the
momentum transfer to the atom by the absorbed and spontaneously emitted
photons. Laser cooling techniques make use of this light forces.
For large detunings spontaneous emission processes can be neglected and the
energy shift can be used to create a conservative trapping potential. This is the
physics that describes optical lattices.
2.1.1 AC Stark Shift
Consider a two level atom, with internal ground state [g` and excited state [e`
and energy diﬀerence ω
o
in a lossless cavity of volume V , interacting with a
monochromatic electromagnetic ﬁeld with frequency ω = 2πν as schematically
shown in Fig.2.1. Assume also that the experiment is performed within a time
smaller than the spontaneous emission rate so that spontaneous emission can be
neglected.
Figure 2.1: AC Stark shift induced by atomlight interaction. The laser frequency
is ω = 2πν which is detuned from the atomic resonance by Λ
The uncoupled Hamiltonian describing the atoms and the electromagnetic ﬁeld
is given by
ˆ
H
o
= ω
o
[e` 'e[ +ω(ˆ a
†
ˆ a + 1/2), (2.1)
where ˆ a is the photon annihilation operator. If the detuning of the laser from
the atomic transition, ∆ = ω − ω
o
, is small [∆[ < ω
o
, then the state with the
atom in the ground state and N photons in the ﬁeld, [0` ≡ [g, N` has similar
energy to the state with the atom in an excited state and N − 1 photons [1` ≡
2.1 Basic theory of optical lattices 11
[e, N −1` , E
1
− E
0
= −∆. The eﬀect of the interactions is to couple these
states. Under the dipole approximation which assumes that the spatial variation
of the electromagnetic ﬁeld is small compared with the atomic wave function, the
coupling Hamiltonian denoted as
ˆ
H
I
in the interaction picture is given by:
ˆ
H
I
= −
d
E (2.2)
=
[e` 'g[ e
iω
o
t
+[g` 'e[ e
−iω
o
t
Ω
∗
(x)
2
ˆ a
†
e
iωt
+
Ω(x)
2
ˆ ae
−iωt
.
Here Ω(x) is the Rabi frequency given by Ω(x) = −2
ω/N)
2∈V
u(x)ε
d, with ε the
unit polarization vector of the ﬁeld,
d the dipole moment of the atom and u(x)
the ﬁeld mode evaluated at the atomic position x,(for plane waves, for example
u(x) = e
−i
kx
). In the rotating wave approximation, valid in the limit [∆[ <ω
o
, the
type of processes with a rapidly oscillating phase, exp(±i(ω
o
+ω)t), are neglected
and only the near resonant frequency processes are considered. The interaction
Hamiltonian is then reduced to
ˆ
H
I
≈
Ω(x)
2
[e` 'g[ ˆ ae
i∆t
+
Ω
∗
(x)
2
[e` 'g[ ˆ a
†
e
−i∆t
. (2.3)
Physically, the resonant process correspond to either the excitation of the atom
along with the emission of a photon or the relaxation of the atom with the absorb
tion of a photon. In the previous line we also assume a large number of photons
and neglect the variation in the coupling constant due to ∆N, i.e N · N −1.
If the detuning is large compared to the Rabi frequency, [∆[ Ω, the eﬀect
of the interactions on the states, [0` and [1`, can be determined with second order
perturbation theory. In this case, the energy shift E
(2)
0,1
is given by
E
(2)
0,1
= ±
[ '1[
ˆ
H
int
[0` [
2
∆
= ±
[Ω(x)[
2
4∆
, (2.4)
with the plus and minus sign for the [0` and [1` states respectively. This energy
shift is the so called acStark shift. Since the atoms are practically always in the
ground state, the energy of the atoms is changed according to the stark shift
[Ω(x)[
2
4∆
, which deﬁnes the optical potential.
Furthermore, if instead of interacting with a monochromatic electromagnetic
ﬁeld, the atoms are illuminated with superimposed counterpropagating laser beams,
the beams interfere and the interference pattern results in a periodic landscape po
tential or optical lattice.
2.1.2 Dissipative interaction
In the above discussion we implicitly assumed that the excited state has an inﬁnite
lifetime. However, in reality it will decay by spontaneous emission of photons. This
12 Chapter 2 Optical lattices
eﬀect can be taken into account phenomenologically by attributing to the excited
state an energy with both real an imaginary parts. If the excited state has a life
time 1/Γ
e
corresponding to a efolding time for the occupation probability of the
state, the corresponding life time for the amplitude will be twice this time. The
energy of the perturbed ground state becomes a complex quantity which we can
write as
E
(2)
0
=
4
[Ω(x)[
2
∆−iΓ
e
/2
= V (x) +iγ
sc
(x), (2.5)
V (x) =
[Ω(x)[
2
∆
4∆
2
−Γ
2
e
≈
[Ω(x)[
2
4∆
, (2.6)
γ
sc
(x) =
2
[Ω(x)[
2
Γ
e
4∆
2
−Γ
2
e
≈
[Ω(x)[
2
Γ
e
8∆
2
. (2.7)
The real part of the energy corresponds to the optical potential whereas the imag
inary part represents the the rate of loss of atoms from the ground state. The sign
of the optical potential seen by the atoms depends on the sign of the detuning.
For blue detuning , ∆ > 0, the sign is positive resulting in a repulsive potential,
and the potential minima correspond to the points with zero light intensity. On
the other hand, in a red detuned light ﬁeld, ∆ < 0, the potential is attractive
and the minima correspond to the places with maximum light intensity. Because
the eﬀective spontaneous emission rate of the atoms increases with the light inten
sity, the spontaneous emission in a red detuned optical lattice will always be more
signiﬁcant than in a blue detuned one.
The proper detuning for an optical lattice depends on the available laser power
I ([Ω[
2
∝ I) and the maximum dissipative scattering rate that can be tolerated.
On one hand, with small detuning it is possible to create larger trap depths for a
given laser intensity since the optical potential scales as V ∼ I/∆. On the other,
the inelastic scattering rate is inversely proportional to the detuning squared and
scales like Γ
e
/∆ V . Therefore, the laser detuning should be chosen as large as
possible within the available laser power in order to minimize inelastic scattering
processes and create a conservative potential.
2.1.3 Lattice geometry
The simplest possible lattice is a one dimensional(1D) lattice lattice. It can be
created by retroreﬂecting a laser beam, such that a standing wave interference
pattern is created. This results in a Rabi frequency Ω(x) = 2Ω
o
sin(kx) which
yields a periodic trapping potential given by
V
lat
(x) = V
o
sin
2
(kx) =
Ω
2
o
∆
sin
2
(kx), (2.8)
where k = 2π/λ is the absolute value of the wave vector of the laser light and V
o
is four times times the depth of a single laser beam without retroreﬂection, due
to the constructive interference of the lasers.
2.2 Single particle physics 13
Periodic potentials in higher dimensions can be created by superimposing more
laser beams. To create a two dimensional lattice potential for example, two or
thogonal sets of counter propagating laser beams can be used. In this case the
lattice potential has the form
V
lat
(x, y) = V
o
cos
2
(ky) + cos
2
(kx) + 2ε
1
ε
2
cos φcos(ky) cos(kx)
. (2.9)
Here k is the magnitude of the wave vector of the lattice light, ε
1
and ε
2
are
polarization vectors of the counter propagating set and φ is relative phase between
them. If the polarization vectors are not orthogonal and the laser frequencies are
the same, they interfere and the potential is changed depending on the relative
phase of the two beams. This leads to a variation of the geometry of the lattice
in a chequerboard like pattern. A simple square lattice with one atomic basis
can be created by choosing orthogonal polarizations between the standing waves.
In this case the interference term vanishes and the resulting potential is just the
sum of two superimposed 1D lattice potentials. Even if the polarization of the two
pair of beams is the same, they can be made independent by detuning the common
frequency of one pair of beams from that the other. Typically a negligible frequency
diﬀerence compared with the optical frequency is required to achieve independence,
thus even in this case to a good approximation the wave vectors can be considered
equal.
A more general class of two 2D lattices can be created from the interference
of three laser beams [34, 33] which in general yields non separable lattices. Such
lattices can provide tighter onsite conﬁnement, better control over the number
of nearest neighbors and signiﬁcantly reduced tunneling between sites compared
with the counter propagating four beam square lattice. In Fig. 3 we show a
variety of possible 2D optical lattice geometries that can be made by three and four
interfering laser beams. Similarly a 3D lattice can be created by the interference
of at least 6 orthogonal sets of counter propagating laser beams.
2.2 Single particle physics
In this section for simplicity we are going to restrict the analysis to a one dimen
sional lattice. Generalization to higher dimensions can be done straightforwardly,
especially if the lattice geometry is separable. The main purpose of this section is
to review the basic aspects that describe the behavior of noninteracting particle
subject to a periodic potential.
2.2.1 Bloch functions
One of the most important characteristics of a periodic potential is the emergence
of a band structure. Consider a one dimensional particle described by the Hamil
tonian H =
ˆ p
2
2m
+ V
lat
(x), where V
lat
x) = V
lat
(x + a). Bloch’s theorem [67, 68]
14 Chapter 2 Optical lattices
y
[
λ
]
(a)
a
1
a
2
−1
0
1
(b)
a
1
a
2
(c)
a
1
a
2
x [λ]
y
[
λ
]
(d)
a
1
a
2
−1 0 1
−1
0
1
x [λ]
(e)
a
1
a
2
−1 0 1
x [λ]
(f)
a
1
a
2
−1 0 1
V
Latt
(r) [E
R
]
−10 −5 0 5
Figure 2.2: Optical lattice potential.(a)(e) potentials for diﬀerent conﬁgurations of
3 beams, (f) potential for the 4 counterpropagating laser beam conﬁguration (The
two pair of light ﬁelds are made independent by detuning the common frequency of
one pair of beams from the other. E
R
is the atomic recoil energy, E
R
=
2
k
2
/2m.
This ﬁgure is a courtesy of P. Blair Blakie [34] )
states that the eigenstates φ
(n)
q
(x) can be chosen to have the form of a plane wave
times a function with the periodicity of the potential:
φ
(n)
q
(x) = e
iqx
u
(n)
q
(x), (2.10)
u
(n)
q
(x +a) = u
(n)
q
(x). (2.11)
Using this ansatz into the Schr¨odinger equation, Hφ
(n)
q
(x) = E
(n)
q
φ
(n)
q
(x), yields
an equation for u
(n)
q
(x) given by:
¸
(ˆ p+q)
2
2m
+V
lat
(x)
¸
u
(n)
q
(x) = E
(n)
q
u
(n)
q
(x) (2.12)
Bloch’s theorem introduces a wave vector q. The quantity q should be viewed as
a quantum number characteristic of the translational symmetry of the periodic
potential, just as the momentum is a quantum number characteristic of the full
translational symmetry of the free space. Even though it is not the same, it turns
out that q plays the same fundamental role in the dynamics in a periodic potential
as the momentum does in the absence of the lattice. To emphasize this similarity
2.2 Single particle physics 15
q is called the quasimomentum or crystal momentum . In general the wave vector
q is conﬁned to the ﬁrst Brilloiun zone, i.e. −π/a < q ≤ π/a.
The index n appears in Bloch’s theorem because for a given q there are many
solutions to the Schr¨odinger equation. Eq. (2.12) can be seen as a set of eigenvalue
problems in a ﬁxed interval, 0 < x < a, one eigenvalue problem for each q. There
fore, each of them, on general grounds has an inﬁnite family of solutions with a
discretely spaced spectrum of modes labelled by the band index n. On the other
hand, because the wave vector q appears only as a parameter in Eq. (2.12), for an
inﬁnite lattice, the energy levels for a ﬁxed n has to vary continuously as q varies.
The description of energy levels in a periodic potentials in terms of a family of
continuous functions E
(n)
q
each with the periodicity of a reciprocal lattice vector,
2π/a, is referred to as the band structure.
In the simple case of a sinusoidal potential, which is the one used in experi
ments, V
lat
(x) = V
0
sin
2
(kx), the band structure can be solved analytically. In this
case the Schr¨odinger Equation is given by
−
d
2
dy
2
φ
(n)
q
(y) +
V
0
4E
R
(2 −2 cos(2y)) φ
(n)
q
(y) =
E
(n)
q
E
R
φ
(n)
q
(y) (2.13)
where E
R
=
2
k
2
/2m is the atomic recoil energy and y = kx. For convenience,
lattice depths are generally speciﬁed in recoil units. Equation (2.13) is just the
Mathieu equation ([69])
d
2
y
dx
2
+(a + 2s cos(2x))y = 0. (2.14)
.
Solutions of the Mathieu equation are generally written in the Floquet forme
iνx
P(x)
where ν is known as the characteristic exponent and a = a(ν, s) is the characteris
tic parameter which is a complicated function of s and ν. In the lattice language
ν correspond to the quasimomentum , s = V
0
/4E
R
and a = E
(n)
q
/E
R
−V
0
/2E
R
.
Fig. 2.3 shows the band structure of a sinusoidal potential for diﬀerent po
tential depths. For V
0
= 0, the particles are free so the spectrum is quadratic
in q. As the potential is increased the band structure appears. For small V
0
the
discontinuity occurs only at the edge of the ﬁrst Brillouin zone qa ±π and the gap
is proportional to V
0
/2. As the depth increases, the band gap increases and the
band width decreases. For very deep lattices the spectrum is almost degenerate in
q and exhibits a dependence on n similar to the one of a particle in a ﬁxed ﬁnite
interval(determined by the period of the potential).
In the absence of a lattice the eigenfunctions of the free system are plane waves.
As the lattice depth is increased the barrier height between adjacent lattices sites
increases and the eigenstates of the system tend to get localized at each lattice
site (regions around the potential minima). In Fig. 2.4 the probability density of
the Bloch wave function φ
(n=0)
q=0
is plotted for diﬀerent lattice depths. It can be
observed how the amplitude of the wave function in between adjacent lattice sites
decreases with increasinglattice depth.
16 Chapter 2 Optical lattices
1 0.5 0 0.5 1
qaΠ
0
2
4
6
8
E
E
R
Vo4E
R
1 0.5 0 0.5 1
qaΠ
0
5
10
15
E
E
R
Vo20E
R
1 0.5 0 0.5 1
qaΠ
0
2
4
6
8
E
E
R
Vo0E
R
1 0.5 0 0.5 1
qaΠ
0
2
4
6
8
E
E
R
Vo0.5E
R
Figure 2.3: Band structure of an optical lattice
2.2.2 Wannier orbitals
Wannier orbitals are a set of orthonormal wave functions that fully describe par
ticles in a band and are localized at the lattice sites. They are deﬁned as:
w
n
(x −x
i
) =
1
√
M
¸
q
e
−iqx
i
φ
(n)
q
(x), (2.15)
where the sum is over the ﬁrst Brillouin zone, M is the total number of lattice sites
and x
i
is the position of the i
th
lattice site. Wannier orbitals are thus a unitary
transformation of the Bloch functions and are formally an equivalent representation
to describe the periodic system. They constitute a more appropriate representation
as the lattice depth is increased and particles get localized at individual lattices
sites. The actual form of a Wannier function may be seen if we assume the
periodic function u
(n)
q
(x) in equation 2.10 to be approximately the same for all
Bloch states in a band. Under this approximation the Wannier function centered
at the origin can be shown to be
w
n
(x) ≈ u
(n)
(x)
sin(kx)
kx
. (2.16)
This looks like u
(n)
(x) at the site center, but spread out with oscillations of
gradually decreasing amplitude. The oscillations are needed to ensure orhogonality
between Wannier functions. In Fig. 2.5 we show the Wannier orbital centered at
the origin site for each of the lattice conﬁgurations shown in Fig.2.2. The lattice
depth in this pictures in deep enough that the Wannier orbitals are well localized.
To observe the small oscillatory tails at the neighboring sites, the plots use a
logscale for the density color map.
2.2 Single particle physics 17
0 1 2 3 4
xa
D
e
n
s
i
t
y
Φ
0
1
x
2
0 1 2 3 4
xa
D
e
n
s
i
t
y
Φ
0
1
x
2
0 1 2 3 4
xa
D
e
n
s
i
t
y
Φ
0
1
x
2
0 1 2 3 4
xa
D
e
n
s
i
t
y
Φ
0
1
x
2
Figure 2.4: Probability density of the Bloch wave function φ
(n=1)
q=0
for diﬀerent
lattice depths. It can be observed the localization of the wave function increases
with the lattice increases.
2.2.3 Tightbinding approximation
The tightbinding approximation deals with the case in which the overlap between
Wannier orbitals at diﬀerent sites is enough to require corrections to the picture
of isolated particles but not too much as to render the picture of localized wave
functions completely irrelevant. Sometimes a very good approximation is only to
take into account overlap between nearest neighbor orbitals. This tightbinding
model is commonly used to solve the problem of a particle in a periodic potential
when also an external potential is applied and it is going to be fundamental when
considering particle interactions.
By expanding the wave function in Wannier orbitals,
ψ(x, t) =
¸
n,i
z
(n)
i
(t)w
n
(x −x
i
), (2.17)
and using it in the Schr¨odinger equation that describes a particle moving in the
potential of a 1D lattice plus a perturbative external potential V (x), we get the
following equations of motion
−i
∂
∂t
z
(n)
i
=
¸
jn
−J
(n
)
ij
δ
nn
z
(n
)
j
(t) +V
(n,n
)
ij
z
(n
)
j
(t), (2.18)
with
J
(n)
ij
= −
dxw
∗
n
(x −x
i
)H
o
w
n
(x −x
j
)dx, (2.19)
V
(n,n
)
ij
= −
dxw
∗
n
(x −x
i
)(x)w
n
(x −x
j
)dx. (2.20)
18 Chapter 2 Optical lattices
y
[
λ
]
(a)
1
1
1 1
1
1
2
2
2
2
2
2
3
3
3
3
3 3
−1
0
1
(b)
1
1 1
1
2
2
3 3
3
3 3
3
(c)
1
1
2
2
3
3
x [λ]
y
[
λ
]
(d)
1
1
2
2
3
3
−1 0 1
−1
0
1
x [λ]
(e)
1
1
2
2
2
2
3
3
−1 0 1
x [λ]
(f)
1 1
1
1
2
2
2 2 3 3
3
3
−1 0 1
log
10
w
0
(r)
2
[λ
−2
]
−15 −10 −5 0
Figure 2.5: Wannier state logdensity distribution corresponding to the lattice
conﬁgurations shown in Fig. 2.2. Dotted white lines indicate the potential energy
contours. The numbers labels the near neighbor sites: 1 the nearest neighbors,
and so on. This ﬁgure is a courtesy of P. Blair Blakie [34] )
If we assume that the external perturbation is not strong enough or sharp enough
to induce interband transitions, we may represent the moving particle quite satis
factory by using Wannier functions of only the ﬁrst band. Moreover, if the lattice
is deep enough such that tunneling to next to nearest neighbors can be ignored,
and the external perturbation is a slowly varying function such that it can be as
sumed constant inside each individual lattice site, i.e. V
(n,n
)
ij
= V (x
i
)δ
ji
δ
n,n
, the
equations of motion reduce to
−i
∂
∂t
z
i
(t) = −J (z
i+1
(t) +z
i−1
(t)) +V (x
i
)z
i
(t) +ε
o
z
i
(t), (2.21)
with
J = −
dxw
∗
0
(x
i
)H
o
w
0
(x −x
i+1
)dx, (2.22)
ε
o
=
dxw
∗
0
(x)H
o
w
0
(x)dx (2.23)
where J is the tunneling matrix element between nearest neighboring lattice sites,
ε
o
is the unperturbed on site energy shift and and z
i
the ﬁrst band coeﬃcients
z
i
(t) = z
(0)
i
(t). Eq. (2.21) is known as the discrete Schr¨odinger equation (DSE) or
tightbinding Schr¨odinger equation.
For deep lattices the localized Wannier orbitals can be approximated by a
Gaussian function. However, because the Gaussian ansatz neglects the small
2.2 Single particle physics 19
oscillations characteristic in the tail of the Wannier function, the tunneling matrix
element J is then underestimated by almost an order of magnitude. For the special
case of a sinusoidal potential V
lat
(x) = V
o
sin
2
(kx) for which Mathieu functions are
the exact solutions an analytic expression for J can be obtained by using tabulated
Mathieu functions in Eq. (2.22) and ﬁtting the data. The best ﬁt we get is:
J/E
R
= α
˜
V
o
β
e
−γ
√
˜
V
o
(2.24)
with α =1.39666, β =1.051 and γ =2.12104 and
˜
V
o
= V
o
/E
R
.
2.2.4 Semiclassical dynamics
Formalism
Solving Eq. (2.21) is presumably not possible for arbitrary V (x
i
). However, we get
general insight to the nature of solutions by an application of the correspondence
principle. It is well known that wavepacket solutions of the Schr¨odinger equation
behave like classical particles obeying the equations of motion derived from the
same classical Hamiltonian. The classical Hamilton equations are
˙ x =
∂H
∂p
, ˙ p = −
∂H
∂x
. (2.25)
Using the correspondence principle and the resemblance of the crystal momentum
or quasimomentum to real momentum, the semiclassical equations of a wave packet
in the ﬁrst band of a lattice can be written as ([67], [68])
˙ x = v
(0)
(q) =
1
dE
(0)
q
dq
, ˙ q = −
dV (x)
dx
. (2.26)
The semiclassical equations of motion describe how the position and wave
vector of a particle evolve in the presence of an external potential entirely in terms
of the band structure of the lattice. If we compare the acceleration predicted by
the model with the conventional newtonian equation, m¨ x = −dV (x)/dx, we can
associate an eﬀective mass induced by the presence of the lattice, m
∗
. This is given
by
1
m
∗
=
1
2
d
2
dq
2
E
(0)
q
. (2.27)
Translationally invariant lattice
If no external potential is applied, V (x) = 0, plane waves are the solutions of DSE:
z
j
(t) = f
(q)
j
e
−it(E
q
+ε
o
)/
, f
(q)
j
=
1
√
M
e
iqaj
with M is the total number of lattice
sites. If also periodic boundary conditions are assumed, the quasimomentum q
is restricted to be an integer multiple of
2π
Ma
. The lowest energy band dispersion
relation in this case is given by
E
q
= −2J cos(qa). (2.28)
20 Chapter 2 Optical lattices
From the above equation it is possible to see the connection between J and the
band width
J =
E
q=
π
a
−E
q=0
/4. (2.29)
Linear external potential and Bloch oscillations
If V (x) is assumed to be a linear potential, V (x) =
Jξ
a
x, it can be shown (see
for example [70],[71],[72]), that in this case the energy spectrum of Eq. (2.21) is
discrete and evenly spaced
(E
s
−E
0
) = Jξs s = 0, 1, 2, 3 . . . . (2.30)
The corresponding eigenfunctions are spatially localized and the amplitudes z
j
(t)
given by :
z
j
(t) = f
(s)
j
e
−iE
s
t/
, (2.31)
f
(s)
j
= J
s−j
(−2/ξ) s = 0, 1, 2, 3, . . . . (2.32)
With J
s
the s
th
order Bessel function of the ﬁrst kind. Because the quantities f
(s)
j
are just displaced Bessel functions, the s
th
eigenstate of the system tends to be
exponentially localized around the s
th
site as ξ is increased. The localization is
almost complete when ξ = 4, i.e. when the potential energy drop over a lattice
period is equal to the zero ﬁeld band width. The localization can be observed in
Fig.2.6 where the ground state amplitudes f
(0)
i
are plotted for diﬀerent values of
ξ.
30 20 10 0 10 20 30
i site
0.6
0.4
0.2
0
0.2
0.4
0.6
f
i
0
Ξ1
30 20 10 0 10 20 30
i site
0.2
0
0.2
0.4
0.6
0.8
f
i
0
Ξ4
30 20 10 0 10 20 30
i site
0.2
0.1
0
0.1
0.2
f
i
0
Ξ0.1
30 20 10 0 10 20 30
i site
0.4
0.2
0
0.2
0.4
f
i
0
Ξ0.5
Figure 2.6: Ground state amplitudes f
(0)
i
as a function of the lattice site i for
diﬀerent values of ξ
An insight on the dynamics of a particle in this linear potential is obtained by
calculating the time dependent probability of ﬁnding it at the site j at time t ,
2.2 Single particle physics 21
[z
j
(t)[
2
, if it was in the state localized at the origin at time zero z
j
(0) = δ
0j
. After
some algebra it can be shown to be given by
P
j
= [z
j
(t)[
2
= J
2
j
4 sin(ξτ/2)
ξ
, (2.33)
with τ = tJ/. The dynamical evolution of P
j,=0
, for a given j consists of an initial
period of growth until a maximum is reached followed by an oscillatory decay. The
sequence is repeated again backwards until a time τ= 2π/ξ at which P
j,=0
vanishes
and a new cycle starts.
In Fig. 2.7, the diﬀusion of a square wave packet occupying at t = 0 the central
20 sites is plotted. It can be seen that the center of mass of the wave function
accelerates and then returns to its initial position after a period τ= 2π/ξ.
40 20 0 20 40
i site
Ψ
i
2
Τ2ΠΞ0.8
40 20 0 20 40
i site
Ψ
i
2
Τ2ΠΞ1.
40 20 0 20 40
i site
Ψ
i
2
Τ2ΠΞ0.4
40 20 0 20 40
i site
Ψ
i
2
Τ2ΠΞ0.6
40 20 0 20 40
i site
Ψ
i
2
Τ2ΠΞ0.
40 20 0 20 40
i site
Ψ
i
2
Τ2ΠΞ0.2
Figure 2.7: Diﬀusion of a square wave packet in the presence of a linear poten
tial.The parameters for the plot are ξ = 0.1.
A simpliﬁed picture of the dynamics of a wave packet in this linear potential can
also be obtained if we use the semiclassical model. If we assume a packet initially
centered at x(t = 0) = 0, with average initial quasimomentum q(t = 0) = q
o
,
the semiclassical model predicts periodic oscillations of the center of mass with
frequency ξ and amplitude 2/ξ.
x(t) =
2
ξ
(1 −cos(ξτ +q
o
a)). (2.34)
22 Chapter 2 Optical lattices
This periodic motion, known as Bloch oscillations, has been observed in systems
of cold atoms [73] and BoseEinstein condensates in optical lattices [38, 36]. On
the other hand, if the external potential is strong enough to cause interband tran
sitions, the tightbinding model does not apply anymore. In this case the atoms
may gain enough energy from the external ﬁeld to tunnel out through the energy
gap into a second band when they reach the Brillioun zone boundary. This phe
nomenon is known as of LandauZener tunneling and it also has been observed in
experiments [40].
Harmonic potential
A case of experimental interest is when, besides the periodic potential, the system
is subject to a harmonic conﬁnement, V (x) =
1
2
mω
2
T
a
2
(x/a)
2
≡ Ω(x/a)
2
, with ω
T
the trapping frequency (note that Ω is unrelated to the Rabi frequency discussed
in section 2.1). The energy spectrum in this case is determined by the equation
E
s
f
(s)
i
= −J(f
(s)
i+1
+f
(s)
i−1
) + Ωi
2
f
(s)
i
. (2.35)
It is possible to show that the solutions f
(s)
i
of Eq. 2.35 are the Fourier coeﬃ
cients of the periodic Mathiue function with period π and the eigenvalues are the
characteristic value of these periodic Mathiue functions [69]. The symmetric and
antisymmetric solutions and eigenvalues are given respectively by
f
(s=2r)
i
=
1
π
2π
0
ce
2r
x,
4J
Ω
cos(2ix)dx r = 0, 1, 2..., (2.36)
f
(s=2r+1)
i
=
1
π
2π
0
se
2r
x,
4J
Ω
sin(2ix)dx,
E
s=2r
=
Ω
4
a
2r
4J
Ω
E
s=2r+1
=
Ω
4
b
2r
4J
Ω
, (2.37)
with ce
2r
(x, 4J/Ω) and se
2r
(x, 4J/Ω) the even and odd periodic solution (with
period π) of the Mathieu equation with characteristic parameter a
2r
(4J/Ω) and
b
2r
(4J/Ω) respectively. The eigenvalues E
s
are complicated functions of
4J
Ω
and
therefore we are only going to consider expressions for certain limits:
• Case J Ω: This is the case where most experiments have been developed,
for example for a 5 E
R
lattice the parameter q =
4J
Ω
varies between 1.7x10
4
for a magnetic trap of 9 Hz to 35 for a trap of 200 Hz. In this regime the
excitation spectrum is given by:
E
s
= −2J +
√
4ΩJ(s +
1
2
) −Ω
(2s + 1)
2
+ 1
32
(2.38)
−Ω
Ω
J
(2s + 1)
3
+3(2s + 1)
2
10
+. . .
2.2 Single particle physics 23
which to a ﬁrst approximation is just the harmonic potential found if no
lattice were present but with an eﬀective frequency ω
∗
and eﬀective mass m
∗
given by
ω
∗
=
√
4ΩJ = ω
t
m
m
∗
, (2.39)
m
∗
=
2
2Ja
2
, (2.40)
It is worth it to emphasize that this result was derived under the tight
binding approximation and therefore it is not valid for very shallow lattices.
The harmonic character of the spectrum can be easily checked if we take the
continuous limit of Eq. (2.35). The diﬀerential equation obtained is just the
harmonic oscillator equation,
(E
s
+ 2J)f
(s)
(x) = −
2
2m
∗
∂f
(s)
(x)
∂x
2
+
1
2
m
∗
ω
∗2
x
2
f
(s)
(x). (2.41)
The average width of the wave function is just a
ho
=
m
∗
ω
∗
.
• Case J <Ω: The single particle spectrum in this case is given by
E
s=2n
≈ E
s=2n−1
=
Ω
4
(2n)
2
+
4J
Ω
2
1
((2n)
2
−1)
+O
4J
Ω
4
+..
n ≥ 3,
E
4
≈ E
3
=
Ω
4
16 +
4J
Ω
2
1
30
+..
,
E
2
=
Ω
4
4 +
4J
Ω
2
5
12
+..
, (2.42)
E
1
=
Ω
4
4 −
4J
Ω
2
1
12
+..
,
E
0
=
Ω
4
−
4J
Ω
2
1
2
+...
.
It can be appreciated that the spectrum is nearly twofold degenerate and it goes
like Ωn
2
to ﬁrst order in (J/Ω) . In this regime then, the trapping potential is
what mostly determines the energy spacing. In Fig. 2.8 the spectrum is shown for
diﬀerent values of
4J
Ω
.
24 Chapter 2 Optical lattices
5 10 15 20
n
th
4
162
321
E
n
J
4J1
5 10 15 20
n
th
40
1624
3208
E
n
J
4J0.1
5 10 15 20
n
th
2
4
E
n
J
4J100
5 10 15 20
n
th
1
17
33
E
n
J
4J10
Figure 2.8: Harmonic conﬁnement plus lattice spectrum: The index n labels the
eigenvalues in increasing energy order.
Chapter 3
The BoseHubbard
Hamiltonian and the superﬂuid
to Mott insulator transition
In chapter 2 we studied the physics of non interacting particles in a periodic system.
In this chapter we go one step ahead and describe the basic properties of a bosonic
many body system with repulsive interactions in a periodic lattice potential.
The simplest non trivial model that describes interacting bosons in a periodic
potential is the Bose Hubbard Hamiltonian. It includes the main physics that
describe strongly interacting bosons, which is the competition between kinetic and
interaction energy. This model has been used to describe many diﬀerent systems
in solid state physics, like short correlation length superconductors, Josephson ar
rays, critical behavior of
4
He and, recently, cold atoms in optical lattices [43]. It
has been studied analytically with many diﬀerent techniques such as mean ﬁeld
approximations [74, 75, 77], renormalization group theories [44] and strong cou
pling expansions [78]. Numerically most of the studies are based on quantum
Monte Carlo methods and density matrix renormalization techniques [79]. The
Bose Hubbard Hamiltonian predicts a quantum phase transition from a superﬂuid
to a Mott insulator state. This transition has been observed experimentally in
atoms conﬁned in a 3D optical lattice [46]. In this chapter we start by review
ing the basic properties of the Bose Hubbard Hamiltonian and outline the most
important aspects that characterize the superﬂuid to Mott insulator transition.
3.1 BoseHubbard Hamiltonian
We begin the analysis by considering the second quantized Hamiltonian that de
scribes interacting bosonic atoms in an external trapping potential plus lattice. In
the grand canonical ensemble it is given by:
26
Chapter 3 The BoseHubbard Hamiltonian and the superﬂuid to
Mott insulator transition
ˆ
H =
dx
ˆ
Φ
†
(x)
−
2
2m
∇
2
+V
lat
(x)
ˆ
Φ(x) + (V (x) −µ)
ˆ
Φ
†
(x)
ˆ
Φ(x) (3.1)
+
1
2
dxdx
t
ˆ
Φ
†
(x)
ˆ
Φ
†
(x
t
)V
at
(x −x
t
)
ˆ
Φ(x)
ˆ
Φ(x
t
),
where
ˆ
Φ
†
(x) is the bosonic ﬁeld operator which creates an atom at the position
x, V
lat
(x) is the periodic lattice potential, V (x) denotes any additional slowly
varying external potential that might be present (such as a magnetic trap), µ is
the chemical potential and acts as a lagrange multiplier to ﬁx the mean number
of atoms in the grand canonical ensemble and V
at
(x) is the interatomic scattering
potential, which is in general a complicated function. However, since for cold gases
the thermal de Broglie wavelength is much larger than the eﬀective extension of the
interaction potential, the only relevant scattering process is the swave scattering,
and the actual interparticle potential plays a minor role. Therefore, to a good
approximation, the interatomic potential can be replaced by an eﬀective contact
interaction:
V
at
(x) ≈
4πa
s
2
m
δ(x), (3.2)
with a
s
is the scattering length and m the mass of an atom. In this thesis we
are going to assume that the scattering length is positive so the interactions are
repulsive.
Similar to the noninteracting situation, where we used Wannier orbitals to ex
pand the single particle wave function, it is convenient to expand the ﬁeld operator
in a Wannier orbital basis. If the chemical potential of the system is less than the
single particle excitation energy to the second band, only the ﬁrst band has to be
considered and the ﬁeld operator can be written as:
ˆ
Φ(x) =
¸
n
ˆ a
n
w
0
(x −x
n
), (3.3)
where w
0
(x) is the Wannier orbital of the lowest vibrational band localized at the
origin, and ˆ a
n
is the annihilation operator at site n which obeys bosonic canonical
commutation relations. The sum is taken over the total number of lattice sites.
If Eq. (3.3) is inserted in
ˆ
H and only tunneling between nearest neighbors is
considered, we obtain the Bose Hubbard Hamiltonian
ˆ
H
BH
= −J
¸
<n,m>
ˆ a
†
n
ˆ a
m
+
1
2
U
¸
n
ˆ a
†
n
ˆ a
†
n
a
n
a
n
+
¸
n
(V
n
−µ)ˆ a
†
n
ˆ a
n
, (3.4)
with
3.2 SuperﬂuidMott insulator transition 27
J = −
dxw
∗
0
(x −x
n
)
ˆ p
2
2m
+V
lat
(x)
w
0
(x +x
n+1
), (3.5)
U =
4πa
s
2
m
dx[w
0
(x)[
4
, (3.6)
V
n
= V (x
n
). (3.7)
The ﬁrst term in the Hamiltonian proportional to J is a measure of the kinetic
energy of the system. J is the hopping matrix element between nearest neighbors
(see also 2.24). Here the notation < n, m > restricts the sum to nearestneighbors
sites. Nexttonearest neighbor tunneling amplitudes are typically two orders of
magnitude smaller than nearestneighbor tunneling amplitudes, and to a good
approximation they can be neglected.
The second term determines the interaction energy of the system. The param
eter U measures the strength of the repulsion of two atoms at lattice site n.The
integral (3.6) is not as sensitive as Eq. (3.5) to the oscillatory tails characteristic
of the Wannier orbitals, and a Gaussian approximation can be used to estimate
it [43]. Under the Gaussian approximation, the ground state wave function cen
tered at the origin of a separable sinusoidal lattice, with lattice depth V
j0
in the
j
th
direction, has the form w
0
(0) ∼
¸
j
exp −
x
j
√
2a
hoj
2
, with a
hoj
=
mω
hoj
and
ω
hoj
=
4E
r
V
j0
2
. Using this ansatz in the integral (3.6) yields an onsite interaction
given by
U =
a
s
√
2π
ω
ho
a
ho
, (3.8)
with the bar indicating the geometric mean. To guarantee the validity of the one
band approximation, the mean interaction energy per particle UN must be smaller
than the energy gap to the ﬁrst vibrational excitation, i.e. UN < ω
ho
. N is the
mean number of atoms. This inequality is readily satisﬁed in practice.
The third term in the Hamiltonian takes into account the energy oﬀset at site
n due to a slowly varying external potential V (x).
The realization of the BoseHubbard Hamiltonian using optical lattices has the
advantage that the interaction matrix element U and the tunneling matrix element
J can be controlled by adjusting the intensity of the laser beams. As the intensity
of the lattice is increased the tunneling rate decreases exponentially and the onside
interaction increases as a power law, V
d/4
0
, (d is the dimensionality of the lattice).
3.2 SuperﬂuidMott insulator transition
At zero temperature the physics described by the BoseHubbard Hamiltonian can
be divided into two diﬀerent regimes. One is the interaction dominated regime
28
Chapter 3 The BoseHubbard Hamiltonian and the superﬂuid to
Mott insulator transition
when J is much smaller than U, and the system is in the Mott insulator phase.
The other is the kinetic energy dominated regime, where tunneling overwhelms the
repulsion and the system exhibits superﬂuid properties. The onset of superﬂuidity
is a consequence of the competition between the kinetic energy, which tries to
delocalize the particles, and the interaction energy, which tries to localize them
and make the number ﬂuctuations small.
In the superﬂuid regime, the kinetic energy term dominates the Hamiltonian.
In this regime, quantum correlations can be neglected and the system can be
described by a macroscopic wave function since the many body state is almost
a product over identical single particle wave functions. There is a macroscopic
welldeﬁned phase and the system is a superﬂuid. Because atoms are delocalized
over the lattice with equal relative phases between adjacent sites, they exhibit an
interference pattern when the lattice is turned oﬀ, as expected from an array of
phase coherent matter wave sources. As interaction increases the average kinetic
energy required for an atom to hop from one site to the next becomes insuﬃcient
to overcome the potential energy cost. Atoms tend to get localized at individual
lattice sites and number ﬂuctuations are reduced. In the Mott insulator phase
the ground state of the system instead consists of localized atomic wave functions
with a ﬁxed number of atoms per site. The lowest lying excitations that conserve
particle number are particlehole excitations (adding plus removing a particle from
the system). This phase is characterized by the existence of an energy gap. The
gap is determined by the energy necessary to create one particlehole pair.
The phase diagram [44],[74][76] that describes the Mott insulator transition
of a translational invariant system with V
n
= 0 exhibits lobelike Mott insulating
phases in the Jµ plane, see Fig. 3.1. Each Mott lobe is characterized by having
a ﬁxed integer density. Inside these lobes the compressibility, ∂ρ/∂µ with ρ the
average density of the system, vanishes. The physics behind this diagram can be
understood as follows: If we start at some point in the Mott insulating phase and
increase µ keeping J ﬁxed, there is going to be a point at which the kinetic energy
of adding an extra particle and letting it hop around will balance the interaction
energy cost. With an extra particle free to move around the lattice phase coherence
is recovered and the system enters the superﬂuid regime. Similarly, by decreasing
µ from a point in the Mott phase, at some point eventually it will be energetically
favorable to remove one particle from the system. The extra mobile hole will induce
also phase coherence and the system will condense in a superﬂuid state. Since the
kinetic energy of the system increases with J the width of the lobes decreases with
J. The distance in the µ direction at ﬁxed J between the upper and lower part of
the lobe is the energy gap. At J = 0 the gap is just equal to U. Also at J = 0 the
intersection points between the lobes with density ρ = n and the ones with density
ρ = n + 1 are degenerate. Because there is no energy barrier to the addition of
an extra particle at these degenerated points the system remain superﬂuid. Mott
insulator phases occur only at integer densities; noninteger density contours lie
entirely in the superﬂuid phase because there is always an extra particle that can
hop without energy cost.
3.2 SuperﬂuidMott insulator transition 29
The phase diagram includes two diﬀerent types of phase transition. One type
takes place at any generic point of the phase boundary, and it is driven by the
energy cost to add or subtract small numbers of particles to the incompressible
Mott state as explained above. On the other hand, the other type only occurs
at ﬁxed integer density and takes place at the tip of the lobes. This transition is
driven at ﬁxed density by decreasing U/J and enabling the bosons to overcome the
on site repulsion. The two kinds of phase transition belong to diﬀerent universality
classes. In the generic one, the parameter equivalent to the reduced temperature
δ = T −T
c
for ﬁnite temperature transitions that measures the distance from the
transition is δ ∼ µ − µ
c
, with µ
c
the chemical potential at the phase boundary.
For the special ﬁxed density on the other hand one must take δ ∼ (J/U) −(J/U)
c
.
As shown in ([44]) provided δ ∼ µ − µ
c
, the compressibility has a singular part
scaling near the transition as δ
−α
, α > 0. This scaling does not apply at the
special ﬁxed density MottSuperﬂuid transition for which δ ∼ (J/U) − (J/U)
c
and diﬀerentiation with respect to delta becomes inequivalent to diﬀerentiation
with respect to the chemical potential. Most of the features of the phase diagram
discussed above can be veriﬁed by simple calculations done using a mean ﬁeld
approximation. The solution of the mean ﬁeld model [74] [76] predicts that the
critical value at the tip of the Mott lobe depends on the density and dimensionality
of the lattice as
(U/J)
c
= t
2n + 1 +
(2n + 1)
2
−1
, (3.9)
where n is the integer density of the lobe, t the number of nearest neighbors,
t = 2d, with d the dimensionality of the system. The parabolic phase boundary of
the lobes in the plane µ vs (J/U) vs is given by
µ/J =
1
2
(2n −1) −Jt/U ±+
(Jt/U)
2
−2(Jt/U)(2n + 1) + 1
.(3.10)
Up to this point we have discussed the basic aspects of the superﬂuidMott
insulator transition in a translationally invariant system. The situation is funda
mentally diﬀerent for a inhomogeneous system with a ﬁxed total number of atoms
and external conﬁnement. This is the case realized in experiments, where besides
the lattice there is a harmonic trap that collects the atoms at the center. In this
case the density of atoms is not ﬁxed since the atoms can redistribute over the
lattice and change the local ﬁlling factor.
To deal with the inhomogeneous case, it is possible to deﬁne an eﬀective local
chemical potential, µ
n
= µ−V
n
, at each lattice site n [43]. If the change in the mean
number of atoms between neighboring sites is small, the system can be treated
locally as an homogeneous system. Because in the inhomogeneous case, the local
chemical potential is ﬁxed by the density, as the ratio U/J is changed the system
can locally cross the boundary between the superﬂuid and Mott insulator phases.
Therefore even for the situation when the local density was not commensurate at
30
Chapter 3 The BoseHubbard Hamiltonian and the superﬂuid to
Mott insulator transition
JUc
JU
µ
U
Phase Diagram
n1
n2
n3
n4
Figure 3.1: Zerotemperature phase diagram. The vertical lines indicate critical
values for diﬀerent ﬁlling factors
the beginning, as the ratio U/J is changed a local phase transition can take place.
For the inhomogeneous case, the gradient in the local chemical potential leads to
a shell structure with Mott insulator regions and superﬂuid regions in between.
In [79], the authors used quantum Monte Carlo simulations to study the ground
state of the one dimensional Bose Hubbard model in a trap. They found that
Mott phases exist in extended domains above a threshold interaction strength,
even without the commensurate ﬁlling required in the translationally invariant
case.
Chapter 4
Mean ﬁeld theory
In this chapter we introduce the mean ﬁeld Discrete Nonlinear Schr¨odinger equa
tions and use it ﬁrst to investigate the behavior of a BEC adiabatically loaded into
an optical lattice moving at constant velocity, and second to model the dynamical
evolution of a BEC initially loaded into every third site of an optical lattice.
4.1 Discrete nonlinear Schr¨odinger equation (DNLSE)
In the weakly interacting regime, quantum ﬂuctuations can be neglected to a good
approximation. In this regime most of the atoms occupy the same condensate
wave function. The macroscopic occupation of a single mode implies that the
commutator of the annihilation and creation operators that create and destroy a
particle in the condensate mode can be neglected with respect to the total number
of atoms in the mode. Therefore, to a good approximation the ﬁeld operator can
be replaced by a cnumber,
ˆ
Φ(x) →ϕ(x). (4.1)
This procedure can be interpreted as giving to the ﬁeld operator a non zero average,
'
ˆ
Φ(x)` = ϕ(x), and thus a well deﬁned phase to system. Because the original
Hamiltonian is invariant under global phase transformation, this deﬁnition of BEC
corresponds to an spontaneous symmetry breaking [11, 12, 80]. The function ϕ(x)
is a classical ﬁeld having the meaning of an order parameter and is often called
the ”condensate wave function”. It characterizes the oﬀdiagonal long range order
present in the one particle density matrix [81].
lim
[x
−x[→∞
'
ˆ
Φ
†
(x
t
)
ˆ
Φ(x)` = ϕ
∗
(x
t
)ϕ(x). (4.2)
Strictly speaking, neither the concept of broken symmetry, nor the one of oﬀ
diagonal long range order can be applied to ﬁnite size systems with well deﬁned
number of particles. However the condensate wave function can still be determined
by diagonalization of the one particle density matrix. The eigenstate with larger
eigenvalue corresponds to ϕ(x).
32 Chapter 4 Mean ﬁeld theory
If we use the approximation (4.1) in the manybody Hamiltonian (3.1) we get
a classical Hamiltonian given by:
H =
dxϕ
†
(x)
−
2
2m
∇
2
+V
lat
(x)
ϕ(x) + (V (x) −µ)ϕ
†
(x)ϕ(x) (4.3)
+
1
2
4πa
s
2
m
dxϕ
†
(x)ϕ
†
(x)ϕ(x)ϕ(x),
The time dependent equations of motion for the condensate wave function can be
obtained from the variational principle by minimizing the quantity i
∂
∂t
−
ˆ
H. The
minimization yields
i
∂ϕ(x)
∂t
=
−
2
2m
∇
2
+V
lat
(x) +V (x) +
4πa
s
2
m
[ϕ(x)[
2
ϕ(x) (4.4)
This equation is known as the GrossPitaevskii equation (GPE). It was derived by
independently by Gross and Pitaevskii [10].
If the chemical potential is small compared to the vibrational level spacing
and the lattice depth is deep enough that a tight binding picture is valid, the
condensate order parameter can be expanded in a Wannier basis keeping only the
lowest band orbitals:
ϕ(x, t) =
¸
n
z
n
(t)w
0
(x −x
n
). (4.5)
Using this ansatz in Eq. (4.4), the GrossPitaevskii equation reduces to the
discrete nonlinear Schroedinger equation (DNLSE), where the z
n
amplitudes sat
isfy:
i
∂z
n
∂t
= −J
¸
<n,m>
z
m
+(V
n
+UN[z
n
[
2
)z
n
, (4.6)
where < n, m > restricts the sum over nearest neighbors, the parameters J and U
are the ones deﬁned in Eq. (3.5) and (3.6) and N is the total number of particles.
The normalization condition for the amplitudes is given by:
¸
n
[z
n
[
2
= 1. (4.7)
Notice the DNLSE can also be derived by assuming the ﬁeld operators present in
the BoseHubbard Hamiltonian to be cnumbers,i.e. ˆ a
n
= z
n
(t). This assumption
leads to a tightbinding classical Hamiltonian, which can be minimized using the
variational principle.
H = −J
¸
<n,m>
z
∗
n
z
m
+
¸
n
V
n
[z
n
[
2
+
U
2
[z
n
[
4
. (4.8)
4.2 Dragging experiment 33
The DNLSE is similar to its linear counterpart, the DSE or tightbinding
Schr¨odinger equation (see 2.21), but with an extra cubic term, which takes into
account mean ﬁeld interaction eﬀects. To see the modiﬁcation that the nonlinear
term introduces in the dispersion relation, we consider a one dimensional optical
lattice with M sites and periodic boundary conditions.
In the translational invariant case with no other external potential present,
V
n
= 0, the solutions of the DNLSE that are similar to those of the noninteracting
case are, z
n
=
√
n
o
exp(−inq) exp (−iE
q
t/), with n
o
= N/M the condensate
density. Using this plane wave ansatz in the DNLSE we get a dispersion relation
given by:
E
q
= −2J cos(qa) +Un
o
(4.9)
In the tightbinding approximation, mean ﬁeld interactions do not modify the
shape of the lowest band: it has still a cosine dependence with bandwith equal to
4J. The only diﬀerence with respect to the single particle case is an energy shift
in the dispersion relation which takes into account the average mean ﬁeld energy
Un
o
.
It is important to mention though, that outside the range of validity of a
tightbinding model, nonlinear eﬀects does not necessarily change the dispersion
relation in a simple way, especially at the edge of the Brillouin zone. It has been
shown in the literature [82, 83] that if the chemical potential is bigger than the
gap vibrational excitation energy to higher bands, the lowest band becomes triple
valued near the Brilloiun zone edge and a loop appears at this point. This loop
reﬂects the existence of new solutions, which only exist when the tightbinding
approximation is not valid. These new solutions have a non zero velocity at the
edge of the Brillouin zone, as opposed to the linear case. The nonzero velocity
carried by the Bloch waves is a manifestation of the superﬂuidity of the system.
For free particles the ﬂow is stopped completely by Bragg scattering from the
periodic potential, but if interactions are strong enough the superﬂuid ﬂow can no
longer be stopped.
4.2 Dragging experiment
In this section we investigate an experiment done at NIST where a BoseEinstein
condensate was prepared in a harmonic trap and adiabatically loaded into a one
dimensional optical lattice moving at constant velocity Ref. [37]. We ﬁrst study the
dynamics in a translationally invariant lattice using a simple single particle picture,
and show how the centerofmass motion of the condensate can be used as a probe
of the lattice band structure. Then we discuss the eﬀects of the trapping potential
and interaction eﬀects on the dynamics. When mean ﬁeld eﬀects are included,
they cause the eﬀective mass of the Bloch state to depend on time, inﬂuencing
the condensate dynamics in the lattice. As a consequence, characteristic nonlinear
eﬀects such as solitons, self focusing eﬀects and dynamical stabilities might be
observed.
34 Chapter 4 Mean ﬁeld theory
4.2.1 Experiment
In the experiment, a condensate prepared in the ground state of a harmonic trap
is loaded in an optical lattice, which is moved at constant velocity. The loading
of the condensate in the optical lattice is done by turning on the lattice, linearly
in time, up to some maximum height. The important part of this turning on is
that it is done suﬃciently slow to be adiabatic with respect to band excitations.
After the maximum is reached, at t = t
r
, the lattice depth is held constant for a
ﬁxed period of time and the system is allowed to expand in the lattice. Finally at
t = t
f
the condensate density distribution along the lattice axis is imaged. The
temporal dependence of the optical lattice depth is shown in Fig. 4.1. In the ﬁgure,
experimental results obtained for the ﬁnal condensate density for diﬀerent lattice
depths are also shown. The general observation is that the deeper the lattice, the
greater the response of the condensate to the lattice motion. In this experiment,
the condensate has a very narrow momentum spread compared to the lattice recoil
momentum.
Figure 4.1: Condensate density in the xdirection at the observation time t
f
. Light
colors areas indicate regions of high condensate density. The dashed line indicates
the ﬁnal location of a comoving point with the optical lattice.
4.2 Dragging experiment 35
4.2.2 Linear free particle model
In this section to a ﬁrst approximation we are going to neglect interactions and
apply a single particle picture to understand the dynamics.
Because the momentum spread of the condensate is very narrow compared with
the lattice recoil momentum at t = 0, the initial wave function can be approximated
by a plane wave with zero momentum. This wave function, in the lattice frame,
which is moving with velocity v
latt
with respect to the lab frame, has the form
ϕ(x
t
) = e
−ik
o
x
, with v
latt
= k
o
/m (we used primes to denote the coordinates
and wave functions in the moving frame). As t increases, the optical potential is
ramped up, the wave function evolves, but due to the fact that the Hamiltonian
is invariant under translation, the quasimomentum is conserved (at least in this
linear model with no other potential than the lattice) during the ramping time and
the average quasimomentum at the end of the loading should remain −k
o
. From
t
r
to t
f
the lattice depth is held constant. During this period of time the average
velocity of the atom, as seen in the lattice frame, is:
v =
1
dE
(n)
q
dq
q=−k
o
, (4.10)
with E
(n)
q
the n
th
band Blochlike atomic dispersion relation evaluated at the ﬁnal
lattice depth. Transforming back to the lab frame, Eq. (4.10) implies that in the
lab frame, during t
r
to t
f
, the average velocity of the atoms is given by:
v
atom
=
k
o
m
+
1
dE
(n)
q
dq
q=−k
o
. (4.11)
Because x(t
f
) is the observable of the experiment, we can calculate it as
δx = 'x(t
f
)` −'x(t
r
)` ≈ v
atom
(t
f
−t
r
). (4.12)
Therefore, by measuring the average position of the atoms for a ﬁxed t
f
− t
i
the experiment measures the atom’s velocity, which is a determined by the band
structure of the lattice.
In Fig. 4.2 we plot the atom’s average velocity in the lab frame vs. the lattice
velocity. The velocity in Eq.(4.11) is calculated using the Bloch dispersion relation
for a three recoil lattice. The tightbinding results (dotted) are also shown. Even
though the tightbinding velocities do not exactly match the velocities calculated
using the Bloch dispersion relation, we can say that for a three recoil lattice the
tightbinding approximation is fair.
We emphasize the following points:
• Because the initial quasimomentum is determined by the lattice velocity,
v
lat
= k
o
/m, to move the lattice at diﬀerent speeds corresponds to loading
the condensate at a diﬀerent Bloch state. How much the atoms are dragged
depends on the derivative of the dispersion relation evaluated at the loaded
quasimonentum.
36 Chapter 4 Mean ﬁeld theory
0 0.5 1 1.5
v
latt
v
B
0
0.5
1
v
a
t
o
m
v
B
Vo3 Er
Bloch
TBA
Figure 4.2: Atom’s average velocity in the lab frame vs. lattice velocity. Line: re
sults from the Bloch dispersion relation, dots: tightbinding approximation results.
v
B
= π/am.
• At zero lattice depth the dispersion relation is quadratic,
1
dE
(n)
q
dq
q=−k
o
=
k
o
/m and therefore Eq. (4.11) predicts no dragging. As the lattice depth
increases the dispersion relation desviates more from freeparticle behavior
and dragging ensues. This is consistent with the experimental results shown
in Fig. 4.1.
• For small lattice velocities ([k
o
a[ < 1, a the lattice constant) the atoms are
almost stationary because then the dispersion relation is more freeparticle
like. On the other hand, as k
o
approaches the k of the Brillouin zone edge,
k
o
= −q
B
, the atoms are dragged the most. At this point, as opposed to the
free free system, Bloch waves carry zero velocity, i.e dE
(n)
q
/dq[
q=−q
B
= 0.
4.2.3 Eﬀect of the magnetic conﬁnement
In the experiment a harmonic conﬁnement was present in addition to the lattice
potential. To model the eﬀect of the magnetic trap on the atoms dynamics we
used a semiclassical model (see chapter 2). The semiclassical model is a very good
approximation whenever it is unnecessary to specify the position of the atom on a
scale comparable with the spread of the wave packet. The validity of the semiclas
sical model requires that the external potential varies slowly over the dimension of
the wave packet and ignores the possibility of inter band transitions. In the exper
iment, the strength of the harmonic trap is less than one percent of the potential
depth and we expect the semiclassical model to be valid. Following this model,
in the lattice frame, the time evolution of the average position and wave vector of
4.2 Dragging experiment 37
the atoms are given by Eq. (2.26):
˙ x = v
(n)
(q) =
1
dE
(n)
q
dq
, ˙ q = −mω
2
(x +v
lat
t), (4.13)
with ω the magnetic trapping frequency. The initial conditions are k(0) = −ko
and x(0) = 0. We solved these equations of motion, using the analytic dispersion
relation for a sinusoidal lattice in terms of Mathieu functions. We used the pa
rameters V
0
= 3E
R
and 1/2mω
2
/E
R
= 0.000025. The results are shown in Fig.
4.3.
0 0.5 1 1.5
v
latt
v
B
0
0.5
1
v
a
t
o
m
v
B
Semiclassical model Vo3 Er
no trap
trap
Figure 4.3: Atom’s average velocity in the lab frame vs. lattice velocity in the
presence of a magnetic trap. v
B
= π/am.
We observe in the plot that when atoms are loaded in the ﬁrst band, for a
range of lattice velocities below half the Brillouin velocity (v
latt
v
B
/2 with
v
B
= π/am), the motion of the atoms with and without trap is almost the same.
For larger velocities, as long as the atoms remain in the ﬁrst band, trapped atoms
move faster than the untrapped ones. However, if the lattice velocities exceeds
the Brillouin velocity, v
B
, the opposite behavior is observed and for an interval of
velocities, v
B
< v
latt
3v
B
/2, trapped atoms slow down compared with the ones
in the homogeneous lattice.
This behavior can be understood as follows: Because of the trap, the atoms feel
a force opposite to the direction of motion. The force not only increases with time
but also with the lattice velocity (see Eq. (4.13)). The average quasimomentum of
the system evolves in the presence of the force, and if at time t = 0 the atoms are
loaded with quasimomentum q(0) = −ko, then at t
f
the ﬁnal quasimomentum is
q(t
f
) = −k
o
+∆(k
o
, t
f
). Here ∆(k
o
, t
f
) is a negative quantity because the force is
a restoring force. Therefore, the ﬁnal velocity of the atoms in the trap , is less or
greater than the velocity of the atoms without the external trap depending on the
38 Chapter 4 Mean ﬁeld theory
sign of the eﬀective mass (sign of the curvature of the dispersion relation) evaluated
at the ﬁnal momentum. If the eﬀective mass is positive,
d
2
E
q
dq
2
> 0, (0 < q < π/2a,
π/a < q < 3π/2a,...), the group velocity is an increasing function of q and the
velocity of the trapped atoms is less than the untrapped atoms’ velocity. On the
other hand, when the eﬀective mass is negative,(π/2a < q < π/a,...), the group
velocity is a decreasing function of q and the trapped atoms speed up with respect
to the untrapped ones. The more pronounced diﬀerence between trapped and un
trapped atom velocities observed in the second band with respect to the ﬁrst one,
is not only because ∆ increases with the lattice velocity but also because of the
discontinuous change of the band structure when crossing the Brillouin zone edge.
Besides the sign diﬀerence in the curvature the second band has a larger band
width than the ﬁrst one.
4.2.4 Interaction eﬀects
To model the eﬀect of the interatomic interactions on the dynamics we used the
tightbinding variational method proposed in Ref. ([84, 85]). In these reference
the authors assume a Gaussian proﬁle wave packet
z
n
(t) =
4
2
γ
2
π
e
[−(n−ξ)
2
/γ
2
+ik(n−ξ)+iδ(n−ξ)
2
/2]
, (4.14)
and used ξ, γ, k, δ as variational parameters. The equations of motion they found
are:
˙
k = −
∂V
∂ξ
, (4.15)
˙
ξ = 2J sin(k)e
−η
, (4.16)
˙
δ = 2J cos(k)e
−η
4
γ
4
−δ
2
+
2NU
√
πγ
3
−
4
γ
∂V
∂γ
, (4.17)
˙ γ = 2J cos(k)e
−η
γδ, (4.18)
with η =
1
2γ
2
−
δ
2
γ
2
8
, V =
2/γ
2
π
dxV (x) exp[−2(x − ξ)
2
/γ
2
] and the time
scaled as t →tE
r
/. Notice that the variable k is just the average quasimomentum
of the wave packet. We can also associate an inverse eﬀective mass given by
m
∗−1
= 2J cos(k)e
−η
.
We numerically integrated the equations of motion assuming no other external
trapping potential besides the lattice, V = 0. In this case the Hamiltonian is
invariant under translations and we expect conservation of the quasimomentum.
This is seen in the above equations, because if V = 0 then dk/dt = 0. The
parameters were chosen according to the experiment: γ(0) = 10, δ(0) = 0 and
NU = 2E
R
. The bandwidth J, was found using Mathieu functions assuming a
3E
R
lattice. We plot Fig. 4.4 v
atom
vs v
latt
evaluated at t = 50. The curve found
is plotted in
When mean ﬁeld eﬀects are included, the dynamics of the system becomes
very rich. Depending on the the initial quasimomentum and width of the packet,
4.2 Dragging experiment 39
0 0.2 0.4 0.6 0.8 1 1.2 1.4
v
latt
v
B
0
0.2
0.4
0.6
0.8
1
v
a
t
o
m
v
B
v
atom
v
latt
linear
non linear
Figure 4.4: Velocity of the atoms predicted by the variational model in the case
when interactions are taken into account (dotted line). The single particle case is
shown in red. The parameters used were γ(0) = 10, δ(0) = 0 and NU = 2E
R
.
a variety of dynamical regimes can be observed. Physically, the diﬀerent regimes
can be understood by realizing that the nonlinear term makes the eﬀective mass
time dependent. The eﬀective mass, m
∗
= e
η
/(2J cos k), not only depends on the
initial quasimomentum but also on the width of the wave packet, and when the
nonlinearity is present η is no longer constant. The diﬀerent dynamical behaviors
include eﬀects such as selftrapping, diﬀusion, and for special cases solitons or
breathers [84].
The selffocusing is a genuine nonlinear eﬀect, characterized by a diverging
eﬀective mass. In particular the selftrapped wave packet cannot translate along
the array. The ﬁnal value for γ
f
, for a packet in the selffocusing regime predicted
by the variational model is,
γ
f
=
UNγ
o
UN −4J
√
πcos(k)γ
o
, (4.19)
With γ
o
= γ(0) the initial width of the packet, which is assumed big compared to
one.
The diﬀusive regime, on the other hand, is characterized by the unbounded
growth of the width of the packet. This is the behavior expected in regular linear
quantum mechanics. The model predicts that the transition between the two
regimes occurs, again assuming γ
o
1, at
(UN)
c
= 4J
√
πcos(k)γ
o
[k[ < π/2
(UN)
c
=
2J
√
π[cos(k)[
γ
o
[k[ > π/2 (4.20)
40 Chapter 4 Mean ﬁeld theory
Using a potential depth of 3E
R
and γ
o
= 10, the predicted dynamical phase
diagram as a function of UN vs v
lat
is shown in Fig. 4.5. It can be seen that for
UN = 2E
R
, the selffocusing regime is expected for lattice velocities greater than
v
lat
= 0.4v
B
. At v
lat
= 0.5v
B
the noninteracting eﬀective mass becomes negative.
0 0.2 0.4 0.6 0.8 1
v
lat
v
B
0
2
4
6
8
U
N
Phase Diagram Vo 3 Er
Diffusion Self Focusing
Figure 4.5: Phase diagram UN vs v
latt
for V
o
= 3E
r
and γ
o
= 10.
Consistently, in Fig. 4.4 we see that for lattice velocities approximately greater
than v
lat
= 0.4v
B
, the velocity of the interacting system becomes greater than the
noninteracting one, and approximately at v
lat
= 0.7v
B
the velocity of the atoms is
just the velocity of the lattice so, the atoms stop moving in the lattice frame. This
regime is a selftrapping regime and nonlinear eﬀects tend to localize the atoms
in the lattice frame. On the other hand, for v
latt
0.4v
B
the dynamics of the
system is similar to the non interacting case. To appreciate the diﬀerent behaviors
exhibited by the system below and above v
B
/2 we plot in Figs. 4.6 and 4.7, the
evolution of the wave packet, for two diﬀerent initial velocities. For v
latt
= −0.7v
B
the wave packet localizes, whereas for v
latt
= −0.15v
B
the wave packet diﬀuses.
4.3 Dynamics of a periodthree patternloaded Bose
Einstein condensate in an optical lattice
In this section we discuss the dynamics of a BoseEinstein condensate initially
loaded into every third site of a one dimensional optical lattice, motivated by
the recent experimental realization of this system by the NIST group [33]. A
condensate loaded in this way is not an eigenstate of the ﬁnal perioda lattice, and
the condensate will continue to evolve in the ﬁnal system. Outside the strongly
4.3 Dynamics of a periodthree patternloaded BoseEinstein
condensate in an optical lattice 41
30 0 30
n
0
0.5
1
z
n
t
2
t30
30 0 30
n
0
0.5
1
1.5
z
n
t
2
t40
30 0 30
n
0
0.5
1
1.5
2
z
n
t
2
t50
30 0 30
n
0
0.5
z
n
t
2
t0
30 0 30
n
0
0.5
z
n
t
2
t10
30 0 30
n
0
0.5
z
n
t
2
t20
Figure 4.6: Localization of the wave packet in the selffocusing regime v
latt
=
−0.7v
B
. Notice the change in the vertical scale in the last row
6030 0 30
n
0
0.5
z
n
t
2
t30
6030 0 30
n
0
0.5
z
n
t
2
t40
6030 0 30
n
0
0.5
z
n
t
2
t50
6030 0 30
n
0
0.5
z
n
t
2
t0
6030 0 30
n
0
0.5
z
n
t
2
t10
6030 0 30
n
0
0.5
z
n
t
2
t20
Figure 4.7: Evolution of the wave packet in the diﬀusive regime v
latt
= −0.15v
B
.
correlated regime, the GPE equation is expected to give a good description of
the condensate dynamics. In our system we assume the lattice is suﬃciently deep
that a tightbinding description is applicable, and the DNLSE is valid. For the
periodic initial condition of equal wavefunction amplitude at every third site and
zero at all others, symmetry arguments can be used to reduce the wavefunction
evolution to a two mode problem (analogous to a double well system with an energy
oﬀset) for which an analytic solution of the dynamics can be given. We show
that for large ratios of the interatomic interaction strength to tunneling energy
the condensate evolves with selfmaintained population imbalance, whereby the
condensate population tends to remain localized in the initially occupied lattice
sites. A similar phenomenon has been studied in doublewell systems [87, 88, 89,
90, 91]. We show that the momentum distribution of an interacting condensate
changes in time in a manner which can be related to the spatial tunneling of
condensate, and would be a suitable experimental observable.
42 Chapter 4 Mean ﬁeld theory
Under the inﬂuence of an external force a Bloch state will exhibit Bloch oscil
lations, as described in chapter 2. To illustrate how a linear potential aﬀects the
motion of a pattern loaded condensate we ﬁnd analytic solutions for the noninter
acting case with periodic initial conditions and show how the dynamics for this
system can be interpreted in terms of the interference of three Bloch states of the
lowest band undergoing Bloch in unison. We present numerical results for more
general (nonperiodic) initial conditions and consider characteristic properties of
the momentum distribution.
4.3.1 Experiment
Figure 4.8: Patterned loading experiment.
In the experiment, a combination of two independently controlled lattices was
used to load the condensate into every third site of a single lattice. Brieﬂy the
procedure consists, as shown in Fig. 4.8, of loading a condensate into the ground
band of lattice with periodicity 3a, so that the condensate is well localized in the
potential minima of this lattice. A second lattice of periodicity a which is parallel
to the ﬁrst lattice, is then ramped up so that the superimposed light potentials
form a superlattice of period 3a. For the ideal case both lattice potentials are
inphase, and the addition of the second lattice will not shift the locations of the
4.3 Dynamics of a periodthree patternloaded BoseEinstein
condensate in an optical lattice 43
potential minima from those of the ﬁrst lattice alone, and the condensate will
remain localized at these positions. Finally, by removing the ﬁrst period3a lattice
on a time scale long compared to band excitations, but short compared to the
characteristic time of transport within the lattice, the condensate will be left in
every third site of the perioda lattice.
4.3.2 Case of no external potential
The major interest of this section is in the tunneling properties of the condensate in
the lattice, and in the noninteracting case the time scale for tunneling is determined
by the hopping matrix element J. For this reason it is convenient to deﬁne a new
dimensionless scale of time τ = Jt/, and dimensionless energy E
n
= V
n
/J and
coupling constant Λ = NU/J. In terms of these new variables the tightbinding
evolution equation (4.6) takes the form
i
∂z
n
∂τ
= −(z
n−1
+z
n+1
) + (E
n
+ Λ[z
n
[
2
)z
n
, (4.21)
• The case of periodic initial conditions: reduction to a two mode system
We treat ﬁrst a model case in which no external potential is present, (E
n
=
0), and in which the initial occupancies of each third site are the same, and in
which the condensate initially has a uniform phase. At τ = 0, the amplitudes
z
n
(τ) are given by
z
3n
(0) =
√
ρ, (4.22)
z
3n+1
(0) = z
3n+2
(0) = 0, (4.23)
where ρ = 3/M with M the total number of lattice sites. Nρ represents
the initial number of atoms per occupied site. This initial condition is ho
mogeneous in the sense that each occupied site has the same amplitude and
phase along the length of the lattice. For an inﬁnite lattice, or one with peri
odic boundary conditions, the amplitudes for all initially occupied sites (z
3n
)
evolve identically in time, and the amplitudes for the initially unoccupied
sites satisfy z
3n+1
(τ) = z
3n+2
(τ) for all τ and all n. This allows us to reduce
the full set of equations (4.21) to a set of two coupled equations
i
∂z
0
∂τ
= −2z
1
+ Λ[z
0
[
2
z
0
, (4.24)
i
∂z
1
∂τ
= −(z
1
+z
0
) + Λ[z
1
[
2
z
1
, (4.25)
where z
3n
≡ z
0
and z
3n+1
= z
3n+2
≡ z
1
for all n. The normalization condi
tion is
[z
0
[
2
+ 2[z
1
[
2
= ρ. (4.26)
44 Chapter 4 Mean ﬁeld theory
The Hamiltonian function, H, of this system is (from Eq. 4.8)
H =
J
ρ
−2(z
∗
0
z
1
+z
0
z
∗
1
) −2[z
1
[
2
+
Λ
2
[z
0
[
4
+Λ[z
1
[
4
(4.27)
=
JΛρ
2
.
By writing z
0
=ψ
0
√
ρ and z
1
=ψ
1
ρ/2, we can transform Eqs. (4.24)(4.25)
to the form
i
∂
∂τ
ψ
0
ψ
1
=
γ[ψ
0
[
2
−
√
2
−
√
2 γ
[ψ
1
[
2
√
2
−1
ψ
0
ψ
1
, (4.28)
where γ = Λρ is the ratio of onsite repulsion to tunneling energies, and the
normalization condition (4.26) is now [ψ
0
[
2
+ [ψ
1
[
2
= 1. The factor of
√
2
diﬀerence in the deﬁnition of ψ
0
and ψ
1
arises because ψ
1
represents the am
plitude of the two initially unoccupied sites. With this factor incorporated,
the matrix appearing in Eq. (4.28) is explicitly Hermitian. We note that
this equation of motion is identical to that for a condensate in a double well
trap in the two mode approximation [92, 93].
We note in passing that a similar reduction, to a system of [m/2] + 1 equa
tions, exists for lattice systems that are loaded such that only every mth
site is initially occupied.
• Solution of the equations of motion
It is convenient to write the lattice site amplitudes appearing in Eqs. (4.24)
and (4.25) as z
0
=fe
iθ
0
√
ρ and z
1
= ge
iθ
1
√
ρ, where f, g, and θ are real. By
introducing the phase diﬀerence φ = θ
0
− θ
1
, Eqs. (4.24), (4.25) and (4.27)
can be recast as
.
f = 2g sin φ, (4.29)
.
g
= −f sin φ, (4.30)
γ
2
= −4fg cos φ +
γ
2
f
4
+ 2g
4
−2g
2
, (4.31)
The analytic solutions of Eqs. (4.29)(4.31), found using a procedure similar
to that presented by Raghavan et al. [87], can be expressed in terms of
Weierstrassian elliptic functions ℘(τ; g
2
, g
3
) [69]. The analytic solutions are
f(τ) =
1 −
24
12℘(τ; g
2
, g
3
) + (9 + 2γ +γ
2
)
, (4.32)
g(τ) =
1 −f(t)
2
2
, (4.33)
4.3 Dynamics of a periodthree patternloaded BoseEinstein
condensate in an optical lattice 45
where the parameters g
2
and g
3
are given by
g
2
=
81 −14γ
2
+4γ
3
+γ
4
/12, (4.34)
g
3
=
729 + 243γ
2
−46γ
3
−15γ
4
+γ
6
/216. (4.35)
The solutions f(τ) and g(τ) are oscillatory functions whose amplitudes and
common period, T(γ), are determined by the parameter γ (see Figs. 4.9
and 4.10). It is useful to qualitatively divide this behavior into two regimes,
separated by γ = 2. Analysis of Eqs. (4.29)  (4.31) shows that f(τ
0
) = g(τ
0
)
for some value of τ
0
when γ ≤ 2, and for γ > 2, f(τ) > g(τ) for all τ.
Tunneling dominated regime
For γ 2, we ﬁnd that the oscillation period is essentially constant (see Fig.
(4.9).
γ (J)
T
(
h
/
J
)
Figure 4.9: Oscillation period (in units of /J) as a function of the interaction
strength γ.
In this case the role of interactions is relatively small, and the behavior can be
approximately understood by taking γ = 0, in which case the matrix of Eq.
(4.28) is constant in time. The equations of motion in this case are equivalent
to those of a twostate Rabi problem [94], where the two levels are coupled
by a Rabi frequency of strength
√
2, which is detuned from resonance by
−1. This system will undergo Rabi oscillations whereby atoms periodically
tunnel from the initially occupied site into the two neighboring sites. Because
the coupling is detuned from resonance the transfer of populations between
wells is incomplete, with [ψ
1
[
2
attaining a maximum value of 8/9. The Rabi
model predicts that the cycling frequency between the levels is equal to the
diﬀerence between the eigenvalues of the matrix of matrix of Eq. (4.28),
which gives the period of oscillation as T
Rabi
= 2π/3 in units of /J (see Fig.
4.9).
Interaction dominated regime
46 Chapter 4 Mean ﬁeld theory
The eﬀect of interactions on the meanﬁeld dynamics is to cause the energies
of the initially occupied sites to shift relative to those of the unoccupied sites.
As γ increases and this energy shift increases relative to the strength of cou
pling between sites, the tunneling between sites occurs at a higher frequency,
but with reduced amplitude. The population of the initially occupied sites
becomes self trapped by the purely repulsive pair interaction, which in the
context of a double well system has been called “macroscopic quantum self
trapping” [93]. This is demonstrated quantitatively in Fig. 4.10 where we
plot the minimum value of f
2
occurring during the oscillation as a function
of γ. In contrast to the tunneling dominated regime, where tunneling peri
odically populates all sites equally, the condensate tends to be localized on
the initially occupied sites in the interactiondominated regime.
(
)
γ (J)
Figure 4.10: Minimum value of f
2
during an oscillation period as a function of γ.
As γ increases the population imbalance between wells increases (see text).
• Momentum space dynamics
Typically the spacing between individual wells in an optical lattice is too
small to resolve the localized density distributions of atoms in neighboring
sites using standard imaging techniques.
The momentum distribution is a more convenient observable which approx
imately corresponds to the expanded spatial distribution of the released
condensate. Here we calculate the momentum dynamics of the condensate
loaded into every third site of an optical lattice, and show how this relates
to the evolution of the spatial amplitudes given in Eqs. (4.32) and (4.33).
In the tightbinding approximation the condensate order parameter ϕ(x, τ)
(4.4) is expressed as a sum over the lattice sites . Because of the periodicity of
the system, the momentum space wavefunction, which we denote as ˜ ϕ(k, τ),
is expressible as the Fourier series
˜ ϕ(k, τ) =
¸
m
z
m
(τ)e
−ikam
χ(k), (4.36)
4.3 Dynamics of a periodthree patternloaded BoseEinstein
condensate in an optical lattice 47
where
χ(k) =
1
√
2π
∞
−∞
e
−ikx
w
0
(x)dx. (4.37)
To compute the momentumdistribution, we invoke the identity
¸
M−1
n=0
e
ikna
=
Mδ
k,2πm/a
, where M is the number of lattice sites, and m an integer. Since
there are only two independent amplitudes in the set ¦z
m
¦, we ﬁnd that
˜ ϕ(k, τ) =
3/ρc
m
(τ)χ
m
δ
k,qm/3
,
χ
m
= χ(qm/3), (4.38)
c
m
(τ) =
1
3ρ
z
0
+z
1
e
−iqm/3
+z
2
e
−i2qm/3
,
where q is the reciprocal lattice vector q = 2π/a. The momentum distri
bution of the system has very sharp peaks of relative amplitude [c
m
[
2
at
momentum k = qm/3, arising from the 3lattice site spatial periodicity of
the condensate wavefunction. In addition, χ
m
describes a slowly varying en
velope determined by the localization of the Wannier states at each lattice
site.
Using the analytic solutions for f and g, and Eq. (4.31), we obtain only two
independent Fourier amplitudes
[c
3n
(τ)[
2
=
1
3
1 +
γ
4
(3f
2
+ 1)(f
2
−1)
, (4.39)
[c
3n+1
(τ)[
2
=
1
3
1 −
γ
8
(3f
2
+ 1)(f
2
−1)
(4.40)
= [c
3n+2
(τ)[
2
. (4.41)
In the reduced zone scheme, where we only consider momenta in the range
k ∈ [−q/2, q/2], the momentum wavefunction then consists of three peaks
corresponding to Bloch states of quasimomenta 0, ±q/3. The identical be
havior of [c
3n+1
[
2
and [c
3n+2
[
2
means that the ±q/3 peaks always have the
same intensity. If interactions between the atoms are ignored (i.e. γ = 0),
the momentum components are constant in time (see Eqs. (4.39) and (4.40)),
even though tunneling occurs between the lattice sites. However, when in
teractions are considered, the momentum intensities explicitly depend on the
occupations of each site and will vary in time when tunneling occurs. The
magnitude of the time variation of the [c
n
[
2
is proportional to γ, but will
reduce for suﬃciently large values of γ, where the selftrapping eﬀect causes
the tunneling between lattice sites to stop (i.e. f
2
≈ 1 at all times). In Fig.
4.11 we show the maximum contrast between the intensity of the Fourier
peaks, deﬁned as ∆
max
≡ ([c
1
[
2
− [c
0
[
2
)
max
, where the value is maximized
by evaluating the [c
n
[
2
at the time when f takes its minimum value (we
note that f(τ) is given by Eq. (4.32)). We note that the maximum contrast
occurs for γ ≈ 3.9; in this case, the zero quasimomentum component c
0
(τ)
48 Chapter 4 Mean ﬁeld theory
vanishes once during each period of oscillation T(γ). For values of γ greater
than 3.9 the contrast between the intensities starts to decrease due to the
reduction in tunneling caused by the nonlinearity induced selftrapping.
0 5 10 15 20
0
0.1
0.2
0.3
0.4
0.5

∆
m
a
x

2
γ (J)
Figure 4.11: Maximum contrast of the Fourier components as a function of γ. The
maximum contrast is deﬁned as ∆
max
≡ ([c
1
[
2
−[c
0
[
2
)
max
with the maximum value
occurring when f
2
is at its minima.
• Application to an inhomogeneous condensate
Here we wish to consider the dynamics for an inhomogeneous condensate,
applicable to a condensate initially prepared in a harmonic trap. For the pat
tern loaded condensate, we use inhomogeneous to refer to the overall spatial
envelope of the period three initial condition. The previous homogeneous
theory we have presented is expected to accurately describe inhomogeneous
cases when the initial pattern of population in every third site extends over
many lattice sites i.e. M 1 so that meanﬁeld energy associated with
each triplet of sites U(n) =
Λ
2
¸
3
i=1
[z
3n+i
[
4
varies slowly across the system.
Taking a particular example we choose a Gaussian envelope to the periodic
arrangement of atoms into every third site, so that the initial state is
z
3n+1
(0) = z
3n+2
(0) = 0, (4.42)
z
3n
(0) = 3
¸
2
πM
2
1/4
exp(−(9n/M)
2
). (4.43)
For the simulations shown here, the parameters used were N = 10
5
, M = 76,
U = 2.11 10
−5
E
R
and J = 0.075 E
R
; these correspond to a condensate
of 10
5
atoms of
87
Rb produced in a magnetic trap with axial and radial fre
quencies of 9 and 12 Hz respectively, and loaded into a lattice with a depth of
4.5E
R
, with E
R
= 2.2kHz. These parameters are typical of the experimental
regime, but also lie in a range in which the homogeneous model is expected to
give a fair description. The total number of occupied wells and the strength
of the on site interatomic interaction were calculated by preserving the value
4.3 Dynamics of a periodthree patternloaded BoseEinstein
condensate in an optical lattice 49
of the chemical potential of the system upon reduction to one spatial dimen
sion and by assuming that each of the localized orbitals in the tightbinding
description are Gaussian. The hopping rate J was estimated by using Math
ieu functions. In Fig. 4.12 we show the evolution of the population of the
central wells (normalized to one) compared with the homogeneous model
with γ taken to be the local mean ﬁeld energy γ
eﬀ
= Λ
2
π
(9/M).
Fig. 4.12 shows the results of numerical integration of the equations of motion
and the approximate analytical solution given by the quasihomogeneous
model described above. We see that the numerical and analytical results
agree well at short times, but diﬀer more as time progresses due to the
diﬀerent meanﬁeld seen by diﬀerent wells.
Figure 4.12: Comparison between the population evolution of the central three
wells for the inhomogeneous condensate and the homogeneous model. Inhomoge
neous condensate: stars for the initially populated well and boxes for the initially
empty wells. Homogeneous model: dashed line represents the initially populated
wells, and the solid line represented the initially unpopulated wells. We used γ
eﬀ
as the local mean ﬁeld energy (see text). The parameters used for the simulation
were J = 0.075E
R
and γ
eﬀ
= 2.64.
To understand the disagreement as time evolves, we show in Fig. 4.13 the
numerical Fourier spectrum for the inhomogeneous case evaluated at several
diﬀerent times. The variation of the intensities of the peaks, which, as shown
below, is related to the spatial tunneling between lattice sites in the presence
of the mean ﬁeld, can be seen in the plot. Initially all occupied sites are in
phase and the three distinctive momentum peaks have a narrow width deter
mined by the intrinsic momentum uncertainty of the condensate envelope.
That is the reason why the homogeneous model ﬁts very well. However, as
time progresses the mean ﬁeld variation across the lattice causes the tun
neling rates to vary with position and leads to momentum peak broadening.
This eﬀect eventually causes the homogeneous model to become an inaccu
rate description of the inhomogeneous system.
We note that momentum space signature for spatially tunneling in the in
50 Chapter 4 Mean ﬁeld theory
t = 1.7ms t = 2.6ms
t = 0ms
t = 0.8ms
momentum momentum
d
e
n
s
i
t
y
(
a
r
b
.
u
n
i
t
s
)
0 q/3 2q/3 0 q/3 2q/3
d
e
n
s
i
t
y
(
a
r
b
.
u
n
i
t
s
)
momentum momentum
0 q/3 2q/3 0 q/3 2q/3
Figure 4.13: Momentum distribution of the inhomogeneous condensate evaluated
at various times for the same parameters as used in Fig.4.12.
teracting system is still present in the inhomogeneous case. This is shown in
Fig. 4.14, where we plot the evolution of the quantities [c
3n
(τ)[
2
, [c
3n+1
(τ)[
2
,
[c
3n+2
(τ)[
2
(calculated from the numerical simulation by partitioning the
numerical Fourier spectrum in three equal non overlapping sections, each
centered around the respective peak and adding the square of the norm of
the Fourier components within each section) vs. the ones calculated with the
homogeneous model, but using an averaged value γ
ave
≡ Λ
¸
n
[z
n
[
4
instead
of γ = Λρ. It can be observed that the predictions of the simple model are
in very good agreement with the numerical results when the three peaks of
the spectrum are well deﬁned. For longer times, the width of the Fourier
peaks increases, until a point when they split. At this point the quantities
[c
3n
(τ)[
2
, [c
3n+1
(τ)[
2
, [c
3n+2
(τ)[
2
are not meaningful anymore. Because the
parameters used for the numerical calculations were chosen to be experimen
tally achievable, and as shown in the plots the model predictions are fair at
least for one period of oscillation, we conclude that the Fourier distribution
can be used as a signature of the meanﬁeld quantum tunneling inhibition.
4.3 Dynamics of a periodthree patternloaded BoseEinstein
condensate in an optical lattice 51
Figure 4.14: Evolution of momentum peak populations. Upper curves: population
of the q = ±2π/3 momentum states. Lower curves: population of the q = 0
momentum state. Inhomogeneous condensate (dotted), homogeneous result (solid
line), where the comparison is made by replacing γ by an average mean ﬁeld energy
γ
ave
≡ Λ
¸
n
[z
n
[
4
= 1.85. Parameters are the same as in Fig. 4.12.
4.3.3 Dynamics with a constant external force
• Homogeneous three state model
In this section we consider the dynamics of a periodically loaded condensate
in the presence of a linear external potential, corresponding to a uniform force
parallel to the lattice. In what follows we assume that the force is suﬃciently
weak that band excitations due to LandauZener tunneling are negligible, so
that a tightbinding picture of the lowest band is suﬃcient to describe the
dynamics. In this case the evolution equation diﬀers from what we considered
in the previous section by the term E
n
in Eq. (4.21) taking the form E
n
=
nξ, where ξ is the potential diﬀerence between lattice sites (in units of the
hopping matrix element J). Taking the initial conditions (4.22)(4.23), and
transforming the Wannier amplitudes as z
3n+j
(τ) =
˜
Ψ
3n+j
(τ)e
−i3nξτ
(j =
0, 1, 2), we obtain the evolution equations
i
∂
˜
Ψ
3n
∂τ
= −(
˜
Ψ
3n−1
e
i3ξτ
+
˜
Ψ
3n+1
)
+Λ[
˜
Ψ
3n
[
2
˜
Ψ
3n
, (4.44)
i
∂
˜
Ψ
3n+1
∂τ
= −(
˜
Ψ
3n+2
+
˜
Ψ
3n
) +ξ
˜
Ψ
3n+1
+Λ[
˜
Ψ
3n+1
[
2
˜
Ψ
3n+1
, (4.45)
i
∂
˜
Ψ
3n+2
∂τ
= −(
˜
Ψ
3n+1
+
˜
Ψ
3n+3
e
−i3ξτ
) + 2ξ
˜
Ψ
3n+1
+Λ[
˜
Ψ
3n+1
[
2
˜
Ψ
3n+1
. (4.46)
Assuming periodic boundary conditions, the periodicity of the initial condi
tions and equations of evolution allow considerable simpliﬁcation from the
52 Chapter 4 Mean ﬁeld theory
full set of M coupled equations. In particular, these assumptions mean that
every third Wannier amplitude evolves identically (i.e.
˜
Ψ
n
=
˜
Ψ
n+3
) and so
the evolution of the system can hence be reduced to the three independent
equations
i
∂
˜
Ψ
0
∂τ
= −(
˜
Ψ
2
e
i3ξτ
+
˜
Ψ
1
) + Λ[
˜
Ψ
0
[
2
˜
Ψ
0
, (4.47)
i
∂
˜
Ψ
1
∂τ
= −(
˜
Ψ
2
+
˜
Ψ
0
) +ξ
˜
Ψ
1
+ Λ[
˜
Ψ
1
[
2
˜
Ψ
1
, (4.48)
i
∂
˜
Ψ
2
∂τ
= −(
˜
Ψ
1
+
˜
Ψ
0
e
−i3ξτ
) + 2ξ
˜
Ψ
2
+Λ[
˜
Ψ
2
[
2
˜
Ψ
2
, (4.49)
where the new amplitudes map onto the original set according to
˜
Ψ
0
↔
¦
˜
Ψ
3n
¦,
˜
Ψ
1
↔ ¦
˜
Ψ
3n+1
¦, and
˜
Ψ
2
↔ ¦
˜
Ψ
3n+2
¦, and obey the normalization
condition
2
¸
j=0
[
˜
Ψ
j
[
2
=
3
M
. (4.50)
The equations of motion (4.47)(4.49) are more diﬃcult to treat analytically
than the case considered in the last section due to the presence of a linear
potential. In this section we derive an analytic solution for a noninteracting
condensate (i.e. Λ=0), which provides valuable insight into the complicated
tunneling dynamics the system exhibits in the absence of nonlinearity, yet
should furnish a good description for dilute condensates satisfying γ < 1.
For the nonlinear regime we present numerical results to illustrate the typical
behavior.
• Analytic solution for linear dynamics
Deﬁning the vector x(t) = (
˜
Ψ
0
(τ),
˜
Ψ
1
(τ),
˜
Ψ
2
(τ)), and using the transforma
tion y(τ) = P(τ)x(τ), where P(τ) is the unitary matrix
P(τ) =
¸
¸
1 1 1
e
−iξτ
e
−i(ξτ−
2π
3
)
e
−i(ξτ+
2π
3
)
e
−i2ξτ
e
−i(2ξτ+
2π
3
)
e
−i(2ξτ−
2π
3
)
¸
, (4.51)
the linear version of Eqs. (4.47)(4.49) can be decoupled, directly yielding
the solutions
˜
Ψ
n
(τ) =
e
−inξτ
√
3M
e
−i∆
0
(τ)
+e
i(
2nπ
3
−∆
1
(τ))
+e
−i(
2nπ
3
+∆
2
(τ))
, (4.52)
4.3 Dynamics of a periodthree patternloaded BoseEinstein
condensate in an optical lattice 53
where we have deﬁned the phase terms
∆
n
(τ) = −
2
ξ
¸
sin(ξτ −
2nπ
3
) + sin(
2nπ
3
)
, (4.53)
for n = 0, 1, 2.
These solutions for the spatial amplitudes
˜
Ψ
i
(τ) can be most easily under
stood by considering a Bloch state decomposition of the condensate wave
function. The nature of our system allows us to construct an analytic form
for the initial wavefunction. Because the system has a three lattice site period
and is assumed to be in the lowest band, the wavefunction can be expressed
as a superposition of three Bloch waves (of the lowest band) which are sym
metrically spaced in quasimomentum. Assuming the condensate initially has
a total crystal momentum of zero, at this time the wavefunction must be of
the form
ϕ(x, 0) = α
0
φ
0
(x) +α
+
φ
q/3
(x) +α
−
φ
−q/3
(x), (4.54)
where φ
k
(x) is a Bloch state with quasimomentum k, and the α are complex
constants determined by the lattice depth, with [α
+
[ = [α
−
[.
The action of an external force on a Bloch state causes it to linearly change
its quasimomentum in time according to
k(τ) = −
ξ
a
τ +k(0). (4.55)
The periodicity of the Bloch dispersion relation in k, and hence of the group
velocity of the Bloch wave, gives rise to the wellknown phenomenon of Bloch
oscillations (see chapter 2). For the case we are considering here, the system
consists of three Bloch states whose quasimomenta will translate in unison
under the action of the external force. During this evolution each state
accumulates phase at a rate determined by the instantaneous Bloch energy,
i.e.
∆
n
(τ) =
τ
0
E(k
n
(s))ds, (4.56)
where k
n
(τ) = −ξτ/a+k
n
(0) is the quasimomentum of Bloch state n at time
t. In the tightbinding approximation the dispersion relation for the Bloch
states has the analytic form
E(k) = −2 cos(ka), (4.57)
for which ∆
n
(τ) can be evaluated, and yields the results given in Eqs. (4.53)).
The wavefunction evolution in the Bloch basis is
ϕ(x, τ) = α
0
φ
−
ξ
a
t
(x)e
−i∆
0
(τ)
+α
+
φ
−
ξ
a
t+
q
3
(x)e
−i∆
1
(τ)
+α
−
φ
−
ξ
a
t−
q
3
(x)e
−i∆
2
(τ)
. (4.58)
54 Chapter 4 Mean ﬁeld theory
From this solution we can obtain solutions for the evolution of the spatial
amplitudes Eqs. (4.52). To do this we expand the Bloch states in terms of
Wannier functions according to, φ
k
(x) =
¸
n
e
ikna
w
0
(x−na) (see chapter 2)
and make use of Eq. (4.5). Note: we take all α =
ρ
3
as determined by the
initial conditions, Eqs.(4.22) and (4.23).
Bloch Oscillations
The evolution of the spatial amplitudes, and in particular the population
in each well is then determined by the interference of the Bloch phases
∆
n
. These functions are all periodic in time with period τ
B
= 2π/ξ (in
units of τ = tJ/). This is the normal Bloch oscillation period, and
gives the time scale over which the quasimomenta of the Bloch states
increases by exactly one reciprocal lattice vector.
Small ξ solution  Non classical transport
To understand the dynamics, we ﬁrst start by considering the case when
ξ is small. For this case, the population in the wells is given by
[
˜
Ψ
0
(τ)[
2
=
1
3M
5 + 4 cos(3τ) +ξ
2
h(τ)
(4.59)
[
˜
Ψ
1
(τ)[
2
=
1
3M
(2 + 3ξ)[1 −cos(3τ) −
ξ
2
2
h(τ)]
(4.60)
[
˜
Ψ
2
(τ)[
2
=
1
3M
(2 −3ξ)[1 −cos(3τ) −
ξ
2
2
h(τ)]
(4.61)
where
h(τ) =
τ
3
2
(6τ +3τ cos(3τ) −4 sin(3τ)) . (4.62)
The above solution shows that when the force is applied, the degeneracy
in the population of the wells represented by
˜
Ψ
1
and
˜
Ψ
2
is lifted. For
ξ > 0, atoms in
˜
Ψ
0
start to tunnel to
˜
Ψ
1
more rapidly than to
˜
Ψ
2
. This
should be compared with the results in the absence of the force, where
˜
Ψ
1
and
˜
Ψ
2
behave identically. Thus, in this weak limit, the eﬀect of a
linear potential is to enhance the tunneling from the initial populated
3n wells to their 3n + 1 neighbors ones, making the system closer to
resonance, in the sense of Eq. (4.28), where the resonance condition
results in the initially populated wells becoming empty at some later
time. It is interesting to note that the system exhibits ”nonclassical”
dynamics whereby the atoms start to tunnel in the direction opposite
the direction of the force (this statement applies even when the external
ﬁeld is not weak).
Resonances
In Fig. 4.15 we show the temporal evolution of the spatial amplitudes
˜
Ψ
i
for a range of values of ξ. For certain choices of ξ the initially occupied
˜
Ψ
0
amplitude periodically disappears  we refer to these as resonances.
4.3 Dynamics of a periodthree patternloaded BoseEinstein
condensate in an optical lattice 55
0
ξ=
5
14
ξ
ξ=3ξ
ξ=
2
7
ξ
ξ= ξ
τ
0
ξ=0
0
ξ=
1
2
ξ
(h/J) τ (h/J)
τ (h/J) τ (h/J)
τ (h/J) τ (h/J)
Figure 4.15: Evolution of the normalized population for diﬀerent values of ξ. One
Bloch period is shown in the plots except ξ = 0 where the period is inﬁnite. Solid
line: [
˜
Ψ
3n
[
2
, dotted line: [
˜
Ψ
3n+1
[
2
, dashdot line: [
˜
Ψ
3n+2
[
2
. The ”nonclassical”
motion can be seen where the 3n+1well populations initially increase more rapidly
that the populations of the 3n+2wells. It can also be seen in the plots that
ξ = ξ
res
max
/2 and ξ = ξ
res
max
are resonant values.
By requiring
˜
Ψ
0
= 0 in Eq. (4.52) we obtain the following conditions
on the phases for these resonances
∆
1
(τ) −∆
0
(τ) = ±
π
3
+ 2πs
, (4.63)
∆
2
(τ) −∆
1
(τ) = ±
π
3
+ 2πm
, (4.64)
where s and m are integers and the same sign choice on the right hand
side must be made for both equations. When these conditions are satis
ﬁed the population is not shared between the adjacent sites, but prefer
entially tunnels to one of the neighbors. In particular, the +sign choice
in Eqs. (4.63) and (4.64) gives the phase condition for complete tunnel
ing into
˜
Ψ
1
. Similarly the −sign case gives the condition for complete
tunneling into
˜
Ψ
2
. Solving for the values of ξ for which Eqs. (4.63) and
56 Chapter 4 Mean ﬁeld theory
0 20 40 60 80 100
n
0.1
0.2
0.3
0.4
0.5
Figure 4.16: The spectrum of values of external force, ordered in decreasing mag
nitude, for which a population resonance occurs i.e. the values of ξ for which
˜
Ψ
0
periodically disappears.
(4.64) hold, one obtains the spectrum
ξ
res
n=(s,m)
= ±
6
√
3
π
(1 + 3s)
(1 + 3s)
2
+ 3(s + 2m+1)
2
, (4.65)
which is shown in Fig. 4.16. For large [s[ or [m[ the resonant values of
ξ are close together, and become spaced further apart as the magnitude
of s decreases.
In the absence of an applied force (i.e. ξ = 0) the phase terms are
timeindependent with ∆
0
= −2, ∆
1
= −1 and ∆
2
= −1 so that the
resonance conditions can never be achieved. When the external force
is applied the phases oscillate at a rate that increases with ξ and an
amplitude that decreases with ξ. The largest value of ξ for which a res
onance can be found is ξ
res
max
≡ 6
√
3/2π, since for values of ξ greater than
this the amplitudes of the phase oscillations ∆
i
are so small that they
cannot satisfy the condition for population resonance. In the regime
ξ > ξ
res
max
, [
˜
Ψ
0
[ exhibits only one (nonzero) minima per Bloch period
and the dynamics of the system is dominated by the Bloch oscillations.
Large Force Limit
For ξ > ξ
res
max
the system exhibits a population imbalance as the Bloch
oscillation suppresses the ability of the system to tunnel between sites.
In the limit ξ ξ
res
max
the population in the wells is described by rapid,
small amplitude oscillations around its initial value,
[
˜
Ψ
0
(τ)[
2
−→
3
M
1 −
8
ξ
2
sin
2
ξτ
2
, (4.66)
[
˜
Ψ
1
(τ)[
2
−→
12
M
1
ξ
2
sin
2
ξτ
2
, (4.67)
[
˜
Ψ
2
(τ)[
2
−→
12
M
1
ξ
2
sin
2
ξτ
2
. (4.68)
4.3 Dynamics of a periodthree patternloaded BoseEinstein
condensate in an optical lattice 57
The case of ξ = 3ξ
res
max
depicted in Fig.4.15 shows an approach to this
behavior.
• Numerical results for the nonlinear case
(h/J)
(h/J)
Figure 4.17: Eﬀects of interactions on generalized Bloch oscillations for the pattern
loaded system. Evolution of [
˜
Ψ
n
[
2
for various interaction strengths. Upper plot:
ξ = 2ξ
res
max
. Lower plot: ξ = 2ξ
res
max
/7.
Because of the diﬃculty of solving analytically the equations of motion the
nonlinear equations were solved numerically. Although the dynamics in this
situation is considerably more complicated than in the linear case, the reso
nance picture still gives us useful guidance concerning the expected behavior.
In general, there exist critical values of ξ and Λ above which no resonances oc
cur. Furthermore, nonlinearity tends to destroy the periodicity of the Bloch
oscillations that is present in the noninteracting case. For example, in Fig.
4.17, we see that for ξ = 2ξ
res
max
, the introduction of interactions brings the
58 Chapter 4 Mean ﬁeld theory
system into resonance, but that for large interaction strength, nonresonant
behavior is restored. For ξ = 2ξ
res
max
/7, on the other hand, the introduction
of interactions eventually draws the system out of resonance.
Concerning the momentum distribution, similarly to the untilted case, inter
atomic interactions induce time variation of the momentum intensities. The
contrast between momentum components vanishes at Λ = 0. As Λ increases,
the contrast increases to a maximum value, then eventually decreases towards
zero when macroscopic imbalance is achieved, analogous to what is seen in
Fig. 4.11 for the untilted case. Because the external ﬁeld together with the
nonlinearity breaks down the periodicity of time evolution, the dynamics of
the momentum distribution is quite complex.
To compare the predictions of the homogeneous model in the presence of a
linear external potential to a more realistic case, we again solved numerically
the discrete nonlinear Schr¨odinger equation for a condensate loaded into
every third lattice site but, instead of being homogeneous, initially with a
Gaussian proﬁle. Very good agreement between the model and the numerical
results was found for short times and modest meanﬁeld energies, if, as in the
untilted case, we use an eﬀective mean ﬁeld energy. In Figs. 4.18 and 4.19
we present a comparison between the evolution of the normalized population
at the central wells found numerically and the prediction of the model for the
parameters: NU = 4.8E
R
, M = 290, J = 0.075 E
R
and ξ = 2. For longer
times, the model predictions start to disagree with numerical simulation due
to the spread of the three Fourier peaks. We observe that the dynamics are
much more sensitive to inhomogeneous eﬀects in the presence of an external
force, and for the more inhomogeneous initial state used in Figs. (4.12 
4.14) (which extends over M = 76 sites), the inhomogeneous result much
more rapidly departs from the homogeneous prediction than the example
presented here.
4.3 Dynamics of a periodthree patternloaded BoseEinstein
condensate in an optical lattice 59
Figure 4.18: Comparison between the evolution of a inhomogeneous condensate
with the homogeneous result. Inhomogeneous condensate: [
˜
Ψ
0
[
2
(boxes), [
˜
Ψ
1
[
2
(stars), and [
˜
Ψ
2
[
2
(diamonds). Homogeneous case: [
˜
Ψ
3n
[
2
(dashdot line), [
˜
Ψ
3n+1
[
2
(solid line), and [
˜
Ψ
3n+2
[
2
(dotted line), where we have taken γ as the local mean
ﬁeld energy. The parameters used were J = 0.075E
R
, ξ = 2 and γ
eﬀ
= 1.59
Figure 4.19: Evolution of the momentum peak populations. q = 0 (dashed line,
squares), q = −2π/3 (dashdot line, stars), q = 2π/3 (solid line, diamonds). Inho
mogeneous condensate (dotted curves), homogeneous model (lines) using the same
parameters as those in Fig. 4.18, but replacing γ by an average value γ
ave
= 1.11
for the homogeneous model.
60 Chapter 4 Mean ﬁeld theory
Chapter 5
Quadratic approximations
As discussed in chapter 3, the BoseHubbard Hamiltonian is an appropriate model,
when the loading process produces atoms in the lowest vibrational state of each
well, with a chemical potential smaller than the energy gap to the ﬁrst vibrationally
excited state. The quartic form of the Hamiltonian makes it very diﬃcult to deal
with it in all the diﬀerent regimes. The aim of this chapter is to rewrite it so that
a systematic approach can be developed in the weakly interacting regime, when
a condensate is present. The basic idea, which was ﬁrst proposed by Bogoliubov
in 1947 [11], is to treat the ﬁeld operator as a cnumber plus a ﬂuctuating term.
The cnumber describes the condensate or the coherent part of the matter ﬁeld.
In the weakly interacting regime, quantum ﬂuctuations are small, and therefore
the dominant terms in the Hamiltonian are quadratic in the ﬂuctuating ﬁeld. The
nonquadratic terms should be of higher order and can be treated perturbatively.
In this chapter, we apply the Bogoliubov approximation to the Bose Hub
bard Hamiltonian and derive the correspondent Bogoliubovde Genes(BdG) equa
tions as developed in Refs. [77, 95]. We also deal with the higher order terms
by considering diﬀerent approximations. Among them we discuss the Hartree
FockBogoliubov (HFB) approximation, the HFBPopov approximation and the
improved Popov approximation where the bare potential is upgraded to the many
body scattering matrix. All of these approximate approaches are valid only in
the weakly interacting regime when quantum ﬂuctuations are small. They will
fail to give a good description as interactions become important and the system
approaches the Mott insulator transition. To test the validity of the approxima
tions we compared them with numerical solutions obtained by diagonalizing the
BoseHubbard Hamiltonian.
In this chapter we also derive an explicit expression for the superﬂuid density
based on the rigidity of the system under phase variations. We show how the
superﬂuid fraction can be thought of as a natural order parameter to describe the
superﬂuid to Mott insulator transition.
Because we are interested in the quantum eﬀects caused by interactions that
drive the quantum phase transition instead of thermal quantum ﬂuctuations, the
analysis in the present chapter is going to be restricted to the zero temperature
62 Chapter 5 Quadratic approximations
case.
5.1 The characteristic Hamiltonian
In the very weakly interacting regime, to a good approximation, the creation and
annihilation operators on each site can be replaced by a cnumber(see chapter
4). To include the small quantum ﬂuctuations in the description of the system
we assume that the ﬁeld operator can be written in terms of a cnumber plus a
ﬂuctuation operator:
ˆ a
n
= z
n
+ ˆ ϕ
n
. (5.1)
Replacing this expression for ˆ a
n
in the BoseHubbard Hamiltonian Eq. (3.4)
leads to :
ˆ
H = H
0
+
ˆ
H
1
+
ˆ
H
2
+
ˆ
H
3
+
ˆ
H
4
, (5.2)
with
H
0
= −J
¸
<n,m>
z
∗
n
z
m
+
¸
n
¸
(V
n
−µ)[z
n
[
2
+
U
2
[z
n
[
4
, (5.3)
ˆ
H
1
= −J
¸
<n,m>
ˆ ϕ
n
z
∗
m
+
¸
n
V
n
−µ +U[z
n
[
2
z
∗
n
ˆ ϕ
n
+h.c., (5.4)
ˆ
H
2
= −J
¸
<n,m>
ˆ ϕ
†
n
ˆ ϕ
m
+
¸
n
(V
n
−µ) ˆ ϕ
†
n
ˆ ϕ
n
+
U
2
¸
n
ˆ ϕ
2†
n
z
2
n
+ ˆ ϕ
2
n
z
2∗
n
+ ( ˆ ϕ
†
n
ˆ ϕ
n
+ ˆ ϕ
n
ˆ ϕ
†
n
)[z
n
[
2
, (5.5)
ˆ
H
3
= U
¸
n
ˆ ϕ
†
n
ˆ ϕ
†
n
ˆ ϕ
n
z
n
+h.c., (5.6)
ˆ
H
4
=
U
2
¸
n
ˆ ϕ
†
n
ˆ ϕ
†
n
ˆ ϕ
n
ˆ ϕ
n
. (5.7)
Where < n, m > restricts the sum over m to nearest neighbors and h.c. stands
for the hermitian conjugate. The terms of the Hamiltonian have been grouped
in equations according to the number of non condensate operators which they
contain.
5.2 Bogoliubovde Gennes (BdG) equations
In this section we neglect higher order terms and focus only on the quadratic
Hamiltonian. The ﬁrst step in the diagonalization is the minimization of the energy
functional H
0
. This requires the condensate amplitudes z
n
to be a stationary
solution of the DNLSE (4.6), which in turn means that the linear Hamiltonian
ˆ
H
1
vanishes. For the time independent situation the DNLSE can be written as
µz
n
= −J
¸
<m,n>
z
m
+ (V
n
+U[z
n
[
2
)z
n
(5.8)
5.2 Bogoliubovde Gennes (BdG) equations 63
The Hamiltonian H
0
gives a description which includes only the contribution from
the condensate. The quadratic Hamiltonian H
2
allows the leading order eﬀects of
the non condensate to be taken into account. By including second order terms in
the Hamiltonian two classes of interactions are included besides the condensate
condensate ones: a) Interactions between one excited atom and one condensate
atom leading to transitions of the form [0j` → [0i` (direct and exchange excita
tions)and b) interactions between two condensate atoms which cause the atoms to
be excited to non condensate states [00` →[ij` (pair excitations). Schematically
they are shown in ﬁg.5.1.
Figure 5.1: Scattering processes included in the quadratic hamiltonian: a) Direct
and exchange excitations, b) Pair excitations
The quadratic Hamiltonian can be diagonalized by ﬁnding a basis of socalled
quasiparticle states which do not interact which each other. Mathematically this
involves a linear canonical transformation of the singleparticle creation and annihi
lation operators ˆ a
†
n
and ˆ a
n
into quasiparticle operators ˆ α
†
s
, ˆ α
s
. The transformation
is known as Bogoliubov transformation and is given by
ˆ ϕ
n
=
¸
s,=0
u
s
n
ˆ α
s
−v
∗s
n
ˆ α
†
s
. (5.9)
In general the spectrum of ﬂuctuations includes a zero mode. This mode is the
Goldstone boson associated with the breaking of global phase invariance by the
condensate. The zero mode is essentially non perturbative and it introduces an
artiﬁcial infrared divergence in low dimensional models. For this reason quadratic
approximations are actually improved if the contribution from this mode is ne
glected all together [96]. A diﬀerent way to deal with the zero mode has been
proposed by Castin and Dum [98], Gardiner [97] and Morgan [16]. Here the the
ory is written in terms of operators which exchange particles between zero and
64 Chapter 5 Quadratic approximations
nonzero modes, conserving the total particle number, and one further operator
which changes total particle number. As shown by [16] both the latter and the
former approaches lead to the same physical predictions, so for simplicity we are
going to ignore zero mode ﬂuctuations and restrict the ﬂuctuation operators to act
only on the excited states.
The Bogoliubov transformation Eq. (5.9) is required to be canonical, which
means that it preserves the commutation relations and leads to bosonic quasipar
ticles. To satisfy this, the amplitudes ¦u
s
n
, v
s
n
¦ are constrained by the conditions:
¸
n
u
∗s
n
u
st
n
−v
∗s
n
v
st
n
= δ
sst
, (5.10)
¸
n
u
s
n
v
st
n
−v
∗s
n
u
∗st
n
= 0. (5.11)
The necessary and suﬃcient conditions that the quasiparticle amplitudes have to
fulﬁll to diagonalize the Hamiltonian are provided by the so called Bogoliubovde
Gennes (BdG) equations
L ´
´
∗
L
u
s
v
s
= ω
B
s
u
s
−v
s
+c
s
z
−z
∗
, (5.12)
with u
s
= (u
s
1
, u
s
2
, ...) , v
s
= (v
s
1
, v
s
2
, ...) and z = (z
1
, z
2
, ...). The matrices L and
´ are given by
L
nm
= −J
¸
<n,l>
δ
nl
δ
lm
+δ
nm
(2U[z
n
[
2
+V
n
−µ) (5.13)
´
nm
= −Uz
2
i
δ
nm
. (5.14)
with δ
nl
the Kronecker delta which is one if n = l and zero otherwise. The param
eters c
s
ensure that the solutions with ω
B
s
= 0 are orthogonal to the condensate
[16]. These parameters are given by
c
s
= U
¸
n
[z
n
[
2
(z
∗
n
u
s
n
−z
n
v
s
n
)
¸
n
[z
n
[
2
(5.15)
The BdG equations without the c
s
parameters have the same quasiparticle energies
as Eq.(5.12), nevertheless the solutions are orthogonal to the condensate only in
the general sense, i.e.
¸
n
(z
∗
n
u
s
n
− z
n
v
s
n
) = 0. Each of the terms,
¸
n
z
∗
n
u
s
n
and
¸
n
z
n
v
s
n
are not necessarily zero. To obtain the desired orthogonal modes one
way to proceed is to solve the BdG equations setting the c
s
to zero and then
to remove from the solutions their projection onto the condensate. If the BdG
equations are satisﬁed, the quadratic Hamiltonian takes the form
ˆ
H
2
=
¸
s,=0
¸
ω
B
s
ˆ α
†
s
ˆ α
s
+
1
2
ω
B
s
−L
ss
. (5.16)
5.3 Higher order terms 65
If the Hamiltonian
ˆ
H
2
, is positive deﬁnite, which is the case when the condensate
amplitudes correspond to a stable state of the system, the eigenvalues of the system
are real. The solutions ω
B
s
come in pairs with positive and negative energies, if
ω
B
s
is a solution for the amplitudes (u
s
, v
s
) then −ω
B
s
is also a solution for the
amplitudes (v
s∗
, u
s∗
). There is always a solution with ω
B
s
= 0 and in this case the
amplitudes must be proportional to the condensate (v
0
, u
0
) ∝ (z, z). We explicitly
exclude the zero mode solution to guarantee that the excitations are orthogonal
to the condensate.
To have a complete description of the quadratic Hamiltonian, we deﬁne the one
body density ﬂuctuation matrix, ρ
nm
and the anomalous average matrix, m
nm
as:
ρ
nm
= 'ˆ a
†
n
ˆ a
m
` −'ˆ a
†
n
`'ˆ a
m
` = ' ˆ ϕ
†
n
ˆ ϕ
m
`, (5.17)
m
nm
= 'ˆ a
n
ˆ a
m
` −'ˆ a
n
`'ˆ a
m
` = ' ˆ ϕ
n
ˆ ϕ
m
`. (5.18)
The averages denotes an ordinary quantum expectation value. At zero tempera
ture, the average is over the ground quasiparticle state given by ˆ α
s
[0` = 0 and the
ρ
nm
and m
nm
can be written in terms of quasiparticle amplitudes as:
ρ
nm
=
¸
s,=0
v
s∗
n
v
s
m
, (5.19)
m
nm
= −
¸
s,=0
u
s
n
v
s∗
m
. (5.20)
Physically, the quantity ρ
nn
represents the non condensate population (or deple
tion) at position n or depletion at position n. If the average number of atoms is
N, the depletion and condensate atoms are related by the equation:
N =
¸
n
ˆ a
†
n
ˆ a
n
=
¸
n
[z
n
[
2
+ρ
nn
. (5.21)
The interpretation to the physical meaning of the anomalous term is postponed to
section 5.7.2.
5.3 Higher order terms
In the preceding section we described the diagonalization of the Hamiltonian con
taining only terms up to quadratic order in the ﬁeld operators. We now want
to go beyond this approximation and consider the corrections introduced by the
cubic and quartic terms. Taking into account higher order terms we are mainly
introducing manybody eﬀects on the scattering.
66 Chapter 5 Quadratic approximations
5.3.1 HartreeFockBogoliubov equations
Third and quartic terms in the Hamiltonian can be included by treating them
in a selfconsistent mean ﬁeld approximation, which relies upon the factorization
approximation: products of many operators are approximated by paring the oper
ators in all possible ways and replacing these pairs by their expectation value [15].
Using the factorization approximation, one can reduce third and quartic products
of ﬂuctuation ﬁeld operators to:
ϕ
†
n
ϕ
n
ϕ
n
→ 2ρ
nn
ϕ
n
+ 2m
nn
ϕ
†
n
, (5.22)
ϕ
†
n
ϕ
†
n
ϕ
n
ϕ
n
→ 4ρ
nn
ϕ
†
n
ϕ
n
+m
nn
ϕ
†
n
ϕ
†
n
+m
∗
nn
ϕ
n
ϕ
n
−(2ρ
2
nn
+[m
nn
[
2
).(5.23)
Eq.(5.23) is justiﬁed by Wick’s theorem [51], which gives 'ϕ
†
n
ϕ
†
n
ϕ
n
ϕ
n
` = 2ρ
2
nn
+
[m
nn
[
2
, while Eq.(5.22) is justiﬁed by analogy [16].
Therefore, using Eq.(5.23) and Eq.(5.22) in
ˆ
H
3
and
ˆ
H
4
( Eqs. (5.6) and (5.7)) one
can reduce them to quadratic forms in terms of the functions ρ
nm
and m
nm
. The
cnumber term in Eq.(5.23) introduces also a energy shift in H
0
:
∆
ˆ
H
0
=
U
2
(2ρ
2
nn
+[m
nn
[
2
) (5.24)
ˆ
H
3
= U
¸
n
(2ρ
nn
ˆ ϕ
n
+m
nn
ˆ ϕ
†
n
)z
n
+h.c., (5.25)
ˆ
H
4
=
U
2
¸
n
(2ρ
nn
ˆ ϕ
†
n
ˆ ϕ
n
+m
nn
ˆ ϕ
n
ˆ ϕ
n
) +h.c.. (5.26)
The corrections from higher order terms yield a modiﬁed quadratic Hamiltonian
which can be also diagonalized. The diagonalization leads to the following equa
tions
µ
HFB
z
n
= −J
¸
<m,n>
z
m
+ (V
n
+U[z
n
[
2
+2Uρ
nn
+Um
nn
)z
n
, (5.27)
L
HFB
´
HFB
´
∗HFB
L
∗HFB
u
s
v
s
= ω
HFB
s
u
s
−v
s
+
c
HFB
−
s
z
−c
HFB
+
s
z
∗
, (5.28)
L
HFB
nm
= −J
¸
<n,l>
δ
nl
δ
lm
+δ
nm
(2U[z
n
[
2
+V
n
−µ
HFB
+ 2Uρ
nn
), (5.29)
´
HFB
nm
= −(Uz
2
i
+Um
nn
)δ
nm
, (5.30)
c
HFB∓
s
=
U
¸
n
[z
n
[
2
(z
∗
n
u
s
n
−z
n
v
s
n
) ∓U
¸
n
(m
∗
nn
z
n
u
s
n
−m
nn
z
∗
n
v
s
n
)
¸
n
[z
n
[
2
.(5.31)
The parameters c
HFB∓
s
are there again to enforce the orthogonality of the quasi
particle modes with the condensate [16].
5.4 The BoseHubbard model and superﬂuidity 67
The above set of equations (5.27) to (5.30) are known as HartreeFockBogoliubov
(HFB) equations. Even though they take into account higher order corrections,
they have the problem that violate the Hugenholtz and Pines theorem [99] which
states that the energy spectrum of a Bose gas is gapless, i.e, there exists an exci
tation with an energy which tends to zero as the momentum tends to zero.
5.3.2 HFBPopov approximation
One way to solve the gap problem is to set the anomalous term ¯ m
nn
to zero
in HFB equations. This procedure is known as HFBPopov approximation. The
HFBPopov equations were ﬁrst introduced by Popov [52], and they are considered
to be a better approximation than the HFB equations because they yield a gapless
spectrum.
The HFBPopov equations have the form:
µ
P
z
n
= −J
¸
<m,n>
z
m
+(V
n
+U[z
n
[
2
+ 2Uρ
nn
)z
n
, (5.32)
L
P
´
P
´
∗P
L
∗P
u
s
v
s
= ω
P
s
u
s
−v
s
+c
s
z
−z
∗
, (5.33)
L
P
nm
= −J
¸
<n,l>
δ
nl
δ
lm
+δ
nm
(2U[z
n
[
2
+V
n
−µ
P
+ 2Uρ
nn
), (5.34)
´
P
nm
= −(Uz
2
i
)δ
nm
, (5.35)
c
s
= U
¸
n
[z
n
[
2
(z
∗
n
u
s
n
−z
n
v
s
n
)
¸
n
[z
n
[
2
. (5.36)
5.4 The BoseHubbard model and superﬂuidity
The concept of superﬂuidity is closely related to the existence of a condensate in the
interacting manybody system. Formally, the onebody density matrix ρ
(1)
(x, x
t
)
has to have exactly one macroscopic eigenvalue which deﬁnes the number of par
ticles in the condensate; the corresponding eigenvector describes the condensate
wave function φ
0
(x) = e
iΘ(x)
[φ
0
(x)[. A spatially varying condensate phase, Θ(x),
is associated with a velocity ﬁeld for the condensate by
v
0
(x) =
m
∇Θ(x) . (5.37)
This irrotational velocity ﬁeld is identiﬁed with the velocity of the superﬂuid ﬂow,
v
s
(x) ≡ v
0
(x) ([100],[101]) and enables us to derive an expression for the superﬂuid
fraction, f
s
. Consider a system with a ﬁnite linear dimension, L, in the e
1
–direction
and a ground–state energy, E
0
, calculated with periodic boundary conditions. Now
we impose a linear phase variation, Θ(x) = θx
1
/L with a total twist angle θ over
68 Chapter 5 Quadratic approximations
the length of the system in the e
1
–direction. The resulting ground–state energy,
E
θ
will depend on the phase twist. For very small twist angles, θ <π, the energy
diﬀerence, E
θ
− E
0
, can be attributed to the kinetic energy, T
s
, of the superﬂow
generated by the phase gradient. Thus,
E
θ
−E
0
= T
s
=
1
2
mNf
s
v
2
s
, (5.38)
where m is the mass of a single particle and N is the total number of particles so
that mNf
s
is the total mass of the superﬂuid component. Replacing the superﬂuid
velocity, v
s
with the phase gradient according to Eq. (5.37) leads to a fundamental
relation for the superﬂuid fraction.
f
s
=
2m
2
L
2
N
E
θ
−E
0
θ
2
=
1
N
E
θ
−E
0
J (∆θ)
2
. (5.39)
where the second equality applies to a one dimensional lattice system on which a
linear phase variation has been imposed. Here the distance between sites is a, the
phase variation over this distance is ∆θ, and the number of sites is M. In this
case, J ≡
2
/(2ma
2
).
Technically the phase variation can be imposed through socalled twisted bound
ary conditions [102]. In the context of the discrete BoseHubbard model it is,
however, more convenient to map the phase variation by means of a unitary trans
formation onto the Hamiltonian [103]. For simplicity we are going to focus in a
one dimensional lattice. The resulting “twisted” Hamiltonian in this case
ˆ
H
θ
=
M
¸
n=1
ˆ n
n
V
n
−J
M
¸
n=1
(e
−i∆θ
ˆ a
†
n+1
ˆ a
n
+e
i∆θ
ˆ a
†
n
ˆ a
n+1
) +
U
2
M
¸
n=1
ˆ n
n
(ˆ n
n
−1) . (5.40)
exhibits additional phase factors e
±i∆θ
— the socalled Peierls phase factors — in
the hopping term [104, 105]. These phase factors show that the twist is equivalent
to the imposition of an acceleration on the lattice for a ﬁnite time. It is interest
ing to note that the present experiments enable us to make a speciﬁc connection
between the formal and operational aspects of the system.
We calculate the change in energy E
θ
−E
0
under the assumption that the phase
change ∆θ is small so that we can write:
e
−i∆θ
· 1 −i∆θ −
1
2
(∆θ)
2
. (5.41)
Using this expansion the twisted Hamiltonian (5.40) takes the following form:
ˆ
H
θ
·
ˆ
H
0
+ ∆θ
ˆ
J −
1
2
(∆θ)
2
ˆ
T =
ˆ
H
0
+
ˆ
H
pert
, (5.42)
where we retain terms up to second order in ∆θ. The current operator
ˆ
J (Note that
the physical current is given by this expression multiplied by
1
) and the hopping
5.4 The BoseHubbard model and superﬂuidity 69
operator
ˆ
T are given by:
ˆ
J = iJ
M
¸
n=1
(ˆ a
†
n+1
ˆ a
n
− ˆ a
†
n
ˆ a
n+1
) (5.43)
ˆ
T = −J
M
¸
n=1
(ˆ a
†
n+1
ˆ a
n
+ ˆ a
†
n
ˆ a
n+1
) . (5.44)
The change in the energy E
θ
− E
0
due to the imposed phase twist can now be
evaluated in second order perturbation theory
E
θ
−E
0
= ∆E
(1)
+ ∆E
(2)
. (5.45)
The ﬁrst order contribution to the energy change is proportional to the expectation
value of the hopping operator
∆E
(1)
= 'Ψ
0
[
ˆ
H
pert
[Ψ
0
` = −
1
2
(∆θ)
2
'Ψ
0
[
ˆ
T[Ψ
0
` . (5.46)
Here [Ψ
0
` is the ground state of the original BoseHubbard Hamiltonian (4.27).
The second order term is related to the matrix elements of the current operator
involving the excited states [Ψ
ν
` (ν = 1, 2, ...) of the original Hamiltonian
∆E
(2)
= −
¸
ν,=0
['Ψ
ν
[
ˆ
H
pert
[Ψ
0
`[
2
E
ν
−E
0
= −(∆θ)
2
¸
ν,=0
['Ψ
ν
[
ˆ
J[Ψ
0
`[
2
E
ν
−E
0
. (5.47)
Thus we obtain for the energy change up to second order in ∆θ
E
θ
−E
0
= (∆θ)
2
−
1
2
'Ψ
0
[
ˆ
T[Ψ
0
` −
¸
ν,=0
['Ψ
ν
[
ˆ
J[Ψ
0
`[
2
E
ν
−E
0
= M(∆θ)
2
D,
D ≡
1
M
−
1
2
'Ψ
0
[
ˆ
T[Ψ
0
` −
¸
ν,=0
['Ψ
ν
[
ˆ
J[Ψ
0
`[
2
E
ν
−E
0
. (5.48)
The quantity D, deﬁned above, is formally equivalent to the Drude weight used
to specify the DC conductivity of charged fermionic systems [106]. The superﬂuid
fraction is then given by the contribution of both the ﬁrst and second order term:
f
s
= f
(1)
s
−f
(2)
s
,
f
(1)
s
≡ −
1
2NJ
'Ψ
0
[
ˆ
T[Ψ
0
`
, (5.49)
f
(2)
s
≡
1
NJ
¸
ν,=0
['Ψ
ν
[
ˆ
J[Ψ
0
`[
2
E
ν
−E
0
.
Here N is the number of atoms in the lattice. In general both the ﬁrst and the
second order term contribute. For a translationally invariant lattice the second
70 Chapter 5 Quadratic approximations
term vanishes (as is going to be shown latter) if one uses a quadratic approximation.
However, in exact calculations the second order term plays a role.
We can further understand this approach to the superﬂuid density by calcu
lating the ﬂow that is produced by the application of the phase twist. To do this
we work out the expectation value of the current operator expressed in terms of
the twisted variables:
ˆ
J
θ
= iJ
M
¸
n=1
(e
−i∆θ
ˆ a
†
n+1
ˆ a
n
−e
i∆θ
ˆ a
†
n
ˆ a
n+1
) . (5.50)
We expand this to ﬁnd the lowest order contributions, i.e.:
ˆ
J
θ
·
ˆ
J +J∆θ
M
¸
n=1
(ˆ a
†
n+1
ˆ a
n
+ ˆ a
†
m
ˆ a
n+1
) =
ˆ
J −
ˆ
T∆θ . (5.51)
We use ﬁrst order perturbation theory on the wave function to obtain the following
expression:
'Ψ(∆θ)[
ˆ
J
θ
[Ψ(∆θ)` = 2∆θ
−
1
2
'Ψ
0
[
ˆ
T[Ψ
0
` −
¸
ν,=0
['Ψ
ν
[
ˆ
J[Ψ
0
`[
2
E
ν
−E
0
(5.52)
= 2NJf
s
∆θ. (5.53)
Thus, the physical current, J
s
, Eq. (5.53) multiplied by
1
, can be expressed as:
J
s
=
1
'Ψ(∆θ)[
ˆ
J
θ
[Ψ(∆θ)` = Nf
s
∆θ
m
∗
a
2
. (5.54)
This is the total ﬂux, j
s
, and we need to divide by M to get the ﬂux density, i.e.
j
s
=
1
M
'Ψ(∆θ)[
ˆ
J
θ
[Ψ(∆θ)` =
∆θ
m
∗
a
Nf
s
aM
= v
s
n
s
. (5.55)
So we see that the Drude formulation of the superﬂuid fraction (5.49) gives an
intuitively satisfying expression for the amount of ﬂowing superﬂuid.
5.5 Expectation values
Using the quadratic approximations we can evaluate expectation values of mean
ingful physical quantities. Useful quantities as the system approach the Mott
insulator state are the number ﬂuctuations and the quasimomentum distribution.
The number ﬂuctuations at the site x
i
are given in the Bogoliubov approxima
tion by
∆n
n
=
'ˆ a
†
n
ˆ a
n
ˆ a
†
n
ˆ a
n
` −'ˆ a
†
n
ˆ a
n
`
2
= [z
n
[
2
¸
s
[u
s
n
−v
s∗
n
[
2
. (5.56)
5.6 Applications 71
The quasimomentum distribution of the atoms released from the lattice is im
portant because it is one of the most easily accessible quantities in the experiments.
The quasimomentum distribution function n
q
is deﬁned as [103]
n
q
=
1
M
¸
n,m
e
iq(x
n
−x
m
)
'a
†
n
a
m
`
=
1
M
¸
¸
n
z
n
e
iqx
n
2
+
¸
n,m
ρ
nm
e
iq(x
n
−x
m
)
¸
, (5.57)
where the quasimomentumq can assume discrete values which are integer multiples
of
2π
a
i
M
i
, where a
i
and M
i
are the lattice spacing and number of lattice sites in the
i direction and M =
¸
i
M
i
is the total number of lattice sites.
Neglecting interaction eﬀects during the expansion, the quasimomentum dis
tribution represents the Fourier transform of the original spatial distribution in
the lattice. In the superﬂuid regime because of the phase coherence of the system,
when it is released from the lattice the gas shows a nice interference pattern. As
the interactions are increased phase coherence is lost, and instead of interference
peaks a incoherent background is observed.
For a one dimensional optical lattice, the superﬂuid fraction in terms of quasi
particles amplitudes and energies can be written as:
f
s
= f
(1)
s
−f
(2)
s
, (5.58)
f
(1)
s
=
M
¸
n=1
f
(1)
sn
=
1
2N
M
¸
n=1
(z
n+1
z
∗
n
+z
∗
n+1
z
n
) +
¸
s
(v
s
n
v
s∗
n+1
+v
s∗
n
v
s
n+1
)
, (5.59)
f
(2)
s
=
J
N
¸
s
¸
n
u
s
n
+v
s
n
(z
n+1
−z
n−1
)
2
ω
s
+ (5.60)
J
N
¸
¸
s,s
¸
n
u
s
n+1
v
s
n
−u
s
n
v
s
n+1
2
ω
s
+ω
s
+δ
ss
¸
n
(u
s
n+1
v
s
n
−u
s
n
v
s
n+1
)
2
2ω
s
¸
.
5.6 Applications
5.6.1 Translationally invariant lattice
To understand manybody eﬀects included in the quadratic approximations we
start by studying the case where no external conﬁnement is present, V
n
= 0.
We assume a d dimensional separable square optical lattice with equal tunneling
matrix element J in all directions and periodic boundary conditions. M is the
total number of wells.
Due to the translational symmetry of the system the condensate amplitudes
are constant over the lattice, z
n
=
√
n
o
. Also, the quasiparticle modes have a
plane wave character and therefore can be related to quasimomentum modes :
u
q
n
=
1
√
M
e
iqx
n
u
q
, v
q
n
=
1
√
M
e
iqx
n
u
q
. (5.61)
72 Chapter 5 Quadratic approximations
Here the vector q denotes the quasimomentum whose components assume discrete
values which are integer multiples of
2π
a
d
√
M
with a is the lattice spacing. The
amplitudes u
q
and u
q
must satisfy the condition [u
q
[
2
− [v
q
[
2
= 1 and can all be
chosen to be real and to depend only on the modulus of the wave vector (u
q
= u
−q
,
v
q
= v
−q
).
The translationally invariant case has the advantage that the quasiparticle modes
are always orthogonal to the condensate and therefore the parameters c
s
are always
zero.
• Bogoliubovde Gennes (BdG) equations and HFB Popov approximation
In the simplest quadratic approximation the DNLSE reduces to
µ = −tJ +n
o
U, (5.62)
where t is the number of nearest neighbors t = 2d.
The BdG equations in the translationally invariant case become the following
2 2 eigenvalue problem
L
qq
−´
q−q
´
q−q
−L
qq
u
q
v
q
= ω
q
u
q
v
q
, (5.63)
with
L
qq
=
q
+n
o
U, ´
q−q
= n
o
U. (5.64)
Here we have introduced the deﬁnition
k
= 4J
¸
d
i=1
sin
2
(
k
i
a
2
).
The quasiparticle energies ω
q
and modes are found by diagonalizing Eq.(5.64):
ω
2
q
= L
2
qq
−´
2
q−q
=
q
2
+2Un
o
q
, (5.65)
u
2
q
=
L
qq
+ω
q
2ω
q
=
q
+n
o
U +ω
q
2ω
q
, (5.66)
v
2
q
=
L
qq
−ω
q
2ω
q
=
q
+n
o
U −ω
q
2ω
q
, (5.67)
u
q
v
q
= −
´
q−q
2ω
q
=
n
o
U
2ω
q
(5.68)
and
n = n
o
+
1
M
¸
q,=0
v
2
q
, (5.69)
with n the total density, n = N/M. The constrain that ﬁxes the number of
particles can be written as
5.6 Applications 73
n
o
= n −
1
M
¸
q,=0
q
+Un
o
2ω
q
−
1
2
. (5.70)
In the homogeneous case, due to the cancellation of the ρ
nn
. terms in the
quasiparticle equations (5.34), the HFBPopov approximation leads to the
same quasiparticle energy spectrum and amplitudes than the BdG equations.
The only diﬀerence is in the chemical potential which has an extra term
2Uρ
nn
µ
(P)
= −tJ +Un
o
+2Uρ
nn
. (5.71)
As opposed to the free particle system, where the single particle energy
dominates at high momenta (it grows as q
2
), the single particle excitations
in the presence of the lattice are always bounded by 4Jd. Therefore, in the
regime Un
o
/J > 1 the interaction term dominates for all quasimomentum
and so ω
q
∼
2Un
o
q
. On the other hand, in the weakly interacting regime,
the most important contribution to the depletion comes from the lowlying
modes. For these modes is also true that ω
q
∼
2Un
o
q
. Thus, to a good
approximation the condensate fraction can be written as:
n
o
≈ g −
Un
o
J
α, (5.72)
with
α = α(d, M) ≡
1
M
¸
q,=0
√
J
2
2
q
, (5.73)
g =
n +
M−1
2M
Un/J 1
n Un/J <1
. (5.74)
In Eq.(5.74) the term
M−1
2M
is a ﬁnite size eﬀect term which is important to
keep for low density systems. α is a dimensionless quantity which depends
only on the dimensionality of the system. In the thermodynamics limit,
M →∞ the sum can be replaced by an integral and we ﬁnd
α(1, ∞) →
1
2
√
2π
ln(cot(q
o
))
q
o
→0
, (5.75)
α(2, ∞) = 0.227293, (5.76)
α(3, ∞) = 0.160287. (5.77)
(5.78)
74 Chapter 5 Quadratic approximations
The infrared divergence in the one dimensional thermodynamic limit is a
consequence of the importance of long wave length correlations in low di
mensional systems. For ﬁnite one dimensional systems however α(1, M) has
a ﬁnite value.
Because Eqs.(5.65)(5.66) are completely determined if n
o
is known, by solv
ing Eq.(5.72) we obtain all necessary information. The solution of the alge
braic equation is:
n
o
≈ g +
α
2
U
2J
−
g
α
2
U
J
+
α
4
U
2
4J
2
. (5.79)
Eq.(5.79) tells us that for very weak interactions almost all the atoms are in
the condensate. As interaction increases the condensate fraction decreases,
but under the BdG (and HFBPopov) approximation it only vanishes when
U/J → ∞. The BdG (and HFBPopov) equations therefore do not predict
any superﬂuid to Mott insulator phase transition.
The calculated quasiparticle amplitudes can be used to get analytic expres
sions for the number ﬂuctuations, the momentum distribution and superﬂuid
fraction. If we restrict our attention to the one dimensional system we have
∆n
n
= n
o
¸
q
q
ω
q
−−−−→
M→∞
2
π
n
o
arctan(
2J
n
o
U
), (5.80)
n
q
= n
o
δ
q,0
+
1
M
[v
q
[
2
δ
q,k
, (5.81)
= n
o
δ
q,0
+
1
M
q
+Un
o
2ω
k
−
1
2
δ
q,k
, (5.82)
f
s
= f
(1)
s
−f
(2)
s
, (5.83)
f
(1)
s
=
M
N
n
o
+
1
M
¸
q
[v
q
[
2
cos(qa)
, (5.84)
f
(2)
s
= 0. (5.85)
Eq. (5.80) has been studied in Ref. [95] to produce results for squeezing.
There the authors show that Eq. (5.80) is consistent with those of other
approaches previously reported in the literature [107, 108, 109, 110].
It is important to emphasize that due to the translational invariance of the
lattice [see Eq. (5.60)], the second order term vanishes in the Bogoliubov
limit.
The expression for the superﬂuid fraction gives a direct insight into the be
havior of the system as atoms are pushed out of the condensate due to
interactions. In Eq. (5.84) the sum involving the Bogoliubov amplitudes v
q
characterizes the diﬀerence between the condensate fraction, which is given
5.6 Applications 75
by the ﬁrst term, and the superﬂuid fraction. For weak interactions and small
depletion, the depletion of the condensate has initially little eﬀect on super
ﬂuidity. As interactions are increased the depleted population spreads into
the central part of the band, (where the cos(qa) term has a negative sign)
and the superﬂuid fraction is reduced. Finally for even larger interaction
strengths, the population in the upper quarter of the band again produces
a positive contribution to the superﬂow and the contribution from the sum
decreases. In a sense the interactions are playing a role akin to Fermi exclu
sion ”pressure” in the case of electron ﬂow in a band. This, however can lead
to perfect ﬁlling and cancellation of the ﬂow. In the case of our Bogoliubov
description we can only see reduction of the ﬂow, not a perfect switching oﬀ
of the superﬂuid. This happens in the Mott insulator state, which cannot be
described by the Bogoliubov approximation.
5 10 15 20 25 30
V
eff
0.5
1
1.5
2
n
n5.33
5 10 15 20 25 30
V
eff
0.2
0.4
0.6
0.8
f
c
n5.33
5 10 15 20 25 30
V
eff
0.2
0.4
0.6
0.8
f
s
,
f
s
2
n5.33
10 20 30 40
V
eff
0.5
1
1.5
2
n
n5
10 20 30 40
V
eff
0.2
0.4
0.6
0.8
f
c
n5
10 20 30 40
V
eff
0.2
0.4
0.6
0.8
f
s
,
f
s
2
n5
Figure 5.2: Comparisons of the exact solution and BdG (and HFBPopov) solu
tions as a function of V
eff
= U/J, for a system with M = 3 and ﬁlling factors
n = 5 and 5.33. Left: number ﬂuctuations (Exact: solid line, BdG(and HFB
Popov): dotted line), middle: condensate fraction (Exact: solid line, BdG(and
HFBPopov): dotted line,analytic (5.79:red line ), right: superﬂuid fraction f
s
(Exact: solid line, BdG(and HFBPopov): dotted line). The exact second order
term (dashed line) of the superﬂuid fraction, f
(2)
s
is also shown in these plots.
The vertical line shown in the plots is an estimation of V
crit
eff
when the system is
commensurate
In Figs. 5.2 and 5.3 we compare the number ﬂuctuations per lattice site,
the condensate fraction and the total and second order superﬂuid fraction
determined from the exact solution of the BoseHubbard Hamiltonian to the
solution obtained from the BdG (and HFBPopov) equations as a function
of the ratio V
eff
= U/J. The systems used for the comparisons are one
dimensional lattices with three wells, M = 3, and commensurate ﬁlling fac
tors n = 5 and 5.33 and n = 50 and 50.33. We were restricted to consider
only three wells due to computational limitations. The size of the matrix
needed in the exact solution for N atoms and M wells scales as
(N+M−1)!
N!(M−1)!
.
76 Chapter 5 Quadratic approximations
50 100150200
V
eff
1
3
5
n
n50.33
50 100150200
V
eff
0.2
0.4
0.6
0.8
f
c
n50.33
50 100150200
V
eff
0.2
0.4
0.6
0.8
f
s
,
f
s
2
n50.33
50 100150200
V
eff
1
3
5
n
n50
50 100150200
V
eff
0.2
0.4
0.6
0.8
f
c
n50
50 100150200
V
eff
0.2
0.4
0.6
0.8
f
s
,
f
s
2
n50
Figure 5.3: Comparisons of the exact solution and BdG(and HFBPopov) solu
tions as a function of V
eff
= U/J, for a system with M = 3 and ﬁlling factors
n = 50 and 50.33. Left: number ﬂuctuations (Exact: solid line, BdG(and HFB
Popov): dotted line), middle: condensate fraction (Exact: solid line, BdG(and
HFBPopov): dotted line, analytic Eq.(5.79): red line ), right: superﬂuid fraction
f
s
(Exact: solid line, BdG(and HFBPopov): dotted line). The exact second order
term (dashed line) of the superﬂuid fraction, f
(2)
s
is also shown in these plots. In
this case the agreement is much better.
However, if the approximate approach works well for these small systems we
expect it to provide a good description of the larger systems prepared in the
laboratory.
Because the second order term of the superﬂuid fraction (second term of
Eq.(5.49) vanishes in the quadratic approximations (see Eq. (5.60)), we
only expect them to give a good description of the superﬂuid fraction in the
region where the second order term is extremely small, provided it predicts
accurately the ﬁrst order term. This is exactly what is observed in the
plots. When the second order term starts to grow, typically above 0.5V
crit
eff
( with V
crit
eff
= (U/J)
c
, see Eq.(3.9)), the BdG (and HFBPopov) equations
starts to fail. An estimate of V
crit
eff
is shown by a vertical line in some of
the ﬁgures. With increasing ﬁlling factor the critical value is shifted towards
larger values of the interaction strength, and the region in which the BdG
(and HFBPopov) equations are accurate gets larger. It is interesting to
note that the number ﬂuctuations predicted by the theory are accurate in
a greater range than the other physical quantities shown. Its predictions
of squeezing agree very well with the exact solutions right up to the point
where the number ﬂuctuations become less than unity.
In the plots for the condensate fraction we also show the analytic approxima
tion given by Eq.(5.79). It can be observed that the analytic solution agrees
well with the numerical solution of the BdG( and HFBPopov) equations.
For the cases with noncommensurate ﬁllings the agreement is signiﬁcantly
5.6 Applications 77
better for all quantities. This is not surprising because when the ﬁlling is
not commensurate there is always a superﬂuid fraction present and the Mott
transition doesn’t occur. As can be seen in the plots for these cases the
second order term is always very small.
• HFB equantions
In the translationally invariant case the HFB Hamiltonian leads to the fol
lowing equations:
L
qq
−´
q−q
´
q−q
−L
qq
u
q
v
q
= ω
q
u
q
v
q
, (5.86)
with
L
qq
=
q
+n
o
U −U ¯ m, ´
q−q
= n
o
U +U ¯ m. (5.87)
The diagonalization of them yields
µ = −tJ +Un
o
+2U¯ n +U ¯ m, (5.88)
v
q
2
= u
q
2
−1 =
L
qq
−2ω
q
2ω
q
=
q
+n
o
U −U ¯ m−ω
q
2ω
q
, (5.89)
u
q
v
q
=
´
q−q
2ω
q
=
n
o
U +U ¯ m
2ω
q
, (5.90)
ω
q
2
=
q
2
+2U(n
o
+ ¯ m)
q
−4U
2
n
o
¯ m, (5.91)
with
¯ n ≡
1
M
¸
q,=0
v
2
q
, (5.92)
¯ m ≡ −
1
M
¸
q,=0
v
q
u
q
. (5.93)
From the above equation it can be seen that the quasiparticle energies pre
dicted under the HFB approximation don’t approach zero as q goes to zero.
As discussed in the previous section, the most important contribution to the
depletion of the condensate in the BdG equations comes from the low ly
ing modes, because of the ω
q
−1
dependence. The ﬁnite value of ω
q
has in
the HFB approximation as q goes to zero explains why the HFB equations
are always going to predict smaller condensate depletion than the gapless
approximation.
In Fig. 5.4 we compare the condensate fraction predicted by the HFB equa
tions with the condensate fraction obtained from the BdG equations, for a
system with M = 3 and ﬁlling factor n = 10. It can be seen that the con
densate fraction is always higher in the HFB solutions and remains constant
at higher values of U/J which is not physically correct.
78 Chapter 5 Quadratic approximations
0 10 20 30 40
V
eff
0.8
0.85
0.9
0.95
1
f
c
n10, M3
HFB
BdG
Figure 5.4: Comparisons of the condensate fraction given by the BdG solutions
and the HFB equations as a function of V
eff
= U/J, for a system with M = 3 and
ﬁlling factor n = 10.
5.6.2 Onedimensional harmonic trap plus lattice
In this section we consider the experimentally relevant case when there is an ex
ternal harmonic magnetic conﬁnement in addition to the lattice potential. For
simplicity we focus our attention on the one dimensional case. We consider the two
most relevant approximations which are the BdG equations and the HFBPopov
approximation. Firstly we study the BdG equations in the weakly interacting
regime and get some insight on the solutions by deriving analytic results using
the so called ThomasFermi approximation [12]. Secondly we study the HFB
Popov approximation and use it to explore the limits of validity of the quadratic
approximations as the interactions are increased.
• BdG equations
In the presence of the harmonic trap the BdG equations take the form:
5.6 Applications 79
ω
B
s
u
s
n
+c
s
z
n
= −J(u
s
n+1
+u
s
n−1
) + (2U[z
n
[
2
−µ + Ωn
2
)u
s
n
−Uz
2
n
v
s
n
, (5.94)
−ω
B
s
v
s
n
−c
s
z
∗
n
= −J(v
s
n+1
+v
s
n−1
) + (2U[z
n
[
2
−µ +Ωn
2
)v
s
n
−Uz
∗2
n
u
s
n
, (5.95)
µz
n
= −J(z
n+1
+z
n−1
) + (U[z
n
[
2
+ Ωn
2
)z
n
, (5.96)
N =
¸
n
([z
n
[
2
+ ˜ n
n
), (5.97)
˜ n
n
=
¸
s
[v
s
n
[
2
, (5.98)
c
s
= U
¸
n
[z
n
[
2
(z
∗
n
u
s
n
−z
n
v
s
n
)
¸
n
[z
n
[
2
. (5.99)
In the parameter regime where U/J < 1, but the number of atoms is large
enough such that NU/Ja
ho
1, where a
ho
=
m
∗
ω
∗
(See Eq. (2.39)) the
Bogoliubov approximation takes a rather simple analytic form. The eﬀect
of increasing the ratio NU/Ja
ho
is to push the atoms outwards, ﬂattening
the central density and increasing the width of the condensate wavefunction.
The quantum pressure, which is proportional to the kinetic energy, takes a
signiﬁcant contribution only near the edges of the wavefunction, and to a
good approximation it can be neglected. This is the socalled ThomasFermi
approximation (TF). This approximation has been very useful to derive an
alytic expressions in magnetic conﬁned condensates without the lattice and
proved to be in agreement with experimental measurements [12, 18, 19].
Under the ThomasFermi approximation one gets a condensate density proﬁle
of the form of an inverted parabola which vanishes at the turning points
R
TF
=
µ/Ω:
[z
n
[
2
=
µ−Ωn
2
U
n < R
TF
0 n > R
TF
(5.100)
The chemical potential is determined by ﬁxing the total number of particles.
To ﬁrst order it is possible to neglect the depletion of the condensate and
assume that
¸
[z
n
[
2
= N. Changing the sum to an integral it can be solved
for µ to get
µ = ΩR
2
TF
,
R
TF
≈
3
4
NU
Ω
1/3
. (5.101)
80 Chapter 5 Quadratic approximations
To solve for the excitation spectrum we use the ThomasFermi condensate
amplitude, Eq.(5.100) into Eqs.(5.94)and (5.95) [18, 111] and use the param
eter J as an expansion parameter in these equations. We deﬁne the variables
F
q
n
= u
q
n
+v
q
n
and G
s
n
= u
s
n
−v
s
n
. For simplicity we are going to set the con
stants c
s
to zero and ignore the orthogonality constraint at the beginning.
We correct this at the end by removing from the modes their projection onto
the condensate. To ﬁrst order in J we get the following set of equations:
2R
2
TF
Ω
1 −
n
2
R
2
TF
G
s
n
= ω
B
s
F
s
n
, (5.102)
−J
F
s
n+1
+F
s
n−1
−F
s
n
¸
¸
¸
1 −
(n+1)
2
R
2
TF
+
1 −
(n−1)
2
R
2
TF
1 −
n
2
R
2
TF
¸
= ω
B
s
G
s
n
. (5.103)
These two equations can be combined to get a simple equation of the form
1 −
(n + 1)
2
R
2
TF
1 −
n
2
R
2
TF
¯
F
s
n+1
−
¯
F
s
n
(5.104)
−
1 −
(n −1)
2
R
2
TF
1 −
n
2
R
2
TF
¯
F
s
n−1
−
¯
F
s
n
¸
= 2
ω
B
s
ω
∗
2
¯
F
s
n
,
with
¯
F
s
n
= F
s
n
/
1 −
n
2
R
2
TF
.
The above equation is just the discretized form of the Legendre equation,
∂
∂x
(1 −x
2
)
∂
∂x
P
n
+n(n +1)P
n
= 0. Therefore the solutions of Eq. (5.104)
are approximately given by :
ω
B
s
= ω
∗
s(s +1)
2
, (5.105)
F
s
n
=
ΩR
TF
(2s +1)
ω
B
s
1 −
n
2
R
2
TF
P
s
n
R
TF
, (5.106)
G
s
n
=
ω
B
s
(2s + 1)
4ΩR
3
TF
1
1 −
n
2
R
2
TF
P
s
n
R
TF
. (5.107)
The larger the size of the condensate the closer are the approximate solutions,
Eq.(5.107), to the exact solutions of Eqs. (5.104).
5.6 Applications 81
Finally, to make the solutions orthogonal to the condensate we have to re
move their projection onto the condensate mode:
u
s
n
→ u
s
n
−c
s
z
n
/ω
B
s
, (5.108)
v
s
n
→ v
s
n
−c
s
z
∗
n
/ω
B
s
. (5.109)
If we calculate the constants c
s
using Eq. (5.99), and orthogonality properties
of the Legendre polynomial we ﬁnd that the only mode that needs to be
corrected is the quadrupole mode, s = 2:
c
s
=
4ω
∗
R
5
Ω
2
15U
δ
s,2
. (5.110)
From Eqs. (5.110) it follows that the desired F and G amplitudes orthogonal
to the condensate are thus given by
F
s
n
→ F
s
n
−2c
s
z
n
/ω
B
s
, (5.111)
G
s
n
→ G
s
n
. (5.112)
Once solved for the quasiparticle amplitudes, it is possible to calculate the
depletion of the condensate,
¯
N =
¸
n,s
[v
s
n
[
2
. It is important to point out
that the term
1 −
n
2
R
2
TF
−1/2
in the G amplitudes makes the sum divergent.
The divergence however is nonphysical and only indicates the failure of the
TF approximation near R
TF
. To perform the sum, we have to exclude the
contribution from the boundary layer of thickness d =
a
4
ho
2R
TF
1/3
where the
TF approximation starts failing [111]. Finally, to ﬁx the number of particles
to N it is necessary to renormalize the condensate wave function and the
chemical potential. This is done by renormalizing the Thomas Fermi radius
and replacing N by N −
¯
N in Eq.(5.101).
In Fig. 5.5 we plot the condensate density proﬁle found by numerically solv
ing the BdG equations and compare it with the TF condensate solution.
The TF approximation reproduces the numerical solution very accurately
except at the edges where the kinetic energy can not be ignored with re
spect to the potential energy. In the same plot we also show the condensate
depletion which is small for the chosen parameters. A small condensate de
pletion guarantees the validity of the quadratic approximation to describe
the manybody system.
In Fig. 5.6 we show comparisons between the quasiparticle energies found
numerically, the TF quasiparticle and excitation spectra of the noninteract
ing system (see Eq. (2.37)). It can be seen in the plot that only the lowest
82 Chapter 5 Quadratic approximations
30 20 10 0 10 20 30
n
0
0.5
1
1.5
2
2.5
#
a
t
o
m
s
TF
depletion
Condensate
Figure 5.5: Comparisons of the condensate wave function found by numerically
solving the BdG equations with the ThomasFermi solution. The parameters used
were U/J = 0.2, Ω/J = 9.5 10
−4
and N = 100. The depletion is also shown in
the plot.
lying modes are well described by the TF approximation. Higher excitation
energies are closer to the noninteracting ones. In the regime where Ω < J
and the system’s size is large enough that discretization eﬀects, neglected
in Eqs. (5.105), are not important, the ﬁrst quasiparticle excitation energy
coincide with the ﬁrst noninteracting excitation energy . This ﬁrst excita
tion is known as the dipole mode and describes the oscillatory motion of the
center of mass when the system is displaced. The dipole quasiparticle am
plitudes (u
s=1
n
, v
s=1
n
), calculated by numerically solving the BdG equations
are shown in Fig. 5.7. In the same graphic the quasiparticle amplitudes
for the s = 2, 3, 4 excitation modes are also depicted. The s = 2 mode is
known as the quadrupole mode and is related to the breathing of the conden
sate when shaken. Higher order modes describe more complicated collective
excitations. In general, quasiparticle excitations other than the dipole are
aﬀected by interactions and they have diﬀerent energy than the noninter
acting excitations. As shown in Fig. 5.6 TF quasiparticle energies lie below
their noninteracting conterparts.
To test the accuracy with which the TF approximation describes the quasi
particle amplitudes, in Fig. 5.8 we compare the product F
s
n
G
s
n
as a function
of the lattice site n using the TF approximation results (Eqs. (5.112) and
(5.111)) to the product found by numerically solving the BdG equations.
We observe good agreement between the two solutions except near the TF
5.6 Applications 83
2 4 6 8 10 12 14 16 18 20
s
th
eigenvalue
0.05
0.1
0.15
0.2
0.25
Ω
s
E
R
UJ0.2 J9.5 x 10
4
N100
Free
TF
BdG
Figure 5.6: Comparisons of the quasiparticle spectrum found by numerically solv
ing the BdG equations(BdG) with the ThomasFermi solution (TF) and the nonin
teracting energies (Free). The parameters used were U/J = 0.2, Ω/J = 9.510
−4
and N = 100.
radius. This behavior is consistent with the fact that near the edge of the
atomic cloud, the kinetic energy becomes comparable to the potential energy
and therefore the TF approximation breaks down.
• HFBPopov approximation
In the translationally invariant system, both the BdG and the HFBPopov
approximations give the same quasiparticle amplitudes and energies. How
ever, when the harmonic conﬁnement is present this is no longer the case.
The HFBPopov approximation shifts the quasiparticle energies depending
on the spatial variation of the non condensate density in the region of the con
densate. In this section we investigate the limits of validity of the HFBPopov
approximation as the parameter U/J is increased. We use the HFBPopov
approximation instead of the BdG approximation because of the relevant
role of the noncondensate atoms into the condensate as the system is driven
to the Mott insulator transition.
In the presence of an harmonic conﬁnement the HFBPopov equations take
84 Chapter 5 Quadratic approximations
302010 0 10 20 30
n
0
u
n s
v
n s
s3
302010 0 10 20 30
n
0
u
n s
v
n s
s4
302010 0 10 20 30
n
0
u
n s
v
n s
s1
u
n
s
v
n
s
302010 0 10 20 30
n
0
u
n s
v
n s
s2
Figure 5.7: Lowlying quasiparticle amplitudes found by numerically solving the
BdG equations. The parameters used were U/J = 0.2, Ω/J = 9.5 10
−4
and
N = 100.
the form:
ω
s
u
s
n
+c
s
z
n
= −J(u
s
n+1
+u
s
n−1
) + (2U([z
n
[
2
+ ˜ n
n
) −µ + Ωn
2
)u
s
n
−Uz
2
n
v
s
n
, (5.113)
−ω
s
v
s
n
−c
s
z
∗
n
= −J(v
s
n+1
+v
s
n−1
) + (2U([z
n
[
2
+ ˜ n
n
) −µ + Ωn
2
)v
s
n
−Uz
∗2
n
u
s
n
, (5.114)
µz
n
= −J(z
n+1
+ z
n−1
) +
U([z
n
[
2
+ 2˜ n
n
) + Ωn
2
z
n
, (5.115)
˜ n
n
=
¸
s
[v
s
n
[
2
, (5.116)
N =
¸
n
([z
n
[
2
+ ˜ n
n
), (5.117)
c
s
= U
¸
n
[z
n
[
2
(z
∗
n
u
s
n
−z
n
v
s
n
)
¸
n
[z
n
[
2
. (5.118)
The HFBPopov are nonlinear equations and therefore it is more complicated
to get analytic approximations. Instead, we solved the HFBPopov equations
numerically by an iterative procedure, similar to the one followed in Ref. [19].
Each cycle of the iteration consists of two steps. In the ﬁrst step we solve
Eq. (5.115) subject to the constraint Eq. (5.117) by using the ˜ n
n
obtained
in the previous cycle. This generates new values for the z
n
. In the second
step we solve for ¦u
s
n
, v
s
n
¦ in Eqs. (5.113) using the ˜ n
n
from the previous
cycle and the newly generated z
n
. The ¦u
s
n
, v
s
n
¦ are used then to update ˜ n
n
.
5.6 Applications 85
20 10 0 10 20
n
0
F
n s
G
n s
s3
TF
BdG
20 10 0 10 20
n
0
F
n s
G
n s
s4
TF
BdG
20 10 0 10 20
n
0
F
n s
G
n s
s1
TF
BdG
20 10 0 10 20
n
0
F
n s
G
n s
s2
TF
BdG
Figure 5.8: Comparisons between the product F
s
n
G
s
n
calculated from the Thomas
Fermi approximation and the exact numerical solution of the BdG equations. The
parameters used were U/J = 0.2, Ω/J = 9.5 10
−4
and N = 100.
Because the HFBPopov is gapless, it is possible to keep the orthogonality
of the excitations to the condensate by solving Eqs. (5.113) with the c
s
set
to zero but removing in each cycle the projection of the calculated ¦u
s
n
, v
s
n
¦
amplitudes onto the condensate. Convergence is reached when the change in
¸
n
[˜ n
n
[
2
from one cycle to the next is smaller than a speciﬁed tolerance.
The parameters chosen for the numerical calculations were Ω = 0.0015E
R
,
with E
R
the one photon recoil energy, which for the case of a rubidium
condensate corresponds to a trap frequency of approximately 90 Hz. We
used a total number of 1000 atoms, N = 1000, and set UN = 1.0E
R
. J
was varied to achieve a range of V
eff
= U/J between 0.01 and 312. The
range was chosen based on a local mean ﬁeld approach [77], which for our
parameters estimates the transition region between V
eff
≈ 640 (at the
center where the local ﬁlling factor is approximately 80) and V
eff
≈ 12 (at
the wings).
The results of the numerical calculations are summarized in Figs. (5.95.13).
In Fig. 5.9 we plot the evolution of the density proﬁle (black boxes), the con
densate population (triangles) and the onsite depletion (empty diamonds)
as V
eff
is increased. In the plots we also show, for comparison purposes, the
ground state density proﬁle for J = 0 (empty boxes). This has the advan
tage that it can be calculated exactly from the Hamiltonian. In general we
observe the reduction of the condensate population and thus the increment
86 Chapter 5 Quadratic approximations
Veff =100
10 5 0 5 10
i
20
40
60
80
Veff =312
10 5 0 5 10
i
20
40
60
80
Veff =3
20
40
60
80
Veff =11
20
40
60
80
Veff =0.01
20
40
60
80
Veff =0.09
20
40
60
80
Figure 5.9: Condensate density (triangles), total density (ﬁlled boxes) and local
depletion (empty diamonds) as a function of the lattice site i for diﬀerent values
of V
eff.
.The site indices i are chosen such that i = 0 corresponds to the center of
the trap. Although these quantities are deﬁned only at the discrete lattice sites
we join them to help visualization. The empty boxes represent the exact solution
for the case J=0.
of the depletion with increasing interaction strength. When the system is in
the superﬂuid regime, most of the atoms are in the condensate, but as J is
decreased the depletion of the condensate becomes very important.
For the chosen parameters, the density proﬁle has a parabolic shape reﬂecting
the conﬁning potential. By comparing the evolution of the density as J is
decreased with the exact solution at J = 0, we can crudely estimate the
validity of the HFBPopov calculations. The density evolves from a Gaussian
type (see plots for V
eff
= 0.01 and 0.09) with smooth edges towards a
ThomasFermi proﬁle with sharp edges adjusting its shape to the J = 0
proﬁle. We can appreciate that around V
eff
= 3 both proﬁles are almost
equal. For lower values of J the HFBPopov density starts to diﬀer from
the J = 0 limit, even though the system is closer to the J = 0 limit. We
can say that beyond this point higher order correlations, neglected by the
theory, begin to be important. The departure of the HFBPopov density
proﬁle from the J = 0 one as J is decreased begins at the edges (see the
panels corresponding to V
eff
= 11 and 100). This is expected if we consider
the onsite depletion. For such values of V
eff
the local depletion in the wings
5.6 Applications 87
corresponds to a considerable percentage of the condensate populations, and
thus the validity of the HFBPopov assumptions starts to be dubious. The
homogeneous results shown in the previous sections corroborate our present
statements for the conﬁned system. For the smallest ﬁlling factor (see Fig.
5.2) the diﬀerences between the homogeneous HFBPopov calculations and
the exact solutions become important for values of V
eff
greater than 20.
For higher values of V
eff
, see plot for V
eff
= 312, the HFBPopov density
predictions diﬀers from the J = 0 solution even at the central wells. At
this point the failure of the method is clear and a fully quantal method is
required.
0 2 4 6
q
0
0.2
0.4
0.6
ω
q
Figure 5.10: Quasiparticle spectrum predicted by the HFBPopov theory for dif
ferent values of V
eff
: Empty diamonds (V
eff
= 0.01), stars (V
eff
= 0.09), crosses
(V
eff
= 3), ﬁlled diamonds (V
eff
= 11), empty boxes (V
eff
= 100) and pentagons
(V
eff
= 312). The letter q labels the quasiparticle energies in increasing order.
The quasiparticle energies are in recoil units.
The HFBPopov quasiparticle spectrum is shown in Fig. 5.10. It can be ob
served how the lower energy eigenvalues evolve from a linear non degenerated
spectrum to an almost degenerated one as J is decreased. It is worth men
tioning that the small energy diﬀerence between the ground and ﬁrst excited
states for high values of V
eff
makes the numerical solution very unstable in
the sense that it is very easy to jump to an excited state when solving for the
condensate wave function. The decrement in the energy spacing predicted by
the HFBPopov theory as the system approaches the transition is very useful
to keep in mind for the experimental realization of the Mott transition. As
the optical lattice depth is ramped up the adiabaticity criteria is harder to
88 Chapter 5 Quadratic approximations
fulﬁll.
10 5 0 5 10
i
0
2
4
6
∆
n
i
Figure 5.11: Number ﬂuctuations in the self consistent HFBPopov approach as
a function of lattice site for V
eff
= 0.01 (boxes), V
eff
= 0.09 (crosses), V
eff
= 3
(circles),V
eff
= 11 (triangles), V
eff
= 100 (stars) and V
eff
= 312 (diamonds). The
maximum value reached by the proﬁle decreases as V
eff
is increased. The empty
boxes shown for each of the curves correspond to the number ﬂuctuations predicted
by the homogeneous HFBPopov model using a local density approximation.
In Fig. 5.11 we plot the results for the number ﬂuctuations found numerically
using the inhomogeneous HFBPopov approach. The number ﬂuctuations
proﬁle reﬂects the condensate proﬁle. We also show the number ﬂuctua
tions evaluated by using a local density approximation (empty boxes). The
latter was calculated by substituting in the number ﬂuctuations expression
(Eq. (5.56)) the ¦u
q
, v
q
¦ amplitudes found for the homogeneous system (Eqs.
(5.66) and (5.67)), but replacing the condensate density in each lattice site
by the one found numerically for the trapped system (see Fig. 5.9). The
complete agreement between the two approaches justiﬁes the validity of the
local density approximation for the estimations of local quantities in con
ﬁned systems. Based on this agreement and the results for the homogeneous
system shown in the previous section, we expect that the inhomogeneous
HFBPopov results for squeezing also agree with the exact solutions right up
to the transition.
In Fig. 5.12 we present the quasimomentum distribution for the same param
eters used in the previous plots. The distribution for the two lowest values
of V
eff
corresponds to the one that characterizes an uncorrelated superﬂuid
phase with a narrow peak at small quasimomenta. As the hopping rate is
decreased we observe that the sharpness of the central peak decreases and
5.6 Applications 89
V
eff
=100
0 1 2 3
aq
0
100
200
n
q
V
eff
=312
0 1 2 3
aq
V
eff
=3
0
100
200
300
n
q
V
eff
=11
V
eff
=0.01
0
100
200
300
400
n
q
V
eff
=0.09
Figure 5.12: Quasimomentum distribution as a function of qa, a the lattice spacing,
q the quasimomentum, for diﬀerent values of V
eff
.
the distribution extends towards large quasimomenta. It is interesting to
note the appearance of a small peak between q = 0.5 and 1 which is most
noticeable for the V
eff
= 3 case. This agrees with numerical solutions of
the BoseHubbard Hamiltonian using Monte Carlo simulations [112]. We
attribute the origin of the small peak to the depletion of the condensate at
the wings. For the parameters when the small peak is present, the most
important contribution to the quasimomentum distribution still comes from
the condensate atoms. The step function like shape of the condensate proﬁle
causes an oscillatory [ sin(x)/x[ shape of the quasimomentum distribution.
As the lattice depth is increased the hopping becomes energetically costly,
the longrange order starts to decrease and the Fourier spectrum becomes
broader.
In Fig. 5.13 we plot the ﬁrst order on site superﬂuid fraction f
(1)
sn
which was
deﬁned in Eq.(5.60). The curves corresponding to V
eff
= 0.01 − 11, which
are in the regime where the HFBPopov is expected to be valid, depict
how as V
eff
is increased the superﬂuid proﬁle decreases faster at the wings
and at the center but no major change is observed in the middle section.
The evolution of the onsite superﬂuidity as the interaction strength is in
90 Chapter 5 Quadratic approximations
10 5 0 5 10
i
0
0.02
0.04
0.06
0.08
0.1
f
s
c
(
1
)
Figure 5.13: First order onsite superﬂuid fraction as a function of the lattice site i
for diﬀerent values of V
eff.
.The site indices i are chosen such that i = 0 corresponds
to the center of the trap. Filled boxes: V
eff.
=0.01, empty boxes: V
eff.
= 0.09,
empty diamonds: V
eff.
= 3, stars: V
eff.
= 11, crosses: V
eff.
= 100 and triangles:
V
eff.
= 312.
creased, exhibiting a domain localized decrement instead of a global one, is
in agreement with the development of uncompressible regions surrounded
by superﬂuid rings predicted for trapped systems [79] as the transition is
approached.
The Mott transition is a quantum phase transition and as for all critical phe
nomena, its behavior depends strongly on the dimensionality of the system.
In the present analysis, due to computational limitations, we considered one
dimensional systems. Experimentally, the Mott transition has been achieved
[46] in a 3 dimensional lattice with ﬁlling factors between 1 and 3. Even
though the HFBPopov approach fails to describe the strong coupling regime
for the one dimensional systems we considered, we showed how the method
is incredibly powerful in describing most of its characteristic features as they
are driven from the superﬂuid regime towards the transition. We expect
the HFBPopov method to give a better description of the transition as the
dimensionality of the system is increased and therefore to be a good model
in an experimental situation.
As shown in previous studies [44], [79] the Mott transition in a ddimensional
homogeneous system has two diﬀerent critical behaviors: one (d+1) XY like,
for systems with ﬁxed integer density as the interaction strength is changed,
and one mean ﬁeldlike exhibited when the transition is induced by changing
5.7 Improved HFBPopov approximation 91
the density. Diﬀerent from the homogeneous case where the Mott transition
is characterized by the global oﬀset of the superﬂuidity, for conﬁned systems,
commensuration is only well deﬁned locally. The inhomogeneity introduced
by the conﬁning potential allows the existence of extended Mott domains
(above a critical interaction strength) surrounded by superﬂuid ones [79],
thus the total superﬂuid fraction doesn’t vanishes in the Mott regime. This
issue, together with the fact that the ﬁnite length scale introduced by the trap
suppresses the long wave ﬂuctuations which are responsible for destroying the
mean ﬁeld [16]
1
, make us believe the critical behavior in conﬁned systems to
be more meanﬁeld like. Because the critical dimension for the latter type
of transition is two [44], [79], we expect that for trapped systems in d = 3,
the range of validity of the HFBPopov extends closer to the transition.
5.7 Improved HFBPopov approximation
The corrections to the quadratic Hamiltonian from higher order terms have essen
tially two major eﬀects. Firstly, they include the eﬀect of the interactions between
condensate and noncondensate atoms on the behavior of the condensate. Secondly,
they take into account the eﬀect of the surrounding atoms on the nature of the
interatomic collisions. We mentioned previously that although the HFB treatment
should represent an advance over the BdG theory it is not useful in practice be
cause it is not gapless. In the ﬁrst part of this section we show, how, by taking
into account the anomalous average ¯ m (Eq. (5.18)) we actually upgrade the bare
interaction potential U to the many body scattering matrix which gives a better
description of interparticle collisions. The problem with the HFB equations is that
not all the interactions are upgraded and the diagonal elements still contain bare
interaction terms. The diﬀerent treatment of the oﬀ diagonal and diagonal terms
is caused by the factorization approximation on which the HFB approximation is
based. The factorization is appropriate for the quartic term in the Hamiltonian
(as it is justiﬁed by Wick’s theorem [16]) but not for the cubic terms.
To solve the problem the ﬁrst approximation that one might think of is to
upgrade by hand the bare interactions in the diagonal terms and replace them
by the manybody scattering matrix. This procedure is known as the improved
Popov approximation [53, 16]. In this section, we discuss the improved Popov
approximation and apply it to a translationally invariant system. We explicitly
show that the improved Popov approximation gives a better description of the
manybody physics as the Mott insulator transition is approached in comparison
with the regular HFBPopov approximation.
1
One obvious consequence of this is that BEC is possible in one and two dimensions in a trap
whereas in the homogeneous, thermodynamic limit it can not occur in fewer than three dimensions
92 Chapter 5 Quadratic approximations
5.7.1 The twobody and manybody scattering matrices
The twobody scattering matrix, T
2b
, describes the scattering of two particles in
vacuum. It is deﬁned as a function of a complex parameter z by the Lippmann
Schwinger equation:[51]
T
2b
(z) = V +V
1
z −
ˆ
H
sp
T
2b
(z), (5.119)
Here V is the interatomic potential and
ˆ
H
sp
is the single particle Hamiltonian.
Although T
2b
is deﬁned for a general complex parameter z, this is physically in
terpreted as the energy of the scattering process. Inserting a complete set of
eigenstates in Eq. (5.119) one obtains:
T
2b
(z) = V +
¸
pq
V [pq`
1
z −(ε
p
+ε
q
)
'pq[T
2b
(z), (5.120)
where ε
p
and ε
q
are single particle energies (eigenvalues of
ˆ
H
sp
), and the kets [pq`
are the corresponding two particle eigenstates and describe the intermediate states
in the collision of two atoms. A diagrammatic representation of T
2b
is shown in
Fig. 5.14. Because of the ladder shape of the diagrams included in the scattering
matrix, they are usually referred as ladder diagrams.
Figure 5.14: Diagrammatic representation of the twobody T
2b
scattering matrix.
In this ﬁgure [kl` designates the initial states, [ij` the ﬁnal states and [np` a set
of intermediate states
T
2b
describes the interaction between two particles in vacuum. In the inter
acting Bose system, binary collisions do not occur in vacuum but in the presence
of other atoms. To describe their inﬂuence on the scattering the concept of the
manybody scattering matrix is introduced. The manybody scattering matrix,
T
mb
, plays the same role as its twobody counterpart, but describes scattering
occurring in manybody systems. For a bosonic gas there are two major eﬀects
5.7 Improved HFBPopov approximation 93
which need to be accounted for as compared to scattering in vacuum. First of all,
the relevant states which enter in the matrix elements should be manybody states
rather than single particle ones. Second, T
mb
should account for Bose enhance
ment of a scattering process. Various forms for the T
mb
have been been proposed
in the literature [16, 53, 54, 113] depending upon the approximations made. The
one that we are going to use is a simplest generalization of the twobody matrix
given by [16, 113]:
T
mb
(z) = V +
¸
pq
V [pq`
1 +n
p
+n
q
z −(ω
p
+ω
q
)
'pq[T
mb
(z), (5.121)
where z is again a complex parameter, ω
p
and ω
q
are quasiparticle energies, n
q
and n
p
are thermal quasiparticle occupation factors which vanish at T = 0, and
[pq` correspond to singleparticle wave functions. The full quasiparticle character
of the intermediate states is not taken into account in this simple deﬁnition. A
more general manybody scattering matrix which includes the full quasiparticle
wave functions has been discussed by Bijlsma and Stoof [53]. However this simpler
T
mb
matrix is the one that naturally appears in the manybody theory, at least at
lowest order.
At this point it is important to highlight that the zero energy of the manybody
scattering matrix is not the same as the zero energy in its twobody counterpart.
Because the twobody matrix is written in terms of single particle energies, in
this case the z is measured relative to the energy of a stationary particle. In the
manybody case T
mb
is deﬁned in terms of quasiparticle energies and therefore z
is measured with respect to the condensate chemical potential. In the dilute gas
limit, where the interparticle distance is large compared with the swave scattering
length, the ladder diagrams included in the T
mb
give the largest contribution to
the modiﬁcation of the bare potential due to interactions.
5.7.2 The anomalous average and the manybody scattering ma
trix
In this section we are going to follow [16, 53] to show how by including the anoma
lous average the manybody scattering matrix is introduced into the theory. For
interpretation purposes we are going to restrict the analysis to the translationally
invariant system.
The anomalous average ¯ m is deﬁned in terms of the quasiparticle amplitudes
u
q
and v
q
according to Eq.(5.93). The product u
q
v
q
, on the other hand, is related
to the oﬀ diagonal matrix elements, see Eq.(5.90). If we combine the two equations
we get an expression for the oﬀ diagonal matrix element ´
q−q
given by
´
q−q
n
o
= U +
U
M
¸
p
1
−(ω
p
+ω
−p
)
´
q−q
n
o
. (5.122)
To make the connection with the manybody scattering matrix, let us take the
inner product of Eq. (5.121), with the states 'q −q[ and [00` and evaluate it at
94 Chapter 5 Quadratic approximations
z = 0. At zero temperature, the thermal occupation factors vanish, n
q
= 0 and we
get:
'q −q[T
mb
(0)[00` = 'q −q[V [00` +
¸
pk
'q −q[V [pk`
1
−(ω
p
+ω
k
)
'pk[T
mb
(0)[00`.
(5.123)
The quantity 'k
1
k
2
[ V [k
3
k
4
` is the momentum representation of the twobody
interaction potential V . In the BoseHubbard Hamiltonian we assume a contact
potential with amplitude U which yields a potential in the momentum represen
tation given by
'k
1
k
2
[ V [k
3
k
4
` =
U
M
δ
k
1
+k
2
,k
3
+k
4
. (5.124)
Using Eq. (5.124) into (Eq. 5.123) we obtain
T
mb
= U +
1
M
¸
p
U
1
−(ω
p
+ω
−p
)
T
mb
, (5.125)
with
'p −p[T
mb
(0)[00` ≡
1
M
T
mb
. (5.126)
Eqs. (5.125) and (5.122) imply that ´
q−q
= n
o
T
mb
, and therefore that by includ
ing the anomalous average in the equations, the oﬀ diagonal matrix element given
by Un
o
in the BdG equations is upgraded to T
mb
n
o
in the HFB equations.
The manybody scattering matrix also appears in the DNLSE since Eq. (5.89)
contains ¯ m:
µ = −tJ +Un
o
+ 2U¯ n +U ¯ m = −tJ +T
mb
n
o
+ 2U¯ n. (5.127)
Finally, to conclude this section we just rewrite the HFB equations in terms of
T
mb
obtaining:
q
+n
o
(2U −T
mb
) −n
o
T
mb
n
o
T
mb
−
q
−n
o
(2U −T
mb
)
u
q
v
q
= ω
q
u
q
v
q
. (5.128)
This allow us to explicitly see the diﬀerent treatment that the theory gives to the
diagonal and oﬀ diagonal terms: The bare potential is completely upgraded to
T
mb
in the oﬀ diagonal terms but is not in the diagonal ones, and in the chemical
potential.
5.7 Improved HFBPopov approximation 95
5.7.3 Improved Popov approximation
The appearance of a gap in the HFB equations is because of the diﬀerent treatment
of the oﬀ diagonal and diagonal terms. We will show in chapter 8 that as we
go beyond the quadratic approximation of the Hamiltonian and include higher
order corrections we start to incorporate the manybody scattering in the diagonal
terms. However, a naive way to correct for the gap problem in the HFB equations
is simple to upgrade by hand the interactions in the diagonal terms and replace
them by the manybody scattering matrix. This way to proceed is known as the
improved Popov approximation [16, 53]. The matrix we have to diagonalize under
the improved Popov approximation is:
q
+n
o
T
mb
−n
o
T
mb
n
o
T
mb
−(
q
+n
o
T
mb
)
u
q
v
q
= ω
q
u
q
v
q
(5.129)
µ = −tJ +T
mb
n
o
+2
¯
T
mb
¯ n (5.130)
Eqs. (5.129) and (5.130) together with Eq. (5.125) and Eq. (5.69) form a closed
set of equations. They are exactly the same than the HFBPopov equations if
U is replaced by T
mb
. The reason why there are two diﬀerent coupling constant
in the equation for the chemical potential T
mb
and
¯
T
mb
, is because they describe
two diﬀerent scattering processes. The second term in the right hand side of Eq.
(5.130) describes the scattering between two atoms in the condensate colliding at
zero momentum, 'q −q[T
mb
(0)[00`, and that is why the coupling constant is T
mb
(see Eq. 5.126). On the other hand, the third term describes the collision between
one condensate atom and one atom out of the condensate and the coupling constant
in this case should be evaluated at diﬀerent energy
¯
T
mb
∝
¸
p
'p0[T
mb
(ω
p
)[0p`.
The extra diﬃculty that the improved Popov equations have compare with the
HFBPopov equations is that T
mb
depends on the quasiparticle energies and all
equations must be solved in a self consistent way. To get simple analytic expressions
we start by analyzing limiting regimes.
• Case n
o
T
mb
> J
In this regime the quasiparticle energy can be approximated by ω
q
2n
o
T
mb
q
.
Using this approximation in the expression for the manybody scattering ma
trix we obtain
T
mb
=
U
1 +
U
2
α
2
n
o
JT
mb
, (5.131)
with α deﬁned in Eq. (5.73). Solving the algebraic equation we get two
possible solutions
T
mb
= U ±
U
2
2Jn
o
α
2
+
α
4
+
4Jn
o
α
2
U
. (5.132)
96 Chapter 5 Quadratic approximations
Because we want the root which increases as U increases we choose the
solution with positive sign.
• Case n
o
T
mb
< J
In general one can show that the condition n
o
T
mb
< J is only satisﬁed if the
quantity T
mb
/J is small. This also implies that the parameter U/J must be
small. In this regime, therefore, we can perturbativelly expand Eq.( 5.131 )
in powers of U/J to get:
T
mb
U
1 +O
U
J
. (5.133)
The fact that Eq. (5.134) has the correct asymptotic behavior in the limit n
o
T
mb
<
J (it approaches to U as U goes to zero) even though it was derived under the
assumption n
o
T
mb
> J, allow us to extend the validity of Eq. (5.132) to all regimes.
Therefore, to a good approximation, we get a general formula for T
mb
given by:
T
mb
= U +
U
2
2Jn
o
α
2
+
α
4
+
4Jn
o
α
2
U
. (5.134)
We have checked numerically that in fact, Eq. (5.134) is a very good approximation
of T
mb
in all regimes.
Having an analytic expression for T
mb
, we can solve for the condensate density to
get:
n
o
= n −
1
M
¸
q,=0
q
+Un
o
2ω
q
+
1
2
,
≈ g −α
T
mb
J
n
o
. (5.135)
In the above equation we used the same approximations that lead to Eq.(5.72),
which was explicitly showed to be good in Figs. 5.2 and 5.3 and the deﬁnition of
g given in Eq. (5.74).
If we solve Eq. (5.135), we ﬁnally obtain an analytic expression for n
o
n
o
≈ g −α
Ug
J
. (5.136)
Surprisingly, the expression for the condensate density that we get after including
the manybody scattering matrix is exactly the same, as the one we get if in the
BdG equations for the condensate, Eq. (5.72), we replace n
o
by g in the right
hand side. Except for the term
M−1
2M
, this corresponds to the lowest order solution
we can get of Eq. (5.72), where instead of solving self consistently the algebraic
equation we replace n
o
by n. The term
M−1
2M
is very small at high densities when
n + 1/2 ≈ n, regime where in fact we only expect a mean ﬁeld treatment to be
valid and it might be interpreted as a ﬁnite size eﬀect. Hereafter we are going
5.7 Improved HFBPopov approximation 97
to refer to the non self consistent solution of the BdG equations as the test ﬁeld
approximation.
Finally, if we use the expression for the condensate density Eq. ( 5.136) in Eq.
(5.134), and assume again that n + 1/2 ≈ n, after some algebraic manipulations
the ﬁnal result we get is
n
o
T
mb
= Un. (5.137)
For us this a striking result. The net eﬀect of including the manybody scat
tering matrix in the theory reduces to replacing Un
o
in the BdG equations with
Un. This result is hard to understand because what it implies is that the lowest
order approximation beyond the simple DNLSE, is the one that reproduces the
best the exact solution as the interactions in the system are increased.
0 5 10 15 20 25 30
UJ
0.7
0.8
0.9
1
f
c
n5 M3
test
Popov
ManyB
Exact
0 5 10 15 20 25 30
UJ
0.7
0.8
0.9
1
f
c
n5.33 M3
test
Popov
ManyB
Exact
Figure 5.15: Comparisons of the condensate fraction predicted from the exact
diagonalization of the BHH (red), the improved Popov approximation (yellow), the
BdG (and HFBPopov) equations (blue) and the test ﬁeld approximation (green)
as a function of V
eff
= U/J, for a system with M = 3 and ﬁlling factors n = 5
and 5.33.
To check the improvement that we get by including the manybody scattering
matrix in the theory, in Figs 5.15 and 5.16 we show comparisons of the condensate
fraction, f
c
= n
o
/n, as a function of U/J calculated from the exact diagonalization
of the manybody Hamiltonian (red) with the improved Popov(yellow) and the
regular BdG (and HFBPopov)(blue) approximations. We also plot the test ﬁeld
approximation results (green). In the plots we used a one dimensional lattice with
M = 3 and densities n = 5 and n = 5.33 in Fig. 5.15 and n = 50 and n = 50.33
in Fig. 5.16.
It is clear in the plots that the improved Popov is a better approximation
than the regular BdG (and HFBPopov) approximation. In particular in the non
commensurate, high density case the agreement between the improved Popov and
the exact solution is very good. It also can be seen in the plots how the improved
Popov reduces to the simple test ﬁeld approximation. The small diﬀerences seen
in the low density case disappear as the density is increased. To corroborate the
98 Chapter 5 Quadratic approximations
0 50 100 150 200 250
UJ
0.7
0.8
0.9
1
f
c
n50. M3
test
Popov
ManyB
Exact
0 50 100 150 200 250
UJ
0.7
0.8
0.9
1
f
c
n50.3 M3
test
Popov
ManyB
Exact
Figure 5.16: Comparisons of the condensate fraction predicted from the exact
diagonalization of the BHH (red), the improved Popov approximation (yellow), the
BdG (and HFBPopov) equations (blue) and the test ﬁeld approximation (green)
as a function of V
eff
= U/J, for a system with M = 3 and ﬁlling factors n = 50
and 50.33.
validity of Eq. (5.137), we plot in Fig. 5.17 T
mb
f
c
, calculated from Eqs. (5.136)
and (5.134), vs U. The curves just overlap in the high density regime n = 50 and
in the low density case n = 5 the disagreement is very small.
0 5 10 15 20 25 30
U
0
5
10
15
20
25
30
T
m
b
f
c
n5
Tmb fc U
0 50 100 150 200 250
U
0
50
100
150
200
250
T
m
b
f
c
n50
Tmb fc U
Figure 5.17: Comparisons between n
o
/nT
mb
= f
c
T
mb
and U as a function of U for
a system with M = 3 and ﬁlling factors n = 5 and 50.
Phase Transition
One of the signatures of the entrance to the Mott insulator phase is a zero conden
sate density. As we mentioned below, the BdG (and HFBPopov) approximation
does not predict the Mott insulator transition because the condensate density only
vanishes in the limit when U/J → ∞. On the other hand, the improved Popov
approximation, does predict the existence of a critical value , V
c
eff
= (U/J)
c
, at
which the condensate density vanishes:
5.8 Conclusions 99
V
c
eff
=
n +
M−1
2M
α
2
−−→
n1
n
α
2
. (5.138)
If we compare the above expression with the equation of the critical point calcu
lated from a mean ﬁeld model starting in the Mott insulator phase (see chapter 3)
given by, V
c
eff
= t
(2n + 1) +
(2n + 1)
2
−1
, we see that in the high density
limit, the only regime where mean ﬁeld theories are expected to be valid, both
equations describe a critical point which scales linear with n and with a propor
tionality constant which depends only on the dimensionality of the system. In the
improved Popov approximation the proportionality constant is 1/α
2
, in the mean
ﬁeld theory starting from the Mott phase it is 8t (remember t is the number of
nearest neighbors). For a one dimensional lattice in the thermodynamic limit the
improved Popov approximation has the infrared divergency problem, α →∞ and
the critical point approaches zero. However, this is consistent with the calculations
done in Ref. [44] where the authors showed using renormalization group techniques
that the upper critical dimension for the transition at constant integer density is
d
c
= 2.
Nevertheless, the phase transition predicted by the improved Popov approxi
mation does not have a clear connection to the real Mott insulator phase transition.
A speciﬁc problem is the fact that the improved Popov approximation does not dis
tinguish between commensurate and incommensurate ﬁllings. This distinction is
crucial to reproduce the characteristic superﬂuid to Mott insulator phase diagram.
Furthermore, there is not clear signature that the system becomes incompress
ible in the Mott phase, and no gap in the excitation spectrum opens up as the
transition is reached.
5.8 Conclusions
In summary, in this chapter we have developed quadratic approximations for de
scribing the approach of a superﬂuid system towards the Mott insulator transition.
We have shown that this method can be used to predict the relevant physical quan
tities over a useful range. However, the quadratic approximations are developed
assuming small ﬂuctuations and therefore as soon as quantum correlations become
important it is clear that a fully quantal method is required.
100 Chapter 5 Quadratic approximations
Chapter 6
The two particle irreducible
eﬀective action (2PI) and the
closed time path (CTP)
formalism
In the last section we focused our attention on quadratic approximations of the
many body Hamiltonian and we used them to describe equilibrium properties of
ultra cold atoms loaded in optical lattices. In this chapter extend our analysis to
non equilibrium systems. The description of the evolution of condensates far from
equilibrium has gained considerable importance in matterwave physics, motivated
by recent experimental achievements such as the colliding and collapsing conden
sates [20, 21, 115, 114], collapses and revivals of the coherent matter ﬁeld [47]
and the NIST patterned loading experiment described in chapter 4. To date most
theoretical descriptions of nonequilibrium dynamics of BEC’s have been based on
the time dependent GrossPitaevskii equation, coupled with extended kinetic theo
ries that describe excitations in dilute weakly interacting systems close to thermal
equilibrium ([15],[58][62],[116]). However, experiments such as those mentioned
above have been able to achieve regimes where the standard mean ﬁeld description
is inapplicable, so new methods are required.
To treat farfromequilibrium dynamics, in this chapter we adopt a closed time
path (CTP) [56] functionalintegral formalism together with a twoparticle irre
ducible (2PI) [57] eﬀective action approach. We retain terms of up to secondorder
in the interaction strength when solving these equations. This method has been
generalized for and applied to the establishment of a quantum kinetic ﬁeld the
ory [117, 118, 119] with applications to problems in gravitation and cosmology
[120, 121], particles and ﬁelds [122, 123], BEC [54, 125] and condensed matter sys
tems [124] as well as addressing the issues of thermalization and quantum phase
transitions [126, 127, 128].
In this chapter we consider under the 2PICTP scheme diﬀerent approxima
tions: the HartreeFockBogoliubov (HFB) approximation, the nexttoleading
102
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
order 1/^ expansion of the 2PI eﬀective action up to secondorder in the in
teraction strength and a secondorder perturbative expansion in the interaction
strength. In chapter 7 we apply the 2PICTP approximations derived here to de
scribe the the patterned loading experiment [33] previously discussed in chapter
4.
6.1 2PI eﬀective action Γ(z, G)
Figure 6.1: Twoloop (upper row) and threeloop diagrams (lower row) contribut
ing to the eﬀective action. Explicitly, the diagram a) is what we call the double
bubble , b) the settingsun and c) the basketball.
The ﬁrst requirement for the study of nonequilibrium processes is a general
initialvalue formulation depicting the dynamics of interacting quantum ﬁelds. The
CTP or SchwingerKeldysh eﬀective action formalism [56] serves this purpose. The
second requirement is to describe the evolution of the correlation functions and
the mean ﬁeld on an equal footing. The two particle irreducible (2PI) formalism
[57] where the correlation functions appear also as independent variables, serves
this purpose. By requiring the generalized (master) CTP eﬀective action [118] to
be stationary with respect to variations of the correlation functions we obtain an
inﬁnite set of coupled (SchwingerDyson) equations for the correlation functions
which is a quantum analog of the BBGKY hierarchy. The 2PI eﬀective action
produces two such functions in this hierarchy. In this section we shall focus on the
2PI formalism, and then upgrade it to the CTP version in the next section.
Our starting point is the one dimensional BoseHubbard Hamiltonian (see chap
ter 3).
ˆ
H = −J
¸
i
(ˆ a
†
i
ˆ a
i+1
+ ˆ a
†
i+1
ˆ a
i
) +
1
2
U
¸
i
ˆ a
†
i
ˆ a
†
i
ˆ a
i
ˆ a
i
+
¸
i
V
i
ˆ a
†
i
ˆ a
i
(6.1)
6.1 2PI eﬀective action Γ(z, G) 103
where ˆ a
i
and ˆ a
†
i
are the bosonic operators that annihilate and create an atom
on the site i. Here, the parameter U denotes the strength of the onsite repulsion
of two atoms on the site i; the parameter V
i
denotes the energy oﬀset of each
lattice site due to an additional slowly varying external potential that might be
present (such as a magnetic trap), and J denotes the hopping rate between adjacent
sites. Because the nexttonearest neighbor amplitudes are typically two orders of
magnitude smaller, tunneling to nexttonearest neighbor sites can be neglected.
The Bose Hubbard Hamiltonian should be an appropriate model when the loading
process produces atoms in the lowest vibrational state of each well, with a chemical
potential smaller than the distance of the ﬁrst vibrationally excited state.
Hereafter we consider for simplicity only a homogeneous lattice with periodic
boundary conditions and no other external potential (V
i
= 0). Once the equations
of motion are derived, it is straightforward to generalize them to higher dimensions
or to include additional external potentials. As usual, we denote the total number
of atoms by N and the number of lattice sites by M.
The classical action associated with the BoseHubbard Hamiltonian (6.1), is
given in terms of the complex ﬁelds a
i
and a
∗
i
by
S[a
∗
i
, a
i
] =
dt
¸
i
ia
∗
i
(t)∂
t
a
i
(t)
+
dt
¸
i
J
a
∗
i
(t)a
i+1
(t) +a
i
(t)a
∗
i+1
(t)
(6.2)
−
dt
¸
i
U
2
a
∗
i
(t)a
∗
i
(t)a
i
(t)a
i
(t).
To simplify our notation we introduce a
a
i
(a = 1, 2) deﬁned by
a
i
= a
1
i
, a
∗
i
= a
2
i
. (6.3)
In terms of these ﬁelds the classical action takes the form
S[a] =
dt
¸
i
1
2
h
ab
a
a
i
(t)∂
t
a
b
i
(t (6.4)
+
dt
¸
i
Jσ
ab
a
a
i+1
(t)a
b
i
(t) −
U
4^
(σ
ab
a
a
i
(t)a
b
i
(t))
2
,
where ^ is the number of ﬁelds,which is two in this case, and summation over
repeated ﬁeld indices a, b = (1, 2) is implied and h
ab
and σ
ab
are matrices deﬁned
as
h
ab
= i
0 −1
1 0
, σ
ab
=
0 1
1 0
. (6.5)
In terms of the familiar Pauli matrices, σ
ab
= σ
x
and h
ab
= −σ
y
.
104
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
After second quantization the ﬁelds a
a
i
are promoted to operators. We denote
the expectation value of the ﬁeld operator, i.e. the mean ﬁeld, by z
a
i
(t) and the
expectation value of the composite ﬁeld by G
ab
ij
(t, t
t
). Physically, [z
a
i
(t)[
2
is the
condensate population and the composite ﬁelds determine the ﬂuctuations around
the mean ﬁeld:
z
a
i
(t) = 'a
a
i
(t)` , (6.6)
G
ab
ij
(t, t
t
) =
T
C
a
a
i
(t)a
b
i
(t
t
)
−'a
a
i
(t)`
a
b
i
(t
t
)
. (6.7)
The brackets denote the expectation value with respect to the density matrix and
T
C
denotes time ordering along a contour C in the complex plane.
All correlation functions of the quantum theory can be obtained from the eﬀec
tive action Γ[z, G], the two particle irreducible generating functional for Green’s
functions parameterized by z
a
i
(t) and the composite ﬁeld G
ab
ij
(t, t
t
). To get an
expression for the eﬀective action we ﬁrst deﬁne the functional Z[J, K] [57] as
Z[J, K] = e
iW[J,K]/
(6.8)
=
¸
a
Da
a
exp
i
¸
S[a] +
dt
¸
i
J
ia
(t)a
a
i
(t) +
1
2
dtdt
t
¸
ij
a
a
i
(t)a
b
j
(t
t
)K
ijab
(t, t
t
)
¸
,
where we have introduced the following index lowering convention
X
a
= σ
ab
X
b
. (6.9)
The functional integral (6.8) is a sum over classical histories of the ﬁeld a
a
i
in the
presence of the local source J
ia
and the non local source K
ijab
. The coherent
state measure is included in Da. The addition of the twoparticle source term is
what characterizes the 2PI formalism.
We deﬁne Γ[z, G] as the double Legendre transform of W[J, K] such that
δW[J, K]
δJ
ia
(t)
= z
a
i
(t), (6.10)
δW[J, K]
δK
ijab
(t, t
t
)
=
1
2
[z
a
i
(t)z
b
i
(t
t
) +G
ab
ij
(t, t
t
)]. (6.11)
Expressing J and K in terms of z and G yields
Γ[z, G] = W[J, K] −
dt
¸
i
J
ia
(t)z
a
i
(t) −
1
2
dtdt
t
¸
ij
z
a
i
(t)z
b
j
(t
t
)K
ijab
(t, t
t
)
−
1
2
dtdt
t
¸
ij
G
ab
ij
(t, t
t
)K
ijab
(t, t
t
). (6.12)
6.1 2PI eﬀective action Γ(z, G) 105
From this equation the following identity can be derived:
δΓ[z, G]
δz
a
i
(t)
= −J
ia
(t) −
dt
t
¸
j
(K
ijad
(t, t
t
))z
d
j
(t
t
), (6.13)
δΓ[z, G]
δG
ab
ij
(t, t
t
)
= −
1
2
K
ijab
(t, t
t
). (6.14)
In order to get an expression for Γ[z, G] notice that by using Eq.(6.8) for W[J, K]
and placing it in Eq.(6.12) for Γ[z, G], it can be written as
exp
i
Γ[z, G]
=
¸
a
Da
a
exp
i
S[a] +
dt
i
J
ia
(t) [a
a
i
(t) −z
a
i
(t)] (6.15)
+
1
2
dt
i
dt
t
j
a
a
i
(t)K
ijab
(t, t
t
)a
b
j
(t
t
) −z
a
i
(t)K
ijab
(t, t
t
)z
b
j
(t
t
)
−
1
2
TrGK
=
¸
a
Da
a
exp
i
S[a] −
dt
i
δΓ[z, G]
δz
a
i
(t)
[a
a
i
(t) −z
a
i
(t)]
−
dt
i
dt
t
j
[a
a
i
(t) −z
a
i
(t)]
δΓ[z, G]
δG
ab
ij
(t, t
t
)
a
b
i
(t
t
) −z
b
i
(t
t
)
+TrG
δΓ[z, G]
δG
¸
,
where Tr means taking the trace. For simplicity we have denoted
dt
¸
i
by
dt
i
.
Deﬁning the ﬂuctuation ﬁeld, ϕ
a
i
= a
a
i
−z
a
i
, we have
Γ[z, G] −TrG
δΓ[z, G]
δG
= −i ln
¸
a
Dϕ
a
exp
i
S[z, G; ϕ]
, (6.16)
S[z, G; ϕ] = S[z +ϕ] −
dt
i
δΓ[z, G]
δz
a
i
(t)
ϕ
a
i
(t) −
dt
i
dt
t
j
ϕ
a
i
(t)
δΓ[z, G]
δG
ab
ij
(t, t
t
)
ϕ
b
i
(t
t
). (6.17)
By introducing the classical inverse propagator iD
−1
(z) given by
iD
ijab
(t, t
t
)
−1
=
δS[z]
δz
a
i
(t)δz
b
j
(t
t
)
(6.18)
= (δ
ij
h
ab
∂
t
+J(δ
i+1j
+δ
i−1j
)σ
ab
) δ(t −t
t
)
−
U
^
(2z
ia
(t)z
ib
(t) +σ
ab
z
c
i
(t)z
ic
(t)) δ
ij
δ(t −t
t
),
the solution of the functional integrodiﬀerential equation (6.16) can be expressed
as
106
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
Γ[z, G] = S[z] +
i
2
Tr lnG
−1
+
i
2
TrD
−1
(z)G+ Γ
2
[z, G] +const. (6.19)
The quantity Γ
2
[z, G] is conveniently described in terms of the diagrams
generated by the interaction terms in S[z + ϕ] which are of cubic and higher
orders in ϕ
S
int
[z +ϕ] = −
U
4^
dt
i
(ϕ
ib
(t)ϕ
b
i
(t))
2
−
U
^
dt
i
ϕ
a
i
(t)z
ia
(t)ϕ
b
i
(t)ϕ
ib
(t). (6.20)
It consists of all twoparticle irreducible vacuum graphs (the diagrams repre
senting these interactions do not become disconnected by cutting two propagator
lines) in the theory with propagators set equal to G and vertices determined by
the interaction terms in S[z +ϕ] .
Since physical processes correspond to vanishing sources J and K, the dynam
ical equations of motion for the mean ﬁeld and the propagators are found by using
the expression (6.19) in equations (6.13) and (6.14) , and setting the right hand
side equal to zero. This procedure leads to the following equations:
h
ab
∂
t
z
b
i
(t) = −J(z
i+1a
(t) +z
i−1a
(t)) (6.21)
+
U
^
(z
id
(t)z
d
i
(t) +G
c
ii c
(t, t))z
ia
(t)
+
U
^
((G
iiad
(t, t) +G
iida
(t, t))z
d
i
(t)
−
δΓ
2
[z, G]
δz
a
i
(t)
,
and
G
−1
ijab
(t, t
t
) = D
ijab
(t, t
t
)
−1
−Σ
ijab
(t, t
t
), (6.22)
Σ
ijab
(t, t
t
) ≡ 2i
δΓ
2
[z, G]
δG
ab
ij
(t, t
t
)
. (6.23)
Equation (6.22) can be rewritten as a partial diﬀerential equation suitable for
initial value problems by convolution with G. This diﬀerential equation reads
explicitly
h
a
c
∂
t
G
cb
ij
(t, t
t
) = −J(G
ab
i+1j
(t, t
t
) +G
ab
i−1j
(t, t
t
)) +
U
^
(z
id
(t)z
d
i
(t))G
ab
ij
(t, t
t
) + (6.24)
2U
^
z
a
i
(t)G
cb
ij
(t, t
t
)z
ic
(t) +i
dt
tt
k
Σ
a
ikc
(t, t
tt
)G
cb
kj
(t
tt
, t
t
) +iδ
ab
δ
ij
δ(t −t
t
).
The evolution of z
a
and G
ab
is determined by Eqs. (6.22) and (6.24) once Γ
2
[z, G]
is speciﬁed.
6.2 Perturbative expansion of Γ
2
(z, G) and approximation schemes 107
6.2 Perturbative expansion of Γ
2
(z, G) and approxima
tion schemes
The diagrammatic expansion of Γ
2
is illustrated in Fig. 6.1, where two and
threeloop vacuum diagrams are shown. The dots where four lines meet represent
interaction vertices. The expression corresponding to each vacuum diagram
should be multiplied by a factor (−i)
l
(i)
s−2
where l is the number of solid lines
and s the number of loops the diagram contains.
The action Γ including the full diagrammatic series for Γ
2
gives the full dy
namics. It is of course not feasible to obtain an exact expression for Γ
2
in a closed
form. Various approximations for the full 2PI eﬀective action can be obtained
by truncating the diagrammatic expansion for Γ
2
. Which approximation is most
appropriate depends on the physical problem under consideration.
6.2.1 The standard approaches
1. Meanﬁeld approximation:
If, in Eq. (6.19), we discard all terms to the right of S[z], we recover the
DNLSE. This gives us the usual meanﬁeld description, in which the system
remains a pure condensate.
2. Oneloop Approximation:
The next approximation consists of discarding Γ
2
altogether. This yields
the socalled oneloop approximation. The oneloop approximation has an
equation for the ﬂuctuations identical to the BdG equations (See chapter 5),
however the equation of motion for the condensate does include the deple
tion and anomalous terms. The presence of this terms is necessary for the
time dependent evolution because they guarantee the conservation of particle
number and energy. The unequal treatment of the condensate and ﬂuctu
ations present in this approximation introduces limitations and makes this
approach not very attractive for studying the non equilibrium dynamics.
3. Timedependent HartreeFockBogoliubov (HFB) approximation:
A truncation of Γ
2
retaining only the ﬁrst order diagram in U, i.e., keeping
only the doublebubble diagram, Fig. 6.1, yields equations of motion for z and
G which correspond to the time dependent HartreeFockBogoliubov (HFB)
approximation. This approximation violates Goldstone’s theorem, but con
serves energy and particle number [132, 15, 133]. It is important to point out
that between all the quadratic approximations discussed in chapter 5 only
the HFB approximation is suitable for describing non equilibrium dynamics
because it is the only one that obeys conservation laws. The HFB equations
can also be obtained by using cumulant expansions up to the secondorder
[136] in which all cumulants containing three or four ﬁeld operators are
neglected. The HFB approximation neglects multiple scattering. It can be
108
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
interpreted as an expansion in terms of Ut/J, (where t is the time of evo
lution) and is good for the description of short time dynamics or weak
interaction strengths. It will be described in more detail in section 6.4.
6.2.2 Higher order expansions
We make a few remarks on the general properties of higher order expansions and
then specialize to two approximations.
• 2PI and Ladder Diagrams
Since the work of Beliaev [63] and Popov [52] it is well known in the literature
that including higher order terms in a diagrammatic expansion corresponds
to renormalizing the bare interaction potential to the fourpoint vertex (see
for example Ref. [53, 54]), thus accounting for the repeated scattering of the
bosons. In particular, in chapter 5 we explicitly showed how by including
the anomalous average in the equations the bare potential was upgraded in
the oﬀ diagonal elements to the many body scattering matrix. In chapter
8 we will also show that by taking into account the twoloop contribution
of the 2PI eﬀective action, the bare potential in the diagonal terms is also
upgraded to the many body scattering matrix up to secondorder in the lad
der expansion. The twoloop contribution includes diagrams topologically
identical to those found by Beliaev, but with the exact propagator instead
of the oneloop propagator. In the dilute gas limit, where the interparticle
distance is large compared with the swave scattering length, the ladder di
agrams give the largest contribution to the fourpoint vertex. To lowest
order in the diluteness parameter, the Tmatrix can be approximated by a
constant proportional to the scattering length (the pseudopotential approx
imation). However this approximation is only valid in the weak interaction
limit and neglects all momentum dependencies which appears in the problem
as higher order terms. In that sense the 2PI eﬀective action approach allows
us to go beyond the weakly interacting limit in a systematic way and to treat
collisions more accurately.
• Nonlocal Dissipation and NonMarkovian Dynamics
Higher order terms lead to nonlocal equations and dissipation. The presence
of nonlocal terms in the equations of motion is a consequence of the fact
that the 2PI eﬀective action really corresponds to a further approximation
of the master eﬀective equation [118]. Though strict dissipation can never
be observed in an energy conserving closed system, characteristic features of
dissipative systems like exponential damping of correlations can be exhibited
once interactions are properly taken into account.
NonMarkovian dynamics is a generic feature of the 2PI formalism which
yields integrodiﬀerential equations of motion. This makes numerical so
lution diﬃcult, but is a necessary price to pay for a fuller account of the
quantum dynamics. Many well acknowledged approaches to the quantum
6.2 Perturbative expansion of Γ
2
(z, G) and approximation schemes 109
kinetics of such systems adopt either explicitly or implicitly a Markovian
approximation [62]. The Markovian approximation assumes that only the
current conﬁguration of the system, but not its history, determines its future
evolution. Markovian approximations are made if one assumes instantaneous
interactions, or in the kinetic theory context that the time scales between
the duration of binary collisions τ
0
and the inverse collision rate τ
c
are well
separated. In the low kinetic energy, weakly interacting regime the time
between collisions (or mean free path) is long compared to the reaction time
(or scattering length): τ
c
>> τ
0
. The long separation between collisions and
the presence of intermediate weak ﬂuctuations, allow for a rapid decay of
the temporal and spatial correlations created between collision partners,
which one can use to justify the Markovian approximation. However,non
Markovian dynamics needs to be confronted squarely in systems such as the
patterned loaded lattice, in which the lattice which conﬁnes the atoms to the
bottom of the wells with enhanced interaction eﬀects, accompanied by the
low dimensionality of the system and farfromequilibrium initial conditions.
That is the rationale for our adoption of the CTP 2PI scheme. Now, for the
speciﬁcs:
1. Secondorder expansion:
A truncation retaining diagrams of secondorder in U consists of the double
bubble, the settingsun and the basketball (see Fig.6.1). By including the
settingsun and the basketball in the approximations we are taking into ac
count two particle scattering processes [121, 119]. Secondorder terms lead
to integro diﬀerential equations which depend on the time history of the
system.
2. Large^ approximation
The 1/^ expansion is a controlled nonperturbative approximation scheme
which can be used to study nonequilibrium quantum ﬁeld dynamics in
the regime of strong interactions[127, 126]. In the large ^ approach the
ﬁeld is modelled by ^ ﬁelds and the quantum ﬁeld generating functional
is expanded in powers of 1/^. In this sense the method is a controlled
expansion in a small parameter, but unlike perturbation theory in the cou
pling constant, which corresponds to an expansion around the vacuum, the
large ^ expansion corresponds to an expansion of the theory about a strong
quasiclassical ﬁeld.
6.2.3 Zero mode ﬂuctuactions
In chapter 5 we neglected zero mode ﬂuctuations due to the non perturbative
character of the zero mode. In the linearized theory, this approach introduces an
artiﬁcial infrared divergence in low dimensional models thus the theory is actually
improved if the contribution from this mode is neglected all together [96]. However,
from a physical point of view the zero mode exists and is quantum in nature. There
110
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
are both fundamental and practical reasons why isolating and subtracting the zero
mode is not as compelling in the dynamical evolution we want to describe in this
chapter. Firstly, because the 2PI formalism goes beyond the linearized approxi
mation, the zero mode does not have the impact it has in the linearized formalism
and it is not clear that subtracting it necessarily leads to a better approximation.
Secondly, the initial state that we are going to assume for our analysis is a coherent
state rather than a proper state of the total particle number as will be described
in chapter 7. Moreover, as the total particle number is not very high, quantum
ﬂuctuations in the total particle number are real, and nonnegligible. Discarding
these ﬂuctuations would spoil the integrity of the formalism. Therefore, in this
chapter we shall not attempt to isolate the contributions from the zero mode. A
full nonperturbative treatment in the future is certainly desirable.
6.3 CTP formalism
In order to describe the nonequilibrium dynamics we will now specify the contour
of integration in Eqs. (6.22) and (6.24) to be the SchwingerKeldysh contour [56]
along the realtime axis or closed time path (CTP) contour. The Schwinger
Keldysh formalism is a powerful method for deriving real and causal evolution
equations for the expectation values of quantum operators for nonequilibrium
ﬁelds. The basic idea of the CTP formalism relies on the fact that a diagonal
matrix element of a system at a given time, t = 0, can be expressed as a product
of transition matrix elements from t = 0 to t
t
and the timereverse (complex
conjugate) matrix element fromt
t
to 0 by inserting a complete set of states into this
matrix element at the later time t
t
. Since each term in the product is a transition
matrix element of the usual or time reversed kind, the standard path integral
representation for each one can be introduced. However, to get the generating
functional we seek, we have to require that the forward time evolution takes place
in the presence of a source J
+
but the reversed time evolution takes place in the
presence of a diﬀerent source J
−
, otherwise all the dependence on the source drops
out.
The doubling of sources, the ﬁelds and integration contours suggest introduc
ing a 2 x 2 matrix notation. This notation has been discussed extensively in the
literature [117, 119]. However we are going to follow Refs. [126] and [127] and
introduce the CTP formalism in our equation of motion by using the composition
rule for transition amplitudes along the time contour in the complex plane. This
way is cleaner notationally and has the advantage that all the functional formal
ism of the previous section may be taken with the only diﬀerence of path ordering
according to the complex time contour C
ctp
in the time integrations.
The twopoint functions are decomposed as
G
ab
ij
(t, t
t
) = θ
ctp
(t, t
t
)G
ab>
ij
(t, t
t
) +θ
ctp
(t
t
, t)G
ab<
ij
(t, t
t
), (6.25)
where
6.3 CTP formalism 111
G
ab>
ij
(t, t
t
) =
ϕ
a
i
(t)ϕ
b
j
(t
t
)
, (6.26)
G
ab<
ij
(t, t
t
) =
ϕ
b
i
(t
t
)ϕ
a
j
(t)
, (6.27)
with ϕ
i
being the ﬂuctuation ﬁeld deﬁned prior Eq. (17) and θ
ctp
(t −t
t
) being the
CTP complex contour ordered theta function deﬁned by
θ
ctp
(t, t
t
) =
θ(t, t
t
) for t and t
t
both on C
+
θ(t
t
, t) for t and t
t
both on C
−
1 for t on C
−
and t
t
on C
+
0 for t on C
+
and t
t
on C
−
. (6.28)
With these deﬁnitions the matrix indices are not required. When integrating over
the second half C
−
, we have to multiply by a negative sign to take into account
the opposite direction of integration.
To show explicitly that the prescription for the CTP integration explained
above does lead to a wellposed initial value problem with causal equations, let us
explicitly consider the integral in Eq. (6.24). The integrand has the CTP ordered
form
Σ(t, t
tt
)G(t
tt
, t
t
) = (6.29)
θ
ctp
(t, t
tt
)θ
ctp
(t
tt
, t
t
)Σ
>
(t, t
tt
)G
>
(t
tt
, t
t
) +θ
ctp
(t, t
tt
)θ
ctp
(t
t
, t
tt
)Σ
>
(t, t
tt
)G
<
(t
tt
, t
t
)
θ
ctp
(t
tt
, t)θ
ctp
(t
tt
, t
t
)Σ
<
(t, t
tt
)G
>
(t
tt
, t
t
) +θ
ctp
(t
tt
, t)θ
ctp
(t
t
, t
tt
)Σ
<
(t, t
tt
)G
<
(t
tt
, t
t
),
where we have omitted the indices because they are not relevant for the discussion.
Using the rule for CTP contour integration we get
dt
tt
Σ(t, t
tt
)G(t
tt
, t
t
) =
t
0
dt
tt
θ(t
tt
, t
t
)Σ
>
(t, t
tt
)G
>
(t
tt
, t
t
) +θ(t
t
, t
tt
)Σ
>
(t, t
tt
)G
<
(t
tt
, t
t
)
+
∞
t
dt
tt
θ(t
tt
, t
t
)Σ
<
(t, t
tt
)G
>
(t
tt
, t
t
) +θ(t
t
, t
tt
)Σ
<
(t, t
tt
)G
<
(t
tt
, t
t
)
−
∞
0
dt
tt
Σ
<
(t, t
tt
)G
>
(t
tt
, t
t
). (6.30)
If t > t
t
, we have
dt
tt
Σ(t, t
tt
)G(t
tt
, t
t
) =
t
0
dt
tt
(Σ
>
(t, t
tt
) −Σ
<
(t, t
tt
))G
>
(t
tt
, t
t
)
−
t
0
dt
tt
Σ
>
(t, t
tt
)(G
>
(t
tt
, t
t
) −G
<
(t
tt
, t
t
)). (6.31)
112
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
On the other hand, if t < t
t
dt
tt
Σ(t, t
tt
)G(t
tt
, t
t
) =
t
0
dt
tt
(Σ
>
(t, t
tt
) −Σ
<
(t, t
tt
))G
<
(t
tt
, t
t
)
−
t
0
dt
tt
Σ
<
(t, t
tt
)(G
>
(t
tt
, t
t
) −G
<
(t
tt
, t
t
)). (6.32)
The above equations are explicitly causal.
It is convenient to express the evolution equations in terms of two independent
twopoint functions which can be associated to the expectation values of the
commutator and the anticommutator of the ﬁelds. We deﬁne, following Ref. [127]
the functions
G
(F)ab
ij
(t, t
t
) =
1
2
G
ab>
ij
(t, t
t
) +G
ab<
ij
(t, t
t
)
(6.33)
G
(ρ)ab
ij
(t, t
t
) = i
G
ab>
ij
(t, t
t
) −G
ab<
ij
(t, t
t
)
, (6.34)
where the (F) functions are usually called statistical propagators and the (ρ) are
called spectral functions [137]. With these deﬁnitions Eq.(6.24) can be rewritten
as:
h
a
c
∂
t
G
(F)cb
ij
(t, t
t
) = −J
G
(F)ab
i+1j
(t, t
t
) +G
(F)ab
i−1j
(t, t
t
)
+
U
^
z
ic
(t)z
c
i
(t))G
(F)ab
ij
(t, t
t
)
+
2U
^
z
a
i
(t)G
(F)cb
ij
(t, t
t
)z
ic
(t)
+
t
0
dt
tt
k
Σ
(ρ)ac
ik
(t, t
tt
)G
(F) b
kj c
(t
tt
, t
t
)
−
t
0
dt
tt
k
Σ
(F)ac
ik
(t, t
tt
)G
(ρ)b
kjc
(t
tt
, t
t
), (6.35)
h
a
c
∂
t
G
(ρ)cb
ij
(t, t
t
) = −J
G
(ρ)ab
i+1j
(t, t
t
) +G
(ρ)ab
i−1j
(t, t
t
)
+
U
^
z
ic
(t)z
c
i
(t))G
(ρ)ab
ij
(t, t
t
+
2U
^
z
a
i
(t)G
(ρ)cb
ij
(t, t
t
)z
ic
(t)
+
t
t
dt
tt
k
Σ
(ρ)ac
ik
(t, t
tt
)G
(ρ)b
kjc
(t
tt
, t
t
).(6.36)
In particular, we deﬁne the normal, ρ, and anomalous, m, spectral and statistical
functions as
G
21(F)
ij
(t, t
t
) ≡ ρ
(F)
ij
(t, t
t
) =
1
2
ϕ
†
i
(t)ϕ
j
(t
t
) +ϕ
j
(t
t
)ϕ
†
i
(t)
, (6.37)
G
21(ρ)
ij
(t, t
t
) ≡ ρ
(ρ)
ij
(t, t
t
) = i
ϕ
†
i
(t)ϕ
j
(t
t
) −ϕ
j
(t
t
)ϕ
†
i
(t)
, (6.38)
G
11(F)
ij
(t, t
t
) ≡ m
(F)
ij
(t, t
t
) =
1
2
ϕ
i
(t)ϕ
j
(t
t
) +ϕ
j
(t
t
)ϕ
i
(t)
, (6.39)
G
11(ρ)
ij
(t, t
t
) ≡ m
(ρ)
ij
(t, t
t
) = i
ϕ
i
(t)ϕ
j
(t
t
) −ϕ
j
(t
t
)ϕ
i
(t)
. (6.40)
6.4 HFB approximation 113
With these relations in place, we now proceed to derive the time evolution
equations for the mean ﬁeld and the twopoint functions from the 2PICTP eﬀec
tive action for the BoseHubbard model under the three approximations described
before.
6.4 HFB approximation
6.4.1 Equations of motion
The HFB equations include the leading order contribution of Γ
2
. They describe
the coupled dynamics of condensate and noncondensate atoms which arise from
the most important scattering processes which are direct, exchange and pair ex
citations. The basic damping mechanisms present in the HFB approximation are
Landau and Beliaev damping associated with the decay of an elementary exci
tation into a pair of excitations in the presence of condensate atoms[133, 134].
However, these kinds of damping
1
found in the HFB approximation (due to phase
mixing, as in the Vlasov equation citeBalescu) are diﬀerent in nature from the
collisional dissipation (as in the Boltzmann equation) responsible for thermaliza
tion processes. Multiple scattering processes are neglected in this approximation.
We expect the HFB equations to give a good description of the dynamics in the
collisionless regime when interparticle collisions play a minor role.
The leading order contribution of Γ
2
is represented by the doublebubble
diagram. Its contribution to Γ
2
is z independent and has an analytic expression
of the form
Γ
(1)
2
[G] = (6.41)
−
U
4^
dt
i
G
a
iia
(t, t)G
b
iib
(t, t) + 2G
iiab
(t, t)G
ab
ii
(t, t)
,
the factor of two arises because the direct and exchange terms are identical.
Using the ﬁrst order expression for Γ
2
in Eqs. (6.22) and (6.24) yields the
following equations of motion.
h
a
b
∂
t
z
b
i
(t) = ζ
z
HFB
, (6.42)
ζ
z
HFB
≡ −J
z
a
i+1
(t) +z
a
i−1
(t)
+
U
^
z
id
(t)z
d
i
(t) +G
d
ii d
(t, t)
z
a
i
(t)
+
2U
^
z
ib
(t)G
ab
ii
(t, t)
, (6.43)
1
we make a distinction between the meaning of the words ‘damping’ and ‘dissipation’, the
former referring simply to the phenomenological decay of some function, the latter with theoretical
meaning, e.g., in the Boltzmann sense.
114
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
h
a
c
∂
t
G
cb
ij
(t, t
t
) = ζ
G
HFB
, (6.44)
ζ
G
HFB
≡ −J(G
ab
i+1j
(t, t
t
) +G
ab
i−1j
(t, t
t
)) +
U
^
(z
id
(t)z
d
i
(t) +G
d
ii d
(t, t))G
ab
ij
(t, t
t
)
+
2U
^
(z
a
i
(t)z
ic
(t)G
cb
ij
(t, t
t
) +G
a
ii d
(t, t)G
db
ij
(t, t
t
)) +iδ
ab
δ
ij
δ
C
(t −t
t
).
In terms of the spectral and statistical functions, Eqs. (6.37) to (6.40), and
setting ^ = 2, the above equations take the form
i∂
t
z
i
(t) = −J(z
i+1
(t) +z
i−1
(t)) +U([z
i
(t)[
2
+ 2ρ
(F)
ii
(t, t))z
i
(t)
+Um
(F)
ii
(t, t)z
∗
i
(t), (6.45)
−i
∂
∂t
ρ
(F)
ij
(t, t
t
) = L
ik
(t)ρ
(F)
kj
(t, t
t
) +´
∗
ik
(t)m
(F)
kj
(t, t
t
), (6.46)
−i
∂
∂t
ρ
(ρ)
ij
(t, t
t
) = L
ik
(t)ρ
(ρ)
kj
(t, t
t
) +´
∗
ik
(t)m
(ρ)
kj
(t, t
t
), (6.47)
i
∂
∂t
m
(F)
ij
(t, t
t
) = L
ik
(t)m
(F)
kj
(t, t
t
) +´
ik
(t)ρ
(F)
kj
(t, t
t
), (6.48)
i
∂
∂t
m
(ρ)
ij
(t, t
t
) = L
ik
(t)m
(ρ)
kj
(t, t
t
) +´
ik
(t)ρ
(ρ)
kj
(t, t
t
), (6.49)
with
L
ij
(t) = −J(δ
i+1j
+δ
i−1j
) + 2Uδ
ij
[z
i
(t)[
2
+ρ
(F)
ii
(t, t)
, (6.50)
´
ij
(t) = Uδ
ij
z
i
(t)
2
+m
(F)
ii
(t, t)
. (6.51)
The time dependent HFB equations are a closed set of selfconsistent equations
that describe the coupled dynamics of the condensate and noncondensate compo
nents of a Bose gas. It can be checked that they preserve important conservation
laws such as the number of particles and energy. The conservation properties of
the HFB equations can also be understood by the fact that these equations can
also be derived using Gaussian variational methods [15]. These methods always
yield a classical Hamiltonian dynamics which guarantees probability conservation.
Because they are local in time they can be decoupled by a mode decomposition.
6.4.2 Mode expansion of the HFB equations
To decouple the HFB equations we apply the well known Bogoliubov transforma
tion to the ﬂuctuation ﬁeld (see chapter 5)
ϕ
j
(t) =
¸
q
u
q
i
(t)ˆ α
q
−v
∗q
i
(t)ˆ α
†
q
, (6.52)
6.4 HFB approximation 115
where (ˆ α
q
, ˆ α
†
q
) are time independent creation and annihilation quasiparticle oper
ators and all the time dependence is absorbed in the amplitudes ¦u
q
i
(t), v
∗q
i
(t)¦.
To ensure that the quasiparticle transformation is canonical, the amplitudes must
fulﬁll the relations
¸
i
u
q
i
(t)u
∗k
i
(t) −v
q
i
(t)v
∗k
i
(t)
= δ
qk
, (6.53)
¸
i
u
q
i
(t)v
k
i
(t) −v
q
i
(t)u
k
i
(t)
= 0. (6.54)
In the zero temperature limit, where the quasiparticle occupation number vanishes,
ˆ α
†
q
ˆ α
k
= 0, the statistical and spectral functions take the form
ρ
(F)
ij
(t, t
t
) =
1
2
¸
q
v
q
i
(t)v
∗q
j
(t
t
) +u
q
j
(t
t
)u
∗q
i
(t)
, (6.55)
ρ
(ρ)
ij
(t, t
t
) = i
¸
q
v
q
i
(t)v
∗q
j
(t
t
) −u
q
j
(t
t
)u
∗q
i
(t)
, (6.56)
m
(F)
ij
(t, t
t
) = −
1
2
¸
q
u
q
i
(t)v
∗q
j
(t
t
) +u
q
j
(t
t
)v
∗q
i
(t)
, (6.57)
m
(ρ)
ij
(t, t
t
) = −i
¸
q
u
q
i
(t)v
∗q
j
(t
t
) −u
q
j
(t
t
)v
∗q
i
(t)
. (6.58)
Notice that at equal time and position due to Eqs.(6.53) and (6.54), ρ
(F)
ii
(t, t)
and m
(F)
ii
(t, t) satisfy
ρ
(F)
ii
(t, t) =
¸
q
[v
q
i
(t)[
2
+
1
2
(6.59)
m
(F)
ii
(t, t) = −
¸
q
u
q
i
(t)v
∗q
i
(t). (6.60)
Replacing Eqs. (6.59)(6.60) into Eqs. (6.45)(6.49) and using the constraints
(6.53) (6.54) we recover the standard time dependent equations for the quasipar
ticle amplitudes
i
∂
∂
z
i
(t) (6.61)
= −J(z
i+1
(t) +z
i−1
(t)) +U([z
i
(t)[
2
+ 2ρ
(F)
ii
(t, t))z
i
(t) +Um
(F)
ii
(t, t)z
∗
i
(t),
116
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
i
∂
∂t
u
q
i
(t) (6.62)
= −J(u
q
i+1
(t) +u
q
i−1
(t)) + 2U([z
i
(t)[
2
+ρ
(F)
ii
(t, t))u
q
i
(t) −U(m
(F)
ii
(t, t) +z
i
(t)
2
)v
q
i
(t),
−i
∂
∂t
v
q
i
(t) = (6.63)
−J(v
q
i+1
(t) +v
q
i−1
(t))) + 2U([z
i
(t)[
2
+ρ
(F)
ii
(t, t))v
q
i
(t) −U(m
∗(F)
ii
(t, t) +z
i
(t)
∗2
)u
q
i
(t).
If we compare Eqs. (6.61)(6.63) with their time independent version, Eqs. (5.27)
to (5.87), we notice that the equations are identical as they should be if we replace
i
∂
∂t
u
q
i
→ω
HFB
u
q
i
, i
∂
∂t
v
q
i
→ω
HFB
v
q
i
and i∂
t
z
i
→µ
HFB
z
i
. The factor of 1/2
diﬀerence between ρ
(F)
ii
and ρ
ii
(see Eq. (5.19)) just leads to a trivial renormaliza
tion of the chemical potential. In the time dependent equations we have included
the zero mode ﬂuctuations, which were neglected in Eqs. (5.27) to (5.87). Hence
we do not need the c
HFB
±
s
variables, which simply keep the orthogonality between
of the condensate and the excitations. Eqs. (6.61) (6.63) correspond to a set of
M(2M +1) coupled ordinary diﬀerential equations, where M is the total number
of lattice sites. They can be solved using standard time propagation algorithms.
Once the time dependent quasiparticle amplitudes are calculated we can derive the
dynamics of physical observables constructed from them as a function of time,
such as the average number of particles in a well n
i
(t), etc.
6.5 Secondorder expansion
6.5.1 Equations of motion
Full secondorder
The secondorder contribution to Γ
2
is described in terms of the settingsun Fig. 6.1b
and the basketball Fig. 6.1c diagrams. The basketball diagram is independent of
the meanﬁeld and is constructed with only quartic vertices. The settingsun
diagram depends on z and contains only threepoint vertices. The secondorder
Γ
(2)
2
eﬀective action can be written as
Γ
(2)
2
[z, G] = (6.64)
i
U
^
2
dt
i
dt
j
z
ib
(t)z
jb
(t
t
)
G
bb
ij
(t, t
t
)G
ijdd
(t, t
t
)G
dd
ij
(t, t
t
) + 2G
bd
ij
(t, t
t
)G
ijdd
(t, t
t
)G
db
ij
(t, t
t
)
+
i
U
2^
2
dt
i
dt
t
j
G
ijbb
(t, t
t
)G
bb
ij
(t, t
t
)G
ijdd
(t, t
t
)G
dd
ij
(t, t
t
)+
2G
ijbb
(t, t
t
)G
bd
ij
(t, t
t
)G
ijdd
(t, t
t
)G
db
ij
(t, t
t
)
.
6.5 Secondorder expansion 117
To simplify the notation, let us introduce the following deﬁnitions [127]:
Π
ij
(t, t
t
) = −
1
2
G
ijab
(t, t
t
)G
ab
ij
(t, t
t
), (6.65)
Ξ
ijab
(t, t
t
) = −D(t, t
t
)G
ijab
(t, t
t
), (6.66)
D(t, t
t
) = z
ib
(t)z
ja
(t
t
)G
ba
ij
(t, t
t
) −Π
ij
(t, t
t
), (6.67)
Λ
b
ij a
(t, t
t
) = −G
cb
ij
(t, t
t
)G
ijca
(t, t
t
), (6.68)
Λ
b
ij a
(t, t
t
) = −G
bc
ij
(t, t
t
)G
ijac
(t, t
t
), (6.69)
Θ
ac
ij
(t, t
t
) = −(z
id
(t)z
jb
(t
t
) +G
ijdb
(t, t
t
))G
ab
ij
(t, t
t
)G
dc
ij
(t, t
t
) + Ξ
ac
ij
(t, t
t
). (6.70)
With the above deﬁnitions we ﬁnd from Eqs. (6.22) and (6.24) the following
equations of motion:
h
a
b
∂
t
z
b
i
(t) = ζ
z
HFB
+ (6.71)
i
2U
^
2
dt
t
j
z
jb
(t
t
)
Π
ij
(t, t
t
)G
ba
ji
(t
t
, t) + Λ
b
ij c
(t, t
t
)G
ac
ij
(t, t
t
)
,
h
a
c
∂
t
G
cb
ij
(t, t
t
) = ζ
G
HFB
+ (6.72)
i
2U
^
2
z
a
i
(t)
dt
tt
k
z
kc
(t
tt
)
Π
ik
(t, t
tt
)G
cb
kj
(t
tt
, t
t
) + Λ
c
ik d
(t, t
tt
)G
db
kj
(t
tt
, t
t
)
+i
2U
^
2
dt
tt
Θ
a
ik d
(t, t
tt
) + Λ
ca
ik
(t, t
tt
)z
ic
(t)z
kd
(t
tt
)
G
db
kj
(t
tt
, t
t
),
where ζ
z
HFB
and ζ
G
HFB
are deﬁned in Eqs. (6.43) and (6.45).
To get explicit expressions of the equations of motion in terms of ρ
(F,ρ)
and
m
(F,ρ)
we introduce the functions
Ω
(F)
ij
[f , g] = f
(F)
ij
(t
i
, t
j
)g
(F)
ij
(t
i
, t
j
) −
1
4
f
(ρ)
ij
(t
i
, t
j
)g
(ρ)
ij
(t
i
, t
j
)
, (6.73)
Ω
(ρ)
ij
[f , g] = f
(F)
ij
(t
i
, t
j
)g
(ρ)
ij
(t
i
, t
j
) +f
(ρ)
ij
(t, t
t
)g
(F)
ij
(t
i
, t
j
). (6.74)
Using the spectral and statistical functions and setting the number of ﬁelds ^ equal
to 2, the equations of motion derived under the full secondorder approximation
can be written as:
118
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
i∂
t
i
z
i
= −J(z
i+1
(t
i
) +z
i−1
(t
i
)) +U([z
i
[
2
+ 2ρ
(F)
ii
)z
i
+Um
(F)
ii
z
∗
i
(6.75)
−2U
2
¸
k
t
i
0
dt
k
z
k
Ω
(ρ)
ik
[ρ, ρ
∗
] +z
k
Ω
(ρ)
ik
[m, m
∗
] +z
∗
k
Ω
(ρ)
ik
[m, ρ]
ρ
(F)
ki
−2U
2
¸
k
t
i
0
z
∗
k
Ω
(ρ)
ik
[ρ, ρ
∗
] +z
∗
k
Ω
(ρ)
ik
[m, m
∗
] +z
k
Ω
(ρ)
ik
[m
∗
, ρ
∗
]
m
(F)
ki
+2U
2
¸
k
t
i
0
dt
k
z
k
Ω
(F)
ik
[ρ, ρ
∗
] +z
k
Ω
(F)
ik
[m, m
∗
] +z
∗
k
Ω
(F)
ik
[m, ρ]
ρ
(ρ)
ki
+2U
2
¸
k
t
i
0
z
∗
k
Ω
(F)
ik
[ρ, ρ
∗
] +z
∗
k
Ω
(F)
ik
[m, m
∗
] +z
k
Ω
(F)
ik
[m
∗
, ρ
∗
]
m
(ρ)
ki
,
−i∂
t
i
ρ
(F)
ij
= (6.76)
−J(ρ
(F)
i+1j
(t
i
, t
j
) +ρ
(F)
i−1j
(t
i
, t
j
)) + 2U([z
i
[
2
+ρ
(F)
ii
)ρ
(F)
ij
+U(m
∗(F)
ii
+z
∗2
i
)m
(F)
ij
−2U
2
¸
k
t
i
0
dt
k
z
i
z
∗
k
Ω
(ρ)
ik
[ρ, ρ] + 2z
i
z
k
Ω
(ρ)
ik
[ρ, m
∗
] + 2z
∗
i
z
∗
k
Ω
(ρ)
ik
[m, ρ]
ρ
(F)
kj
−2U
2
¸
k
t
i
0
dt
k
Ω
(ρ)
ik
[ρ, ∆] + 2z
∗
i
z
k
Ω
(ρ)
ik
[ρ, ρ
∗
] + Ω
(ρ)
ik
[m, m
∗
]
ρ
(F)
kj
−2U
2
¸
k
t
i
0
dt
k
2z
i
z
∗
k
Ω
(ρ)
ik
[m
∗
, ρ] +z
i
z
k
Ω
(ρ)
ik
[m
∗
, m
∗
] + 2z
∗
i
z
k
Ω
(ρ)
ik
[ρ
∗
, m
∗
]
m
(F)
kj
−2U
2
¸
k
t
i
0
dt
k
Ω
(ρ)
ik
[m
∗
, Υ] +z
∗
i
z
∗
k
2Ω
(ρ)
ik
[ρ, ρ
∗
] + 2Ω
(ρ)
ik
[m, m
∗
]
m
(F)
kj
+2U
2
¸
k
t
j
0
dt
k
z
i
z
∗
k
Ω
(F)
ik
[ρ, ρ] + 2z
i
z
k
Ω
(F)
ik
[ρ, m
∗
] + 2z
∗
i
z
∗
k
Ω
(F)
ik
[m, ρ]
ρ
(ρ)
kj
+2U
2
¸
k
t
j
0
dt
k
Ω
(F)
ik
[ρ, ∆] + 2z
∗
i
z
k
Ω
(F)
ik
[ρ, ρ
∗
] + Ω
(F)
ik
[m, m
∗
]
ρ
(ρ)
kj
+2U
2
¸
k
t
j
0
dt
k
2z
i
z
∗
k
Ω
(F)
ik
[m
∗
, ρ] +z
i
z
k
Ω
(F)
ik
[m
∗
, m
∗
] + 2z
∗
i
z
k
Ω
(F)
ik
[ρ
∗
, m
∗
]
m
(ρ)
kj
+2U
2
¸
k
t
j
0
dt
k
Ω
(F)
ik
[m
∗
, Υ] + 2z
∗
i
z
∗
k
Ω
(F)
ik
[ρ, ρ
∗
] + Ω
(F)
ik
[m, m
∗
]
m
(F)
kj
,
6.5 Secondorder expansion 119
−i∂
t
i
ρ
(ρ)
ij
= (6.77)
−J(ρ
(ρ)
i+1j
(t
i
, t
j
) +ρ
(ρ)
i−1j
(t
i
, t
j
)) + 2U([z
i
[
2
+ρ
(F)
ii
)ρ
(ρ)
ij
+U(m
∗(F)
ii
+z
∗2
i
)m
(ρ)
ij
−2U
2
¸
k
t
i
t
j
dt
k
z
i
z
∗
k
Ω
(ρ)
ik
[ρ, ρ] + 2z
i
z
k
Ω
(ρ)
ik
[ρ, m
∗
] + 2z
∗
i
z
∗
k
Ω
(ρ)
ik
[m, ρ]
ρ
(ρ)
kj
−2U
2
¸
k
t
i
t
j
dt
k
Ω
(ρ)
ik
[ρ, ∆] + 2z
∗
i
z
k
Ω
(ρ)
ik
[ρ, ρ
∗
] + Ω
(ρ)
ik
[m, m
∗
]
ρ
(ρ)
kj
−2U
2
¸
k
t
i
t
j
dt
k
2z
i
z
∗
k
Ω
(ρ)
ik
[m
∗
, ρ] +z
i
z
k
Ω
(ρ)
ik
[m
∗
, m
∗
] + 2z
∗
i
z
k
Ω
(ρ)
ik
[ρ
∗
, m
∗
]
m
(ρ)
kj
−2U
2
¸
k
t
i
t
j
dt
k
Ω
(ρ)
ik
[m
∗
, Υ] + 2z
∗
i
z
∗
k
Ω
(ρ)
ik
[ρ, ρ
∗
] + Ω
(ρ)
ik
[m, m
∗
]
m
(ρ)
kj
,
i∂
t
i
m
(F)
ij
= (6.78)
−J(m
(F)
i+1j
(t
i
, t
j
) +m
(F)
i−1j
(t
i
, t
j
)) + 2U([z
i
[
2
+ρ
(F)
ii
)m
(F)
ij
+U(m
(F)
ii
+z
2
i
)ρ
(F)
ij
−2U
2
¸
k
t
i
0
dt
k
2z
∗
i
z
k
Ω
(ρ)
ik
[m, ρ
∗
] +z
∗
i
z
∗
k
Ω
(ρ)
ik
[m, m] + 2z
i
z
∗
k
Ω
(ρ)
ik
[ρ, m]
ρ
(F)
kj
−2U
2
¸
k
t
i
0
dt
k
Ω
(ρ)
ik
[m, Υ] + 2z
i
z
k
Ω
(ρ)
ik
[ρ, ρ
∗
] + Ω
(ρ)
ik
[m, m
∗
]
ρ
(F)
kj
−2U
2
¸
k
t
i
0
dt
k
z
∗
i
z
k
Ω
(ρ)
ik
[ρ
∗
, ρ
∗
] + 2z
∗
i
z
∗
k
Ω
(ρ)
ik
[ρ
∗
, m] + 2z
i
z
k
Ω
(ρ)
ik
[m
∗
, ρ
∗
]
m
(F)
kj
−2U
2
¸
k
t
i
0
dt
k
Ω
(ρ)
ik
[ρ
∗
, ∆] + 2z
i
z
∗
k
Ω
(ρ)
ik
[ρ, ρ
∗
] + Ω
(ρ)
ik
[m, m
∗
]
m
(F)
kj
+2U
2
¸
k
t
j
0
dt
k
2z
∗
i
z
k
Ω
(F)
ik
[m, ρ
∗
] +z
∗
i
z
∗
k
Ω
(F)
ik
[m, m] + 2z
i
z
∗
k
Ω
(F)
ik
[ρ, m]
ρ
(ρ)
kj
+2U
2
¸
k
t
j
0
dt
k
Ω
(F)
ik
[m, Υ] + 2z
i
z
k
Ω
(F)
ik
[ρ, ρ
∗
] + Ω
(F)
ik
[m, m
∗
]
ρ
(ρ)
kj
+2U
2
¸
k
t
j
0
dt
k
z
∗
i
z
k
Ω
(F)
ik
[ρ
∗
, ρ
∗
] + 2z
∗
i
z
∗
k
Ω
(F)
ik
[ρ
∗
, m] + 2z
i
z
k
Ω
(F)
ik
[m
∗
, ρ
∗
]
m
(ρ)
kj
+2U
2
¸
k
t
j
0
dt
k
Ω
(F)
ik
[ρ
∗
, ∆] + 2z
i
z
∗
k
Ω
(F)
ik
[ρ, ρ
∗
] + Ω
(F)
ik
[m, m
∗
]
m
(ρ)
kj
,
120
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
i∂
t
i
m
(ρ)
ij
= (6.79)
−J(m
(ρ)
i+1j
(t
i
, t
j
) +m
(ρ)
i−1j
(t
i
, t
j
)) + 2U([z
i
[
2
+ρ
(F)
ii
)m
(ρ)
ij
+U(m
(F)
ii
+z
2
i
)ρ
(ρ)
ij
+2U
2
¸
k
t
i
t
j
dt
k
2z
∗
i
z
k
Ω
(ρ)
ik
[m, ρ
∗
] +z
∗
i
z
∗
k
Ω
(ρ)
ik
[m, m] + 2z
i
z
∗
k
Ω
(ρ)
ik
[ρ, m]
ρ
(ρ)
kj
+2U
2
¸
k
t
i
t
j
dt
k
Ω
(ρ)
ik
[m, Υ] +z
i
z
k
2Ω
(ρ)
ik
[ρ, ρ
∗
] + 2Ω
(ρ)
ik
[m, m
∗
]
ρ
(ρ)
kj
+2U
2
¸
k
t
i
t
j
dt
k
z
∗
i
z
k
Ω
(ρ)
ik
[ρ
∗
, ρ
∗
] + 2z
∗
i
z
∗
k
Ω
(ρ)
ik
[ρ
∗
, m] + 2z
i
z
k
Ω
(ρ)
ik
[m
∗
, ρ
∗
]
m
(ρ)
kj
+2U
2
¸
k
t
i
t
j
dt
k
Ω
(ρ)
ik
[ρ
∗
, ∆] + 2z
i
z
∗
k
Ω
(ρ)
ik
[ρ, ρ
∗
] + Ω
(ρ)
ik
[m, m
∗
]
m
(ρ)
kj
,
with
∆
(F,ρ)
ij
= Ω
(F,ρ)
ij
[ρ, ρ
∗
] + 2Ω
(F,ρ)
ij
[m, m
∗
] , (6.80)
Υ
(F,ρ)
ij
= 2Ω
(F,ρ)
ij
[ρ, ρ
∗
] + Ω
(F,ρ)
ij
[m, m
∗
] . (6.81)
In the above equations we adopted the notation z
k
meaning z
k
(t
k
) for the
condensate and m
kj
and ρ
kj
meaning m
kj
(t
k
, t
j
) and ρ
kj
(t
k
, t
j
) respectively for
the two point functions. These notation is also going to be used in the equations
for the 1/^ expansion.
2PI1/^ expansion
The 2PI eﬀective action is a singlet under O(^) rotations. It can be shown that
all graphs contained in an O(^) expansion can be built from the irreducible
invariants[127]: z
2
, Tr(G
n
) and Tr(zzG
n
) , with n < ^. The factors of ^ in
a single graph contributing to the same 1/^ expansion then have two origins: a
factor of ^ from each irreducible invariant and a factor of 1/^ from each vertex.
The leading order large ^ approximation scales proportional to ^, the next to
leading order (NLO) contributions are of order 1 and so on. At leading order only
the ﬁrst term of Eq. (6.42) contributes. At the next to leading order level, if we
truncate up to secondorder in the coupling strength, the doublebubble is totally
included but only certain parts of the settingsun and basketball diagrams are: the
ﬁrst term in both of the integrals of Eq. (6.64),
Γ
(2)1/A
2
[z, G] = i
U
^
2
dt
i
dt
j
z
ib
(t)z
jb
(t
t
)
G
bb
ij
(t, t
t
)G
ijdd
(t, t
t
)G
dd
ij
(t, t
t
)
+
i
U
2^
2
dt
i
dt
t
j
G
ijbb
(t, t
t
)G
bb
ij
(t, t
t
)G
ijdd
(t, t
t
)G
dd
ij
(t, t
t
)
. (6.82)
6.5 Secondorder expansion 121
The equations of motion under this approximation are the ones obtained for the
full secondorder expansion but with Λ = Λ = 0, and Θ = Ξ.
In terms of the spectral and statistical functions the equation of motion can
be written as:
i∂
t
i
z
i
= −J(z
i+1
(t
i
) +z
i−1
(t
i
)) +U([z
i
[
2
+ 2ρ
(F)
ii
)z
i
+Um
(F)
ii
z
∗
i
(6.83)
−U
2
¸
k
t
i
0
dt
k
Π
(ρ)
ik
z
k
ρ
(F)
ki
+z
∗
k
m
(F)
ki
+U
2
¸
k
t
i
0
dt
k
Π
(F)
ik
z
k
ρ
(ρ)
ki
+z
∗
k
m
(ρ)
ki
,
−i∂
t
i
ρ
(F)
ij
= (6.84)
−J(ρ
(F)
i+1j
(t
i
, t
j
) +ρ
(F)
i−1j
(t
i
, t
j
)) + 2U([z
i
[
2
+ρ
(F)
ii
)ρ
(F)
ij
+U(m
∗(F)
ii
+z
∗2
i
)m
(F)
ij
−U
2
¸
k
t
i
0
dt
k
z
i
z
∗
k
Ω
(ρ)
ik
[ρ, ρ] +z
i
z
k
Ω
(ρ)
ik
[ρ, m
∗
] +z
∗
i
z
∗
k
Ω
(ρ)
ik
[m, ρ]
ρ
(F)
kj
−U
2
¸
k
t
i
0
dt
k
Ω
(ρ)
ik
[ρ, Π] +z
∗
i
z
k
2Ω
(ρ)
ik
[ρ, ρ
∗
] + Ω
(ρ)
ik
[m, m
∗
]
ρ
(F)
kj
−U
2
¸
k
t
i
0
dt
k
z
i
z
∗
k
Ω
(ρ)
ik
[m
∗
, ρ] +z
i
z
k
Ω
(ρ)
ik
[m
∗
, m
∗
] +z
∗
i
z
k
Ω
(ρ)
ik
[ρ
∗
, m
∗
]
m
(F)
kj
−U
2
¸
k
t
i
0
dt
k
Ω
(ρ)
ik
[m
∗
, Π] +z
∗
i
z
∗
k
Ω
(ρ)
ik
[ρ, ρ
∗
] + 2Ω
(ρ)
ik
[m, m
∗
]
m
(F)
kj
+U
2
¸
k
t
j
0
dt
k
z
i
z
∗
k
Ω
(F)
ik
[ρ, ρ] +z
i
z
k
Ω
(F)
ik
[ρ, m
∗
] +z
∗
i
z
∗
k
Ω
(F)
ik
[m, ρ]
ρ
(ρ)
kj
+U
2
¸
k
t
j
0
dt
k
Ω
(F)
ik
[ρ, Π] +z
∗
i
z
k
2Ω
(F)
ik
[ρ, ρ
∗
] + Ω
(F)
ik
[m, m
∗
]
ρ
(ρ)
kj
+U
2
¸
k
t
j
0
dt
k
z
i
z
∗
k
Ω
(F)
ik
[m
∗
, ρ] +z
i
z
k
Ω
(F)
ik
[m
∗
, m
∗
] +z
∗
i
z
k
Ω
(F)
ik
[ρ
∗
, m
∗
]
m
(ρ)
kj
+U
2
¸
k
t
j
0
dt
k
Ω
(F)
ik
[m
∗
, Π] +z
∗
i
z
∗
k
Ω
(F)
ik
[ρ, ρ
∗
] + 2Ω
(F)
ik
[m, m
∗
]
m
(F)
kj
,
122
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
−i∂
t
i
ρ
(ρ)
ij
= (6.85)
−J(ρ
(ρ)
i+1j
(t
i
, t
j
) +ρ
(ρ)
i−1j
(t
i
, t
j
)) + 2U([z
i
[
2
+ρ
(F)
ii
)ρ
(ρ)
ij
+U(m
∗(F)
ii
+z
∗2
i
)m
(ρ)
ij
−U
2
¸
k
t
i
t
j
dt
k
z
i
z
∗
k
Ω
(ρ)
ik
[ρ, ρ] +z
i
z
k
Ω
(ρ)
ik
[ρ, m
∗
] +z
∗
i
z
∗
k
Ω
(ρ)
ik
[m, ρ]
ρ
(ρ)
kj
−U
2
¸
k
t
i
t
j
dt
k
Ω
(ρ)
ik
[ρ, Π] +z
∗
i
z
k
2Ω
(ρ)
ik
[ρ, ρ
∗
] + Ω
(ρ)
ik
[m, m
∗
]
ρ
(ρ)
kj
−U
2
¸
k
t
i
t
j
dt
k
z
i
z
∗
k
Ω
(ρ)
ik
[m
∗
, ρ] +z
i
z
k
Ω
(ρ)
ik
[m
∗
, m
∗
] +z
∗
i
z
k
Ω
(ρ)
ik
[ρ
∗
, m
∗
]
m
(ρ)
kj
−U
2
¸
k
t
i
t
j
dt
k
Ω
(ρ)
ik
[m
∗
, Π] +z
∗
i
z
∗
k
Ω
(ρ)
ik
[ρ, ρ
∗
] + 2Ω
(ρ)
ik
[m, m
∗
]
m
(ρ)
kj
,
i∂
t
i
m
(F)
ij
= (6.86)
−J(m
(F)
i+1j
(t
i
, t
j
) +m
(F)
i−1j
(t
i
, t
j
)) + 2U([z
i
[
2
+ρ
(F)
ii
)m
(F)
ij
+U(m
(F)
ii
+z
2
i
)ρ
(F)
ij
−U
2
¸
k
t
i
0
dt
k
z
∗
i
z
k
Ω
(ρ)
ik
[m, ρ
∗
] +z
∗
i
z
∗
k
Ω
(ρ)
ik
[m, m] +z
i
z
∗
k
Ω
(ρ)
ik
[ρ, m]
ρ
(F)
kj
−U
2
¸
k
t
i
0
dt
k
Ω
(ρ)
ik
[m, Π] +z
i
z
k
Ω
(ρ)
ik
[ρ, ρ
∗
] + 2Ω
(ρ)
ik
[m, m
∗
]
ρ
(F)
kj
−U
2
¸
k
t
i
0
dt
k
z
∗
i
z
k
Ω
(ρ)
ik
[ρ
∗
, ρ
∗
] +z
∗
i
z
∗
k
Ω
(ρ)
ik
[ρ
∗
, m] +z
i
z
k
Ω
(ρ)
ik
[m
∗
, ρ
∗
]
m
(F)
kj
−U
2
¸
k
t
i
0
dt
k
Ω
(ρ)
ik
[ρ
∗
, Π] +z
i
z
∗
k
2Ω
(ρ)
ik
[ρ, ρ
∗
] + Ω
(ρ)
ik
[m, m
∗
]
m
(F)
kj
+U
2
¸
k
t
j
0
dt
k
z
∗
i
z
k
Ω
(F)
ik
[m, ρ
∗
] +z
∗
i
z
∗
k
Ω
(F)
ik
[m, m] +z
i
z
∗
k
Ω
(F)
ik
[ρ, m]
ρ
(ρ)
kj
+U
2
¸
k
t
j
0
dt
k
Ω
(F)
ik
[m, Π] +z
i
z
k
2Ω
(F)
ik
[ρ, ρ
∗
] + Ω
(F)
ik
[m, m
∗
]
ρ
(ρ)
kj
+U
2
¸
k
t
j
0
dt
k
z
∗
i
z
k
Ω
(F)
ik
[ρ
∗
, ρ
∗
] +z
∗
i
z
∗
k
Ω
(F)
ik
[ρ
∗
, m] +z
i
z
k
Ω
(F)
ik
[m
∗
, ρ
∗
]
m
(ρ)
kj
+U
2
¸
k
t
j
0
dt
k
Ω
(F)
ik
[ρ
∗
, Π] +z
i
z
∗
k
2Ω
(F)
ik
[ρ, ρ
∗
] + Ω
(F)
ik
[m, m
∗
]
m
(ρ)
kj
,
6.5 Secondorder expansion 123
i∂
t
i
m
(ρ)
ij
= (6.87)
−J(m
(ρ)
i+1j
(t
i
, t
j
) +m
(ρ)
i−1j
(t
i
, t
j
)) + 2U([z
i
[
2
+ρ
(F)
ii
)m
(ρ)
ij
+U(m
(F)
ii
+z
2
i
)ρ
(ρ)
ij
+U
2
¸
k
t
i
t
j
dt
k
z
∗
i
z
k
Ω
(ρ)
ik
[m, ρ
∗
] +z
∗
i
z
∗
k
Ω
(ρ)
ik
[m, m] +z
i
z
∗
k
Ω
(ρ)
ik
[ρ, m]
ρ
(ρ)
kj
+U
2
¸
k
t
i
t
j
dt
k
Ω
(ρ)
ik
[m, Π] +z
i
z
k
Ω
(ρ)
ik
[ρ, ρ
∗
] + 2Ω
(ρ)
ik
[m, m
∗
]
ρ
(ρ)
kj
+U
2
¸
k
t
i
t
j
dt
k
z
∗
i
z
k
Ω
(ρ)
ik
[ρ
∗
, ρ
∗
] +z
∗
i
z
∗
k
Ω
(ρ)
ik
[ρ
∗
, m] +z
i
z
k
Ω
(ρ)
ik
[m
∗
, ρ
∗
]
m
(ρ)
kj
+U
2
¸
k
t
i
t
j
dt
k
Ω
(ρ)
ik
[ρ
∗
, Π] +z
i
z
∗
k
2Ω
(ρ)
ik
[ρ, ρ
∗
] + Ω
(ρ)
ik
[m, m
∗
]
m
(ρ)
kj
,
with
Π
(F,ρ)
ij
= Ω
(F,ρ)
ij
[ρ, ρ
∗
] + Ω
(F,ρ)
ij
[m, m
∗
] . (6.88)
We end this section by emphasizing that the only approximation introduced
in the derivation of the equations of motion presented here is the truncation up
to secondorder in the interaction strength. These equations depict the nonlin
ear and nonMarkovian quantum dynamics, which we consider as the primary
distinguishing features of this work. It supersedes what the secondorder kinetic
theories currently presented can do, their going beyond the HFB approximation
notwithstanding. For example Ref. [62] presents a kinetic theory approach that
includes binary interactions to secondorder in the interaction potential but uses
the Markovian approximation. In Ref. [61] the authors gave a nonMarkovian gen
eralization to the quantum kinetic theory derived by Walser et. al.[58] by including
memory eﬀects. However in that work symmetry breaking ﬁelds, z and anomalous
ﬂuctuations, m, are neglected.
6.5.2 Conservation laws
For a closed (isolated) system the mean total number of particles N and energy
are conserved quantities as they are the constants of motion for the dynamical
equations.
Particle number conservation is a consequence of the invariance of the Hamil
tonian under a global phase change. The mean total number of particles is given
by
ˆ
N
=
¸
i
ˆ a
†
i
ˆ a
i
=
¸
i
[z
i
[
2
+ρ
(F)
ii
−
1
2
(6.89)
= N.
124
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
The kinetic equation for N is then
d
dt
ˆ
N(t)
=
¸
i
2Re
z
i
(t)
∂
∂t
z
∗
i
(t)
(6.90)
+ lim
t→t
∂
∂t
ρ
(F)
ii
(t, t
t
) +
∂
∂t
t
ρ
∗(F)
ii
(t
t
, t)
= 0.
All three approximations we have considered, namely, HFB, 1/^ expansion
and full secondorder expansion, conserve the particle number. This can be shown
by plugging in the kinetic equation of
ˆ
N(t)
(Eq. 6.90) the equation of mo
tion for the mean ﬁeld and the normal statistical propagator Eqs. (45),(46),
Eqs.(6.76),(6.77) and Eqs (6.85),(6.86)), and cancelling terms. It is important
to note that even though total population is always conserved there is always a
transfer of population between condensate and non condensate atoms.
While number conservation can be proved explicitly, to prove total energy
conservation is not obvious as the Hamiltonian cannot be represented as a linear
combination of the relevant operators. It is clear that the exact solution of a
closed system is unitary in time and hence disallows any dissipation. However,
the introduction of approximation schemes that truncate the inﬁnite hierarchy of
correlation functions at some ﬁnite order with causal boundary conditions may
introduce dissipation [118].
To discuss energy conservation we can use the phiderivable criteria [138] which
states that nonequilibrium approximations in which the self energy Σ is of the
form δΦ/δG, with Φ a functional of G, conserve particle number, energy and
momentum. All the approximations we consider in this paper are phi derivable
and thus they obey energy, particle number and momentum conservation laws.
For HFB, Φ = Γ
(1)
2
, for the full secondorder expansion, Φ = Γ
(1)
2
+ Γ
(2)
2
and for
the secondorder next to leading order 1/^ expansion, Φ = Γ
(1)
2
+ Γ
(2)1/A
2
. See
Eqs. (6.23), (6.42), (6.64) and (6.82). For a detailed discussion of the complete
next to leading order 1/^ expansion see Refs. [126, 127] and references therein.
6.6 Conclusions
In summary, we have presented a new approach for the description of the nonequi
librium dynamics of a BoseEinstein condensate and ﬂuctuations in a closed quan
tum ﬁeld system. The formalism allows one to go beyond the well known HFB
approximation and to incorporate the nonlinear and nonMarkovian aspects of the
quantum dynamics as manifest in the dissipation and ﬂuctuations phenomena.
The 2PI eﬀective action formalism provides a useful framework, where the mean
ﬁeld and the correlation functions are treated on the same footing selfconsistently
and which respects conservations of particle number and energy. The CTP for
malism ensures that the dynamical equations of motion are also causal. In their
6.6 Conclusions 125
current form the scattering terms are nonlocal in time and thus hard to estimate
analytically, and their calculation is numerically demanding. However, this sys
tematic approach can be used as a quantitative means to obtain solutions in diﬀer
ent regimes and make comparisons with kinetic theory results where a Markovian
approximation is assumed. We postpone these discussions to chapter 8.
126
Chapter 6 The two particle irreducible eﬀective action (2PI) and the
closed time path (CTP) formalism
Chapter 7
Nonequilibrium dynamics of a
patterned loaded optical
lattice: Beyond the mean ﬁeld
approximation
In this chapter the twoparticle irreducible (2PI) closedtimepath (CTP) eﬀective
action formalism derived in chapter 6 is used to describe the nonequilibrium dy
namics of a Bose Einstein condensate (BEC) selectively loaded into every third
site of a onedimensional optical lattice.
In chapter 4 we used a mean ﬁeld approach to describe the dynamics of this
system. However, we show here that even in the case when the kinetic energy
is comparable to the interaction energy, interatomic collisions play a crucial role
in determining the quantum dynamics of the system, and therefore a mean ﬁeld
approach is only accurate for short times. This result is demonstrated by com
parison between the mean ﬁeld solutions and the exact numerical time evolution
of the initial state using the Bose Hubbard Hamiltonian for systems with small
numbers of atoms (N ∼ 10) and lattice sites (M = 2 or 3). The exact numerical
solution is also used to test the validity of the various methods derived under the
2PICTP approximation.
We show that because the secondorder 2PI approximations include multi
particle scattering in a systematic way, they are able to capture damping eﬀects
exhibited in the exact solution, which the mean ﬁeld and the HFB approxima
tions fail to reproduce. However, our numerical results also show that all of the
approximations fail at late times, when interaction eﬀects become signiﬁcant.
7.1 Mean ﬁeld dynamics
In this section we review the basic features of the mean ﬁeld results obtained in
chapter 4, in order to provide the context for the subsequent discussion of the
128
Chapter 7 Nonequilibrium dynamics of a patterned loaded optical
lattice: Beyond the mean ﬁeld approximation
dynamics in the 2PICTP formalism.
7.1.1 Dynamical evolution
By making the mean ﬁeld ansatz in the BoseHubbard Hamiltonian, we replace
the ﬁeld operator ˆ a
i
by a cnumber z
i
(t). The amplitudes z
i
(t) satisfy the DNLSE
(Eq. (4.6)).
In section 4.3.2, we treated a model case in which the initial occupancies of each
third site are the same, and in which the condensate initially has a uniform phase.
Thus at τ = 0, (τ ≡
tJ
), the amplitudes z
i
(τ) are given by z
3i
(0) =
3/M,
z
3i+1
(0) = z
3i+2
(0) = 0, where M is the total number of lattice sites. For an
inﬁnite lattice, or one with periodic boundary conditions, the amplitudes for all
initially occupied sites z
3i
(τ) = z
0
(τ) evolve identically in time, and the amplitudes
for the initially unoccupied sites satisfy z
3i+1
(τ) = z
3i+2
(τ) = for all τ. This allows
us to reduce the full set of equations to a set of two coupled equations for z
0
(τ)
and z
1
(τ).
The solutions [z
0
(τ)[ and [z
1
(τ)[ are oscillatory functions whose amplitudes
and common period, T(γ), are determined by the parameter γ ≡ 3
NU
MJ
= 3λ. The
mean ﬁeld dynamical behavior can be qualitatively divided into two regimes:
The tunneling dominated regime (γ < 1): In this regime the oscillation
period is essentially constant, the role of interactions is relatively small, and the
equations of motion are equivalent to those of a twostate Rabi problem. This
system will undergo Rabi oscillations whereby atoms periodically tunnel from the
initially occupied site into the two neighboring sites. For γ = 0 the period of
oscillation is
2π
3
.
Interaction dominated regime: The eﬀect of interactions on the mean ﬁeld
dynamics is to cause the energies of the initially occupied sites to shift relative to
those of the unoccupied sites. As γ increases the tunneling between sites occurs
at a higher frequency, but with reduced amplitude. The population of the ini
tially occupied sites becomes eﬀectively selftrapped by the purely repulsive pair
interaction.
7.1.2 Comparisons with the exact solution
To check the validity of the mean ﬁeld approximation, we made comparisons with
the exact many body solution for N = 6 atoms and M = 3 wells. We use a
modest number of atoms and lattice sites for the comparisons, due to the fact that
the Hilbert space needed for the calculations increases rapidly with the number of
atoms and wells.
7.1 Mean ﬁeld dynamics 129
Exact solution
For the exact solution we used an initial state given by
[ϕ(0)` = (e
−N/2
e
√
Nˆ a
†
o
[0`) ⊗[0` ⊗[0` . (7.1)
The initial state represents a coherent state with an average of N atoms in the
initially populated well and zero atoms in the others. We chose this state because
in the experiment the loading of the atoms was done slow enough with respect to
band excitations but fast with respect to many body excitations that at time t = 0
most of the atoms in the initially populated wells were condensed.
The fully quantal solution was found by evolving the initial state in time with
the BoseHubbard Hamiltonian, so that [ϕ(t)` = e
−
i
ˆ
Ht
[ϕ(0)`. To do the numerical
calculations we partitioned the Hilbert space in subspaces with a ﬁxed number
of atoms and propagated independently the projections of the initial state on the
respective subspaces. A subspace with N
n
number of atoms and M wells is spanned
by
(N
n
+M−1)!
N
n
!(M−1)!
states. This procedure could be done because the Hamiltonian
commutes with the number operator
¸
i
ˆ a
i
†
ˆ a
i
, and therefore during the dynamics
the diﬀerent subspaces never get mixed. The number of subspaces used for the
numerical evolution were such that no change in plots of the dynamical observables
was detected by adding another subspace. Generally for N atoms in the initial
state, this condition was achieved by including the subspaces between N − 4
√
N
and N +4
√
N atoms.
Numerical comparisons
In Fig. 7.1 we plot the average population per well
ˆ a
†
i
(t)ˆ a
i
(t)
and the conden
sate population per well ['ˆ a
i
(t)`[
2
and compare them with the mean ﬁeld predic
tions, i.e. [z
i
(t)[
2
, for three diﬀerent values of γ. The salient features observed in
these comparisons are:
1. Weakly interacting regime (γ = 0.2):
In this regime the DNLSE gives a good description of the early time dy
namics. We observe in Fig. 7.1 that the total population per well predicted
by the mean ﬁeld solution agrees with the exact solution and also that the
condensate population remains large for the time under consideration. We
expect the semiclassical approach to be valid for time scales less than the
inverse energy level spacing. In Ref. [129] the authors show the validity of
the semiclassical approach when τ < τ
cl
∼
N
Mγ
in the case of two lattice sites.
This time scale is in good agreement with the numerical results shown in in
Fig.7.1. After τ
cl
quantum eﬀects become important.
2. Intermediate regime (γ = 2):
Quantum ﬂuctuations lead to a nontrivial modulation of the classical oscil
lations. In this regime the ratio between interaction and kinetic energy is
130
Chapter 7 Nonequilibrium dynamics of a patterned loaded optical
lattice: Beyond the mean ﬁeld approximation
small enough to allow the atoms to tunnel but not too small to make in
teraction eﬀects negligible. Mean ﬁeld results are accurate only for a short
time. In this regime, the exact solution exhibits damped oscillations of the
atomic population. Quantum scattering eﬀects are crucial, even for rather
early times.
To understand the dynamics in the weak and intermediate regimes, we have
to focus on the coherent properties of the system. Even though interactions
can be strong, the ground state is indeed superﬂuid. If we look at the initial
coherence of the system, determined by
ˆ a
†
i
(0)ˆ a
j
(0)
i,=j
, it can be seen to
be zero due to the patterned loading. However, this is no longer the case
for t > 0 , and nonzero correlations are developed in the dynamics. The
dynamical restoration of the phase coherence which tends to distribute atoms
uniformly among the lattice sites and to damp the oscillations characterizes
the dynamics in the superﬂuid regime. In Ref. [129], the authors show, not
for a patterned loaded initial state but for an initial Mott state also with zero
initial phase coherence, how the phase coherence is restored dynamically.
3. Strongly correlated regime (γ = 12):
The system exhibits macroscopic quantum selftrapping of the population.
Qualitatively, both the mean ﬁeld and the exact solutions agree, in the sense
that both predict selftrapping of atoms in the initially populated wells, due
to interactions. However, the fast decrease of the condensate population and
its subsequent revivals (as found in the exact solutions) give us an idea of the
importance of correlation eﬀects beyond mean ﬁeld. For a uniform loaded
lattice, the collapse and revivals of the condensate in this strong interacting
regime and the importance of quantum eﬀects have been experimentally
observed[47].
Even though there is minimal initial coherence between adjacent sites due
to the patterned loading procedure we are still preparing the system in a
superﬂuid state in the initially populated well. At time t = 0 we have a
condensate fraction of order one. However, the ground state of the system is
not superﬂuid. It is expected then that, after some time, the phase is going
to randomize and this will lead to the collapse of the condensate population.
After the collapse, the system will remain for a while with zero condensate
population. However, it can not remain zero forever because we are dealing
with a closed quantum system, with ﬁnite recurrence time. Therefore at
some time t
rev
we expect the condensate to revive again. The collapses and
revivals of the condensate population in the strong interacting regime can
be easily estimated by considering the energy spectrum. In this regime the
energy eigenstates of the system are almost number (Fock) states and the
energy spectrum is almost quadratic, E
n
≈ n(n−1)U/2. The dynamics of the
system is described by the interference of the diﬀerent nparticle Fock states
that span the coherent state of the initially populated well. At integer values
of t
rev
= (U/h)
−1
, the phase factors add to an integer value of 2π, leading
7.2 2PICTP approximations 131
to a revival of the initial state. This time scale agrees with the one estimated
in Ref.[130] for a more general situation. That reference, also shows how the
collapse time, t
coll
, depends on the variance of the initial atomic distribution;
it is given by: t
coll
∼ t
rev
/(2πσ). If the initial state is a coherent state, the
initial distribution is Poissonian and t
coll
is given by t
coll
∼ /(
√
NU). For
the parameters used in the strongly correlated regime, γ = 12 and N = 6,
we observe that the estimated collapses and revival times are in agreement
with what is shown in Fig. 7.1.
7.2 2PICTP approximations
It was shown in the previous section that to describe the dynamics of the patterned
loaded optical lattice, approximations beyond the standard mean ﬁeld theory are
required. In this section we proceed to test the validity of the diﬀerent 2PICTP
approximations derived in chapter 6, explicitly, the time dependent HFB approxi
mation, the full secondorder approximation and the 1/^ expansion up to second
order in the interaction strength. Because in all these approximations we use
U/J as an expansion parameter we will focus our calculations on the intermediate
regime, where the ratio U/J is small enough that truncation up to secondorder
makes sense but not too small so that interaction eﬀects still have to be taken into
account.
We start by describing the initial conditions chosen for the numerical calcula
tions, then we outline the numerical algorithms used, and ﬁnally we discuss the
results.
7.2.1 Initial conditions and parameters
To model the patterned loading, the initial conditions assumed for the numerical
solutions were z
i
(0) = Nδ
i0
, ρ
(F)
ij
(0, 0) =
1
2
δ
ij
, ρ
(ρ)
ij
(0, 0) = −iδ
ij
and m
(F)
ij
(0, 0)) =
m
(ρ)
ij
(0, 0) = 0. Here z
i
(t) are the condensate amplitudes and ρ
(F)
ij
(t
i
, t
j
), ρ
(ρ)
ij
(t
i
, t
j
),
m
(F)
ij
(t
i
, t
j
) and m
(ρ)
ij
(t
i
, t
j
) are the statistical and spectral normal and anomalous
propagators as deﬁned in chapter 6. The initial conditions correspond to an initial
coherent state with N atoms in the initially populated well.
To study the kinetic energy dominated regime we chose for the simulations
three diﬀerent sets of parameters:. The ﬁrst set is chosen to be in the very weak
interacting regime, M = 3, N = 6, J = 1 and U/J = 1/30. With this choice we
wanted to show the validity of a mean ﬁeld approach to describe this regime and
the corrections introduced by the higher order approximations. The second set of
parameters is M = 3, N = 8, J = 1 and U/J = 1/3. In this regime the kinetic
energy is big enough to allow tunneling but the eﬀect of the interactions is crucial
in the dynamics.
At the mean ﬁeld level (using the DNLSE) for a given number of wells, the only
relevant parameter for describing the dynamics of the system is the ratio UN/J.
This is not the case in the exact solution where both UN/J and N are important.
132
Chapter 7 Nonequilibrium dynamics of a patterned loaded optical
lattice: Beyond the mean ﬁeld approximation
The larger the initial population N, the larger the initial population in the coherent
matter ﬁeld, and therefore we expect better agreement of the truncated theories
with the exact solution with larger N. To study the dependence of the dynamics
on the total number of atoms, the third set of parameters in our solutions is chosen
to be M = 2, J = 1/2 and NU/J was ﬁxed to 4, but the number of atoms was
changed from 20 to 80. To increase the number of atoms in the exact calculations
we had to reduce the number of wells to two due to the fact that the dimension of
the Hilbert space scales exponentially with N and M.
7.2.2 Numerical algorithm for the approximate solution
The time evolution equations obtained in chapter 6 are nonlinear integrodiﬀerential
equations. Though the equations are very complicated, they can be solved on a
computer. The important point to note is that all equations are causal in time,
and all quantities at some later time t
f
can be obtained by integration over the
explicitly known functions for times t ≤ t
f
.
For the numerical solution we employed a time discretization t = na
t
, t
t
= ma
t
,
with n and m integers and took the advantage that, due to the presence of the
lattice, the spatial dimension is discrete (indices i and j). The discretized equations
for the time evolution of the matrices ρ
(F,ρ)
ijnm
, m
(F,ρ)
ijnm
and z
in
advance time stepwise
in the ndirection for ﬁxed m. Due to the symmetries of the matrices only half of
the (n, m) matrices have to be computed and the values ρ
ρ
ijnn
= −i, m
ρ
ijnn
= 0 are
ﬁxed for all time due to the bosonic commutation relations. As initial conditions
one speciﬁes ρ
(F,ρ)
ij00
, m
(F,ρ)
ij00
and z
i0.
To ensure that the discretized equations retain the conservation properties
present in the continuous ones one has to be very careful in the evolution of the
diagonal terms of ρ
(F)
iinn
and take the limit m →n in a proper way:
ρ
(F,ρ)
ijn+1n+1
−ρ
(F,ρ)
ijnn
= (7.2)
ρ
(F,ρ)
ijn+1n
−ρ
(F,ρ)
ijnn
±
ρ
∗(F,ρ)
jin+1n
−ρ
∗(F,ρ)
jinn
,
m
(F,ρ)
ijn+1n+1
−m
(F,ρ)
ijnn
= (7.3)
m
(F,ρ)
ijn+1n
−m
(F,ρ)
ijnn
±
m
(F,ρ)
jin+1n
−m
(F,ρ)
jinn
,
with the positive sign for the statistical propagators, (F)
t
s, and negative sign for
the spectral functions, (ρ)
t
s. We used the fourth order RungeKutta algorithm to
propagate the local part of the equations and a regular one step Euler method
to iterate the non local parts. For the integrals we used the standard trapezoidal
rule. Starting with n = 1, for the time step n + 1 one computes successively all
entries with m = 0......, n, n+1 from known functions evaluated at previous times.
The time step a
t
was chosen small enough so that convergence was observed,
that is, further decreasing it did not change the results. The greater the parameter
UN/J, the smaller is the time step required. The main numerical limitation of
7.2 2PICTP approximations 133
the 2PI approximation is set by the time integrals, which make the numerical
calculations time and memory consuming. However, within a typical numerical
precision it was usually not necessary to keep all past values of the two point
functions in the memory. A characteristic time, after which the inﬂuence of the
early time in the late time behavior is negligible, is given by the inverse damping
rate. This time is described by the exponential damping of the twopoint correlator
at time t with the initial time[127]. In our numerics we extended the length of the
employed time interval until the results did not depend on it. In general, it was less
than the inverse damping rate. We used for the calculations a single PII 400 MHz
workstation with 260 Mb of memory. For a typical run 12 days of computational
time were required.
7.2.3 Results and discussions
In Figs. 7.2 to 7.6 we show our numerical results. We focus our attention on the
evolution of the condensate population per well, [z
i
(t)[
2
, the total atomic popu
lation per well, [z
i
[
2
+ ρ
(F)
ii
(t, t) −
1
2
, the depletion per well or atoms out of the
condensate, ρ
(F)
ii
(t, t) −
1
2
, and the total condensate population,
¸
i
[z
i
(t)[
2
. The
total population is also explicitly shown in the ﬁgures to emphasize number con
servation.
The quasimomentum distribution of the atoms released from the lattice is
important because it is one of the most easily accessible quantities from an ex
periment. (see sec.5.5) The quasimomentum distribution function n
k
is deﬁned
as
n
k
(t) =
1
M
¸
i,j
e
ik(i−j)
a
†
i
(t)a
j
(t)
, (7.4)
where the quasimomentum k can assume discrete values which are integral mul
tiples of
2π
Ma
, with M the total number of lattice sites and a the lattice spacing.
The basic features of the plots can be summarized as follows:
The very weakly interacting regime: In Fig. 7.2 the dynamics of the atomic
population per well resembles the Rabi oscillation phenomenon. Notice that even
though there are three wells, periodic boundary conditions enforce equal evolution
of the initially empty ones. In this regime damping eﬀects remain very small for the
time depicted in the plots. The numerical simulations show a general agreement
between the diﬀerent approaches and the exact solution. The eﬀect of including
higher order terms in the equations of motion is to introduce small corrections
which improve the agreement with the exact dynamics. This shows up in the plots
of the condensate population and depletion, where the small diﬀerences can be
better appreciated. The secondorder 1/^ expansion gives an improvement over
the HFB and the complete secondorder perturbative expansion almost matches
the exact solution perfectly. In the duration depicted in the plots of Fig. 7.2 the
total condensate constitutes an important fraction of the total population. Re
garding the quasimomentum distribution, we observe that similarly to the spatial
134
Chapter 7 Nonequilibrium dynamics of a patterned loaded optical
lattice: Beyond the mean ﬁeld approximation
distribution where the initial conﬁguration and periodic boundary conditions re
duce the three well system to a double well one, they enforce equal evolution of
the ±
2π
3
quasimomentum intensities. The k = 0 and ±
2π
3
intensities oscillate with
the same frequency as the atomic population per well, both are also well described
by the approximations in consideration.
The intermediate regime: We can see the eﬀect of the interactions in the
dynamics. They modulate the oscillations in the population per well and scatter
the atoms out of the condensate.
1. In Figs. 7.3 we plot the numerical solution for the parameters M = 3, N =
8, J = 1 and U/J = 1/3. In contrast to the case of the very weak interacting
regime, it is only at early times that any of these approximations is close to
the exact solution. Even though none of them are satisfactory after the ﬁrst
oscillation the HFB approximation is the only one that fails to capture the
exponential decrease of the condensate population. This is expected, because
even though this approximation goes beyond mean ﬁeld theory and takes
into account the most important scattering eﬀects, it includes the eﬀects
of collisions only indirectly through energy shifts, and breaks down outside
the collisionless regime where multiplescattering eﬀects are important. In
contrast, the exponential decay of the condensate is present in the second
order approximations. Non local parts of the selfenergy included in them
encode scattering eﬀects responsible for damping. It is important to point out
that, even though we observe the collapse of the condensate population, the
total population is always conserved: as the condensate population decreases,
the noncondensate population increases.
2. Comparing the two secondorder approaches we observe that the full second
order expansion gives a better description of the dynamics than the 1/^
solution only in the regime when the perturbative solutions are close to the
exact dynamics. As soon as the third order terms start to be important the
large 1/^ expansion gives a better qualitative description. This behavior is
better appreciated in ﬁgs. 7.4 to 7.6 as the number of atoms is increased (see
discussion below).
We observe as a general issue in this regime that, regardless of the fact that
the secondorder solutions capture the damping eﬀects, as soon as the con
densate population decreases to a small percentage of the total population,
they depart from the exact dynamics: the secondorder approaches predict
faster damping rates. The overdamping is more severe in the dynamics of
the population per well than in the condensate dynamics. The failure can be
understood under the following lines of reasoning. At zero temperature con
densate atoms represent the most ”classical” form of a matter wave. When
they decay, the role of quantum correlations becomes more important. At
this point the higher order terms neglected in the secondorder approxima
tions are the ones that lead the dynamical behavior. Thus, to have a more
7.2 2PICTP approximations 135
accurate description of the dynamics after the coherent matter ﬁeld has de
cayed one needs a better treatment of correlations.
Damping eﬀects are also quite noticeable in the quantum evolution of the
quasimomentum intensities. Similarly to what happens to the spatial ob
servables, the HFB approximation fails completely to capture the damping
eﬀects present in the evolution of the Fourier intensities whereas the second
order approaches overestimate them.
3. In Figs. 7.4 to 7.6 we explore the eﬀect of the total number of atoms on the
dynamics. In the plots we show the numerical solutions found for a double
well system with ﬁx ratio UN/J = 4 and three diﬀerent values of N : N =
20, 40 and 80. We present the results obtained for the evolution of the atomic
population per well in Fig. 7.4, the condensate population per well and total
condensate population in Fig. 7.5 and the quasimomentum intensities in
Fig. 7.6. To make the comparisons easier we scaled the numerical results
obtained for the three diﬀerent values of N by dividing them by the total
number of atoms. In this way for all the cases we start with an atomic
population of magnitude one in the initially populated well.
In the exact dynamics we see that as the number of atoms is increased the
damping eﬀects occur at slower rates. This feature can be noticed in the
quantum dynamics of all of the observables depicted in the plots 7.4 to 7.6.
The decrease of the damping rates as the number of atoms is increased is not
surprising because by changing the number of atoms we aﬀect the quantum
coherence properties of the system. As discussed above the collapse time
of the condensate population is approximately given by t
coll
∼
t
rev
2π
√
N
. The
revival time is proportional to U
−1
and varies with N for ﬁxed UN/J as
t
rev
∝
N
J
, thus t
coll
∝
√
N increases with N as observed in the numerical
calculations. Besides damping rates, the qualitative behavior of the exact
quantum dynamics is not much aﬀected as the number of atoms is increased
for a ﬁxed UN/J.
The improvement of the 2PI approximations as N is increased, as a result of
the increase in the initial number of coherent atoms is in fact observed in the
plots. Even though the problem of underdamping in the HFB approximation
and the overdamping in the secondorder approaches are not cured, as the
number of atoms is increased, we do observe a better matching with the full
quantal solution. The 1/^ expansion shows the fastest convergence. Perhaps
this issue can be more easily observed in the quasimomentum distribution
plots, Fig. 7.6. The better agreement of the 1/^ expansion relies on the fact
that even though the number of ﬁelds is only two in our calculations the 1/^
expansion is an expansion about a strong quasiclassical ﬁeld conﬁguration.
136
Chapter 7 Nonequilibrium dynamics of a patterned loaded optical
lattice: Beyond the mean ﬁeld approximation
7.3 Conclusions
In this work we have used the CTP functional formalism for 2PI Green’s func
tions to describe the nonequilibrium dynamics of a condensate loaded in an optical
lattice on every third lattice site. We have carried out the analysis up to second
order in the interaction strength.This approximation is introduced so as to make
the numerical solution manageable, but it is suﬃcient to account for dissipative
eﬀects due to multiparticle scattering that are crucial even at early times. Our
formulation is capable of capturing the salient features of the system dynamics in
the regime under consideration, such as the decay of the condensate population
and the damping of the oscillations of the quasimomentum and population per
well unaccounted for in the HFB approximation. However, at the point where
an important fraction of the condensate population has been scattered out, the
secondorder approximations used here predict an overdamped dynamics. To im
prove on this a better treatment of higher correlations is required. One might try
to include the full next to leading order large ^ expansion without the truncation
to secondorder as done in Ref. [127] but it is not obvious that this will lead to the
required improvement. Alternatively, one may try to adopt a stochastic approach,
but the challenge will be shifted to the derivation of a noise term (which is likely
to be both colored and multiplicative) which contains the eﬀects of these higher
correlations and the solution of the stochastic equations.
Even though, as is clear that the secondorder 2PI approximations fail to cap
ture the fully correlated dynamics in the system, it has been proved to work at
intermediate times when correlations are not negligible and standard mean ﬁeld
techniques fail poorly. Because of their success in describing moderately correlated
regimes, the secondorder approximations could become a useful tool for describing
other experimental situations.
7.3 Conclusions 137
γ = 0.2
0 2 4 6 8 10
t
0
2
4
6
#
o
f
a
t
o
m
s
γ = 2
0
2
4
6
#
o
f
a
t
o
m
s
Conden.
DNLSE
Popul.
γ = 12
0
2
4
6
#
o
f
a
t
o
m
s
Figure 7.1: Comparisons between the exact and the DNLSE solutions for six atoms
and three wells. The time is given in units of /J. Top panel, strongly correlated
regime (γ = 12); middle panel, intermediate regime (γ = 2); bottom panel, weakly
interacting regime (γ = 0.2). The solid line is the DNLSE prediction for the
population per well: [z
0
(t)[
2
and [z
1,2
(t)[
2
, the triangles are used to represent
the exact solution for the population per well calculated using the Bose Hubbard
Hamiltonian: 'ˆ a
†
0
ˆ a
0
`, 'ˆ a
†
1,2
ˆ a
1,2
`. The pentagons show the condensate population
per well calculated from the exact solution: ['ˆ a
0
`[
2
and ['ˆ a
1,2
`[
2
. Due to the
symmetry of the initial periodic conditions the curves for the i = 1 and 2 wells are
the same in all depicted curves .
138
Chapter 7 Nonequilibrium dynamics of a patterned loaded optical
lattice: Beyond the mean ﬁeld approximation
0 2 4 6 8 10
t
0
0.2
0.4
0.6
0.8
D
e
p
l
e
t
i
o
n
/
w
e
l
l
0 2 4 6 8 10
t
1.9
1.95
2
2.05
m
o
m
e
n
t
u
m
0
2
4
6
P
o
p
u
l
a
t
i
o
n
/
w
e
l
l
0
2
4
6
c
o
n
d
e
n
s
a
t
e
/
w
e
l
l
5
5.5
6
T
o
t
a
l
C
o
n
d
e
n
s
a
t
e
HFB
1/N
2nd
Exact
5
5.25
5.5
5.75
6
T
o
t
a
l
N
u
m
b
e
r
Figure 7.2: Comparisons between the exact solution(solid line), the HFB approxi
mation (boxes), the secondorder large ^ approximation (pentagons) and the full
2PI secondorder approximation(crosses) for the very weak interacting regime. The
parameters used were M = 3, N = 6, J = 1 and U/J = 1/30. The time is given
in units of /J. In the plots where the population, condensate and depletion per
well are depicted the top curves correspond to the initially populated well solutions
and the lower to the initially empty wells. Notice the diﬀerent scale used in the
depletion plot. In the momentum distribution plot the upper curve corresponds
to the k = ±2π/3 intensities and the lower one to the k = 0 quasimomentum
intensity.
7.3 Conclusions 139
k = 0
0 2 4 6 8 10
t
0
1
2
3
q

m
o
m
e
n
t
u
m
k = q/3
0 2 4 6 8 10
t
2.5
3
3.5
2 nd Population
0
1
2
3
#
o
f
a
t
o
m
s
2 nd Condensate
0 2 4 6 8 10
t
2 nd Depletion
0
1
2
3
1 st Population
0
2
4
6
8
#
o
f
a
t
o
m
s
1 st Condensate
1 st Depletion
0
2
4
6
8
Total Number
0
2
4
6
8
#
o
f
a
t
o
m
s
Total Condensate
HFB
1N
2nd
Exact
Figure 7.3: Comparisons for the case M = 3, N = 8, J = 1 and U/J = 1/3.
The time is given in units of /J. In the plots the abbreviation 1st is used for the
initially occupied well and 2nd for the initially empty wells. In the quasimomentum
plots k = 2π/a is the reciprocal lattice vector with a the lattice spacing.
140
Chapter 7 Nonequilibrium dynamics of a patterned loaded optical
lattice: Beyond the mean ﬁeld approximation
P2, N20
0 2 4 6 8 10
t
0
0.2
0.4
0.6
0.8
1
(
#
a
t
o
m
s
)
/
N
P2, N40
0 2 4 6 8 10
t
P2, N80
0 2 4 6 8 10
t
P1, N20
0
0.2
0.4
0.6
0.8
1
(
#
a
t
o
m
s
)
/
N
P1, N40 P1, N80
HFB
1/N
2nd
Exact
Figure 7.4: Comparison between the evolution of the atomic population per well
for M = 2, J = 1/2, NU/J = 4 and N = 20, 40 and 80. Time is in units of /J. In
the plots P1 stands for the fractional atomic population in the initially populated
wells and P2 for the population in the initially empty wells. The number of atoms
N is explicitly shown in each panel.
7.3 Conclusions 141
CT, N20
0 2 4 6 8 10
t
0
0.2
0.4
0.6
0.8
1
(
#
a
t
o
m
s
)
/
N
CT, N40
0 2 4 6 8 10
t
CT, N80
0 2 4 6 8 10
t
C2, N20
0
0.2
0.4
0.6
0.8
1
(
#
a
t
o
m
s
)
/
N
C2, N40 C2, N80
C1, N20
0
0.2
0.4
0.6
0.8
1
(
#
a
t
o
m
s
)
/
N
C1, N40
C1, N80
HFB
1/N
2nd
Exact
Figure 7.5: Time evolution of the condensate population per well and the total
condensate population,for the same parameters as Fig. 7.4. Time is in units of
/J. In the plots C1 stands for the fractional condensate population in the initially
populated well, C2 for the fractional condensate population in the initially empty
one and CT for the total condensate fraction.
142
Chapter 7 Nonequilibrium dynamics of a patterned loaded optical
lattice: Beyond the mean ﬁeld approximation
ko, N20
0 2 4 6 8 10
t
0.2
0.3
0.4
0.5
0.6
0.7
q

m
o
m
e
n
t
u
m ko, N40
0 2 4 6 8 10
t
ko, N80
0 2 4 6 8 10
t
k1, N20
0.3
0.4
0.5
0.6
0.7
0.8
q

m
o
m
e
n
t
u
m
k1, N40 k1, N80
HFB
1N
2nd
Exact
Figure 7.6: Dynamical evolution of the quasimomentum intensities. The param
eters used were M = 2, J = 1/2, NU/J = 4 and N = 20, 40 and 80. Time is
in units of /J. In the plots ko denotes the k = 0 quasimomentum component
and k1 the k = π/a one (a the lattice spacing). The plots are scaled to set the
integrated quasimomentum density to one for all N .
Chapter 8
From the 2PICTP
approximations to kinetic
theories and local equilibrium
solutions
In chapter 6 we used the CTP functional formalism for 2PI Green’s functions
to derive dynamical equations of motion. We have carried out the analysis up
to secondorder in the interaction strength. In chapter 7 we used the formalism
to investigate the nonequilibrium dynamics of a condensate loaded on every third
site of an optical lattice. We showed that the formalism allowed us to go beyond
the HFB approximation and to incorporate the nonlinear and nonMarkovian as
pects of the quantum dynamics as manifested in the dissipation and ﬂuctuation
phenomena.
However, in the present form the equations of motion are complicated nonlocal
nonlinear equations from which we can hardly get any physical information regard
ing the system behavior except by numerical solution. Moreover, the numerical
solutions are complicated enough that application to real system with many lattice
sites are presently beyond the scope of standard computation capabilities. Never
theless, when the perturbation induces disturbances in the system of wave length
longer than thermal wave lengths and frequencies much smaller than characteristic
particle energies then the system is in a regime where standard kinetic theories
give a good description of the dynamics. The purpose of this chapter is then to
simplify the complicated 2PICTP equations and show that they in fact reproduce,
in the slowly varying regime, standard kinetic theories and equilibrium solutions
for weakly interacting gases well known in the literature since the late 1950’s.
144
Chapter 8 From the 2PICTP approximations to kinetic theories and
local equilibrium solutions
8.1 Rewriting the 2PICTP secondorder equations
In order to make the comparisons with standard approaches we will start by getting
rid of the matrix indices that were useful to derive the equations of motion but
which complicate the notation. We deﬁne the quantities
z(t
i
) ≡ z
a
i
(t), (8.1)
G(t
i
, t
t
j
) ≡ −iG
ij
a
b
(t, t
t
), (8.2)
H(t
i
, t
t
j
) ≡ −iz
a
i
(t)z
j
b
(t
t
), (8.3)
G
>
(t
i
, t
t
j
) ≡ −iG
>
ij
a
c
(t, t
t
), (8.4)
G
<
(t
i
, t
t
j
) ≡ −iG
<
ij
a
c
(t, t
t
) = −iG
>
ji
c
a
(t
t
, t). (8.5)
Here we used the notation introduced in Eqs. (6.26) and (6.27). With these deﬁni
tions we now explicitly separate the single particle, the HFB and the secondorder
contributions in the equations of motion for the condensate and the propagators
(Eq. (6.22) and Eq. (6.24)).
¸
k
dt
tt
D
−1
o
(t
i
, t
tt
k
) −S
HFB
(t
i
, t
tt
k
)
z (t
tt
k
) =
¸
k
dt
tt
S(t
i
, t
tt
k
)z(t
tt
k
),(8.6)
¸
k
dt
tt
D
−1
o
(t
i
, t
tt
k
) −Σ
HFB
(t
i
, t
tt
k
)
G(t
tt
k
, t
t
j
) =
¸
k
dt
tt
Σ(t
i
, t
tt
k
)G(t
tt
k
, t
t
j
)
−δ
ij
δ
C
(t −t
t
), (8.7)
where D
−1
o
(t
i
, t
t
j
) is the inverse free particle propagator given by:
D
−1
o
(t
i
, t
t
j
) ≡ (iδ
ij
σ
z
∂
t
+J(δ
i+1j
+δ
i−1j
) −δ
ij
V
i
) δ(t −t
t
), (8.8)
with σ being the Pauli matrices:
σ
z
=
1 0
0 −1
, σ
x
=
0 1
1 0
. (8.9)
In D
−1
o
(t
i
, t
t
j
) we have allowed the presence of an external potential V
i
. The label
HFB stands for the HFB contribution. Using the deﬁnition (8.1)(8.5) in Eqs.
(6.43) and (6.45) they can then be rewritten as:
Σ
HFB
(t
i
, t
t
j
) ≡ (8.10)
i
U
^
Tr
H(t
i
, t
t
j
) +G(t
i
, t
t
j
)
I +2
H(t
i
, t
t
j
) +G(t
i
, t
t
j
)
δ(t −t
t
)δ
ij,
,
S
HFB
(t
i
, t
t
j
) ≡ i
U
^
Tr
H(t
i
, t
t
j
)I +G(t
i
, t
t
j
)
+ 2G(t
i
, t
t
j
)
δ(t −t
t
)δ
ij
. (8.11)
8.1 Rewriting the 2PICTP secondorder equations 145
where I is the identity matrix. To evaluate the secondorder contribution, we use
Eqs. (6.31) and (6.32) but instead of setting the initial time to zero we choose it to
be −∞ and in this way we recover the standard Kadanoﬀ and Baym integration
contours [137]:
¸
k
∞
−∞
dt
tt
D
−1
o
(t
i
, t
tt
k
) −S
HFB
(t
i
, t
tt
k
)
H(t
tt
k
, t
t
j
) −
t
−∞
dt
tt
γ(t
i
, t
tt
k
)H(t
tt
k
, t
t
j
)
= 0, (8.12)
¸
k
∞
−∞
dt
tt
H(t
i
, t
tt
k
)
D
−1
o
(t
tt
k
, t
t
j
) −S
HFB
(t
tt
k
, t
t
j
)
+
t
−∞
dt
tt
H(t
i
, t
tt
k
)γ(t
tt
k
, t
t
j
)
= 0, (8.13)
¸
k
∞
−∞
dt
tt
D
−1
o
(t
i
, t
tt
k
) −Σ
HFB
(t
i
, t
tt
k
)
G
(≷)
(t
tt
k
, t
t
j
) = (8.14)
¸
k
t
−∞
dt
tt
Γ(t
i
, t
tt
k
)G
(≷)
(t
tt
k
, t
t
j
) −
¸
k
t
−∞
dt
tt
Σ
(≷)
(t
i
, t
tt
k
)A(t
tt
k
, t
t
j
),
¸
k
∞
−∞
dt
tt
G
(≷)
(t
i
, t
tt
k
)
D
−1
o
(t
tt
k
, t
t
j
) −Σ
HFB
(t
tt
k
, t
t
j
)
= (8.15)
¸
k
t
−∞
dt
tt
A(t
i
, t
tt
k
)Σ
(≷)
(t
tt
k
, t
t
j
) −
¸
k
t
−∞
dt
tt
G
(≷)
(t
i
, t
tt
k
)Γ(t
tt
k
, t
t
j
).
In the above equations, Eq. (8.15) is the hermitian conjugate of Eq. (8.14), Eq.
(8.13) is the hermitian conjugate of Eq. (8.12) and we have introduced the spectral
functions
γ(t
i
, t
t
j
) ≡ (S
>
(t
i
, t
t
j
) −S
<
(t
i
, t
t
j
)), (8.16)
Γ(t
i
, t
tt
j
) ≡ (Σ
>
(t
i
, t
tt
j
) −Σ
<
(t
i
, t
tt
j
)), (8.17)
A(t
i
, t
tt
j
) ≡ (G
>
(t
i
, t
tt
j
) −G
<
(t
i
, t
tt
j
)). (8.18)
Notice that A(t
i
, t
tt
j
) is just the spectral function deﬁned in Eq.(6.34) multiplied
by a minus sign, however, we adopted the notation γ, Γ and A to be consistent
with standard Kadanoﬀ and Baym notation [137].
If we use the full secondorder expansion, Eqs.(6.71 and 6.72), it can be shown
that the matrices S
(≷)
and Σ
(≷)
are given by
S
(≷)
(t
i
, t
t
j
) ≡ −
1
2
2U
^
2
G
(≷)
(t
i
, t
t
j
)Tr
G
(≷)
(t
i
, t
t
j
)G
(≶)
(t
t
j
, t
i
)
+ 2G
(≷)
(t
i
, t
t
j
)G
(≶)
(t
t
j
, t
i
)G
(≷)
(t
i
, t
t
j
)
, (8.19)
146
Chapter 8 From the 2PICTP approximations to kinetic theories and
local equilibrium solutions
Σ
(≷)
(t
i
, t
t
j
) ≡ −
1
2
2U
^
2
(8.20)
H(t
i
, t
t
j
)Tr
G
(≷)
(t
i
, t
t
j
)G
(≶)
ik
(t
t
j
, t
i
)
+ 2H(t
i
, t
t
j
)G
(≶)
(t
t
j
, t
i
)G
(≷)
(t
i
, t
t
j
)
+2G
(≷)
(t
i
, t
t
j
)
H(t
t
j
, t
i
)G
(≷)
(t
i
, t
t
j
) +G
(≶)
(t
t
j
, t
i
)H(t
i
, t
t
j
) +G
(≶)
(t
t
j
, t
i
)G
(≷)
(t
i
, t
t
j
)
G
(≷)
(t
i
, t
t
j
)Tr
H(t
t
j
, t
i
)G
(≷)
(t
i
, t
t
j
) +G
(≶)
(t
t
j
, t
i
)H(t
i
, t
t
j
) +G
(≶)
(t
t
j
, t
i
)G
(≷)
(t
i
, t
t
j
)
¸
.
It is convenient to decompose the above equations in their matrix components.
To do that we introduce the deﬁnitions
G
>
(t
i
, t
t
j
) = −i
¯ ρ
ij
m
ij
m
∗
ji
ρ
ji
, (8.21)
G
<
(t
i
, t
t
j
) = −i
ρ
ij
m
ji
m
∗
ij
¯ ρ
ji
. (8.22)
Using Eqs. (8.21) and (8.22) into the selfenergy equations we get
Σ
HFB
(t
i
, t
t
j
) =
2U
^
2[z
i
[
2
+ρ
ii
+ ¯ ρ
ii
z
2
i
+m
ii
z
∗2
i
+m
∗
ii
2[z
i
[
2
+ρ
ii
+ ¯ ρ
ii
δ(t −t
t
)δ
ij
, (8.23)
S
HFB
(t
i
, t
t
j
) =
2U
^
[z
i
[
2
+ρ
ii
+ ¯ ρ
ii
m
ii
m
∗
ii
[z
i
[
2
+ρ
ii
+ ¯ ρ
ii
δ(t −t
t
)δ
ij
, (8.24)
S
>
11
(t
i
, t
t
j
) =
−i8U
2
^
2
¯ ρ
ij
(2m
ij
m
∗
ji
+ ¯ ρ
ij
ρ
ji
), (8.25)
S
>
12
(t
i
, t
t
j
) =
−i8U
2
^
2
m
ij
(m
ij
m
∗
ji
+ 2¯ ρ
ij
ρ
ji
), (8.26)
S
<
11
(t
i
, t
t
j
) =
−i8U
2
^
2
ρ
ij
(2m
ji
m
∗
ij
+ ¯ ρ
ji
ρ
ij
), (8.27)
S
<
12
(t
i
, t
t
j
) =
−i8U
2
^
2
m
ji
(m
ij
m
∗
ji
+ 2¯ ρ
ji
ρ
ij
), (8.28)
8.2 Boltzmann equations 147
Σ
>
11
(t
i
, t
t
j
) =
−i8U
2
^
2
(ρ
ji
¯ ρ
2
ij
+ 2m
ij
¯ ρ
ij
m
ji
∗
+2¯ ρ
ij
m
ji
∗
z
i
z
j
+ (8.29)
¯ ρ
2
ij
z
j
z
i
∗
+2ρ
ji
¯ ρ
ij
z
i
z
j
∗
+ 2m
ij
m
ji
∗
z
j
∗
z
i
+2m
ij
¯ ρ
ij
z
i
∗
z
j
∗
),
Σ
>
12
(t
i
, t
t
j
) =
−i8U
2
^
2
(2ρ
ji
m
ij
¯ ρ
ij
+ 2ρ
ji
¯ ρ
ij
z
i
z
j
+m
ij
2
m
ji
∗
+ (8.30)
2m
ij
z
i
z
j
m
ji
∗
+ 2m
ij
¯ ρ
ij
z
j
z
i
∗
+ 2ρ
ji
m
ij
z
i
z
j
∗
+m
ij
2
z
i
∗
z
j
∗
),
Σ
<
11
(t
i
, t
t
j
) =
−i8U
2
^
2
(ρ
ij
2
¯ ρ
ji
+ 2ρ
ij
m
ji
m
ij
∗
+2ρ
ij
z
i
z
j
m
ij
∗
+ρ
2
ij
z
j
z
i
∗
+
2ρ
ij
¯ ρ
ji
z
i
z
j
∗
+ 2m
ji
z
i
m
ij
∗
z
j
∗
+ 2ρ
ij
m
ji
z
i
∗
z
j
∗
), (8.31)
Σ
<
12
(t
i
, t
t
j
) =
−i8U
2
^
2
(2ρ
ij
m
ji
¯ ρ
ji
+ 2ρ
ij
¯ ρ
ji
z
i
z
j
+m
ji
2
m
ij
∗
+ (8.32)
2m
ji
z
i
z
j
m
ij
∗
+ 2ρ
ij
m
ji
z
j
z
i
∗
+ 2m
ji
¯ ρ
ji
z
i
z
j
∗
+m
ji
2
z
i
∗
z
j
∗
),
and
S
(≷)
22
(t
i
, t
t
j
) = S
(≷)
11
(t
i
, t
t
j
) ¦ρ
ij
→ ¯ ρ
ij
¦ , (8.33)
S
(≷)
21
(t
i
, t
t
j
) = S
(≷)
12
(t
i
, t
t
j
) ¦m
ji
→m
ij
∗
¦ , (8.34)
Σ
(≷)
22
(t
i
, t
t
j
) = Σ
(≷)
11
(t
i
, t
t
j
) ¦z
i
→z
j
, ρ
ij
→ ¯ ρ
ij
¦ , (8.35)
Σ
(≷)
21
(t
i
, t
t
j
) = Σ
(≷)
12
(t
i
, t
t
j
) ¦z
i
→z
j
∗
, m
ji
→m
ij
∗
¦ . (8.36)
The above expressions for the selfenergy exactly agree with the ones used in Refs.
[59, 60, 62]. In Refs. [59, 62] the authors used these equations as their starting
point before applying the Markovian approximation.
8.2 Boltzmann equations
From previous sections it can be observed that the equations of motion at second
order are quite involved: they are nonlinear and nonlocal integro diﬀerential equa
tions, not readily solvable in closed form. To progress further we need to introduce
approximations based on physical considerations. One of them is to recognize two
time scales in the system, one related to quantum processes (microscopic) which
determines the degree of quantummechanical entanglement of the system and one
related to the statistical and kinetic behavior (macroscopic) determined by the
range of the interactions among particles. In a classical point of view the two
diﬀerent scales in the system can be understood as the time (or length) scale sep
aration between the duration of a collision event (or scattering length) and the
inverse collision rate (or the mean free path). Close to equilibrium (when the ex
ternal potential induces disturbances much longer than the lattice spacing and
frequencies much smaller than the characteristic particle energies) and in the
weakly interacting regime, we expect that a reasonable assumption would be to
consider the kinetic scale larger than the quantum one, which in a classical picture
corresponds to assume that be the time between collisions (or mean free path) is
long compared to the reaction time (or scattering length).
148
Chapter 8 From the 2PICTP approximations to kinetic theories and
local equilibrium solutions
Using this approximation it is possible to recast the quantum dynamics into the
much simpler forms of kinetic theory. In contrast to normal systems were there is
not a condensate, in condensed systems making the scale separation requires ﬁrst
to do a gauge transformation which makes it easier to identify (and coarsegrain
away) the fast variations induced by the rapid change of the condensate phase.
Following Ref. [139] we introduce the gauge transformation
z
i
(t) = e
iθ
i
(t)
n
o
(t
i
), (8.37)
G
(≷)
(t
i
, t
t
j
) = e
iθ
i
(t)σ
z ˜
G
(≷)
(t
i
, t
t
j
)e
−iθ
j
(t
)σ
z
, (8.38)
where
n
o
(t
i
) and θ
i
(t) are real. In equilibrium, ∆θ
i
≡ (θ
i+1/2
− θ
i−1/2
)/a is
related to the superﬂuid velocity and the time derivative of the phase to the
chemical potential. Extending these identiﬁcations to the nonequilibrium system
we deﬁne the chemical potential and superﬂuid velocity as
µ
i
= −∂
t
θ
i
−
2
4Ja
2
v
s
i
2
−V
i
, (8.39)
v
s
i
= 2J∆θ
i
a
2
. (8.40)
After the gauge transformation the variables n
o
,
˜
G , v
s
and µ are expected to be
slowly varying functions of
R = (i +j)/2, T = (t +t
t
)/2, (8.41)
and peaked about the zeros of
r = (i −j), τ = (t −t
t
), (8.42)
necessary conditions to derive Boltzmanntype kinetic equations.
The equations of motion are invariant under the phase transformation if we
replace D
−1
o
by
˜
D
−1
o
:
˜
D
−1
o
(t
i
, t
t
j
) = (δ
ij
(iσ
z
∂
t
−∂
t
θ
i
) −δ
ij
V
i
+ (8.43)
J(e
iσ
z
∆θ
i+1/2
δ
i+1j
+e
−iσ
z
∆θ
i−1/2
δ
i−1j
)
δ(t −t
t
).
Since hereafter we will use the gaugetransformed functions exclusively, in the
following the tildes will be dropped to simplify the notation. To obtain the kinetic
equations we Fourier transform the functions G and n
o
with respect to r and τ .
The Fourier transform reads
G
(≷)
(t
i
, t
t
j
) = G
(≷)
T +
τ
2
R+r/2
,
T −
τ
2
R−r/2
≡ −i
1
2πM
¸
q
dωe
(iqr−iωτ)
G
(≷)
(Rq; T, ω), (8.44)
H(t
i
, t
t
j
) = H
T +
τ
2
R+r/2
,
T −
τ
2
R−r/2
≡ −i
1
2πM
¸
q
dωe
(iqr−iωτ)
H(Rq; T, ω), (8.45)
8.2 Boltzmann equations 149
Neglecting secondorder variations of n
o
(Rq; T, ω) we can approximate it as
H(Rq; T, ω) = 2πM (I +σ
x
) n
o
(R, T)δ(ω)δ
q0
. (8.46)
In Eq.(8.46), the quantity n
o
(R, T) is just related to the condensate density of
atoms at the space time point (Ra, T). In Eq.(8.44), the twopoint function
G
(<)
11
(Rq; T, ω) corresponds to the well known Wigner distribution function [140]. It
can be interpreted as the density of noncondensed particles with quasimomentum q
and energy ω at the lattice site R and time T. On the other hand, G
(>)
11
(Rq; T, ω)
is essentially the density of states available to a particle that is added to the system
at (Ra, T) with quasimomentum q and energy ω. As opposed to a normal system,
the presence of the condensate gives non zero values to the oﬀ diagonal terms of
the functions G
(≷)
12
(Rq; T, ω). We refer to them as the anomalous contributions to
the respective two point functions.
The generalized Boltzmann equations can be obtained as the Fourier transform
of the equations of motion for the case in which the variations in R and T are very
slow. In particular when the inverse propagator D
−1
o
and the self energies vary very
little as Ra is changed by a characteristic excitation wavelength or T is changed
by an inverse excitation energy. Following Ref.[139] and assuming a (q, ω, R, T)
dependence of the variables, which is not explicitly written for simplicity, we get
the following equations:
D
−1
o
−'S +
i
2
γ
H = −
i
2
D
−1
o
, H
+
i
2
['S, H] +
1
4
[γ, H] , (8.47)
H
D
−1
o
−'S −
i
2
γ
= −
i
2
H, D
−1
o
+
i
2
[H, 'S] −
1
4
[H, γ] , (8.48)
D
−1
o
−'Σ +
i
2
Γ
G
(≷)
− Σ
(≷)
'G+
i
2
A
= −
i
2
D
−1
o
, G
(≷)
+
i
2
'Σ, G
(≷)
+
i
2
Σ
(≷)
, 'G
+
1
4
Γ, G
(≷)
−
1
4
Σ
(≷)
, A
, (8.49)
G
(≷)
D
−1
o
−'Σ −
i
2
Γ
−
'G−
i
2
A
Σ
(≷)
= −
i
2
G
(≷)
, D
−1
o
+
i
2
G
(≷)
, 'Σ
+
i
2
'G, Σ
(≷)
−
1
4
G
(≷)
, Γ
+
1
4
A, Σ
(≷)
, (8.50)
with
D
−1
o
(Rq; T, ω) ≡ (σ
z
(ω −v
s
2J sin(qa)) + (2J cos(qa) +µ
R
(T)) I) . (8.51)
Because all the quantities are slowly varying functions of R and T, in Eqs (8.47
8.50) we approximated the discretized equation to continuous diﬀerential equa
tions. The brackets denote the generalized Poisson brackets deﬁned as:
[A, B] =
∂A
∂ω
∂B
∂T
−
∂A
∂T
∂B
∂ω
+∂
R
A∂
q
B −∂
q
A∂
R
B. (8.52)
150
Chapter 8 From the 2PICTP approximations to kinetic theories and
local equilibrium solutions
In the equations we have also introduced the following functions:
'S(Rq; T, ω) = S
HF
(Rq; T, ω) +'S
B
(Rq; T, ω), (8.53)
'Σ(Rq; T, ω) = Σ
HF
(Rq; T, ω) +'Σ
B
(Rq; T, ω), (8.54)
'S
B
(Rq; T, ω) = P
dω
t
2π
γ(Rq; T, ω)
ω −ω
t
, (8.55)
'Σ
B
(Rq; T, ω) = P
dω
t
2π
Γ(qR, ω
t
)
ω −ω
t
, (8.56)
'G = P
dω
t
2π
A(Rq; T, ω)
ω −ω
t
. (8.57)
with P denoting the Cauchy principal value and γ(Rq; T, ω), Γ(Rq; T, ω), S
HF
(Rq; T, ω),
Σ
HF
(Rq; T, ω) and A(Rq; T, ω) understood as Fourier transforms of the functions
γ(t
i
, t
t
j
), Γ(t
i
, t
t
j
), S
HF
(t
i
, t
t
j
), Σ
HF
(t
i
, t
t
j
) and A(t
i
, t
t
j
) respectively.
If we deﬁne the statistical variables as:
F =
G
>
+G
<
2
, (8.58)
Π =
Σ
>
+ Σ
<
2
. (8.59)
Eq. (8.49)and Eq. (8.50), can be rewritten in term of statistical and spectral
functions as:
D
−1
o
−'Σ +
i
2
Γ
F − Π
'G+
i
2
A
= (8.60)
−
i
2
¸
D
−1
o
−'Σ +
i
2
Γ, F
−
¸
Π, 'G+
i
2
A
,
F
D
−1
o
−'Σ −
i
2
Γ
−
'G−
i
2
A
Π = (8.61)
−
i
2
¸
F, D
−1
o
−'Σ −
i
2
Γ
−
¸
'G−
i
2
A, Π
,
D
−1
o
−'Σ
A−Γ'G = −
i
2
¸
D
−1
o
−'Σ, A
−[Γ, 'G]
¸
, (8.62)
A
D
−1
o
−'Σ
−'GΓ = −
i
2
¸
A, D
−1
o
−'Σ
−['G, Γ]
¸
. (8.63)
Eqs. (8.47), (8.48) and (8.60) (8.63) are our passage to the Boltzmann equa
tions. They describe the state of the gas at a given time. In contrast to the HFB
equations, they include collisional integrals to describe binary interactions.
To progress further we can introduce even more simpliﬁcations based on phys
ical assumptions. The ordinary Boltzmann equation emerges from the approxima
tion in which the self energies that appear on the left side of Eqs.(8.47), (8.48) and
8.2 Boltzmann equations 151
(8.60) (8.63) are handled diﬀerently from those which appear on the right. These
two appearances of the selfenergy play a diﬀerent physical role in the descrip
tion of the dynamics [137]. The self energies on the right hand side describe the
dynamical eﬀects of collisions, i.e., how the collisions transfer particles from one
energymomenta conﬁguration to another. On the other hand, the self energies on
the left describe the quantum kinetic eﬀects due to interactions, i.e. how inter
action eﬀects change the energy momentum dispersion relations from that of free
particles to a more complicated spectrum. Because these two eﬀects are physically
so diﬀerent, we can treat the left and the right hand sides in diﬀerent ways.
In the derivation of the ordinary Boltzmann equations, one completely neglects
all the kinetic eﬀects in the secondorder self energies (the dependence on T and R
in the secondorder self energy terms on the right hand side) and retain dynamical
eﬀects (T and R dependence on the left hand side). In this way, we get the familiar
Boltzmann equations which describe the particles as free particles in between col
lisions. This is a reasonable assumption in dilute weakly interacting gases in which
the duration of a collision is very short compared to the essentially interactionfree
dynamics between isolated collisions. Neglecting kinetic eﬀects in the secondorder
self energies, Eqs.(8.47), (8.48) and (8.60)(8.63) can be approximated to
D
−1
o
−'S +
i
2
γ
H = −
i
2
D
−1
o
−S
HF
, H
, (8.64)
H
D
−1
o
−'S −
i
2
γ
= −
i
2
H, D
−1
o
−S
HF
, (8.65)
D
−1
o
−'Σ +
i
2
Γ
F −Π
'G+
i
2
A
= −
i
2
D
−1
o
−Σ
HF
, F
, (8.66)
D
−1
o
−'Σ
A−Γ'G = −
i
2
D
−1
o
−Σ
HF
, A
, (8.67)
F
D
−1
o
−'Σ −
i
2
Γ
−
'G−
i
2
A
Π = −
i
2
F, D
−1
o
−Σ
HF
, (8.68)
A
D
−1
o
−'Σ
−'GΓ = −
i
2
A, D
−1
o
−Σ
HF
. (8.69)
If we take the trace of the sum and the diﬀerence of each one of the above equations
with its hermitian conjugate, they can be simpliﬁed to :
Tr
¸
D
−1
o
−'S
H
¸
= 0, (8.70)
Tr
¸
D
−1
o
−'Σ
F −Π'G
¸
= 0, (8.71)
Tr
¸
D
−1
o
−'Σ
A−Γ'G
¸
= 0, (8.72)
Tr
D
−1
o
−S
HF
, H
= −Tr(γH), (8.73)
Tr
D
−1
o
−Σ
HF
, F
= −Tr(ΓF −ΠA). (8.74)
Tr
D
−1
o
−Σ
HF
, A
= 0 (8.75)
152
Chapter 8 From the 2PICTP approximations to kinetic theories and
local equilibrium solutions
Moreover, if we deﬁne the operator rM = M
12
+ M
∗
21
and apply it again to
the sum and the diﬀerence of each one of the equations (8.60) to (8.63) with its
transpose we also get:
Re
r
¸
D
−1
o
−'S
H
¸
=
1
2
Im
r
D
−1
o
−S
HF
, H
+(γH)
, (8.76)
Re
r
¸
D
−1
o
−'Σ
F −Π'G
¸
=
1
2
Im
r
D
−1
o
−Σ
HF
, F
+r(ΓF −ΠA)
, (8.77)
Re
r
¸
D
−1
o
−'Σ
A−Γ'G
¸
=
1
2
Im
r
D
−1
o
−Σ
HF
, A
, (8.78)
Im
r
¸
D
−1
o
−'S
H
¸
= −
1
2
Re
r
D
−1
o
−S
HF
, H
+(γH)
, (8.79)
Im
r
¸
D
−1
o
−'Σ
F −Π'G
¸
= −
1
2
Re
r
D
−1
o
−Σ
HF
, F
+r(ΓF −ΠA)
, (8.80)
Im
r
¸
D
−1
o
−'Σ
A−Γ'G
¸
= −
1
2
Re
r
D
−1
o
−Σ
HF
, A
, (8.81)
with Re and Im denoting the real and imaginary parts. To close this set of equa
tions we need an equation of motion for the superﬂuid velocity which can be found
from the deﬁnitions Eq.(8.39) and Eq.(8.40) to be:
2Ja
2
∂v
s
∂T
= −
∂
∂R
(µ +V ) +
4Ja
2
v
s
2
. (8.82)
Eqs.(8.708.81) together with Eq.(8.82) form a closed set of equations that
describe the state of the gas at a given time. Equations (8.708.72) and (8.76
8.78) are usually called gap equations. They describe the quantum properties of
the gas which is evolving according to the Boltmaznn equations (8.738.75) and
(8.798.81). Under the derived formalism the Boltzmann and gap equations form
a coupled set of equations which replaces the original dynamics. The equations
have to be solved self consistently for any analysis.
8.3 Equilibrium properties for a homogeneous system
In this section we will show how the nonequilibrium Boltzmann equations lead, in
a special case, to the linear equilibrium solutions discussed in chapter 5 upgraded
with secondorder corrections in U not included in the quadratic approximations.
There are two situations in which we expect an equilibrium solution to come from
the Boltzmann equations. Firstly, when the system has never been disturbed it
remains in its equilibrium state. Secondly, when the system has had suﬃcient time
to relax after an applied perturbation.
At equilibrium, in the absence of any external potential, the functions G
≷
and
H are completely independent of R and T. In this case the generalized Poisson
brakets terms are zero and Eqs.(8.72, 8.75) and (8.78, 8.75) imply that:
A
D
−1
o
−'Σ
−('G)Γ = 0. (8.83)
8.3 Equilibrium properties for a homogeneous system 153
Because 'G(q, ω) is determined by A(q, ω) as indicated in Eq. (8.57), Eq. (8.83)
is satisﬁed when A(q, ω) is given by
A(q, ω) = i
¸
D
−1
o
−'S +
i
2
Γ
−1
−
¸
D
−1
o
−'S −
i
2
Γ
−1
¸
, (8.84)
and the function 'G(q, ω) given by
'G(q, ω) = P
dω
t
2π
A(q, ω
t
)
ω −ω
t
=
1
2
¸
D
−1
o
−'S +
i
2
Γ
−1
+
¸
D
−1
o
−'S −
i
2
Γ
−1
¸
. (8.85)
From Eqs.(8.73), (8.79),(8.74) and (8.80)) we also get that at equilibrium
γ = 0, (8.86)
ΓF −ΠA = 0. (8.87)
Eqs. (8.86) and (8.87) are just the mathematical statement of detailed balance.
They just represent the physical condition that at equilibrium the net rate of
change of the density of particles with momentum q and energy ω is zero. Since
it is always possible to write [137]
F(q, ω) =
n
q
(ω) +
1
2
A(q, ω), (8.88)
then Eq.(8.87) can only be satisﬁed if
Π(q, ω) =
n
q
(ω) +
1
2
Γ(q, ω), (8.89)
is satisﬁed. Detailed study of the structure of the self energy indicates that n
q
(ω)
is related to the BoseEinstein distribution, n
q
(ω) =
1
e
βω
−1
with β interpreted as
the local inverse temperature in energy units [137, 139].
Since H contains delta functions in momentum and energy at equilibrium, we get
from Eq. (8.70):
µ = −2J +'S
11
(0, 0) +'S
12
(0, 0). (8.90)
8.3.1 Quasiparticle formalism
In the non interacting case the diagonal terms of A(q, ω) are just delta functions
with peaks at values of ω that matches the possible energy diﬀerence which re
sults from adding a single particle with quasimomentum q to the system. In the
154
Chapter 8 From the 2PICTP approximations to kinetic theories and
local equilibrium solutions
many body system the energy spectrum is suﬃciently complex so that the diag
onal elements of A(q, ω) are not delta functions but instead continuous functions
of ω. However, there are always sharp peaks in A. These sharp peaks represent
the coherent and long lived excitations, which behave like weakly interacting par
ticles. These excitations are called quasiparticles. From Eq. (8.84) it is possible
to see that the quasiparticle decay rate is determined by Γ. The quasiparticle
approximation is obtained by considering Γ very small for small values of ω. This
assumption implies that D
−1
≡ D
−1
o
− 'Σ −
i
2
Γ is essentially real with only an
inﬁnitesimal imaginary part. The zeros of D
−1
about which A is very sharply
peaked are identiﬁed with the quasiparticle energies ω
q
.
Using the assumption of a very small Γ, and the identity lim
→0
1
ω−ω
+i
= P
1
ω−ω
−
iπδ(ω −ω
t
), it is possible to write the matrix components of D
−1
as:
D
−1
(q, ω) = ω
1 0
0 −1
−
L
qq
(q, ω) ´
q−q
(q, ω)
´
∗
q−q
(q, ω) L
∗
qq
(−q, −ω)
, (8.91)
with
L
qq
(ω) = −2J cos qa −µ +Σ
HFB
11
(q, ω) +
dω
t
2π
Γ
11
(q, ω
t
)
ω −ω
t
+i
, (8.92)
´
q−q
(ω) = Σ
HFB
12
(q, ω) +
dω
t
2π
Γ
12
(q, ω
t
)
ω −ω
t
+i
. (8.93)
The quasiparticle amplitudes u
q
and v
q
are the solutions to the eigenvalue problem
L
qq
(q, ω
q
) ´
q−q
(q, ω
q
)
´
∗
q−q
(q, ω
q
) L
∗
qq
(−q, −ω
q
)
u
q
v
q
= ω
q
1 0
0 −1
u
q
v
q
, (8.94)
and satisfy the normalization condition [u
q
[
2
−[v
q
[
2
= 1. In the absence of vortices
it is always possible to ﬁnd an ensemble in which the amplitudes (u
q
, v
q
) are
purely real and u
q
= u
−q
, v
q
= v
−q
. In terms of the quasiparticle amplitudes,
the matrix elements of the spectral function A, Eq.(8.84), are given by:
A
11
(q, ω) = −2Im
¸
u
2
q
ω −ω
q
+i0
+
−
v
2
q
ω −ω
q
+i0
−
¸
= 2π
u
2
q
δ(ω −ω
q
) −v
2
q
δ(ω +ω
q
)
, (8.95)
A
12
(q, ω) = 2Im
¸
u
q
v
q
ω −ω
q
+ i0
+
−
v
q
u
q
ω −ω
q
+i0
−
= −2πu
q
v
q
[δ(ω −ω
q
) −δ(ω +ω
q
)] , (8.96)
A
22
(q, ω) = −A
11
(−q, −ω), (8.97)
A
21
(q, ω) = A
∗
12
(q, ω). (8.98)
Finally, using the deﬁnitions of F and A, we can express the matrix compo
nents: ρ
q
(ω), ¯ ρ
q
(ω) and m
q
(ω) deﬁned as the Fourier transform of ρ
ij
, ¯ ρ
ij
and m
ij
respectively (see Eqs. (8.21) and (8.22)) in terms of quasiparticle amplitudes:
8.3 Equilibrium properties for a homogeneous system 155
ρ
q
(ω) = 2π
u
2
q
n
q
(ω)δ(ω −ω
q
) +v
2
q
(1 +n
q
(ω))δ(ω +ω
q
)
, (8.99)
¯ ρ
q
(ω) = 2π
u
2
q
(1 +n
q
)(ω)δ(ω −ω
q
) +v
2
q
n
q
(ω)δ(ω +ω
q
)
, (8.100)
m
q
(ω) = 2πu
q
v
q
[n
q
(ω)δ(ω −ω
q
) −(1 +n
q
(ω))δ(ω +ω
q
)] . (8.101)
8.3.2 HFB approximation
Under the HFB approximation the matrix 'Σ and 'S are just given by Σ
HFB
and S
HFB
. In terms of the quasiparticle amplitudes and setting ^ = 2, they can
be written as:
Σ
HFB
= U
2 (n
o
+ ¯ n) n
o
+ ¯ m
n
o
+ ¯ m 2 (n
o
+ ¯ n)
, (8.102)
S
HFB
= U
n
o
+2¯ n ¯ m
¯ m n
o
+ 2¯ n
. (8.103)
with
¯ n = =
1
M
¸
q
(1 +n
q
(ω
q
))v
2
q
+u
2
q
n
q
, (8.104)
¯ m =
1
M
¸
q
u
q
v
q
(2n
q
(ω
q
) + 1) . (8.105)
In the HFB approximation, Eq.(8.94) and Eq. ( 8.90) then yield:
−2J cos(qa) −µ +2U(n
o
+ ¯ n) U (n
o
+ ¯ m)
U (n
o
+ ¯ m) −2J cos(qa) −µ +2U(n
o
+ ¯ n)
u
q
v
q
= ω
q
u
q
−v
q
,
(8.106)
µ = −2J +Un
o
+ 2U¯ n +U ¯ m. (8.107)
As a ﬁnal step, to ﬁx the total density to n, the constraint
n = n
o
+ ¯ n, (8.108)
has to be satisﬁed.
For a given density and temperature Eqs. (8.106) (8.108) form a closed set
of equations. At zero temperature, they reduce to the HFB equations derived in
chapter 5 using the quadratic approximation.
156
Chapter 8 From the 2PICTP approximations to kinetic theories and
local equilibrium solutions
8.3.3 Secondorder and Beliaev approximations
When secondorder terms are taking into account the matrixes L
qq
and M
q−q
become energy dependent. For simplicity we restrict the calculations to the zero
temperature case when n
q
= 0. In terms of the quasiparticle amplitudes the
contributions to the selfenergy at secondorder are given by
M
q−q
(q, ω) = Un
o
+U ¯ m+
2U
2
M
n
o
¸
k
2/
k
B
q−k
+ 2(
k
/
p
+ 2(
k
B
q−k
+ 3(
k
(
q−k
ω −ω
k
−ω
q−k
+i
−
2B
k
/
q−k
+ 2(
k
/
p
+ 2(
k
B
q−k
+3(
k
(
q−k
ω +ω
k
+ω
q−k
−i
(8.109)
+
2U
2
M
2
¸
k,p
2/
k
B
p
(
q−k−p
+(
k
(
p
(
q−k−p
ω −ω
k
−ω
p
−ω
q−k−p
+i
−
2B
k
/
p
(
q−k−p
+(
k
(
p
(
q−k−p
ω +ω
k
+ω
p
+ω
q−k−p
−i
,
L
qq
(q, ω) = −2J cos qa −µ +2Un
o
+ 2U¯ n +
2U
2
n
o
M
¸
k
/
k
/
q−k
+ 2/
k
B
q−k
+ 4(
k
/
q−k
+ 2(
k
(
q−k
ω −ω
k
−ω
q−k
+i
−
B
k
B
q−k
+ 2(B
k
/
q−k
) + 4(
k
B
q−k
+ 2(
k
(
q−k
ω +ω
k
+ω
q−k
−i
(8.110)
+
2U
2
M
2
¸
k,p
/
k
/
p
B
q−k−p
+2/
k
(
p
(
q−k−p
ω −ω
k
−ω
p
−ω
q−k−p
+i
−
B
k
B
p
/
q−k−p
+ 2B
k
(
p
(
q−k−p
ω +ω
k
+ω
p
+ω
q−k−p
−i
,
µ = −2J +Un
o
+ 2U¯ n +U ¯ m (8.111)
−
2U
2
M
2
¸
k,p
2/
k
B
p
(
k+p
+ 2B
k
/
p
(
k+p
+ 2(
k
(
p
(
k+p
ω
k
+ω
p
+ω
k+p
−
2U
2
M
2
¸
k,p
2/
k
(
p
(
k+p
+/
k
/
p
B
k+p
+ 2B
k
(
p
(
k+p
+B
k
B
p
/
k+p
ω
k
+ω
p
+ω
k+p
,
where the quantities /, B and ( are deﬁned as
/
k
= u
2
k
, B
k
= v
2
k
, (
k
= −u
k
v
k
. (8.112)
The inclusion of secondorder terms modiﬁes the structure of the HFB equa
tions. The matrix that we need to diagonalize to ﬁnd the quasiparticle energies
depends now on the quasiparticle mode in consideration. This means that a sep
arate nonlinear problem must be solved for every quasiparticle state, whereas the
solution of the HFB equations yields the whole quasiparticle spectrum. The ma
trix which is to be diagonalized also become intrinsically nonlocal and to solve
for a quasiparticle state with quasimomentum q we have to sum over all diﬀerent
8.3 Equilibrium properties for a homogeneous system 157
quasimomenta. Finally, the diagonal elements are no longer equal as was always
the case in all the quadratic approximations considered in chapter 5.
If we omit the secondorder terms containing no condensate amplitudes, the
equations that we get are the tightbinding version of the ones originally derived
by Beliaev [63]:
M
q−q
(q, ω) = Un
o
+U ¯ m +
2U
2
M
n
o
¸
k
2/
k
B
q−k
+ 2(
k
/
p
+ 2(
k
B
q−k
+3(
k
(
q−k
ω −ω
k
−ω
q−k
+i
−
2B
k
/
q−k
+ 2(
k
/
p
+ 2(
k
B
q−k
+ 3(
k
(
q−k
ω +ω
k
+ω
q−k
−i
, (8.113)
L
qq
(q, ω) = 4J sin
2
(qa/2) +Un
o
−U ¯ m+ (8.114)
2U
2
n
o
M
¸
k
/
k
/
q−k
+ 2/
k
B
q−k
+ 4(
k
/
q−k
+ 2(
k
(
q−k
ω −ω
k
−ω
q−k
+i
−
B
k
B
q−k
+2(B
k
/
q−k
) + 4(
k
B
q−k
+2(
k
(
q−k
ω +ω
k
+ω
q−k
−i
, (8.115)
µ = −2J +Un
o
+ 2U¯ n +U ¯ m. (8.116)
As discussed in chapter 5, the HFB approximation has the problem that it is
not gapless. It was shown by Beliaev that when secondorder Beliaev contributions
are included the gap problem disappears. The reason is that as ¯ m introduces the
manybody scattering matrix in the oﬀdiagonal terms, secondorder corrections
introduce the manybody scattering matrix in the diagonal terms. This can be seen
by considering the secondorder Beliaev corrections in the particle approximation,
u
q
→1, v
q
→0. In this case the secondorder terms in Eq.(8.115) become
L
(2)
qq
(q, ω) = −
2U
2
n
o
M
¸
k
1
ω −ω
k
−ω
q−k
−i
. (8.117)
Comparison with Eq. (5.121) shows that L
(2)
qq
(q, ω) is the second order contribution
to the many body scattering matrix.
When both HFB and secondorder corrections are included the quasiparticle
energies not only are shifted with respect to the BdG quasiparticle energies but
they also become complex. The imaginary part that the quasiparticle energies
acquire comes from the poles of the secondorder terms and it is associated with
a damping rate. The physical meaning is that when the energy denominator in
the secondorder terms vanishes a real decay process is energetically allowed. The
damping mechanism in which a quasiparticle decay into two of lower energy is
known as Beliaev damping and was calculated by Beliaev in the case of a uniform
Bose superﬂuid [63].
158
Chapter 8 From the 2PICTP approximations to kinetic theories and
local equilibrium solutions
8.4 Conclusions
In summary, we showed in this chapter that the complicated non local non Marko
vian secondorder solutions derived under the 2PICTP formalism, actually reduce
to the standard kinetic theory solutions when the scale separation assumption is
made. We also showed that at equilibrium the full secondorder 2PI equations re
produce the secondorder corrections to the self energy well known since Beliaev.
Chapter 9
Characterizing the Mott
Insulator Phase
Besides the superﬂuid regime, the other regime where analytic solutions become
relatively simple is the Mott insulator phase. Deep in the Mott insulator regime
the kinetic energy term in the Hamiltonian, which delocalizes the atoms, can be
treated as a perturbation. In this chapter we use perturbation theory to describe
the basic properties of the Mott phase. We study both a translationally invariant
system when no other external potential is present and the case when there is
an external harmonic conﬁnement, as is the case in real experimental situations.
Speciﬁcally, the trapped system is studied assuming a parameter regime where
multioccupancy in inhibited which is the relevant regime for the lattice based
quantum computation proposals.
9.1 Commensurate translationally invariant case
9.1.1 Perturbation theory
In this section we assume a commensurately ﬁlled one dimensional lattice with
integer ﬁlling factor N/M = g, periodic boundary conditions and a parameter
regime where U/J (U/J)
c
with (U/J)
c
the critical point, so that we are deep
in the insulating phase. In this parameter regime, it is fair to consider the kinetic
term of BoseHubbard Hamiltonian as a perturbation:
ˆ
H =
ˆ
H
(0)
+δ
ˆ
H, (9.1)
ˆ
H
(0)
=
1
2
U
¸
n
ˆ a
†
n
ˆ a
†
n
ˆ a
n
ˆ a
n
, (9.2)
δ
ˆ
H = −J
¸
n
(ˆ a
†
n
ˆ a
n+1
+ ˆ a
†
n+1
ˆ a
n
). (9.3)
The unperturbed Hamiltonian includes only the onsite interaction term, which
is diagonal in a number Fock state basis, and to zeroth order the ground state
160 Chapter 9 Characterizing the Mott Insulator Phase
[Φ
0
` is the Fock state with g atoms in every lattice site. We refer to this state
as the target state [T`. The lowest lying excitations are described by the Fock
states, [Ψ
nm
`, that have g + 1 particles at site n, g − 1 particles at site m, and
exactly g particles in every other site. We call these states one particlehole (1ph)
excitations . There are M(M−1) 1ph excitations and because of the translational
symmetry of the system they are degenerate at zeroth order. All of them have an
energy U above the ground state. This energy gap is one of the most important
characteristics of the Mott insulator phase.
Φ
(0)
0
≡ [T` ≡ [g, g, ...., g, g` , (9.4)
Φ
(0)
i
≡ [Ψ
nm
`
n,=m
i = 1, . . . , M(M −1) (9.5)
=
g, ..., g + 1
. .. .
n
, ..., g −1
. .. .
m
, .., g
¸
.
The unperturbed energies are
E
(0)
0
=
U
2
Mg(g −1), (9.6)
E
(0)
i
= U +E
(0)
0
i = 1, . . . , M(M −1). (9.7)
At ﬁrst order the kinetic energy term mixes the target state with 1ph excita
tions with the particle and the hole at adjacent sites. The ﬁrst order ground state
wavefunction is
Φ
(1)
0
=
[T` +
J
U
2Mg(g +1)[S`
, (9.8)
with [S` the normalized translationally invariant state with a particle and hole at
adjacent sites:
[S` ≡
M
¸
n=1
[Ψ
nn+1
` +[Ψ
nn−1
`
√
2M
. (9.9)
Notice the factor M that appears in the ﬁrst order correction. It gives a restriction
on the validity of a perturbative treatment as the number of lattice sites increases.
In order for perturbation theory to be valid, the parameter
Jg
U
√
M has to be small.
The ﬁrst order correction to the ground state energy vanishes. The second order
correction is given by
E
(2)
0
= −
2J
2
M
U
g(g + 1). (9.10)
Because of the degeneracy of the unperturbed states, in order to ﬁnd ﬁrst order
corrections to the M(M−1) low lying excited states we must diagonalize the kinetic
9.1 Commensurate translationally invariant case 161
energy Hamiltonian within the 1ph subspace. If we span the eigenstates as a linear
combinations of 1ph excitations,
Φ
(1)
i
=
M
¸
n,m=1,m,=n
C
i
nm
[Ψ
nm
` , (9.11)
the necessary and suﬃcient conditions that the coeﬃcients C
i
nm
have to fulﬁll in
order to diagonalize the kinetic energy operator are given by the following equations
−(g + 1)(C
i
n+1m
+C
i
n−1m
) −g(C
i
nm+1
+C
i
nm−1
) =
¯
E
i
C
i
nm
, (9.12)
with E
(1)
i
= U −J
¯
E
i
. Besides Eq. (9.12), the amplitudes have to fulﬁll two other
equations:
C
i
nn
= 0, (9.13)
C
i
n+Mm
= C
i
nm+M
= C
i
nm
. (9.14)
Eq. (9.12) is analogous to the tightbinding Schr¨odinger equation of a two dimen
sional square lattice in the x − y plane, with the x direction associated with the
position of the extra particle and the y direction with the position of the hole. The
diﬀerent weights g + 1 and g can be understood in the 2Dlattice model as diﬀer
ent eﬀective masses in the two directions. Eq. (9.14) imposes periodic boundary
conditions, whereas the constraint C
i
nn
= 0 takes into account the requirement
that the extra particle and the hole at the same site annihilate each other. It can
be thought of as a hard wall in the x = y axis. The solutions are not straightfor
ward due to the fact that the eﬀective mass diﬀerence breaks the lattice symmetry
around the x = y axis and makes the hard wall constraint hard to fulﬁll.
We now describe a procedure for solving the eigenvalue equations and evalu
ating limiting cases where it is possible to ﬁnd analytic solutions. Without any
constraint, a general solution of Eq. (9.12) has the form
C
nm
∝ e
inz
e
imt
, (9.15)
with
¯
E
i
= −2 ((g + 1) cos (z) +g cos (t)) . (9.16)
To satisfy the constraint C
nn
= 0 we use the linear character of the problem and
look for a linear combination of solutions which have the same eigenvalue
¯
E
i
but
which also fulﬁll the hard wall constraint. We look then for a general solution of
the form:
C
nm
∝ sin (r(n −m)) e
iml
e
ins
, (9.17)
and determine the free parameters, l, r, s by forcing the solution to satisfy Eq.
(9.12). This procedure leads to the following equation:
(g + 1) cos (r +s) +g cos (l −r) = (g + 1) cos (s −r) +g cos (l +r) . (9.18)
162 Chapter 9 Characterizing the Mott Insulator Phase
Eq. (9.18) is satisﬁed if we choose s to be given by:
s = Arc sin
g
g + 1
sin l
. (9.19)
The tricky part is trying to satisfy also the periodic boundary conditions, Eq.(9.14),
because in general solutions of the form Eq. (9.17) are not necessarily compatible
with the periodic boundary conditions in the two variables n and m. However,
there are some limiting cases in which both conditions are satisﬁed. Such cases
are discussed in the remainder of the section.
• Case l = 0
If we choose l = 0 and use Eq. (9.19) and Eq. (9.17), we can ﬁnd M − 1
solutions of Eq. (9.12) which satisfy all the boundary conditions and are
linearly independent. These are
C
r
nm
=
√
2
M
sin
πr
M
[n −m[
, r odd
√
2
M
sin
2πr
M
(n −m)
, r even
(9.20)
E
(1)
r
= U −2J(2g +1) cos
πr
M
, r = 1, ..M −1 (9.21)
Inside the 1ph subspace all the translationally invariant states are spanned
by the states whose amplitudes C
r
nm
are the r ∈ odd solutions of Eq. (9.20).
Because in the absence of an external conﬁnement the many body Hamilto
nian is translationally invariant, the ground state must also be and it is only
coupled through the Hamiltonian to the subspace spanned by these transla
tionally invariant states. The dispersion relation, Eq. (9.21), agrees to ﬁrst
order in J to the meanﬁeld solution found in [77]. We want to point out
that this dispersion relation only describes the spectrum of M − 1 of the
M(M −1) low tying excitations.
When l = 0 the derivation of analytic solutions is more elaborate. However,
in the limiting cases of high ﬁlling factor, g 1, it is still easy to ﬁnd an
analytic solution.
• Case g 1
In this limiting situation s ≈ l, (see Eq.(9.19)) and the periodic boundary
conditions are satisﬁed if s, l and r are integer multiples of
π
M
. The eigen
values and an orthonormal set of modes in the high ﬁlling factor regime can
be chosen to be:
9.1 Commensurate translationally invariant case 163
E
(1)
rR
= U −2J(2g + 1) cos
πr
M
cos
πR
M
, (9.22)
C
r,R,=0
nm
=
2
M
sin(
πr
M
[n −m[) sin(
πR
M
(n +m) +α
rR
),
2
M
sin(
πr
M
(n −m)) sin(
πR
M
(n +m) +β
rR
),
C
r,0
nm
=
√
2
M
sin
πr
M
[n −m[
, r odd
√
2
M
sin
πr
M
(n −m)
, r even
(9.23)
with r = 1, . . . M − 1 and R = 0, . . . M − 1. The notation R
t
restricts
the values of R to the ones where R + r is an odd number and R
tt
to the
values where R + r is even. The constants α
rR
= π(r − R + 1)/4 and
β
rR
= π(r + R − 1 + M)/4 guarantee the orthogonality of the eigenmodes.
In Fig. 9.1 we show a contour plot of the two dimensional 1ph band.
0
Π
4
Π
2
3 Π
4
Π
kx
0
Π
4
Π
2
3 Π
4
Π
k
y
P.h. Band
Figure 9.1: Contour plot of the two dimensional band of the 1ph excitations to
ﬁst order in perturbation theory. In the plot the brighter the color the higher the
energy. The labels are ky = 2π/MR and kx = 2π/MR.
To check the range of validity of our analytical solutions, in Fig.9.2 we plot
comparisons between the ﬁrst order energy shifts calculated by the exact diago
nalization of the BoseHubbard Hamiltonian but restricting the Hilbert space to
the 1ph subspace and Eq. (9.22). We label the eigenvalues in ascending order.
164 Chapter 9 Characterizing the Mott Insulator Phase
0 20 40 60 80 100
n
th
6
4
2
0
2
4
6
E
n
1
g1
analytic
ph subspace
0 20 40 60 80 100
n
th
15
10
5
0
5
10
15
E
n
1
g4
analytic
ph subspace
Figure 9.2: Comparisons between the ﬁrst order corrections to the 1ph excita
tions calculated by diagonalizing the BoseHubbard Hamiltonian inside the 1ph
subspace and the analytic solution Eq. (9.22). The number of sites used for the
plot was M = 11. Energies are in units of J.
The parameters used for the comparisons were M = 11 and ﬁlling factors g = 1
and 4. We notice that even though the analytic expression was calculated under
the high ﬁlling factor assumption, for values of g = 4, the agreement between
the two solutions is very good. For the case g = 1 the two spectra do not agree
perfectly, however, we can say that the analytic solution is fair even in this low
density regime.
In Fig. 9.2 we checked for the validity of the analytic solution Eq. (9.22) when
the Hilbert space is restricted to the 1ph subspace. However, to restrict the low
lying excitations to the 1ph excitations is only good as long as ﬁrst order pertur
bation theory is valid. This implies that the parameter
Jg
U
√
M is small. In Fig.
9.3, we explore the range of validity of perturbation theory by plotting the spec
trum for diﬀerent values of U/J and ﬁlling factors. The solutions presented in the
plots are the eigenenergies found by the exact diagonalization of the Hamiltonian
(red), by diagonalization in the restricted 1ph subspace (green) and the analytic
solution, Eq. (9.22) (blue).
We observe in the plots that for low ﬁlling factors (g = 1) the analytic solution
is not as good as it is for high ﬁlling factors in reproducing the 1ph subspace
spectrum. Nevertheless, at low densities the validity of perturbation theory holds
for a larger range of U/J values. On the other hand, when the ﬁlling is high
the analytic solution gives a pretty good description of the 1ph subspace but the
parameter U/J must be approximately g times higher than the unit ﬁlled case to
ensure that higher order corrections are negligible. The number of wells chosen for
the plots was small because of the exponentially scaling of the Hilbert space. As
the number of wells is increased, the ratio of U/J required for perturbation theory
to be valid becomes larger.
Using the perturbation theory results, it is possible to calculate expectation
values of physical observables that are relevant to the experiments. For example,
9.1 Commensurate translationally invariant case 165
0 5 10 15 20 25 30
n
th
24
26
28
30
32
E
n
E
0
g1 K6 UJ10
ph sub
anal
exact
0 5 10 15 20
n
th
20
25
30
35
40
45
50
E
n
E
0
g4 K5 UJ27
ph sub
anal
exact
0 5 10 15 20 25 30
n
th
40
42
44
46
48
50
E
n
E
0
g1 K6 UJ45
ph sub
anal
exact
0 5 10 15 20
n
th
90
95
100
105
110
115
E
n
E
0
g4 K5 UJ100
ph sub
anal
exact
Figure 9.3: Comparisons between the energy eigenvalues calculated by the exact
diagonalization of the Hamiltonian (red dots), the diagonalization of the Hamil
tonian in the restricted 1ph subspace (green triangles) and the analytic solution
(blue crosses), Eq. (9.22). The index n labels the eigenvalues in increasing order of
energy. The ﬁlling factor g, the number of wells M and the ratio U/J is indicated
in each plot. Energies are in units of J.
the quasimomentum distribution (Eq. (5.57)) to ﬁrst order in perturbation theory
is given by:
n
q
= g +
4J
U
g(g + 1) cos(qa). (9.24)
Here q is the quasimomentum q =
2πj
aM
, j = 0, 1, . . . M−1 and a the lattice spacing.
The interference pattern after ballistic expansion is closely related to the quasi
momentum distribution [103]. In the superﬂuid regime the intensity of the prin
cipal interference peak is proportional to the occupation number of the q = 0
quasimomentum component, which describes the number of particles in the con
densate. The washing out of the interference peaks in the Mott phase is linked
to the redistribution of the population from the condensate to states with higher
quasimomenta. When the system is in the Mott insulator regime, instead of a
macroscopically occupied state, at zeroth order in J/U, the momentum distribu
tion is ﬂat and all the quasimomentum states of the lowest band are uniformly
occupied. This feature is indicated in Eq. (9.24). At ﬁrst order in J/U, we
166 Chapter 9 Characterizing the Mott Insulator Phase
take into account the nearest neighbors’ remaining coherence, always present in
the Mott ground state, which creates a small cosinusoidal modulation of the ﬂat
quasimomentum background.
In Fig. 9.4, we plot the quasimomentum distribution as a function of U/J for a
unit ﬁlled lattice with six wells. The exact solution is depicted with solid lines and
the ﬁrst order perturbative results with dots. The decrease of the zero momentum
population and the tendency towards a ﬂat distribution for large U/J ratios is
observed in the plot.
10 20 30 40 50 60 70 80
UJ
0.5
1
1.5
2
n
q
g1
q0
qΠ3
q2Π3
qΠ
Figure 9.4: Quasimomentum distribution as a function of U/J for a unit ﬁlled lat
tice with six wells (M = 6). The exact solutions are displayed with solid lines and
the ﬁrst order perturbative results with dotted lines. Due to the lattice symmetry
n
4π/3
= n
2π/3
and n
5π/3
= n
π/3
.
Another characteristic feature of the Mott insulator phase is the reduction
of the number ﬂuctuations. The ﬁrst non vanishing number ﬂuctuations (see
Eq. (5.56)) are quadratic in J/U and given by:
∆n = 4g(g +1)
J
2
U
2
. (9.25)
We omitted the site index in ∆n because of the translational symmetry. Notice
that due to the fact that the true Mott ground state is not a Fock state, the
number ﬂuctuations decrease as the ratio U/J increases, but they are not exactly
zero. This is a problem for the lattice based quantum computer proposals and we
will try to correct for it in chapter 11.
9.2 Harmonic conﬁnement plus lattice 167
We conclude this section by using the ﬁrst order perturbation theory results,
to crudely estimate the critical transition point. Quantum phase transitions such
as the superﬂuid to Mott insulator transition can be understood based on the
phenomena of level crossing: as a characteristic parameter is changed (in this case
U/J), at some critical point a state that was an excited state becomes the new
ground state of the system. If we use this criteria to calculate the critical point
and use Eq. (9.22) we get
(U/J)
c
≈ 2(2g + 1) ∼ 4g (9.26)
This result only applies for the one dimensional case. It is a factor of two
smaller than the calculated value from mean ﬁeld theory (Eq. 3.9) but it is in
better agreement with numerical Monte Carlo simulations and strong coupling
expansions which predict a critical point for a unit ﬁlled lattice of (U/J)
c
≈ 4.65
[78, 79].
9.2 Harmonic conﬁnement plus lattice
In this section we consider a one dimensional optical lattice in the presence of a
magnetic conﬁnement with oscillation frequency ω
T
. We assume that the magnetic
trap has its minimum at the lattice site n = 0. The magnetic conﬁnement intro
duces a characteristic energy scale Ω = ma
2
ω
T
/2, so that V
n
= Ωn
2
(see Eq.(6.1)).
We ﬁrst focus on the case in which the system has an odd number of atoms.
For the analysis we are also going to consider a parameter regime where
U >
Ω(N −1)
2
4
, (9.27)
ΩN > J. (9.28)
The ﬁrst condition expresses the requirements that the on site interaction energy
U must be bigger than the trapping energy of the most externally trapped atom,
so multiple atom occupation in any well is inhibited. The second states the re
quirement that the kinetic energy must be smaller than the potential energy cost
for an atom at the edge of the atomic cloud to tunnel to the next unoccupied site.
Thus it is energetically costly for the atoms to tunnel.
In this parameter regime the BoseHubbard Hamiltonian can be split into two
parts: an unperturbed part which includes the interaction and potential energy
and which is diagonal in a Fock state representation, and a kinetic energy part
which is going to be treated as a perturbation.
ˆ
H =
ˆ
H
(0)
+δ
ˆ
H, (9.29)
ˆ
H
(0)
= Ω
¸
n
n
2
ˆ a
†
n
ˆ a
n
+
1
2
U
¸
n
ˆ a
†
n
ˆ a
†
n
ˆ a
n
ˆ a
n
, (9.30)
δ
ˆ
H = −J
¸
n
(ˆ a
†
n
ˆ a
n+1
+ ˆ a
†
n+1
ˆ a
n
). (9.31)
168 Chapter 9 Characterizing the Mott Insulator Phase
With the constraints Eq. (9.27) and Eq. (9.28) we are guaranteed that to zeroth
order in perturbation theory, the ground state of the system consists of a unit ﬁlled
central core comprising N wells in the central region of the trap, surrounded by
external empty sites, which in solid state language can be called a sea of holes. If
the number of atoms is odd, due to the the assumed trap symmetry, at zero order
in perturbation theory the ground state is unique and given by:
Φ
(0)
0
=
0, .., 0, 1, 1, ...., 1, 1
. .. .
N sites
, 0, .., 0
¸
. (9.32)
The unperturbed ground state energy is
E
(0)
0
=
Ω
24
N(N
2
−1). (9.33)
The zeroth order low lying excitations consist of 1ph excitations and another kind
of excitations which we refer as nhh excitations. In the following we proceed to
describe them in more detail:
• One particle hole excitation (1ph)
The 1ph excitations are described by the Fock states,
Ψ
ph
nm
, with an extra
particle at site m and a hole at the site n, both inside the central core.
Φ
(0)
iph
≡
Ψ
ph
mn
(9.34)
=
0, .., 0, 1, ..., 2
....
m
, ..., 0
....
n
, .., 1, 0, .., 0
¸
.
As opposed to the translationally invariant system, the presence of the trap
breaks the degeneracy and the 1ph excitations are not degenerate at zero
order. Their energy is given by the interaction energy cost U to create a
1ph pair plus the potential energy cost due to the trap:
E
ph(0)
nm
= E
(0)
0
+U + Ω(m
2
−n
2
). (9.35)
with n and m integers, n, m = −
N−1
2
, . . . ,
N−1
2
.
• Hole hopping excitations (nhh)
These excitations are described by the states with a maximum of one particle
per well and holes inside the inner central core. They have to be included in
the trapped system because of the reservoir of holes surrounding the central
9.2 Harmonic conﬁnement plus lattice 169
core which brings an extra degree of delocalization. Because the nhh excita
tions are only due to the transfer of holes to the inner core, they don’t have
the interaction energy cost U for having two particles at the same site and
therefore they can be lower in energy than the 1ph excitations. We label the
1hh, states with only one hole inside the central core, by
Ψ
hh
mn
. Explicitly
they are given by:
Φ
(0)
ihh
≡
Ψ
hh
mn
(9.36)
=
0, .., 1
....
n
, .., 0, 1, ..., 0
....
m
, .., 1, 0, .., 0
¸
.
The zero order energy cost to create 1hh excitations is:
E
hh(0)
nm
= E
(0)
0
+ Ω(m
2
−n
2
), (9.37)
with n and m integers such that [n[ > (N −1)/2 and [m[ < (N −1)/2
At ﬁrst order the kinetic energy term mixes the unperturbed ground state with
the 1ph excitation which have the extra particle and the hole at adjacent sites
and with the 1hh excitations were the most external trapped atom tunnels to the
ﬁrst available vacant site. The ground state wave function at ﬁrst order is then
given by
Φ
(1)
0
=
Φ
(0)
0
+ 2J
(N−3)/2
¸
n=−(N−1)/2
Ψ
ph
nn+1
+
Ψ
ph
−n−(n+1)
√
2(U −Ω(1 + 2n))
+
√
2J
ΩN
Ψ
hh
nn+1
+
Ψ
hh
−n−(n+1)
√
2
. (9.38)
The ﬁrst order wave functions of the excitations directly coupled to the unper
turbed ground state are
Φ
(1)
±nph
=
Ψ
ph
nn±1
−
√
2J
Φ
(0)
0
(U −Ω(1 ±2n))
, (9.39)
Φ
(1)
±hh
=
Ψ
hh
±L±(L+1)
−J
Φ
(0)
0
(ΩN)
, (9.40)
with L = (N −1)/2.
In Fig. 9.5 we plot comparisons between the analytic (blue) and exact (red)
excitation spectra. The index n labels the eigenenergies in increasing energy order.
170 Chapter 9 Characterizing the Mott Insulator Phase
The parameters used for the plots were U/J = 40 and 60, Ω/J = 1.875, N = 9 and
M = 11. The analytical spectrum was calculated to zeroth order in perturbation
theory. The convergence of perturbative results to the exact ones as the interaction
parameter U/J is increased can be seen in the plots. The lines indicate the energies
of the states that are coupled to the zero order ground state. The ﬁrst two lines
correspond to the two degenerate 1hh excitation states
Ψ
hh
±L±(L+1)
with energy
ΩN, the others correspond to 1ph excitations
Ψ
ph
nn±1
with energies (U −Ω(1 ±
2n)).
0 50 100 150 200
n
0
20
40
60
E
n
J
UJ60.
0 50 100 150 200
n
0
10
20
30
40
50
E
n
J
UJ40.
Figure 9.5: Comparisons between the exact (red) and the perturbative (blue)
spectra for a trapped system with N = 9, M = 11 and Ω/J = 1.875 deep in the
Mott regime. The lines indicate the energies of the states that are coupled directly
to the ground state.
When the number of atoms is even, the trap symmetry is broken and the
ground state becomes degenerate. In this case, at zeroth order, the degenerate
ground states are:
9.3 Conclusions 171
Φ
(0)
0±
=
1
√
2
Φ
(0)
L
±
Φ
(0)
R
, (9.41)
Φ
(0)
L
=
0, ..., 0, 1, 1, 1, ...., 1, 1
. .. .
N−1 sites
, 0, .., 0
¸
,
Φ
(0)
R
=
0, .., 0, 1, 1, ...., 1, 1
. .. .
N−1 sites
, 1, 0, .., 0
¸
.
When the system has an even number of atoms, there are always going to be a
nonzero number ﬂuctuation at the trap edge. Because the states
Φ
(0)
L
and
Φ
(0)
R
are not coupled by the BoseHubbard Hamiltonian at ﬁrst order, the previous anal
ysis we did for the system with an odd number of atoms can be straightforwardly
extended to the even case.
The constraint ΩN > J used for the analysis above can be an important
experimental restriction if the total number of trapped atoms is large. In the
regime where ΩN < J, but still U > Ω(N/2)
2
, the energy splitting between 1ph
excitations induced by the trap is small and degenerate perturbation theory must
be used at ﬁrst order. An analytic treatment becomes really diﬃcult, nevertheless,
it is possible to have a qualitative analysis of the system. In this case, the system
can be divided in two parts: a central unit ﬁlled core comprising K < N wells
at the center of the trap and a superﬂuid layer surrounding the central core with
maximum one atom per lattice site where the mobility of the holes is high. The
size K of the unit ﬁlled central core is determined by the strength of the magnetic
conﬁnement, K ∼ J/Ω. The properties of the unit ﬁlled core can be described to
a good approximation by the commensurate homogeneous results. On the other
hand, the atoms surrounding the central core are almost free to tunnel and they
have superﬂuid like properties.
9.3 Conclusions
In summary, in this chapter we have described the basic properties of the Mott
insulating phase both in translationally invariant and harmonically trapped sys
tems. All the results derived here are going to be used as the starting point for
the calculations in the following chapters.
172 Chapter 9 Characterizing the Mott Insulator Phase
Chapter 10
Bragg Spectroscopy
To date, the primary observable used to study ultra cold atoms in an optical lattice
has been the momentum distribution of the system, observed after ballistic expan
sion. In particular, this type of measurement has been used to reveal the phase
coherence between sites in the lattice and has been a really useful technique to
characterize the superﬂuid phase. However, the disappearance of the interference
pattern is not a conclusive diagnostic of the Mott phase. Recent analysis shows
that this is more correctly related to the degree of condensate depletion. Indeed,
for this reason, the diagnostic tool used in the experiments by Greiner et al. [46],
to prove the achievement of the Mott phase, was to apply a potential gradient to
the lattice and show the presence of a gap in the excitation spectrum.
Moreover, because the usual procedure for producing the Mott insulator state is
to begin with a magnetically trapped BoseEinstein condensate (with no discernible
thermal component), and slowly load it into an optical lattice, ﬁnite temperature
eﬀects or nonadiabatic ramping of the lattice may lead to the production of an
imperfect Mott phase. As more elaborate experiments are undertaken with optical
lattices in the strongly correlated regime it will be crucially important to diagnose
the properties of the states produced.
The results we present here suggest that Bragg spectroscopy can not only give
information about the excitation spectrum, which is a crucial diagnostic of the
Mott insulator phase as characterized by the opening of a gap, but that it can also
be used in the Mott regime to estimate the temperature of the system. Speciﬁcally,
we study the linear response to Bragg spectroscopy of cold atoms loaded in a one
dimensional optical lattice based on a perturbative analysis. We obtain analytical
expressions for the dynamical structure factor and use them to calculate the energy
deposited into the system. We also test the accuracy of our approximations by
comparing them with numerical solutions obtained by diagonalizing the Bose
Hubbard Hamiltonian for moderate number of atoms and wells. The calculations
are done for translationally invariant lattices and in the presence of an harmonic
external potential. First we start by reviewing the basic formalism that describes
Bragg spectroscopy.
174 Chapter 10 Bragg Spectroscopy
10.1 Formalism
The application of Bragg spectroscopy to study cold atomic systems was ﬁrst
pioneered by the NIST and MIT groups [65, 66]. Their scheme involved using a
pair of interfering laser ﬁelds to Bragg scatter atoms into a higher momentum state.
Since those experiments Bragg spectroscopy has been established as a versatile
tool for probing BoseEinstein condensates in a wide range of situations (e.g see
[141, 142, 143, 144]).
Figure 10.1: Bragg spectroscopic scheme considered in this paper. The lattice
potential with atoms loaded into the ground band is perturbed by a shallow running
wave perturbation (the Bragg potential).
Twophoton Bragg spectroscopy in a lattice is performed by superimposing a
periodic travelling wave potential on the lattice potential. This could be arranged
by modulating the lattice amplitude, as done by St¨oferle et al. [145]. The more
general situation arises when the Bragg potential is produced by an additional
pair of independent laser ﬁelds, chosen to produce a shallow potential that is not
necessarily commensurate with lattice potential (see Fig. 10.1). Speciﬁcally, the
atomic sample is illuminated with two additional laser beams with wave vectors
of magnitudes k
1
and k
2
in the direction of the lattice and frequency diﬀerence
ω. The intersecting beams create a periodic potential parallel to the lattice with
travelling intensity modulation given by V
o
cos(qx − ωt). The diﬀerence in wave
vectors of the beams deﬁnes the Bragg momentum q = k
1
−k
2
and the diﬀerence
in frequency the Bragg energy ω. Atoms exposed to these beams can undergo
stimulated light scattering by absorbing a photon from one of the beams and
emitting a photon into the other.
We consider a one dimensional optical lattice with M lattice sites, suﬃciently
deep that the tightbinding approximation is valid, and assume that we can restrict
the dynamics of the atoms to the lowest vibrational band. This is guaranteed if
the loading is done slow enough to avoid band excitations,i.e. if the energy of the
Bragg perturbation ω is less than the energy gap to the second band and if the
momentum transfer q is contained within the ﬁrst Brillioun zone. A detailed anal
ysis of the validity of the ﬁrst band approximation is found in Ref.[148]. There the
authors, using a mean ﬁeld approach combined with Bogoliubov analysis, extended
the existing theory of Bragg spectroscopy of magnetically trapped BEC [146, 147]
to lattice systems.
10.1 Formalism 175
As shown in chapter 3, under the singleband approximation the system is de
scribed by the BoseHubbard Hamiltonian, which we will denote as
ˆ
H
o
. Hereafter
we assume, unless otherwise speciﬁed, a harmonic magnetic conﬁnement such that
V
n
= Ωn
2
. In the tightbinding approximation, the Hamiltonian describing the
Bragg perturbation is given by
ˆ
H
B
=
1
2
V
o
(ˆ ρ
†
q
e
−iωt
+ ˆ ρ
q
e
iωt
). (10.1)
The density ﬂuctuation operator ˆ ρ
†
q
is deﬁned as
ˆ ρ
†
q
=
M−1
¸
m,n=0
I
n,m
q
ˆ a
†
m
ˆ a
n
, (10.2)
with
I
n,m
q
=
dxe
iqx
w
∗
0
(x −am)w
0
(x −an), (10.3)
where w
0
(x) is the Wannier orbital centered at the origin of the lowest vibra
tional band. To calculate I
n,m
q
we approximate the Wannier orbitals by Gaus
sians localized at the bottom of the potential wells , φ
0
(x − an) ∼ exp[−(x −
an)
2
/2a
2
ho
]/(π
1/4
√
a
ho
), where the width a
ho
can be estimated by minimizing the
ground state energy. For a lattice potential of the form E
R
V
lat
sin
2
(πx/a), with
E
R
the recoil energy of the atoms, a
ho
scales like a
ho
∼ (V
lat
)
−1/4
a/π. Ne
glecting the overlap between Wannier orbitals located in diﬀerent sites we get
I
n,m
q
∼ I
q
δ
n,m
exp(ia(n + m)q/2), where δ
n,m
is the Kronecker delta function.
Then
I
q
= exp
−
1
√
V
lat
qa
2π
2
. (10.4)
Notice that I
q
approaches one for small qa and deep lattices. Using the approxi
mate expression for I
n,m
q
we can rewrite the density operator as:
ˆ ρ
†
q
= I
q
M−1
¸
m=0
ˆ a
†
m
ˆ a
m
e
iqma
. (10.5)
To simplify the notation and to have a better physical understanding of the
excitations induced by the Bragg perturbation, it is convenient to introduce the
quasimomentum ﬁeld operator
ˆ
b
k
which annihilates an atom with quasimomentum
k, and is deﬁned as the Fourier transform of the ﬁeld operator ˆ a
n
, i.e.,
ˆ
b
k
=
1
√
M
¸
M
n=1
e
ikna
ˆ a
n
. The quasimomentum k can assume only discrete values which
are integer multiples of
2π
Ma
.
In the quasimomentum basis the operator ˆ ρ
†
q
can be written as
ˆ ρ
†
q
= I
q
M−1
¸
k=0
ˆ
b
†
k+q
ˆ
b
k
, (10.6)
176 Chapter 10 Bragg Spectroscopy
where we have assumed that q can be also expressed as an integer multiple of
2π
Ma
. By writing the Bragg perturbation potential in the quasimomentum basis it
can be seen that the two means by which momentum is imparted to the atomic
sample,is either by promoting an atom from quasimomentum k to k + q or by
demoting a particle with momentum k to k −q.
10.2 Observables
Useful experimental observables in Bragg spectroscopy are related to the density
density response function. Here we focus our attention on the imparted momentum
and energy, which can be measured by timeofﬂight techniques. In this section
we elaborate on the form of these observables.
In the tightbinding approximations the total momentum is given by
ˆ
P =
¸
k
k
ˆ
b
†
k
ˆ
b
k
. (10.7)
From the Heisenberg equations of motion we get
d
ˆ
P
dt
=
i
ˆ
H
,
ˆ
P
=
i
ˆ
H
B,
ˆ
P
+
ˆ
H
o,
ˆ
P
(10.8)
= −
iqV
o
2
(ˆ ρ
†
q
e
−iωt
− ˆ ρ
q
e
iωt
) −2Ω
ˆ
X. (10.9)
where
ˆ
H is the total Hamiltonian,
ˆ
H =
ˆ
H
o
+
ˆ
H
B
, and
ˆ
X is the center of mass
position given by
ˆ
X =
¸
n
(na)ˆ a
†
n
ˆ a
n
. The second term in the momentum rate
equation is just related to the force felt by atoms due to the presence of an harmonic
external conﬁnement. On the other hand, the ﬁrst term takes into account the rate
of change of the momentum due to the Bragg perturbation. If the perturbation is
applied for a time short compared to the oscillator period t
osc
=
2π
√
4JΩa
the second
term in the righthandside is small and can be neglected in the calculations.
The energy change rate is
d
ˆ
H
o
dt
=
i
ˆ
H
B,
ˆ
H
o
=
i
ˆ
H
B,
ˆ
H
, (10.10)
= −
1
2
V
o
e
−iωt
dˆ ρ
†
q
dt
+e
iωt
dˆ ρ
q
dt
.
Because the energy of a closed system is conserved, as opposed to the momentum
rate equation, the energy change rate is independent of the magnetic potential and
only depends on the applied Bragg perturbation.
10.3 Linear response 177
10.3 Linear response
The ﬁrst order variation δ'
ˆ
O(t)` of the thermal average of an observable
ˆ
O(t) with
respect to its mean value due to a weak perturbation
ˆ
H
B
applied to a system at
t = t
o
, is given by the Kubo formula [51]
δ
ˆ
O(t)
=
i
t
t
o
dτ
ˆ
H
B
(τ),
ˆ
O(t)
, (10.11)
where the operators
ˆ
O(t) and
ˆ
H
B
(τ) are in the Heisenberg representation with
respect to the time independent hamiltonian
ˆ
H
o
, and the expectation value is
taken using the equilibrium state of the system.
For the particular case where the observables are the momentum and energy
rates, we get after some algebra :
δ
d
ˆ
P
dt
¸
= q
V
o
2
2
(10.12)
t
t
o
dτ
e
−iω(τ−t)
ˆ ρ
†
q
(τ), ˆ ρ
q
(t)
−e
iω(τ−t)
ˆ ρ
q
(τ), ˆ ρ
†
q
(t)
+e
−iω(τ+t)
ˆ ρ
†
q
(τ), ˆ ρ
†
q
(t)
−e
iω(τ+t)
'[ ˆ ρ
q
(τ), ˆ ρ
q
(t)]`
−2Ωδ
ˆ
X
,
δ
d
ˆ
H
dt
¸
= −
i
V
o
2
2
(10.13)
t
t
o
dτ
e
−iω(τ−t)
d
dt
ˆ ρ
†
q
(τ), ˆ ρ
q
(t)
−e
iω(τ−t)
d
dt
ˆ ρ
q
(τ), ˆ ρ
†
q
(t)
+e
−iω(τ+t)
d
dt
ˆ ρ
†
q
(τ), ˆ ρ
†
q
(t)
−e
iω(τ+t)
d
dt
'[ ˆ ρ
q
(τ), ˆ ρ
q
(t)]`
.
We now introduce the functions S(q, ω) and R(q, ω) given by
ˆ ρ
†
q
(τ)ˆ ρ
q
(t)
=
e
−iω(τ−t)
S(q, ω)dω, (10.14)
ˆ ρ
†
q
(τ)ˆ ρ
†
q
(t)
=
e
iω(τ+t)
R(q, ω)dω. (10.15)
In terms of these functions it can be seen that since
ˆ ρ
†
q
(τ), ˆ ρ
†
q
(t)
=
ˆ ρ
q
(τ), ˆ ρ
q
(t)
=
0 Eq. (10.13) and Eq. (10.14) reduce to:
δ
d
ˆ
P
dt
¸
= 2q
V
o
2
2
dω
t
S(q, ω
t
) −S(−q, −ω
t
)
sin ((ω −ω
t
)t)
(ω −ω
t
)
−2Ωδ
ˆ
X
,
178 Chapter 10 Bragg Spectroscopy
δ
d
ˆ
H
dt
¸
=
2
V
o
2
2
dω
t
ω
t
S(q, ω
t
) −S(−q, −ω
t
)
(10.16)
sin ((ω −ω
t
)t)
(ω −ω
t
)
.
where we have set t
o
= 0. The quantity S(q, ω), the Fourier transform of density
correlations, is the so called dynamic structure factor and is given by
S(q, ω) =
1
Z
¸
ij
e
−βE
i
['i[ ˆ ρ
q
[j`[
2
δ(ω −ω
ij
) (10.17)
≡
¸
ij
S
ij
(q)δ(ω −ω
ij
). (10.18)
where [i` and E
i
are eigenstates and eigenenergies of the unperturbed Hamiltonian,
e
−βE
i
is the usual Boltzmann factor with β = 1/k
B
T where k
B
is Boltzmann’s
constant and T the temperature, Z is the canonical partition function, ω
ij
=
E
j
−E
i
and S
ij
(q) ≡
1
Z
e
−βE
i
['i[ ˆ ρ
q
[j`[
2
.
Notice that because atoms could be scattered by absorbing a photon from
either one of the laser beams the response of the system is determined not only
by S(q, ω) but by the combination S(q, ω
t
) − S(−q, −ω
t
). If the perturbation
is applied for a time long with respect to the frequency of the applied ﬁeld the
momentum and energy rates approach the golden rule results.
Performing the frequency integration we get
δ
d
ˆ
P
dt
¸
= 2q
V
o
2
2
¸
ij
S
ij
(q)
sin ((ω −ω
ij
)t)
ω −ω
ij
(10.19)
−S
ij
(−q)
sin ((ω +ω
ij
)t)
ω +ω
ij
−2Ωδ
ˆ
X(t)
,
δ
d
ˆ
H
dt
¸
= 2
V
o
2
2
¸
ij
ω
ij
S
ij
(q)
sin ((ω −ω
ij
)t)
ω −ω
ij
(10.20)
+S
ij
(−q)
sin ((ω +ω
ij
)t)
ω +ω
ij
).
The total momentum and energy transfer after applying the Bragg perturbation
can be obtained by integrating the above equations. We get as a ﬁnal result that
10.4 Zerotemperature regime 179
δ
ˆ
P
= q
V
o
2
¸
ij
S
ij
(q)
sin ((ω −ω
ij
)τ/2)
ω −ω
ij
2
(10.21)
−S
ij
(−q)
sin ((ω +ω
ij
)τ/2)
ω +ω
ij
2
−2Ω/a
τ
0
δ
ˆ
X(t)
dt
δ
ˆ
H
=
V
o
2
¸
ij
ω
ij
S
ij
(q)
sin ((ω −ω
ij
)τ/2)
ω −ω
ij
2
(10.22)
+S
ij
(−q)
sin ((ω +ω
ij
)τ/2)
ω +ω
ij
2
,
where τ is the duration of the perturbation.
10.4 Zerotemperature regime
10.4.1 Bogoliubov approach
In this section we use the zero temperature Bogoliubov approximation for atoms in
an optical lattice to study the dynamical structure factor in the superﬂuid regime.
Because quadratic approximations were discussed in detail in in chapter 5, we refer
the reader to this chapter for details.
Under the Bogoliubov approximation the ﬁeld operator at lattice site n is
written as a complex number z
n
plus small ﬂuctuations
ˆ
δ =
¸
M
s,=0
(u
s
n
´ α
s
−v
∗s
n
´ α
†
s
),
with ¦u
s
n
, v
s
n
¦ the quasiparticle amplitudes. Using this ansatz in Eq.(10.17) we get
an expression for the dynamical structure factor given by
S
T=0
(q, ω) = S
0
(q)δ(ω) +
¸
i
S
i
(q)δ(ω −ω
B
i
), (10.23)
S
0
(q) ≡
M
¸
n=1
[z
n
[
2
e
iqan
2
, (10.24)
S
i
(q) ≡
¸
n
(z
n
u
∗i
n
−z
∗
n
v
∗i
n
)e
iqan
2
. (10.25)
Translationally invariant system
To understand manybody eﬀects included in the Bogoliubov approximation we
start by studying the case of a translationally invariant lattice (V
n
= 0) with
N atoms, M wells and periodic boundary conditions. Using the results for the
quasiparticle amplitudes and energies calculated using the improved Popov ap
proximation, which was shown to be the best quadratic approximation we get:
180 Chapter 10 Bragg Spectroscopy
S(q, ω) = N
2
δ(ω) +Mn
o
ε
q
ω
B
q
δ(ω −ω
B
q
/) (10.26)
= N
2
δ(ω) +Mn
o
S(q)δ(ω −ω
B
q
/), (10.27)
with n
o
given by Eq. (5.136), ε
k
= 4J sin
2
(ak/2) and ω
B
k
=
ε
2
k
+ 2nUε
k
. Here n
is the total density n = N/M. Notice that due to the translational invariance of
the system, the quasimomentum is a good quantum number and only quasiparticle
states that have the same quasimomentum as the transferred momentum q are
excited.
For small q, the structure factor behaves like
J
2Un
q which is just the free
particle expression with the bare particle mass m replaced by the eﬀective mass,
m
∗
=
2
/2Ja
2
(a the lattice spacing). Suppression of S(q) in the phonon regime is
a direct consequence of the importance of quantum correlations in the long wave
length limit. On the other hand, away from the phonon regime, the presence of
the optical lattice changes the behavior of S(q) in a drastic way. As opposed to
the free case where at high transferred momenta ω
B
q
∼ q
2
and S(q) approaches
one, the periodic potential modiﬁes the single particle dispersion relation, which
becomes always bounded by 4J. The term proportional to 2nUε
k
dominates even
at high momenta and manybody eﬀects play a crucial role.
In Fig.10.2 we show the energy δ'
ˆ
H` imparted to the system as a function of
the frequency ω of the Bragg probe, for diﬀerent values of and U/J and of the
momentum q. The blue curve corresponds to the solution using the Bogoliubov
approximation and the red one to the solution obtained by the exact diagonal
ization of the BoseHubbard Hamiltonian. The number of lattice sites and wells
used in the plots were M = N = 9. For U/J = 0.001 the Bogoliubov and exact
curves perfectly overlap. The agreement is expected as the noninteracting Hamil
tonian is quadratic and therefore exactly matches the Bogoliubov Hamiltonian.
The almost non interacting regime may be experimentally achieved by varying
the onsite interaction energy U independently of J, by means of a Feshbach res
onance that changes the scattering length [20, 21]. For U/J = 1 the agreement
is good only for the low momentum q = 2π/9. The agreement for low momenta
can be understood by the fact that for low q the Bragg spectroscopy is probing
the long wavelength modes. These modes have a phononlike dispersion relation
and are almost not aﬀected by the presence of the lattice. In general Bogoliubov
quasiparticle states correspond to solutions of the approximate Hamiltonian with
a planewave character which are only valid in the weakly interacting regime. As
quantum correlations become important, the exact eigenstates of the interacting
system do not necessarily have the simple plane wave character,(see Ref. Lieb
and Liniger [149]), especially in the commensurate ﬁlling situation where for a
critical value of U/J the system exhibits the superﬂuidMott insulator transition.
However the unit ﬁlled commensurate case is the worse scenario, and as discussed
in chapter 5, for non commensurate ﬁllings or commensurate systems with larger
10.4 Zerotemperature regime 181
ﬁlling factor we expect the validity of the Bogoliubov approximation to extend
over a larger range of U/J values.
As a ﬁnal remark, it is important to point out that the small oscillations in
Fig. 10.2 are due to the square shape of the applied Bragg pulse. These oscillations
would disappear if instead a pulse with Gaussian proﬁle is applied.
0 2 4 6 8 10 12
ΩJ
0
20
40
60
80
∆
E
V
o
2
J
UJ2. q2Π9
0 2 4 6 8 10 12
ΩJ
0
200
400
600
∆
E
V
o
2
J
UJ2. q8Π9
0 2 4 6 8
ΩJ
0
20
40
60
80
100
∆
E
V
o
2
J
UJ1. q2Π9
0 2 4 6 8
ΩJ
0
200
400
600
800
∆
E
V
o
2
J
UJ1. q8Π9
0 1 2 3 4 5 6
ΩJ
0
20
40
60
80
100
∆
E
V
o
2
J
UJ0.001 q2Π9
0 1 2 3 4 5 6
ΩJ
0
200
400
600
800
∆
E
V
o
2
J
UJ0.001 q8Π9
Figure 10.2: Imparted energy: Comparisons between the the exact and Bogoliubov
approximation for N = M = 9. The horizontal axis is in units of . For the plot
we used Jτ/ = 10. See text Eq. (10.23).
10.4.2 Inhomogeneous system
Using the quasiparticle amplitudes derived under the T.F. approximation (see
section 5.6.2), we get an expression for the dynamical structure factor given by
S(q, ω) = 9N
2
sin(x) −xcos(x)
x
3
2
δ(ω)
+
ω
∗
8U
¸
i
(i(i +1)(2i + 1))[
¯
P
i
(x)[
2
δ(ω −ω
∗
i(i + 1)/2)(10.28)
where x ≡ qaR
TF
and
¯
P
n
(x) =
1
−1
P
n
(u)e
iux
du.
An alternative way to calculate the dynamical structure factor is to use the so
called local density approximation (LDA). This approximation is valid for large
condensates, where the density proﬁle varies in a smooth way and the system
behaves locally as an uniform gas. Using a LDA the dynamical structure factor
can be written as
182 Chapter 10 Bragg Spectroscopy
S
LDA
(q, ω) =
¸
m
z(m)e
iqam
δ(ω) +ε
s
¸
m
[z(m)[
2
ω
B
q
(m)
δ(ω −ω
B
q
(m)). (10.29)
Here ω
B
q
(m) is the local homogeneous Bogoliubov dispersion relation with n re
placed by the local density in the trap. The LDA ignores the Doppler eﬀect
associated with the spreading of the momentum distribution of the condensate
and therefore it is not valid for large momentum transfers.
To test the validity of the HFBPopov approximation to describe the energy
imparted to the system by Bragg spectroscopy in the trapped case we use a system
with 11 lattice sites and 9 atoms. Unfortunately for such a small number of atoms
and wells the assumptions under which the analytic Thomas Fermi expressions
were developed, i.e. large number of atoms and smooth variations of the density
proﬁle, do not apply. Therefore for the case with M = 11, N = 9 we restrict
ourselves to comparisons between numerical solutions of the HFBPopov equations
and solutions found by the exact diagonalization of the manybody Hamiltonian.
We solve numerically Eqs.(5.113) (5.117) to obtain the quasiparticle energy and
amplitudes. We then use them to calculate the imparted energy. In the presence of
the trap, the number of occupied wells for a ﬁxed number of atoms depends on the
trap frequency, particles are not necessarily uniformly distributed throughout the
lattice, and therefore commensurability is only meaningful locally. This explains
why in contrast to the commensurate homogeneous system, the agreement between
the two solutions is good up to U/J = 5 for the chosen parameters, as shown in
Fig. 10.3. The failure of the HFBPopov approximation for U/J ∼ 10 can also be
observed in the density plots, as shown in Fig. 10.4.
The multiple peaks depicted in the plots are due to the external trapping po
tential, which not only discretizes the spectrum but also changes the plane wave
character of the quasiparticle amplitudes. Therefore, for a given transfer momen
tum q diﬀerent quasiparticle modes can be excited. In Fig.10.3, we explicitly show
with a line the position of the quasiparticle energies.Similar to the homogeneous
case, the lower the transfer momentum q, the better is the agreement between the
HFBPopov results and the exact solution.
For small systems such as the one in consideration discretization eﬀects can
not be ignored. On the other hand, if the inverse of the transfer momentum is
bigger than the size of the condensate wave function, then to ﬁrst approximation
the discretization in the excitation spectrum can be safely ignored and the system
can be treated as locally uniform. Instead of distinguishable peaks the response
curve becomes smooth.
To explore the validity of Eq.(10.28) derived under the ThomasFermi approx
imation we plot in Fig. 10.5 S(q, ω) vs. ω for diﬀerent transfer momenta. The
parameters used for the plot were U/J = 0.2, Ω/J = 9.510
−4
and N = 100. For
these parameters we believe the HFBPopov should give a good description of the
system. In chapter 5 we showed that in the regime when the TF approximation is
valid, it gives a very good description of the low lying excitations. However it fails
10.5 Mott insulator regime 183
0 5 10 15 20 25
ΩJ
0
2.5
5
7.5
10
12.5
15
∆
E
V
o
2
J
UJ10 q2Π11
0 5 10 15 20 25
ΩJ
0
20
40
60
80
100
120
∆
E
V
o
2
J
UJ10 q6Π11
0 5 10 15 20
ΩJ
0
5
10
15
20
25
30
∆
E
V
o
2
J
UJ5 q2Π11
0 5 10 15 20
ΩJ
0
20
40
60
80
100
120
140
∆
E
V
o
2
J
UJ5 q6Π11
0 2 4 6 8 10 12 14
ΩJ
0
10
20
30
40
∆
E
V
o
2
J
UJ0.1 q2Π11
0 2 4 6 8 10 12 14
ΩJ
0
50
100
150
200
∆
E
V
o
2
J
UJ0.1 q6Π11
Figure 10.3: Comparisons of the imparted energy vs. Bragg frequency curves
calculated from the exact diagonalization of the manybody Hamiltonian (red) and
the from the HFBPopov approximation(blue) for a trapped system with N = 9,
M = 11 and Ω/J = 1.875. The horizontal axis is in units of . The lines depicted
in the plot are located at the diﬀerent quasiparticle energies excited by the Bragg
perturbation.
to describe high energy modes. To illustrate this issue we compare in Fig. 10.6
the TF excitation spectrum with the one found by solving the HFBPopov Equa
tions. The quasiparticle energies are labelled in increasing energy order. When the
ThomasFermi approximation is used in S(q, ω), we expect then to obtain a fair
description of it only if lowlying modes are probed. Lowlying modes are probed
if the inverse of the transfer momentum q
−1
is big compared to the size of the con
densate wave function. This is explicitly shown in Fig. 10.5, where only for small
values of q the two solutions agree. For higher Bragg momenta, qR
TF
> 1, higher
modes contribute to S(q/ω) and the TF approximation gives a poor description
of S(q, ω).
10.5 Mott insulator regime
In this section we derive expressions for the zero temperature dynamical structure
factor in the Mott regime for both the homogeneous and trapped systems and
compare them with the solutions obtained by exact diagonalization of the Bose
Hubbard Hamiltonian.
184 Chapter 10 Bragg Spectroscopy
4 2 0 2 4
n
0
1
2
3
4
5
6
D
e
n
s
i
t
y
UJ0.1
Exact
Bog.
4 2 0 2 4
n
0
0.5
1
1.5
2
2.5
D
e
n
s
i
t
y
UJ5.05
Exact
Bog.
4 2 0 2 4
n
0
0.5
1
1.5
2
D
e
n
s
i
t
y
UJ10.
Exact
Bog.
Figure 10.4: Density proﬁle calculated from the exact diagonalization of the many
body Hamiltonian (red) and from the HFBPopov approximation (blue) for a
trapped system with N = 9, M = 11 and Ω/J = 1.875. n labels the lattice
site.
10.5.1 Commensurate homogeneous system
In this section we assume a commensurately ﬁlled lattice with ﬁlling factor N/M =
g, periodic boundary conditions and a parameter regime where U J. Using the
eigenvalues and eigenstates calculated in section 9.1 with ﬁrst order perturbation
theory in the high ﬁlling factor limit, we get an expression for the dynamical
structure factor given by
S
(T=0)
(q, ω) = N
2
δ(ω)δ
q0
+
J
2
g(g +1)
U
2
(10.30)
¸
r,R
δ(ω −E
(1)
rR
)
M
¸
m=1
e
iqam
C
rR
mm+1
+C
rR
mm−1
−C
rR
m+1m
−C
rR
m−1m
2
= N
2
δ(ω)δ
q0
+
32J
2
g(g + 1)
U
2
sin
2
qa
2
¸
r
t
sin
πr
M
2
δ(ω −E
(1)
r q
)(10.31)
with qa = 2π¯ q/M, and the prime in the sum meaning that it is constrained over
the r’s with ¯ q +r even. Is important to emphasize that only the states with R = 0
10.5 Mott insulator regime 185
0.2 0.4 0.6 0.8
ΩJ
0
2.5
5
7.5
10
12.5
15
S
q
,
Ω
q 0.17 Π
0.2 0.4 0.6 0.8
ΩJ
0
5
10
15
20
S
q
,
Ω
q 0.24 Π
0.2 0.4 0.6 0.8
ΩJ
0
1
2
3
4
5
S
q
,
Ω
q 0.034 Π
0.2 0.4 0.6 0.8
ΩJ
0
2
4
6
8
10
12
14
S
q
,
Ω
q 0.1 Π
Figure 10.5: Dynamical structure factor vs. ω/J in the ThomasFermi (blue)
and HFBPopov (red) approximations. The horizontal axis is in units of . The
system parameters are U/J = 0.2, Ω/J = 9.5 10
−4
and N = 100.
have a dispersion relation which agrees to ﬁrst order in J to the meanﬁeld solution
found in Ref. [151]. However, these states give zero contribution to the sum of Eq.
(10.31).
In Fig. 10.7 we compare the imparted energy as a function of the Bragg fre
quency calculated using the expression of the dynamical structure factor Eq. (10.31)
to results obtained by the exact diagonalization of the BoseHubbard Hamiltonian
with parameters N = M = 9 and U/J = 45. We show the response for two dif
ferent Bragg momenta q = 2π/9 and 8π/9. In contrast to the superﬂuid regime,
where Bragg spectroscopy excites only the quasiparticle state with quasimomen
tum q, in the Mott regime we observe M − 1 peaks centered around U. The
diﬀerent peaks are due to the two dimensional character of the 1ph dispersion
relation. The Bragg momentum q ﬁxes one quantum number R but the other can
take M − 1 diﬀerent values. In the analytic solution due to the constraint in Eq.
(10.31), ¯ q +r even, only (M −1)/2 of the possible M −1 peaks are present. The
constraint is a consequence of the extra symmetry introduced in the high ﬁlling fac
tor solution where similar ”eﬀective masses” are assumed. Nevertheless, Fig. 10.7
shows that even in the situation with g = 1, the peaks captured by the analytic
solution are the most relevant ones. As g is increased, as long as the perturbation
analysis is valid, we expect even better agreement. The interval of frequencies
over which the system responds has an approximate width of 4J(2g +1) cos(qa/2).
The cos(qa/2) dependence indicates that as q approaches π the response width is
minimized. This behavior can be observed in Fig. 10.7, when we plot the response
186 Chapter 10 Bragg Spectroscopy
5 10 15 20
n
th
eigenvalue
0.2
0.4
0.6
0.8
1
1.2
1.4
Ω
n
J
UJ0.2 J9.5 10
4
N100
TF
BdG
Figure 10.6: Eigenvalues of the quasiparticle spectrum plotted in increasing energy
order. The vertical scale is in units of
for q = 8π/9. It is important to point out that the presence of distinguishable
peaks in Fig. 10.7 is a result of the ﬁnite system size. As the number of lattice
sites is increased, more peaks are contained in the interval 4J(2g + 1) cos(qa/2).
The separation between one peak and the next decreases and instead of individual
peaks a continuous envelope is approached.
39 43 47 50
ΩJ
10
20
30
∆
E
V
o
2
J
UJ45. q2Π9
Perturb.
Exact
39 43 47 51
ΩJ
0
30
60
90
120
∆
E
V
o
2
J
UJ45. q8Π9
Figure 10.7: Transfer energy vs. Bragg frequency for a zero temperature homoge
neous system with M = N = 9 and diﬀerent transfer momenta q. The horizontal
axis is in units of . The perturbation time was set to Jτ/ = 10. Note the change
of the vertical scale between the two panels.
10.5.2 Inhomogeneous system
In this section we consider a one dimensional optical lattice in the presence of a
magnetic conﬁnement. We assume that the magnetic trap has its minimum at the
lattice site n = 0. In the regime where U >
Ω(N−1)
2
4
and ΩN > J assuming an odd
number of atoms, we can use the eigenvalues and eigenmodes derived to ﬁrst order
10.5 Mott insulator regime 187
10 20 30 40 50 60
ΩJ
0.5
1
1.5
2
∆
E
V
o
2
J
UJ45. q2Π11
Perturb
Exact
10 20 30 40 50 60
ΩJ
4
8
12
∆
E
V
o
2
J
UJ45. q10Π11
Figure 10.8: Transfer energy vs. Bragg frequency for a zero temperature trapped
system with parameters U/J=45, Ω/J = 1.857, M = 11 and N = 9 and diﬀerent
transfer momenta q. The horizontal axis is in units of . The perturbation time
was set to Jτ/ = 10. Note the change of the vertical scale between the two
panels.
in perturbation theory in section 9.2 in the expression for the dynamical structure
factor to get:
S
(T=0)
(q, ω) = N
2
δ(ω)δ
q0
+
8J
2
sin
2
(qa/2)
(NΩ)
2
δ(ω −NΩ)
+
L−1
¸
m=−L
16J
2
sin
2
(qa/2)
(U −Ω(1 + 2m))
2
δ(ω −U + Ω(1 + 2m)), (10.32)
where L = (N − 1)/2. In Fig. 10.8 we plot the imparted energy as a function
of the Bragg frequency ω and Bragg momenta q = 2π/11, 10π/11. The system
parameters are Ω = 1.8J, U/J = 45, M = 11 and N = 9. We observe nine
peaks both in the perturbative and exact solutions. The peak at lower frequency,
NΩ ≈ 18J, is described by the ﬁrst term in Eq. (10.32) and corresponds to the
two degenerated 1hh excitations created when a hole tunnels into one of the most
externally occupied sites. The other N − 1 peaks, separated by Ω, are described
by the second term in Eq. (10.32) and correspond to the 1ph excitations with
the particle and hole at adjacent lattice sites. To explicitly show the sin(qa/2)
dependence of the response of the system, we plot the imparted energy for two
diﬀerent transfer momenta q, one close to zero the other close to π. Even in the
presence of the trap S(q, ω) tends to 0 as qapproaches an integer value of 2π.
If the number of atoms is even, the trap symmetry is broken and the ground
state becomes degenerated. To ﬁrst order in perturbation theory the dynamical
structure factor in this case is given by
188 Chapter 10 Bragg Spectroscopy
S
(T=0)
(q, ω) = 2N
2
δ(ω)δ
q0
+8J
2
sin
2
(q/2)
δ(ω −(N + 1)Ω)
((N + 1)Ω)
2
+
δ(ω −(N −1)Ω)
((N −1)Ω)
2
+ 4
N/2−1
¸
m=N/2
δ(ω −(U −Ω(1 + 2m))
(U −Ω(1 + 2m))
2
¸
. (10.33)
In this case instead of a single excitation peak at ω = NΩ we have two almost
degenerate peaks at ω = (N ±1)Ω. Apart from this, the response of the system
is analogous to the odd number case.
All the previous analysis was done for the case ΩN > J. This can be a signif
icant constraint if the total number of atoms is large. In the opposite case when
ΩN < J but still U > Ω(N/2)
2
, the energy splitting between one particle hole ex
citations is small and degenerate perturbation theory must be used at ﬁrst order.
In this case the system can be divided in a central unit ﬁlled core, which to a good
approximation has a response which can be described by the commensurate homo
geneous results, surrounded by a superﬂuid region whose excitations are mainly
due hole hopping excitations. We expect that the response to Bragg spectroscopy
of the system in this parameter regime has peaks localized around U, which probes
the one particle hole excitation band analogous to the commensurate translation
ally invariant case. We also expect lower frequency peaks around ΩN associated
with hole hopping excitations inside the superﬂuid region that surrounds the unit
ﬁlled central core.
Be aware that the parameter regime chosen for all the plots in the Mott phase
lies in the regime where ﬁrst order perturbation theory is valid and therefore peaks
at 2U due to higher order excitations are suppressed even in the exact numerical
results.
10.6 Finite temperature
In most of the experiments the way to load ultracold atoms in an optical lattice
is by ﬁrst forming a BoseEinstein condensate in a weak magnetic trap and then
adiabatically turning on the lattice by slowly ramping up the intensity of the laser
beams. Ideally this kind of process should end up with the atoms in the manybody
ground state.However, that the BoseEinstein condensate used as a starting point
is not produced exactly at T = 0, and the ﬁnal temperature is not trivially related
to the initial one. It has been shown [150] that even for an ideal non interacting
gas there are diﬀerent regimes where the atomic sample can be signiﬁcantly heated
or cooled by adiabatically changing the lattice depth. The role of interactions is
and how manybody eﬀects aﬀect the ﬁnal temperature of the sample are not yet
well understood.
Besides this issue, there remains the point of how long one must wait to be
truly adiabatic with respect to manybody excitations. Interactions are essential
for establishing equilibrium in the system, and understanding them in detail will
10.6 Finite temperature 189
be really important to determine the time scales for adiabatic loading. For a deep
lattice the tunneling rate seems to be the most restrictive time scale for maintaining
adiabaticity. If adiabaticity were determined by the tunneling time /J, it would
be very hard to be satisﬁed experimentally because J decreases exponentially with
the lattice depth.
Because of these issues, it would be very interesting from an experimental
point of view to have a mechanism for determining the temperature of the system.
What we show in this section is that Bragg spectroscopy could give information
about the temperature, at least deep in the Mott regime. As we did for the zero
temperature case, we start by describing the superﬂuid case, then the deep Mott
insulator case, and ﬁnally showing numerical calculations in the regime where the
Mott transition takes place.
• Superﬂuid Regime
At ﬁnite temperature both quantum and thermal ﬂuctuations contribute to
the depletion of the condensate. We expect the Bogoliubov approach to be a
good description of the manybody system for moderate lattice depths and
small thermal depletion. At ﬁnite temperature the occupation number of a
quasiparticle state is determined by Bose statistics, n
s
=
ˆ α
†
s
ˆ α
s
with n
s
the Bose distribution function n
s
=
1
e
βω
B
s −1
. Taking into account thermal
depletion, the expression for the dynamical structure factor is
S(q, ω) = S
0
(q)δ(ω) +
¸
i
(n
i
+1)S
i
(q)δ(ω −ω
B
i
)
+
¸
i
n
i
S
i
(−q)δ(ω +ω
B
i
) (10.34)
with S
0
(q) and S
i
(q) deﬁned in Eqs. (10.24) and (10.25).
Even though the ﬁnite temperature dynamic structure factor is in fact dif
ferent from the zero temperature one, observables such as the imparted mo
mentum and imparted energy do not depend on S(q, ω) but on S(q, ω) −
S(−q, −ω). This diﬀerence is independent of the temperature, at least at the
order of approximation at which Eq.(10.34) was calculated [15, 147, 148].
Therefore, Bogoliubov analysis shows that, provided the temperature is suf
ﬁcient low and the interactions are weak enough that Eq.(10.34) is valid,
Bragg spectroscopy is not a good method for determining the temperature
of the gas.
• Mott Regime
In the homogeneous Mott phase all the 1ph excitations have an energy sep
aration of order U from the ground state but splitting between them of order
J. If the temperature of the system is k
B
T U/5, 1ph excitations begin to
190 Chapter 10 Bragg Spectroscopy
contribute to the dynamical structure factor (Eq. (10.17)) and it starts to
develop frequency peaks at frequencies resonant with the energy diﬀerence
between two 1ph excitations. These peaks survive even when the diﬀerence,
S(q, ω) −S(−q, −ω), is taken and therefore when Bragg spectroscopy is per
formed a small frequency response is observed. In the presence of the trap the
analysis is more complicated because also nhh excitations have to be taken
into account. The relevant temperature scale in this case is k
B
T ΩN/5
and the small frequency peaks probe the energy diﬀerence between two nhh,
two 1ph or one nhh and one 1ph excitations.
In Figs. 10.9 and 10.10 we plot the imparted energy as a function of the
Bragg frequency calculated from the exact diagonalization of the BoseHubbard
Hamiltonian for diﬀerent temperatures and diﬀerent ratios of U/J. In Fig. 10.9
we use a homogeneous system with N = M = 9 and in Fig. 10.10 a trapped
system with N = 9, M = 11 at diﬀerent temperatures. The response is consistent
with the previous analysis: a response almost independent of the temperature in
the superﬂuid regime and the appearance of low energy Bragg peaks in the Mott
regime. We observe some dependence on the temperature in the intermediate
region but it is not as pronounced as the one observed deep in the Mott regime.
10.7 Conclusions
In recent experiments [145], Bragg spectroscopy was performed using a setup where
the laser beams for the Bragg perturbation were the same as those used to create
the lattice potential, and the response was observed. Previous linear response
analysis done for translationally invariant systems [148, 151]) found no scattering
when the Bragg momentum equals the lattice momentum. In ref. [151], the authors
attributed the signal observed in the experiments to nonlinear response or to eﬀects
of inhomogeneity or ﬁnite system size. Our linear response calculations considering
small trapped systems still show no scattering at q = 2π/a and therefore indicate
nonlinear response as the most plausible explanation for the experimental results.
The validity of a linear response treatment can be checked by verifying the linear
dependence of the Bragg signal upon the intensity of the Bragg beams.
In summary, we have studied the linear response of cold atoms loaded in a
one dimensional optical lattice to Bragg spectroscopy and showed that it can be
used to probe the excitation spectrum of the system, with and without harmonic
conﬁnement. In the superﬂuid regime we showed the validity of the Bogoliubov
approximation to describe the very weakly interacting regime and its breakdown
as quantum correlations become important. A new theory beyond the simple Bo
goliubov approximation is required to describe regimes beyond the very weakly
interacting one. In the Mott insulator phase we showed how Bragg spectroscopy
can be used to get information about the excitation spectrum: Bragg peaks are
centered around the characteristic Mott excitation gap, contained in an interval
whose width is proportional to the 1ph excitation band width and have an av
erage height which is maximized when the Bragg momentum approaches π/a. In
10.7 Conclusions 191
0 5 10 15 20
ΩJ
0
5
10
15
20
25
30
∆
E
V
o
2
J
2
UJ15.3 q8Π9
0.2
0.4
0.8
2.
3.4
KTJ
0 10 20 30 40 50
ΩJ
0
2
4
6
8
10
∆
E
V
o
2
J
2
UJ34.7 q8Π9
0.5
0.9
2.
4.
7.5
KTJ
0 2 4 6 8 10
ΩJ
0
20
40
60
80
100
120
140
∆
E
V
o
2
J
2
UJ1.02 q8Π9
0.02
0.04
0.08
0.2
0.33
KTJ
0 2 4 6 8 10 12
ΩJ
0
20
40
60
80
∆
E
V
o
2
J
2
UJ5.1 q8Π9
0.04
0.08
0.2
0.3
0.67
KTJ
Figure 10.9: Transfer energy as a function of the temperature for diﬀerent U/J
parameters. The plots are for a homogeneous system with N = M = 9. The
horizontal axis in in units of .
the trapped case, Bragg peaks at lower energy reveal information about 1hh ex
citations. Finally, we also discussed how Bragg spectroscopy can be an important
experimental tool to determine the temperature of the system in the Mott insulator
phase.
192 Chapter 10 Bragg Spectroscopy
0 5 10 15 20 25 30 35
ΩJ
0
10
20
30
40
∆
E
V
o
2
J
2
UJ22.5 q6Π11
0.2
0.4
0.7
1.
3.
KTJ
0 10 20 30 40 50 60
ΩJ
0
1
2
3
4
5
6
∆
E
V
o
2
J
2
UJ45. q6Π11
0.3
0.7
1.
3.
5.
KTJ
0 1 2 3 4 5
ΩJ
0
10
20
30
40
50
∆
E
V
o
2
J
2
UJ0.1 q6Π11
0.08
0.2
0.3
0.6
KTJ
0 2 4 6 8 10 12 14
ΩJ
0
5
10
15
20
25
30
35
∆
E
V
o
2
J
2
UJ5.05 q6Π11
0.07
0.1
0.3
0.6
1.
KTJ
Figure 10.10: Transfer energy as a function of the temperature for diﬀerent U/J
parameters. The plots are for an trapped system with M = 11, N = 9 and
Ω/J = 1.875. The horizontal axis in in units of .
Chapter 11
Scalable register initialization
for quantum computing in an
optical lattice
The goal of building a quantum computer has spurred tremendous progress in
coherent control and measurement of small quantum systems. Neutral atoms in
optical lattices have been proposed as a suitable candidate for quantum computing
implementation. The appeal of this system stems from the defectfree nature of
the lattice potential, and the long coherence times of the constituent atoms. In
order to fully realize the promised computational speedup of a quantum device,
the underlying system should be scalable to a large number of information carriers
or qubits. Indeed, the ﬁrst two criteria delineated by DiVincenzo [152] for scalable
quantum computation are:
• A scalable physical system with well characterized qubits
• The ability to initialize the state of the qubits to a simple ﬁducial state.
In many systems, the ﬁrst criterion can be met by increasing the number of
storage components for the qubits, e.g. in solid state systems the number of dopant
qubits in the bulk material could be increased, in optical lattices the number of
trapped atoms could be increases and in ion systems large scale microtrap arrays
have been proposed [153]. There are two main approaches to satisfy the second
criterion [152]. One is to allow the system to interact with the environment and
“naturally” cool to its ground state and thereafter use this state as the initial
state. The other is to actively cool each qubit by projective measurement to a
ﬁducial state [0`. A problem arises with these approaches when the ground state
of the many body Hamiltonian is not a suitable initial state. This is the case for
bosonic qubits embedded in systems with periodic conﬁnement, for example in
Josephson junction arrays [154], neutral atoms trapped in electromagnetic micro
traps [155] or optical lattices [46]. For these systems, the underlying dynamics is
BoseHubbardlike and the ground state contains residual coherences described by
194
Chapter 11 Scalable register initialization for quantum computing in
an optical lattice
nonzero number ﬂuctuations in each mode. As shown in Eq.(9.8), these ﬂuctua
tions scale as gJ/U
√
M, so if not corrected, the dynamics will impart a constraint
to scalability. In this chapter we show how this can be corrected in two steps,
ﬁrst by introducing an inhomogeneity to the lattice using a quadratic trapping
potential and second by projecting out components of the manybody wavefunc
tion with multiply occupied lattice sites by selective measurements on a molecular
photoassociative transition. Provided the measurement strength is suﬃciently
large, the system does not evolve out of the restricted basis and the measurement
can maintain a unit ﬁlled register for the duration of quantum computation. This
strategy allows the Mott insulator transition to become a robust mechanism for
register initialization.
11.1 Homogeneous dynamics
It was recognized early on that the Mott insulator transition might be an eﬃ
cient way to initialize a register of atomic qubits in an optical lattice for use in
quantum information processing. A key advantage of loading from a BEC is the
availability of an initially high phase space density which can be frozen to the Mott
insulator state with atoms occupying most lattice sites. For the homogeneous sys
tem (V
i
= 0), only commensurate ﬁllings give rise to a Mott insulator transition.
For the purposes of quantum computation, one particle per well is desirable. In
practice, this is diﬃcult to achieve directly because the precise number of atoms
is unknown and the lattice strength is not perfectly uniform on the boundaries.
There are proposals to prepare unit ﬁlled lattices using dissipative techniques in
volving ﬁlling the lattice atom by atom [156] or using Raman sideband cooling
[157]. Additionally, it has been shown that one can repair imperfect ﬁlling from a
BEC via an adiabatic transfer mechanism between two sublevels of each atom [48].
While these techniques can initialize the lattice, any mechanism used to prepare a
register of qubits in the unit ﬁlled state will suﬀer a degradation in ﬁdelity, deﬁned
as the population in the unit ﬁlled state. The degradation comes from the coupling
between the desired unit ﬁlled state and undesired states that pertain to the exact
Mott ground state.
In this section we derive the time dependent ﬁdelity in a homogeneous com
mensurately ﬁlled system prepared at time t = 0 in the unit ﬁlled Fock state [T`
(See Eq.(9.4)). While our results can be generalized to higher dimensions, we con
sider only a one dimension system. We assume an idealized homogeneous lattice
with periodic boundaries and unit ﬁlling, g = N/M = 1, where N is the number of
atoms and M is the number of wells. The regime of interest is the strong coupling
limit, where ﬁrst order perturbation theory is valid. This regime is experimentally
achievable in an optical lattice because the tunneling decreases exponentially with
the trap depth.
To study the time dependent ﬁdelity at ﬁrst order in perturbation theory, in
principle we have to consider all the M(M−1) eigenmodes that span the one par
ticle hole subspace. However, by translational symmetry only the translationally
11.1 Homogeneous dynamics 195
invariant states are coupled to [T`, and only they have to be considered for the
time evolution. As shown in chapter 9, the translationally invariant states inside
the one particle hole subspace are the M/2 states in the subspace, given by:
Φ
(1)
r
=
M
¸
n,m,=n
C
r
nm
[Ψ
nm
` ,
C
r
nm
=
√
2
M
sin
π(2r −1)[n −m[
M
, (11.1)
E
(1)
r
= U −6J cos
(2r −1)π
M
, r = 1, . . . (M)/2. (11.2)
The state vector at any time can be written as:
[Ψ(t)` = c
0
(t) [T` +
]M/2
¸
r=1
c
r
(t)
Φ
(1)
r
with c
0
(0) = 1. (11.3)
By solving the Schr¨odinger equation to ﬁrst order in perturbation theory, the
time dependent ﬁdelity to be in the unit ﬁlled state F(t) = [c
0
(t)[
2
can be esti
mated to be:
F(t) = 1 −
]M/2
¸
r=0
[c
r
(t)[
2
, (11.4)
[c
r
(t)[
2
≈ 64
J
E
(1)
r
2
sin
2
π(2r −1)
M
sin
2
E
(1)
r
t/
2
. (11.5)
If the number of wells M 1, the sum in Eq. (11.4) can be approximated by an
integral. In this case we get an expression for the ﬁdelity of the form:
F(t) · 1 −64
J
2
M
U
2
π
π
0
dxsin
2
(x) sin
2
(U −6J cos(x))t
2
= 1 −8
J
U
2
M
1 −
cos(Ut/)J
1
(6tJ/)
3Jt/
, (11.6)
where J
n
is the n
th
Bessel function of the ﬁrst kind.
To test the validity of the approximations, we ﬁrst compare in Fig.11.1 the per
centage ﬁdelity given by the Eq. (11.4) (red) and the numerical solution calculated
by the exact diagonalization of the BoseHubbard Hamiltonian (dots). Due to the
exponential scalability of the Hilbert space we restrict the calculations to only 5
wells. We also show comparisons with the analytic solution, Eq. (11.6) (blue). The
on site interaction strength used was U/J = 100.
We observe a very nice agreement between the perturbative and the exact
solutions. The agreement conﬁrms the validity of restricting the evolution to the
196
Chapter 11 Scalable register initialization for quantum computing in
an optical lattice
0 0.5 1 1.5 2 2.5
t J
99.2
99.4
99.6
99.8
100
%
F
i
d
e
l
i
t
y
M5, g1 ,
MJU0.02
Figure 11.1: Comparisons between the time evolution of the percentage ﬁdelity
calculated by diagonalizing the BoseHubbard Hamiltonian (dots), the perturba
tive solution inside the one particle hole subspace (red) and the analytic solution
Eq. 11.6 (blue).
one particle hole subspace. On the other hand, because of the small number of
wells used for the simulations, we do not expect Eq. (11.6) to be a very good
approximation. Nevertheless, we observe that for short times the analytic solution
is a fair description of the dynamics.
In Fig.11.2 we compare the analytic solution, Eq. (11.6) (blue) with the pertur
bative solution given by Eq. (11.4) (red) for a larger system with M = 31. Even
though for this large number of wells an exact solution is not available, we ensure
the validity of perturbation theory by choosing the same
√
MJ/U than the one in
Fig.11.1. In this M = 31 case, which lies in a regime where we expect Eq. (11.6)
to be valid, the agreement between the blue and red curves is very good.
In general the behavior of F(t) consists of fast oscillations with frequency equal
to U/, modulated by longer oscillations with frequency determined by the zeros
of J
1
. For short times, Jt/ <1, the ﬁdelity is
F(t) ≈ 1 −16
J
U
2
M sin
2
Ut/
2
, (11.7)
which corresponds to the Rabi oscillations of an eﬀective two level system spanned
by [T` and [S`, with [S` the translationally invariant state directly coupled to the
ﬁrst order ground state (see Eq. (9.9)). For later times the coupling to other
states becomes important. Notice that the time average of the ﬁdelity over many
oscillation periods is 'F(t)` = 1 −8(J/U)
2
M. Consequently, the deviation from a
11.2 Dynamics in presence of the external trap 197
0 0.5 1 1.5 2
t J
99.2
99.4
99.6
99.8
100
%
F
i
d
e
l
i
t
y
M31, g1 ,
MJU0.02
Figure 11.2: Percentage ﬁdelity as a function of time calculated using the pertur
bative solution 11.4 (red) and the analytic solution Eq. 11.6 (blue).The number
of sites used for the plot is M = 31
perfectly prepared register, described by 1−'F(t)`, is twice as bad as if the system
were prepared in the stationary ground state of the BoseHubbard Hamiltonian
(Eq. (9.8)). This indicates that a lattice ﬁlled with a commensurate number of
atoms in the Mott insulator state may create a more robust quantum computer
register than one prepared dissipatively.
11.2 Dynamics in presence of the external trap
As mentioned before, in practice it is diﬃcult to prepare an optical lattice with
exactly one atom per well. To arrange the desired conﬁguration, we propose to
use an inhomogeneous lattice with open boundaries created by a weak quadratic
magnetic trap. We require fewer number of atoms than available sites, N < M.
The addition of the trap acts to collect atoms near the potential minimum and
leaves empty wells (holes) at the edges. In experiments where the optical lattice is
loaded from a BEC, the external trap is already present to conﬁne the condensate.
For simplicity we assume a one dimensional trap with oscillation frequency given
by ω
T
. The magnetic conﬁnement introduces a characteristic energy scale Ω =
m/2a
2
ω
T
, so that V
n
= Ωn
2
(see Eq. (6.1)). To inhibit multiple atom occupation
in any well in the ground state conﬁguration, we require the on site interaction
energy U to be larger than the trapping energy of the most externally trapped
atoms, U >
Ω(N−1)
2
4
.
198
Chapter 11 Scalable register initialization for quantum computing in
an optical lattice
Figure 11.3: Schematic of an inhomogeneous lattice ﬁlled with N qubits with an
onsite interaction energy U. An externally applied trapping potential of strength
V
j
= Ωj
2
, e.g. due to a magnetic ﬁeld, acts to ﬁll gaps in the central region of the
trap. The center subspace { of the lattice deﬁnes the quantum computer register
containing K < N qubits.
We deﬁne our register as the subspace { comprising K < N wells in the
center region of the the trap (see Fig. 11.3). The barrier space ﬂanking { will
act to suppress percolation of holes from the edges to the center. The esti
mated probability for holes in { due to tunneling through the barrier is p
h
≈
¸
(N+1)/2
j=(K−1)/2
(J/Ω(2j + 1))
2
= (J/2Ω)
N−K+4
(Γ[K/2]/Γ[N/2 + 1])
2
, which is negli
gible provided the barrier region is suﬃciently large and J/KΩ < 1.
Hereafter, we restrict our attention to the dynamics of the reduced state of the
register obtained by the tracing over spatial modes outside the register subspace
of the entire many body wavefunction. The degree of inhomogeneity is quantiﬁed
by the ratio Ω/J. For 0 < Ω/J < 1, the energy splitting between Fock states
describing particlehole pairs is small and the model of homogeneous dynamics is
valid.
For Ω/J 1, to ﬁrst order in perturbation theory, the state space is spanned
by the unit ﬁlled state in the register subspace, [T`
1
, and the 2K nearest neighbor
particlehole pairs [S
±
j
`:
[T`
1
= [1, 1, 1, . . . , , 1, 1, 1`
. .. .
K sites
, (11.8)
[S
+
j
`
1
= [1, . . . , 2
....
j
, 0
....
j+1
, . . . , 1`. [j[ < (K −1)/2, (11.9)
[S
−
j
`
1
= [1, . . . , 2
....
j−1
, 0
....
j
, . . . , 1`. (11.10)
Here we have introduced coordinates with the site j = 0 coincident with the trap
minimum. For each j the states [S
±
j
`
1
are distinguished by the two energetically
distinct orientations of a doubly occupied site and its neighboring hole with ener
gies, E(S
±
j
) = U(1 ∓
Ω
U
(2j −1)). We deﬁne the zero of energy coincident with the
state [T`
1
.
11.2 Dynamics in presence of the external trap 199
With an eye to the projective measurement, it is important to understand the
free dynamics when the trap is present and the register is initialized in the pure
state [T`
1
inside the register,
1
'T[Ψ(0)`
1
= 1. Here [Ψ(0)`
1
is the projection
of the initial many body state [Ψ(0)` in the register subspace, i.e. [Ψ(0)`
1
=
[{`'{[Ψ(0)` with [{`'{[ the projection operator that traces over spatial modes
outside the register. We proceed to estimate the ﬁdelity to be in the unit ﬁlled
state in the register: F
1
(t) = [
1
'T[{`'[{[Ψ(t)`[
2
≡ [
1
'T[Ψ(t)`
1
[.
The quantity F
1
(t) is generally diﬃcult to compute because atoms couple into
and out of the register. Instead, we will ﬁrst solve the simpler problem of the
ﬁdelity to be in the target state of a commensurately ﬁlled inhomogeneous lattice
with K sites. We will use this ﬁdelity to estimate F
1
(t). In the commensurately
ﬁlled inhomogeneous system, to ﬁrst order in J/U, the dynamics can be restricted
to the Hilbert space spanned by the state vectors ¦[T`
1
, [S
±
j
`
1
¦, and therefore the
time dependent state can be written as :
[Ψ(t)`
com
= c
T
(t)[T`
1
+
¸
j,±
e
−iE(S
±
j
)t/
c
S
±
j
(t)[S
±
j
`
1
. (11.11)
The evolution of the amplitudes ¦c
T
(t), c
S
±
j
(t)¦ is dictated by the following equa
tions of motion:
i
∂
∂t
c
T
(t) = −J
√
2
¸
j,±
e
−iE(S
±
j
)t/
c
S
±
j
(t), (11.12)
i
∂
∂t
c
S
±
j
(t) = −J
√
2e
−iE(S
±
j
)t/
c
T
(t). (11.13)
Using the fact that J/U <1, for times Jt we can set the value of c
T
(t) in Eq.
(11.13) to 1. Integrating the equation we obtain
[c
S
±
j
(t)[
2
= 8
J
U
2
sin
2
E(S
±
j
)t
2
, (11.14)
[c
T
(t)[
2
= 1 −8
J
U
2
¸
j,±
c
S
±
j
(t)
2
. (11.15)
After performing the sum, the ﬁdelity F
com
(K, t) ≡ [c
T
(t)[
2
is :
F
com
(K, t) ≈ 1 −8
J
2
U
2
K −cos(Ut/)
1 +
sin(Ω(K −1)t/)
sin(Ωt/)
. (11.16)
Notice that for Ωt →0, we recover the short time ﬁdelity in the homogeneous case,
Eq. (11.7).
Once F
com
(K, t) is calculated, we can bound the ﬁdelity F
1
(t) inside the reg
ister. Provided N > K the following inequalities on the time averaged ﬁdelities
hold:
'F
com
(N, t)` ≤ 'F
1
(t)` ≤ 'F
com
(K, t)`. (11.17)
200
Chapter 11 Scalable register initialization for quantum computing in
an optical lattice
The lower bound arises because the probability to be in the unit ﬁlled state of
a large commensurately ﬁlled lattice, given an initial state which is unit ﬁlled, is
always less than or equal to the probability to be unit ﬁlled over a smaller subspace
of K < N of a non commensurately ﬁlled lattice whose register is prepared in the
unit ﬁlled state. This inequality holds provided the probability for holes to tunnel
into the register { is small over relevant time scales. The upper bound is a
consequence of the fact that unit ﬁlling in the register state is degraded because
particles can tunnel in and out of the register. Therefore its ﬁdelity is less than
that of a commensurately ﬁlled lattice of the same size K prepared in the unit
ﬁlled state.
0 0.5 1 1.5 2
t J
99.2
99.4
99.6
99.8
100
%
F
c
o
m
K
5
,
t
g1 ,UJ100, J2.7
Figure 11.4: Comparisons between the time evolution of the percentage ﬁdelity
F
com
(K = 5, t) calculated by evolving the initial target state using the Bose
Hubbard Hamiltonian (dots), the restricted one particle hole basis (red) and the
analytic solution Eq. 11.16 (blue). In the plot we assumed a commensurate unit
ﬁlled lattice with ﬁve sites and inﬁnitely high boundaries.
In Fig. 11.4 we compare this analytic solution, Eq. 11.16 (blue), with the solutions
found by propagating the initial state using the exact BoseHubbard Hamiltonian
(dots) and by using a restricted basis set of dimension K(K − 1) + 1 consisting
of the target state and all particlehole pairs (red). Inﬁnitely high boundaries
were assumed for the numerical solutions. The parameters used for the plot were
U/J = 100,Ω/J = 2.7 and K = 5. We observe complete agreement between the
exact points and the red curve, which justiﬁes the restriction of the dynamics to
the oneparticle hole subspace. On the other hand, the analytic approximation
accurately reproduces the exact dynamics only for times t 0.5/J. This is not
unexpected because the analytic solution only takes into account the states with
11.3 Measurement 201
0 0.5 1 1.5 2
t J
99.2
99.4
99.6
99.8
100
%
F
c
o
m
K
3
1
,
t
g1 UJ250 J0.675
Figure 11.5: Percentage ﬁdelity as a function time calculated using the approxi
mated solution 11.16 (blue) and the ﬁdelity found numerically by restricting the
dynamics to the one particle hole subspace. We assumed inﬁnity high boundaries
and K = 31.
no separation between the atomic pair and the hole which are the ones coupled to
ﬁrst order in J to the ground state and sets c
T
(t) in Eq. (11.13) to one. In Fig.
11.5 we plot the ﬁdelity for a lattice with more sites, K = 31. Because the Hilbert
space is too large to have an exact solution, we decrease J to get the same J
√
K/U
ratio as the one used in Fig. 11.4. In this way we can be sure of the validity of
the solution constrained to the oneparticle hole subspace which is plotted in red.
We observe for this parameter regime that the analytic solution (blue) gives a very
good description of the quantum dynamics. This is consistent with the fact that
as the ratio J/U is decreased, the role in the dynamics of other modes diﬀerent
from the [S
±
j
`
1
modes becomes less important.
11.3 Measurement
In Sec. 11.2 we showed that the system dynamics can be restricted to a set of
two level couplings ¦[T`
1
→ [S
±
j
`¦. We now sketch how to perform a continuous
measurement to drive the register into the unit ﬁlled “target” state [T`
1
. The full
details are contained in [158]. The idea is to apply an external control ﬁeld that is
resonant with a coupling between the “faulty” states ¦[S
±
j
`¦ and a set of excited
states ¦[M
±
j
`
1
¦. These states are described by K−2 atoms trapped in the lattice
and a dipoledipole molecular S+P state at site j+(1∓1)/2. The bound molecular
202
Chapter 11 Scalable register initialization for quantum computing in
an optical lattice
T
{S
±
j
}
√
2J
0
−
g
(P
3/2
)
{M
±
j
}
∆
γ
M
S +P
3/2
E(S
±
j
) + V
c

Ω
M
Figure 11.6: Schematic of the relevant couplings in the problem. The unit ﬁlled
state [T` describing a target quantum register and the states [S
±
j
` having one
doubly occupied lattice site and a neighboring hole are coupled to ﬁrst order in J.
A catalysis laser resonantly couples the ground states [S
±
j
` to the excited states
[M
±
j
` describing a bound molecule at the doubly occupied site. The bound states
quickly decay and give the possibility of monitoring population in the “faulty”
register states [S
±
j
`.
state is chosen such that the catalysis laser is far oﬀ resonance from other bound
states and repulsive potentials, see Fig. 11.6. If the population in the excited states
is easily measurable, for instance by the emission of photons during decay, then the
presence of population in the particlehole states can be monitored. It is vital that
the coupling ﬁeld be able to spectroscopically resolve the measurement transition
without exciting the target state. This is possible if the atoms in multiply occupied
wells see a shifted excited state, E(M
±
j
) = E(S
±
j
) +E −U, where for example E
is the energy of a dipoledipole molecular state. The “bare” energy Hamiltonian
of the system is:
H
0
+H
BH
=
¸
j,±
E(S
±
j
)[S
±
j
`
11
'S
±
j
[ +E(M
±
j
)[M
±
j
`
11
'M
±
j
[
−
√
2J
¸
j,±
([S
±
j
`
11
'T[ +[T`
11
'S
±
j
[).
(11.18)
When the catalysis laser is turned on, for the bound states of interest, such
as the long range bound states of the 0
−
g
(P
3/2
) potential [159], the detuning from
atomic resonances is several thousands of linewidths, meaning the atomic satura
tion is low. In this case, the excited atomic states can be adiabatically eliminated
and each atom in a singly occupied well experiences a light shift equal to V
c
. Be
cause there are K singly occupied wells in the [T` state its total single atom light
shift is equal to KV
c
. The [S
±
j
` states have K − 2 singly occupied wells giving
a corresponding light shift of (K − 2)V
c
. The diﬀerential single atom light shift
between these states is then 2[V
c
[. Therefore, when the control Hamiltonian with
11.3 Measurement 203
Figure 11.7: Population in the unit ﬁlled register state [T`
1
during continuous
measurement of the register beginning in the BoseHubbard ground state [Ψ
g
`. The
plots show dynamics appropriate to tunneling in one dimension with U/J = 500.
(a) Quantum trajectories corresponding to a null measurement result for three
diﬀerent register sizes K. The time scale to saturate the target state is independent
of the number of qubits: t
sat
≈ κ
−1
. (b) Long time dynamics for K = 501,
N = 551 and ﬁnite detector eﬃciencies η. The population in [T`
1
for η = 1 is
indistinguishable from one. Also shown is the oscillatory dynamics at fundamental
frequency U described by Eq. (11.16) if the measurement is turned oﬀ after the
target state is reached. The arrow indicates ρ
T,T
(0).
Rabi frequency Ω
M
is turned on resonant, the total Hamiltonian in the rotating
frame is:
H
I
=
¸
j,±
(2[V
c
[ +E(S
±
j
))[S
±
j
`
11
'S
±
j
[ + ([2V
c
[ +E(S
±
j
) −U)[M
±
j
`
11
'M
±
j
[
−
√
2J([S
±
j
`
11
'T[ +[T`
11
'S
±
j
[) +Ω
M
/2([M
±
j
`
11
'S
±
j
[ +[S
±
j
`
11
'M
±
j
[)),
(11.19)
Any population in the bound molecular states will decay at a rate γ
M
≈ 2Γ,
where Γ is the single atom decay rate. For molecular photoassociation by red
detuned light, the decay products are typically ground state molecular species or
“hot” atoms meaning the atoms escape the trapped ground states described by Eq.
(11.10). It is possible then to model the system according to a trace nonpreserving
master equation:
˙ ρ = −i/[H
I
, ρ] −γ
M
/2
¸
j,±
([M
±
j
`
11
'M
±
j
[ρ +ρ[M
±
j
`
11
'M
±
j
[). (11.20)
204
Chapter 11 Scalable register initialization for quantum computing in
an optical lattice
Assuming low saturation of the excited states, the dynamics in the ground state is
˙ ρ
S
±
j
,T
= −iρ
S
±
j
,T
(E(S
±
j
) +[V
c
[)/ +i(ρ
T,T
−ρ
S
±
j
,S
±
j
)
√
2J/ −ρ
S
±
j
,T
κ
˙ ρ
T,T
= i(ρ
S
±
j
,T
−ρ
T,S
±
j
)
√
2J/
˙ ρ
S
±
j
,S
±
j
= −i(ρ
S
±
j
,T
−ρ
T,S
±
j
)
√
2J/ −2ρ
S
±
j
,S
±
j
κ.
(11.21)
These equations describe the BoseHubbard coupled states with a decay in
population of each state with an atomic pair at a rate 2κ = Ω
M
2
γ
M
/(4(U/)
2
+
(γ
M
/2)
2
), and decay of coherences between each of these states and state [T` at a
rate κ.
This type of evolution characterized by measurement induced phase damping
was studied extensively by Gagen and Milburn [160]. We now show that our system
can satisfy the conditions for this eﬀect and in particular can be driven to the [T`
state by monitoring the environment for a signature of decay from the molecular
bound state.
For the inhomogeneous system, the state [T` couples to 2K distinguishable
states [S
±
j
`. However, we can deﬁne an eﬀective Rabi frequency between the state
[T` and the subspace spanned by [S
±
J
`. This frequency is close to the coupling
matrix element between the state [T` and the state [S` in the homogeneous system,
namely 2
√
KJ. The coherences between the two subspaces decay at a rate κ,
and the population in the subspace ¦[S
±
j
`¦ decays at a rate 2κ. The “good”
measurement regime as derived in [160] is then:
Ω
M
/γ
M
<1 < κ/2
√
KJ. (11.22)
The left side inequality ensures that the excited states [M
±
j
` are weakly pop
ulated (equivalent to the condition for adiabatic elimination of these states). The
right side inequality ensures that measurement is suﬃciently strong to damp co
herences on the time scale that they develop due to tunneling.
When the environment is monitored, for instance by looking for photon scatter
ing from the bound molecular state, the evolution of ground states can be modelled
using quantum trajectories. The success or failure of the preparation is conclusive
with failure probability p
fail
= 1 −ρ
T,T
(0). Real experiments will be constrained
to ﬁnite detector eﬃciencies η. For η = 0, corresponding to nonselective measure
ment, the system dynamics evolve according to Eq. (11.21). In the case of ﬁnite
detector eﬃciencies, we can express the approximate ﬁdelity to be in the target
state is F(η, t) = ρ
T,T
(t) = η + (1 − η)ρ
η=0
T,T
(t). Simulations of successful register
preparation and maintenance by measurement with diﬀerent eﬃciencies are shown
in Fig. 11.7.
It is necessary to keep the measurement on during a computation to maintain
high ﬁdelity in the unit ﬁlled state. If instead the catalysis ﬁeld is turned oﬀ after
the target state is reached, the system will freely evolve as shown in Sec. 11.2.
To illustrate this we show in Fig. 11.7 the evolution of the ﬁdelity described by
Eq. (11.16) if the measurement is turned oﬀ after the target state is reached.
11.4 Conclusions 205
11.4 Conclusions
In summary, we have shown that eﬀorts to prepare a register of atomic qubits in an
optical lattice suﬀer from errors inherent in the underlying many body dynamics.
By considering the free dynamics in both homogeneous and trapped systems we
have shown that there is a loss of ﬁdelity of initialization of the unit ﬁlled regis
ter which scales with the number of qubits because it is not a stationary state.
Because our analysis is based on ﬁrst order perturbation theory we established
the parameter regime where our analytic approach is valid by comparing it with
exact numerical results for moderate numbers of atoms and wells. To make the
Mott insulator transition a robust mechanism for initialization, we have suggested
one approach to correct for the mentioned problems by performing a continuous
measurement on the system. While our discussion has focused on one dimensional
dynamics, the method is also applicable to higher dimensions, which is the relevant
regime for scalability.
206
Chapter 11 Scalable register initialization for quantum computing in
an optical lattice
Chapter 12
Conlusions
This thesis gives a global understanding of the basic physics that describes bosonic
cold atoms in optical lattices, starting from the superﬂuid regime and going into
the strongly correlated Mott insulator regime. In the following I am going to
summarize what I consider are the most relevant contributions of this work.
• The equilibrium properties of lattice systems in the superﬂuid regime were
studied by using standard mean ﬁeld techniques and quadratic approxima
tions.
• An explicit expression for the superﬂuid density based on the rigidity of the
system under phase variations was derived. This expression enabled us to
explore the connection between the quantum depletion of the condensate
and the quasimomentum distribution on the one hand, and the superﬂuid
fraction on the other. Also, the superﬂuid fraction was shown to be a natural
order parameter to describe the superﬂuid to Mott insulator transition.
• A functional eﬀective action approach, the 2PICTP formalism, capable of
dealing with nonequilibrium situations that require a treatment beyond mean
ﬁeld theory, was studied. Using the patterned loading system, this formalism
was shown to be a powerful tool to go beyond the HFB approximation and
to incorporate the nonlocal and nonMarkovian aspects characteristic of the
quantum dynamics.
• It was shown that the complicated nonlocal and nonMarkovian solutions
derived using the 2PICTP formalism reduce to the standard kinetic theory
equations when the system dynamics admits twotime separation.
• Bragg spectroscopy was shown to be a suitable experimental tool to char
acterizing the Mott insulator phase and to estimate the temperature of the
system deep in the Mott insulator regime.
• It was shown that the use of a lattice with a spatial inhomogeneity created
by a quadratic magnetic trapping potential, together with a selective mea
208 Chapter 12 Conlusions
surement of atomic pairs allow for the Mott insulator transition to become
a robust mechanism for quantum register initialization.
In the recent years there has been spectacular progress in experimental and the
oretical studies of atoms loaded into an optical lattice. However, there are still
many ideas and possible experiments that could help us to address fundamen
tal questions in solid state physics, atomic physics, quantum optics and quantum
information. I want to ﬁnish this work by mentioning some of them.
One can use for example multicomponent ultracold atoms in optical lattices
together with Feshbach resonances to realize Hamiltonians other than the pure
BoseHubbard one. Also by loading fermionic atoms into an optical lattice the
Hubbard Hamiltonian could be realized. This could help to answer many theo
retical questions still open in fermionic systems, such as the BECBCS crossover
and the basic physics behind high temperature superconductivity. From the per
spective of atomic and molecular physics an interesting experiment could be the
use of the Mott insulator state with two atoms per site as a mean to create a
molecular condensate. The creation of vortices in individual lattice sites could
also allow the study of the integer and fractional quantum Hall eﬀects in ultracold
gases. By introducing noise in a controlled way in optical lattice, one can also
study disordered periodic systems, which are very important in condensed matter
physics. And ﬁnally and perhaps one of the most challenging tasks is to develop
the necessary techniques and protocols that allow neutral atoms in optical lattices
as a mean to implement a quantum computer.
Bibliography
[1] A. Einstein, Sitzungsbericht der Preussischein Akademie der Wissenschaften,
p. 3,(1925).
[2] S. N. Bose Z. Phys 26, 178(1924).
[3] F. London, Nature 141, 643 (1938); Phys. Rev 54, 947 (1938).
[4] E. C. Svensson and V. F. Sears, Progress of low temperature physics, Vol XI,
ed. D.F. Brewer, (North Holland,Amsterdam,187).
[5] I. F. Silvera and J. T. M. Walraven, Phys. Rev. Lett 44, 164 (1980).
[6] M. H. Anderson, J. R. Ensher, M. R. Matthewes, C. E. Wieman and E. A.
Cornell Science 269, 198 (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] C. C. Bradley, C. A. Sackett, J. J. Tollet and R. Hulet, Phys. Rev. Lett. 78,
985 (1997).
[9] C. J. Foot, Contemp. Phys. 32, 369 (1991).
[10] E. P. Gross, Nuovo Cimiento, 20, 454 (1961); L. P. Pitaevskii, Sov. Phys.
JETP 13, 451 (1961).
[11] N. N. Bogoliubov, J. Phys. (USSR) 11, 23 (1947).
[12] F. Dalfovo, S. Giorgini, L. P. Pitaevskii and S. Stringari, Theory of Bose
Einstein condensation in trapped gases, Rev. Mod. Phys. 71, 463 (1999).
[13] D. A. W. Hutchinson et al., J. Phys. B: At. Mol. Opt. Phys. 57 3825 (2000).
[14] N.P. Proukakis, S.A. Morgan, S. Choi and K. Burnett, Phys. Rev. A 58, 2435
(1998).
[15] H. Shi and A. Griﬃn, Phys. Rep. 304, 1 (1998) and A. Griﬃn, Phys. Rev. B,
53, 9341 (1995).
210 BIBLIOGRAPHY
[16] S. A. Morgan A gapless theory of BoseEinstein Condensation in Dilute Gases
at Finite Temperature, D. Phil. Thesis, Oxford University (1999), S. A. Mor
gan, J. Phys. B: At. Mol. Opt. Phys. 57, 3847 (2000).
[17] D. S. Jin, J. R. Ensher, M. R. Matthews, C. E. Wieman and E. A. Cornell,
Phys. Rev. Lett. 77, 420 (1996).
[18] S. Stringari Phys. Rev. Lett. 77, 2360 (1996); F. Dalfovo, S. Giorgini, M.
Guilleumas, L. Pitaevskii, and S. Stringari, Phys. Rev. A 56, 3840 (1997).
[19] M. Edwards, R. J. Dodd, C. W. Clark, P. A. Ruprecht and K. Burnett, Phys.
Rev. A 53, R1950 (1996); M. Edwards, P. A. Ruprecht, K. Burnett, R. J.
Dodd and C. W. Clark, Phys. Rev. Lett. 77, 1671 (1996); M. Rusch and K.
Burnett, Phys. Rev. A 59, 3851 (1999); D. A. W. Hutchinson, E. Zaremba and
A. Griﬃn, Phys. Rev. Lett. 78, 1842 (1997); L. P. Pitaevskii and S. Stringari,
Phys. Lett. A 235, 398 (1997); R. J. Dodd, M. Edwards, C. W. Clark and K.
Burnett, Phys. Rev. A, 57, R32 (1998).
[20] S. L. Cornish, N. R. Claussen, J. L. Roberts, E. A. Cornell, and C. E. Wieman
Phys. Rev. Lett. 85, 1795 (2000).
[21] Elizabeth A. Donley, Neil R. Claussen, Simon L. Cornish, Jacob L. Roberts,
Eric A. Cornell, Carl E. Wieman, Nature 412, 295 (19 Jul 2001).
[22] P. O. Fedichev, M. W. Reynolds, and G. V. Shlyapnikov, Phys. Rev. Lett.
77, 2921 (1996).
[23] Phillip L. Gould, George A. Ruﬀ, and David E. Pritchard, Phys. Rev. Lett.
56, 827 (1986).
[24] P. J. Martin, Bruce G. Oldaker, Andrew H. Miklich, and David E. Pritchard,
Phys. Rev. Lett. 60, 515 (1988).
[25] C.S. Adams, M. Sigel and J. Mlynek, Phys. Rep 240, 143 (1994).
[26] A. Hemmerich and T. W. H¨ansch, Phys. Rev. Lett. 70, 410 (1993).
[27] A. Hemmerich, M. Weidem¨ uller, T. Esslinger, C. Zimmermann, and T. Hn
sch,Phys. Rev. Lett. 75, 37 (1995).
[28] M. Weidemller, A. Hemmerich, A. G¨orlitz, T. Esslinger, and T. W. H¨ansch,
Phys. Rev. Lett. 75, 4583 (1995).
[29] G. Grynberg, B. Lounis, P. Verkerk, J. Y. Courtois, and C. Salomon, Phys.
Rev. Lett. 70, 2249 (1993).
[30] Vladan Vuletic, Cheng Chin, Andrew J. Kerman, and Steven Chu, Phys. Rev.
Lett. 81, 5768 (1998).
BIBLIOGRAPHY 211
[31] Andrew J. Kerman, Vladan Vuletic, Cheng Chin, and Steven Chu, Phys. Rev.
Lett. 84, 439442 (2000).
[32] Marshall T. DePue, Colin McCormick, S. Lukman Winoto, Steven Oliver,
and David S. Weiss, Phys. Rev. Lett. 82, 2262 (1999).
[33] S. Peil, J. V. Porto, B. L. Tolra et. al, Phys. Rev. A 67, 051603 (2003).
[34] P. Blair Blakie and Charles W. Clark, J. Phys. B: At. Mol. Opt. Phys. 37
1391 (2004).
[35] M. Greiner, I. Bloch, O. Mandel, T. W. H¨ansch, and T. Esslinger, Phys. Rev.
Lett. 87, 160405 (2001).
[36] J. H. Denschlag, J. E. Simsarian, H. H¨aﬀner, C. McKenzie, A. Browaeys,
D. Cho, K. Helmerson, S. L. Rolston, and W. D. Phillips, J. Phys. B 35, 3095
(2002).
[37] F. Ferlaino, P. Maddaloni, S. Burger, F. S. Cataliotti, C. Fort, M. Modugno,
and M. Inguscio Phys. Rev. A 66, 011604 (2002).
[38] O. Morsch, J. H. Muller, M. Cristiani, D. Ciampini, and E. Arimondo, Phys.
Rev. Lett. 87, 140402 (2001).
[39] B. Anderson and M. Kasevich, Science 282, 1686 (1998).
[40] O. Morsch, M. Cristiani, J. H. Muller, D. Ciampini, and E. Arimondo, Phys.
Rev. A 66, 021601 (2002).
[41] S. Burger, F. S. Cataliotti, C. Fort, F. Minardi, M. Inguscio, M. L. Chiofalo,
and M. P. Tosi, Phys. Rev. Lett. 86, 4447 (2001).
[42] W. K. Hensinger, H. H¨aﬀner, A. Browaeys, N. R. Heckenberg, K. Helmerson,
C. McKenzie, G. J. Milburn, W. D. Phillips, S. L. Rolston, H. Rubinsztein
Dunlop, Nature 412, 52 (2001).
[43] D. Jaksch, C. Bruder, C. W. Gardiner, J.I. Cirac and P. Zoller, Phys. Rev.
Lett. 81, 3108 (1998).
[44] M. P. A. Fisher, P. B. Weichman, G. Grinstein and D. S. Fisher, Phys. Rev.
B 40, 546 (1989).
[45] C. Orzel, A. K. Tuchman, M. L. Fenselau, M. Yasuda, and M. A. Kasevich,
Science 23, 2386 (2001).
[46] M. Greiner, O. Mandel, T. Esslinger, T. W. H¨ansch, and I. Bloch, Nature
415, 39 (2002).
[47] Markus Greiner, Olaf Mandel, Theodor W. H¨ansch and Immanuel Bloch,
Nature 419, 51 (2002).
212 BIBLIOGRAPHY
[48] P. Rabl et al., Phys. Rev. Lett. 91, 110403 (2003).
[49] I. Deutsch et al., Phys. Rev. A 57, 1972 (1998).
[50] T. Calarco et al., J. Mod. Opt. 47, 2137 (2000).
[51] A. L. Fetter and J. D. Walecka,Quantum Theory of Many Particle Systems,
(McGrawHill, New York, 1971).
[52] V. N. Popov, Functional Integrals in Quantum Field Theory and Statistical
Physics (Reidel, Dordrecht, 1983).
[53] M. Bijlsma and H. T. C. Stoof, Phys. Rev. A 55, 498 (1997); U. Al Khawaja,
J. O. Andersen, N. P. Proukakis, and H. T. C. Stoof Phys. Rev. A 66, 013615
(2002).
[54] H. T. C. Stoof,in Coherent Matter Waves, proceedings of the summer Les
Houches Lectures, session 72,1999, (SprinderVerlag, Berlin, 2001).
[55] Emil Lundl and Jorgen Rammer, Phys. Rev. A 66, 033607 (2002).
[56] J. S. Schwinger, J. Math. Phys. 2, 407 (1961); L. V. Keldysh, Zh. Eksp. Teor.
Fiz. 47 , 1515 (1964) [Engl. trans. Sov. JETP 20 1018 (1965)]; V. Korenman,
Ann. Phys. 39, 72 (1966); G. Zhou, Z. Su, B. Hao and L. Yu, Phys. Rep. 118,
1 (1985).
[57] J. Luttinger and J. Ward, Phys. Rev. 118, 1417 (1960); C. de Dominicis
and P. Martin, J. Math. Phys. 5, 14 (1964); H.D. Dahmen and G. Jona
Lasino, Nuovo Cimiento, 52A, 807 (1962); J.M. Cornwall, R. Jackiw and E.
Tomboulis, Phys. Rev. D 10 2428 (1974).
[58] R. Walser, J. Williams, M. J. Holland et al., Phys. Rev. A 59, 3878(1999); R.
Walser, J. Cooper and M. Holland, Phys. Rev. A 63, 013607 (2000).
[59] J. Wachter, R. Walser, J. Cooper and M. Holland, Phys. Rev. A 64,
053612(2001).
[60] M. ImamoviTomasovi and A. Griﬃn, in Progress in Nonequilibrium Green’s
Functions, edited by M. Bonitz (World Scientiﬁc, Singapore, 2000), p. 404;
M. ImamoviTomosovi and A. Griﬃn, Phys. Rev. A 60, 494 (1999).
[61] S. G. Bhongale, R. Walser and M. J. Holland Phys. Rev. A 66, 043618 (2002).
[62] J. Wachter, R. Walser, J. Cooper, and M. Holland Phys. Rev. A 65,
039904 (2002); J. Wachter, R. Walser, J. Cooper, M. Holland, arXiv:cond
mat/0212432.
[63] S. T. Beliaev, Zh. Eksp. Teor. Fiz. 34, 433 (1958) [Sov. Phys. JETP 7, 299
(1958)].
BIBLIOGRAPHY 213
[64] B. Paredes et al., Nature 429, 277 (2004).
[65] J. Stenger et al., Phys. Rev. Lett. 82, 4569 (1999).
[66] D. StamperKurn et al., Phys. Rev. Lett. 83, 2876 (1999).
[67] N. W. Ashcroft and N. D. Mermin, Solid State Physics, W. B. Saunders
Company (1976).
[68] J. M. Ziman, Principles of the Theory of Solids, Cambridge University Press,
(1964).
[69] I. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions (Na
tional Bureau of Standards, 1964).
[70] Hidestoshi Fukuyama, Phys. Rev. B, 8, 5579 (1973).
[71] G. Wannier, Phys. Rev. 17, 432 (1960).
[72] Gerhard Fasol, Spectroscopy of Semiconductor Microstructures, (NATO ASI
Series, 1989, p. 333)
[73] M. BenDahan, E. Peik, J. Reichel, Y. Castin, and C. Salomon, Phys. Rev.
Lett. 76, 4508 (1996).
[74] W. Krauth, et al. Phys. Rev. B 45, 3137 (1992).
[75] K. Seshadri et al, Europhys. Lett. 22, 257 (1993).
[76] L. Amico and V. Penna, Phys. Rev. B 62, 1224 (2000).
[77] D. van Oosten, P. van der Straten, and H. T. C. Stoof, Phys. Rev. A 63,
053601 (2001).
[78] J. K. Freericks and H. Monien, Phys. Rev. B 53, 2691 (1996).
[79] G. G. Batrouni et al. Phys. Rev. Lett. 65, 1765 (1990); R. T. Scalettar et
al., Phys. Rev. Lett. 66, 3144 (1991); P. Niyaz et al., Phys. Rev. B. 44, 7143
(1991); R. T. Scalettar et al., Phys. Rev. Lett. 74, 2527 (1995); G. G. Batrouni
et al., Phys. Rev. Lett. 89, 117203 (2002).
[80] P. W. Anderson, Rev. Mod. Phys. 38, 298 (1966); A. J. Leggett, Rev. Mod.
Phys. 73, 307 (2001); A. J. Leggett and F. Sols, Found. Phys. 31, 353 (1991).
[81] V. L. Ginzburg and L. D. Landau, Zh. Eksp. Theor. Fiz. 20, 1064 (1950); O.
Penrose and L. Onsager, Phys. Rev. 104, 576 (1956).
[82] Jared C. Bronski, Phys. Rev. Lett. 86, 1402 (2001).
[83] B. Wu, R. B. Diener, and Q. Niu, Phys. Rev. A 65, 025601 (2002).
[84] A. Trombettoni and A. Smerzi, J. Phys. B 34, 4711 (2001).
214 BIBLIOGRAPHY
[85] A. Trombettoni and A. Smerzi, Phys. Rev. Lett. 86, 2353 (2001).
[86] D. I. Choi and Q. Niu, Phys. Rev. Lett. 82, 2022 (1999).
[87] S. Raghavan, A. Smerzi, S. Fantoni, and S. R. Shenoy, Phys. Rev. A 59, 620
(1999).
[88] I. Marino, S. Raghavan, S. Fantoni, S. R. Shenoy, and A. Smerzi., Phys. Rev.
A 60, 487 (1999).
[89] S. Raghavan, A. Smerzi, and V. M. Kenkre, Phys. Rev. A 60, 1787 (1999).
[90] A. Smerzi and S. Raghavan, Phys. Rev. A 61, 063601(2000).
[91] S. Giovanazzi, A. Smerzi, and S. Fantoni, Phys. Rev. Lett. 84, 4521 (2000).
[92] G. J. Milburn, J. Corney, E. M. Wright, and D. F. Walls, Phys. Rev. A 55,
4318 (1997).
[93] A. Smerzi, S. Fantoni, S. Giovanazzi, and S. R. Shenoy, Phys. Rev. Lett. 79,
4950 (1997).
[94] C. CohenTannoudji, B. Diu, and F. Lalo¨e, Quantum Mechanics, Vol. 2 (John
Wiley and Sons, 1977).
[95] K. Burnett, M. Edwards, C. W. Clark and M. Shotter, J. Phys. B: At. Mol.
Opt. Phys. 35, 1671 (2002).
[96] J. Andersen, U. Al Khawaja and H. T. C. Stoof, Phys. Rev. A 66, 013615
(2002).
[97] C. W. Gardiner, Phys. Rev. A 56, 1414 (1997).
[98] Y. Castin and R. Dum, Phys. Rev. A 57, 308 (1998).
[99] N. M. Hugenholtz and D. Pines, Phys. Rev. 116, 489 (1959); J. Goldstone,
Nuovo Cimento 19, 154 (1961).
[100] A. J. Legett, Rev. Mod. Phys. 71, S318 (1999).
[101] E. M. Lifshitz and L. Pitaevskii, Statistical Physics part 2, Pergamon (Ox
ford, 1980).
[102] M. E. Fisher, M. N. Barber, and D. Jasnow, Phys. Rev. A 8, 1111 (1973).
[103] R. Roth and K. Burnett, Phys. Rev. A 67, 031602 (2003).
[104] D. Poilblanc, Phys. Rev. B 44, 9562 (1991).
[105] B. S. Shastry and B. Sutherland, Phys. Rev. Lett. 65, 243 (1990).
[106] B. S. Shastry and B. Sutherland, Phys. Rev. B 47, 7995 (1993).
BIBLIOGRAPHY 215
[107] J. Javanainen and M. Yu. Ivanov, Phys. Rev. A 60, 2351(1999).
[108] J. Javanainen, Phys. Rev. A 60, 4902 (1999).
[109] M. Shotter, J. Phys. B: At. Mol. Opt. Phys. 35, 30193028 (2002).
[110] G. Baym, J. Phys. B: At. Mol. Opt. Phys. 34, 45414550 (2001).
[111] F. Dalfovo, L. Pitaevskii, and S. Stringari, Phys. Rev. A 54, 4213 (1996); A.
L. Fetter and D. L. Feder Phys. Rev. A 58, 3185 (1998).
[112] V. A. Kashurnikov, N. V. Prokofev, and B. V. Svistunov Phys. Rev. A 66,
031601 (2002).
[113] M. D. Lee, S. A. Morgan, M. J. Davis, and K. Burnett Phys. Rev. A 65,
043617 (2002).
[114] S. Inouye et al, Nature 392, 151 (1998).
[115] C. Regal et al. Nature 424, 47 (2003).
[116] C. W. Gardiner and P. Zoller, Phys. Rev. A 61, 033601 (2000); C.W. Gar
diner, J.R. Anglin, T. I. A. Fudge, arXiv:condmat/0112129; C.W. Gardiner
and M. J. Davis, arXiv:condmat/0308044.
[117] E. Calzetta and B. L. Hu, Phys. Rev. D 37, 2878(1988)
[118] E. Calzetta and B. L. Hu, in Heat Kernel Techniques and Quantum Gravity,
Vol. 4 of Discourses in Mathematics and Its Applications, edited by S. A.
Fulling, Texas (A&M University Press, College Station, TX, 1995) or hep
th/9501040; E. Calzetta and B. L. Hu, Phys. Rev. D 61, 025012 (2000).
[119] E. Calzetta, B. L. Hu, S. A. Ramsey, Phys. Rev. D 61 125013 (2000).
[120] E. Calzetta and B. L. Hu, Phys. Rev. D 35, 495(1987).
[121] S.A. Ramsey and B. L. Hu, Phys. Rev. D 56 661 (1997).
[122] E. Calzetta and B. L. Hu, Phys. Rev. D40, 656 (1989).
[123] D. Bodeker, Nucl Phys. B 559, 502 (1999); 566, 402 (2000); J. Blaizot and
E. Iancu, Phys. Rep. (2001).
[124] J. Rammer, Quantum Transport theory, Perseus books (Massachusetts,
1998).
[125] D. Boyanovsky, Ann. Phys, 300, 1 (2002).
[126] F. Cooper, S. Habib, Y. Kluger, E. Mottola, J.P. Paz and P. Anderson, Phys.
Rev. D 50, 2848 (1994); Bogdan Mihaila, John F. Dawson and F. Cooper,
Phys. Rev. D 56, 5400 (1997); Bogdan Mihaila, John F. Dawson and F.
Cooper, Phys. Rev. D 63, 096003(2001); Bogdan Mihaila et al, Phys. Rev. D
62, 125015(2000).
216 BIBLIOGRAPHY
[127] G. Aarts et al, arXiv:hepph/0201308; J. Berges, Nucl. Phys. A 699, 847
(2002); J. Berges and J. Cox, Phys. Lett. B 517, 369 (2001); G. Aarts and J.
Berges, Phys. Rev. D 64, 105010 (2001).
[128] F. Lombardo, F. D. Mazzitelli and R. Rivers, arXiv:hepph/0204190.
[129] A. Polkovnikov, S. Sachdev, and S. M. Girvin, Phys. Rev. A 66, 053607
(2002).
[130] A. Smerzi and S. Raghavan, Phys. Rev. A 61, 063601(2000); A. J. Leggett
and F. Sols, Foundations of Physics, 21, 353 (1991); F. Sols, Physica B 194,
1389 (1994).
[131] E. M. Wright , T. Wong, M. J. Collett, S. M. Tan, and D. F. Walls, Phys.
Rev. A 56, 591(1997).
[132] P. Hohenberg and P. Martin, Ann. Phys. (NY) 34, 291 (1965).
[133] S. Giorgini, Phys. Rev. A 57 2949(1997); S. Giorgini, Phys. Rev. A 61
063615(2000).
[134] B. L. Hu and D. Pavon, Phys. Lett. B 180, 329 (1986); H. E. Kandrup, Phys.
Rev. D 37, 3505 (1988); Phys. Rev. D 38, 1773 (1988); S. Habib, Y. Kluger,
E. Mottola and J. P. Paz, Phys. Rev. Lett. 76, 4660 (1996); Y. Kluger, E.
Mottola and J. M. Eisenberg, Phys. Rev. D 58, 125015 (1998).
[135] R. Balescu, Equilibrium and Nonequilibrium Statistical Mechanics, John Wi
ley (New York, 1975).
[136] N. P. Proukakis and K. Burnett, J. Res. Natl. Inst. Stand. Technol. 101 457
(1996); N. P. Proukakis and K. Burnett, Phys. Rev. A 57 1230 (1998); T.
K¨ohler and K. Burnett ,Phys. Rev. A 65 033601 (2002).
[137] L.P. Kadanoﬀ and G. Baym, Quantum Statistical Mechanics: Green’s Func
tion Methods in Equilibrium and Nonequilibrium Problems, Addison Wesley
(New York, 1962).
[138] G. Baym, Phys. Rev. 127, 1391 (1962), B. Vanderheyden and G. Baym, J.
Stat. Phys. 93, 843 (1998); M. Bonitz Progress in Nonequilibrium Green’s
functions, World Scientiﬁc (Singapore, 2000).
[139] J. W. Kane and L. P. Kadanoﬀ, J. Math. Phys. 6, 1902 (1965).
[140] E. Wigner, Phys. Rev. 40, 749 (1932).
[141] S. Richard, F. Gerbier , J.H. Thywissen, M. Hugbart, P. Bouyer and A.
Aspect, Phys. Rev. Lett. 91, 010405 (2003).
[142] J. Steinhauer N. Katz, R. Ozeri , N. Davidson, C. Tozzo and F. Dalfovo,
Phys. Rev. Lett. 91, 060404(2003).
BIBLIOGRAPHY 217
[143] J.M. Vogels, K. Xu, C. Raman, J. R. AboShaeer and W. Ketterle, Phys.
Rev. Lett. 88, 060402(2002).
[144] R. Geursen, N.R. Thomas, A.C. Wilson, Phys. Rev. A 68, 043611(2003).
[145] T. St¨oferle, Henning Moritz, Christian Schori, Michael K¨ohl and Tilman
Esslinger, Phys. Rev. Lett. 92, 130403 (2004).
[146] P. B. Blakie, R. J. Ballagh and C. W. Gardiner, Phys. Rev. A 65, 033602
(2002).
[147] A. Brunello, F. Dalfovo, L. Pitaevskii, S. Stringari and F. Zambelli, Phys.
Rev. A 64, 063614 (2001).
[148] C. Menotti, M. Kr¨amer, L. Pitaevskii and S. Stringari, Phys. Rev. A, 67,
053609 (2003).
[149] Elliott H. Lieb and Werner Liniger, Phys. Rev. 130, 1605 (1963); Elliott H.
Lieb, Phys. Rev. 130, 1616 (1963).
[150] P. B. Blakie and J. V. Porto Phys. Rev. A 69, 013603 (2004).
[151] D. van Oosten et al., condmat/0405492.
[152] D. P. DiVincenzo, Fortschr. Phys. 48, 771 (2000).
[153] D. Kielpinski, C. Monroe, and D. J. Wineland, Nature 417, 709 (2002).
[154] A. van Oudenaarden and J. E. Mooij, Phys. Rev. Lett. 76, 4947 (1996).
[155] P. Kr¨ uger, et al., Phys. Rev. Lett. 91, 233201 (2003).
[156] D. J. Han, M. T. DePue, and D. S. Weiss, Phys. Rev. A 63, 023405 (2001).
[157] S. E. Hamann et al., Phys. Rev. Lett. 80, 4149 (1998).
[158] G. K. Brennen, G. Pupillo, A. M. Rey, C. W. Clark, C. J. Williams, submit
ted to Phys. Rev. Lett., arXiv:condmat/0312069.
[159] A. Fioretti et al., Eur. Phys. J. D 15, 189 (2001).
[160] M.J. Gagen and G.J. Milburn, Phys. Rev. A 47, 1467 (1993).
There are men who ﬁght one day and are good. There are men who ﬁght one year and are better. There are some who ﬁght many years and they are better still. But there are some that ﬁght their whole lives, these are the ones that are indispensable. Berthold Brecht
The manybody techniques discussed in these chapters systematically include higher order quantum corrections. The ﬁrst part comprises chapters 1 to 3. Clark National Institute of Standards and Technology This thesis covers most of my work in the ﬁeld of ultracold atoms loaded in optical lattices.Abstract Title of Dissertation: Ultracold bosonic atoms in optical lattices Ana Maria Rey. the so called two particle irreducible eﬀective action (2PI) together with the closed time path (CTP) formalism to study farfromequilibrium dynamics. In them I adopt an eﬀective action formalism. Charles W. 2004 Dissertation directed by: Prof. The second part deals with the superﬂuid weakly interacting regime where standard mean ﬁeld techniques can be applied. which we show are crucial for a . chapter 4 introduces the discrete nonlinear Schr¨dinger equation o (DNLSE) and uses it to model some experiments. not included in the quadratic approximations of the Hamiltonian. Kirkpatrick Physics Dr. Even though this thesis is a theoretical work all the chapters are linked either with experiments already done or with ongoing experimental eﬀorts. This thesis can be divided into four diﬀerent parts. Speciﬁcally. It makes a route through the physics of cold atoms in periodic potentials starting from the simple noninteracting system and going into the manybody physics that describes the strongly correlated Mott insulator regime. In chapter 5 I go one step further and include the small quantum ﬂuctuations neglected in the DNLSE by studying quadratic approximations of the BoseHubbard Hamiltonian. In these chapters. This is covered in chapters 4 and 5. Chapters 6 to 8 can be grouped as the third part of the thesis. after a brief introduction to the ﬁeld of optical lattices I review the fundamental aspects pertaining to the physics of systems in periodic potentials. Doctor of Philosophy. Theodore R.
. the response of the system to Bragg spectroscopy.correct description of the quantum dynamics outside the very weakly interacting regime. chapter 9 to 11 are devoted to study the Mott insulator phase. Finally. Even though small these are not ideal for the use of neutral atoms in optical lattice as a tool for quantum computation. and propose a mechanism to correct for the residual quantum coherences inherent to the Mott insulator ground state. In these chapters using perturbation theory I study the Mott insulator ground state and its excitation spectrum.
He was my other advisor at NIST. He has always made himself available for help and advice. Bei Lok Hu. He was an invaluable source of knowledge and assistance. in particular Blair Blakie. Ted Kirkpatrick for serving as my University of Maryland supervisor. First and foremost I would like to thank Charles Clark for giving me the invaluable opportunity to join the NIST group.Acknowledgements I want to use this lines to thank and acknowledge all the people that made this thesis possible. Besides colleagues they have been wonderful friends and helped me any time I needed help or advice. I worked with him during the summer of 2002 when he came to NIST. I am also grateful to Prof. Discussions with Bei Lok. Especially I am greatly indebted with Blair who was not only my oﬃce mate when he was at NIST and helped me to sort out all the problems but also for his support and great ideas even now that he is in New Zealand. I also want to thank Prof. It has been a pleasure to work with and learn from such an extraordinary person. love for physics and dedication he gave me not only very good advice and guidance but also all the encouragement I needed to feel the work I was doing was worthwhile and important. Keith Burnett. Marianna Safranova. His ideas and suggestions were really valuable for the development of this thesis. Besides an incredible person. Jamie Williams and Mark Edwards. I cannot leave out Esteban Calzetta and Albert Roura. Guido Pupillo and Gavin Brennen also deserve a special mention. Zach Dutton. Because of their great ideas and critical thinking the work described in chapters 9 to 11 was possible. Nicolai Nygaard. especially with regard to the 2PICTP formalism discussed in chapters 6 to 8. I also owe special gratitude to everybody in the Physics Laboratory at NIST and colleagues in the Radiation Physics Building. Through his intelligent ideas. he is also a bright physicist. willing to help me all the time and providing me with very useful ideas. Esteban and Albert regarding the application of the 2PICTP formalism to the patterned loaded system made possible the publication of the paper. There has never been an occasion when I’ve knocked on his door and he hasn’t given me time. I can not forget to thank Prof. He was the one who gave me the idea of using the superﬂuid . I also want to thank Nicolai not only for providing me all the LaTex style ﬁles for writing this thesis but also for all his time reviewing the manuscript. I would also like to acknowledge help and support from Carl Williams.
. He is the most important person in my life. my sister and my brother. my father. Finally.viii properties to study the approach to the Mott insulator phase. I have no words to express the gratitude I have to my husband Juan Gabriel. I owe my deepest thanks to my family . I really owe all my achievements to them. They have always stood by me in good and bad times. program and without his love and support the completion of this work would not have been possible.my mother. We came together to pursue the PhD.
4. 27 4 Mean ﬁeld theory 4. . . . . . . . . .1 AC Stark Shift . . . . . . . . . . . . . . . .1. . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . . . .3 Tightbinding approximation 2. . . . . . . .2. . . .2. conden.2 Single particle physics . . . . . .1. . . .3. . . . . . 4. . .1. . 4. . . 4. 31 31 33 34 35 36 38 40 42 43 51 . . . . .3 Dynamics with a constant external force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . . .1 Basic theory of optical lattices . . . . . . . . . . .4 Interaction eﬀects . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . .3 Lattice geometry . .2. . . . . . . . . . .1 Experiment . . . . . 25 3. . . . . . . . . . .1 Discrete nonlinear Schr¨dinger equation (DNLSE) . . .2 Linear free particle model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Dragging experiment . . . . . . . . . . . 2. .1 Bloch functions . . . . . 2. 4. . . .3. . . .2 SuperﬂuidMott insulator transition . . . . . 2. . . . . . . .3 Dynamics of a periodthree patternloaded BoseEinstein sate in an optical lattice . . . . . . . . . . . . . . . . . . . . . . . . .2 Case of no external potential . . o 4. .2. . .2 Wannier orbitals . . .2 Dissipative interaction . . . . .1 BoseHubbard Hamiltonian . . . . . 3 The BoseHubbard Hamiltonian and the superﬂuid to Mott insulator transition 25 3. . . . . . . .4 Semiclassical dynamics . . . xii 1 9 9 10 11 12 13 13 16 17 19 . . . . . . . . . . . . 4. . .2. . . . . . . . . . . . . . .2. . . . . . 2. . .1 Experiment . . . . . . . . . . . . .3 Eﬀect of the magnetic conﬁnement . . . . . . . . . . . . . . . . . .2. . . .2. . . . . . . . . . . .Table of Contents List of Figures 1 Introduction 2 Optical lattices 2. . . . . . . . . . . . .
. . . . . . .7. . . . . .7. . . . . . . . .4. 5. . . . . . . . 123 6. .2 Mode expansion of the HFB equations . . . . . .2 Bogoliubovde Gennes (BdG) equations . . . . . . . .2 HFBPopov approximation . . . . . . . . G) and approximation schemes . . . . . . . . .2. . . .3. . . . . . . . . . . . . . . . . . . . . . . . . 7. . . . . 107 6. . 5. . . .8 Conclusions . . 131 . . . . . .1 Equations of motion . . . . 116 6. . .4 The BoseHubbard model and superﬂuidity .3. . .1 Mean ﬁeld dynamics .1 Translationally invariant lattice . . . . . . . . . . . . . . . . . .3 Improved Popov approximation . . . . . . . . . . . . . . . . . . 109 6. . . . . . .7. . . . . . . . . . . . . . .6. . . . .2. 7. . . . . . . . . .2 Higher order expansions . . . . . . .5. . . . . . . .6.2 Conservation laws . 7. .1 Dynamical evolution . . . 7. 6 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism 101 6. . . . . . . . . . . 5. . . . . . . .2 Numerical algorithm for the approximate solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128 . .2 The anomalous average and the manybody scattering matrix 5. .4. 124 7 Nonequilibrium dynamics of a patterned loaded optical Beyond the mean ﬁeld approximation 7. . . . . . . . . . . .3 CTP formalism . . . . . . . . . . . . . . . . . 127 . . . .1 2PI eﬀective action Γ(z. . .3 Conclusions . . . . . . . 5. . . . . . . . . . . . . . . . 131 . . . . . . . . . . . . . . . . .2. . . 110 6. . . . .2 Onedimensional harmonic trap plus lattice . . . . 7. . . . . . . . . . . .x TABLE OF CONTENTS 61 62 62 65 66 67 67 70 71 71 78 91 92 93 95 99 5 Quadratic approximations 5.5 Expectation values . . . 5. .3 Higher order terms . . . . . . . . . . 128 .2. G) .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 6. . 5. . . . . . . . . 5.1 The standard approaches . . .3 Results and discussions .1 Initial conditions and parameters . . . . . . . .6 Conclusions . . . .1. . . . . . . . . . . . 5. . . .1 The characteristic Hamiltonian . . .2. . . . . 5. . .5 Secondorder expansion . . . . . . . . . . . . 116 6. . . . . . . . . . . . . . . . . . . 7. . 107 6. . . . . . . . 108 6. 102 6. . . . . . . . 5. .6 Applications . . . . . . . . 133 . . . .1 The twobody and manybody scattering matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 2PICTP approximations . . . . . . . . . . . . . . . . 5. . . . . . . . . . . . . . . . . . . . . . .2 Comparisons with the exact solution . . . . . . . 132 . . .1 Equations of motion . . . . . . . . . . . . . . . . . . 5. . . . . . . . . . . . . . . . . . . . . . . .5. . . . . .1 HartreeFockBogoliubov equations . . . 113 6. . . .2 Perturbative expansion of Γ2 (z. . 7.4 HFB approximation . . . 5. . . . . . . . . . . . . . .1. . . . 113 6. . . . . . . .3 Zero mode ﬂuctuactions . . 136 . . . . . . . . .7 Improved HFBPopov approximation . . . lattice: 127 . . . .
. . .4 Conclusions . . . . . . . . . . . . . . . .1 Commensurate translationally invariant case . . . . . . . . . . . . . . . . . 9. . . 11 Scalable register initialization for quantum tical lattice 11. . . . . . . . .3. . .1 Quasiparticle formalism . . 144 8. . . . . . . . . 10. . . . .2 Boltzmann equations . 10 Bragg Spectroscopy 10. . . . . . . . . . . . . . .3 Secondorder and Beliaev approximations . . . . . . . . . . . . . . . . . . . . . . . . .4 Conclusions . . . . . . . . . . . . . . 11. . . . . . . . . . .5 Mott insulator regime . . . . .2 Harmonic conﬁnement plus lattice . . . . . . 10. . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. . .3. . . . . .2 Inhomogeneous system . . . . . . . . . . . . . . . .1 Rewriting the 2PICTP secondorder equations . . . . . .3 Conclusions . 158 9 Characterizing the Mott Insulator Phase 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10. . . .TABLE OF CONTENTS xi 8 From the 2PICTP approximations to kinetic theories and local equilibrium solutions 143 8. . . . . . . . . . . . . . . . . . . . 156 8. 9. 155 8. . . . . 153 8. . . . . . . . . . . . . . . . . . . . . . . .3 Linear response . .2 Observables .4. . . . . . . . . . . . . . . . . . . . 197 . . .1 Perturbation theory . . . . . . . . . . . . . . . . . 205 207 . . . . . . 10. . . . . . . . . . . .2 Dynamics in presence of the external trap . . . . . . . . . . . . . . . . . . . . . . . 152 8. . . . . . . 11. .2 HFB approximation . . . . . .4 Zerotemperature regime . . . . . . . . .1 Bogoliubov approach . . . .3 Measurement . . . . . . . . . 10. . . . . . . . . . . . . . . . . . . . . 201 . . . . . 10. . . . . . . . . . 12 Conlusions . . . . . 10. .1. . .3 Equilibrium properties for a homogeneous system . . . . . 11. . . . . . . .2 Inhomogeneous system . . . . . . . . .5. . . . . . . . . . . . . . . . . . . . . . . . . . .5. . . . . . . 194 . . .3. . . . . . . . . . . . . . . . . . .7 Conclusions . 147 8. . . . . . . . . . . . 10. 159 159 159 167 171 173 174 176 177 179 179 181 183 184 186 188 190 computing in an op193 . . . .6 Finite temperature . . . . 10. . .1 Formalism . .1 Commensurate homogeneous system 10. . . . . . .1 Homogeneous dynamics . . . . . .
xii TABLE OF CONTENTS .
. . . . and so on. . . . . . . .2 2. . . Optical lattice potential. . Ground state amplitudes as a function of the lattice site i for diﬀerent values of ξ . . . . . . . .1. .2. . . . . . . . . . . . . 2. . . . . Condensate density in the xdirection at the observation time tf . . . . . (0) fi 10 2. . . . . . . . . . . . . The dashed line indicates the ﬁnal location of a comoving point with the optical lattice. It can be observed the localization of the wave function increases with the lattice increases. . . Zerotemperature phase diagram.4 14 16 17 2. . . . . . . The laser frequency is ω = 2πν which is detuned from the atomic resonance by Λ . . . . . . . . . . . . . .7 2. . . . .1 34 . . . . . . ER = 2 k 2 /2m. . . . . . .List of Figures 2. Light colors areas indicate regions of high condensate density. . . . . . .(a)(e) potentials for diﬀerent conﬁgurations of 3 beams. . . . . . .8 3. ER is the atomic recoil energy. . . . . Wannier state logdensity distribution corresponding to the lattice conﬁgurations shown in Fig. . . . . This ﬁgure is a courtesy of P. . . . . . . .3 2. . The vertical lines indicate critical values for diﬀerent ﬁlling factors . . Blair Blakie [34] ) . .The parameters for the plot are ξ = 0. . . Band structure of an optical lattice . . . . . . . . . . . . . . Harmonic conﬁnement plus lattice spectrum: The index n labels the eigenvalues in increasing energy order. This ﬁgure is a courtesy of P. . .5 18 20 21 24 30 2. . . . . . . . . . . . . . . . . . . .1 4. . . . . . . . Blair Blakie [34] ) . . . . . . . Diﬀusion of a square wave packet in the presence of a linear potential. . . . . . (n=1) Probability density of the Bloch wave function φq=0 for diﬀerent lattice depths. .6 2. . . . . . The numbers labels the near neighbor sites: 1 the nearest neighbors. . . Dotted white lines indicate the potential energy contours. . . . . . . . . . . . . . . (f) potential for the 4 counterpropagating laser beam conﬁguration (The two pair of light ﬁelds are made independent by detuning the common frequency of one pair of beams from the other. . .1 AC Stark shift induced by atomlight interaction. .
. . . We used γeﬀ as the local mean ﬁeld energy (see text). . . lattice velocity in the presence of a magnetic trap. . . . . . . . Atom’s average velocity in the lab frame vs.8 4. . . . Oscillation period (in units of /J) as a function of the interaction strength γ.5 4. .10 Minimum value of f 2 during an oscillation period as a function of γ. . . . .15vB . . . .11 Maximum contrast of the Fourier components as a function of γ. . . . where the comparison is made by replacing γ by an average mean ﬁeld energy γave ≡ Λ n zn 4 = 1. 4. .075ER and γeﬀ = 2. . .13 Momentum distribution of the inhomogeneous condensate evaluated at various times for the same parameters as used in Fig. . . .7vB . . . . . . . . 46 48 49 50 51 . . . .3 4. .xiv 4. Localization of the wave packet in the selffocusing regime vlatt = −0. . Inhomogeneous condensate (dotted). . . 4. Homogeneous model: dashed line represents the initially populated wells. Notice the change in the vertical scale in the last row . . . . . . . . The single particle case is shown in red. . . . . .12. . . . . . . 4. . The parameters used for the simulation were J = 0. 36 37 4. vB = π/am. .6 4. . . . . . . . . . . .2 LIST OF FIGURES Atom’s average velocity in the lab frame vs. . . As γ increases the population imbalance between wells increases (see text). . . . . . . . . . Line: results from the Bloch dispersion relation. . . . lattice velocity.85. . . . .14 Evolution of momentum peak populations. . . and the solid line represented the initially unpopulated wells. . . . . . . . . . . The maximum contrast is deﬁned as ∆max ≡ (c1 2 − c0 2 )max with the maximum value occurring when f 2 is at its minima. . . . Lower curves: population of the q = 0 momentum state. . . .12 Comparison between the population evolution of the central three wells for the inhomogeneous condensate and the homogeneous model. . . . Parameters are the same as in Fig. .7 4. 42 45 4. Upper curves: population of the q = ±2π/3 momentum states. . Inhomogeneous condensate: stars for the initially populated well and boxes for the initially empty wells. . . . . . .9 Evolution of the wave packet in the diﬀusive regime vlatt = −0. . . . . . .4 39 40 41 4. dots: tightbinding approximation results. . . . . . . . . . . . . . 4. Phase diagram U N vs vlatt for Vo = 3Er and γo = 10. . . . . homogeneous result (solid line). . . . δ(0) = 0 and N U = 2ER . . . . . . . The parameters used were γ(0) = 10. vB = π/am.12. . . . . . . . . . .64. . . . . 4. . 41 Patterned loading experiment. Velocity of the atoms predicted by the variational model in the case when interactions are taken into account (dotted line). . . . .4. . . . . .
. . . q = 2π/3 (solid line.19 Evolution of the momentum peak populations. . BdG(and HFBPopov): dotted line). q = 0 (dashed line.18 Comparison between the evolution of a inhomogeneous condensate ˜ with the homogeneous result.16 The spectrum of values of external force. .11 for the homogeneous model. .analytic (5. . . homogeneous model (lines) using the same parameters as those in Fig. . . and Ψ2 2 (diamonds). . . . .LIST OF FIGURES 4. . . . . . 4. . . Left: number ﬂuctuations (Exact: solid line. stars). . . . . . . . . . . . diamonds). . . . Homogeneous case: ˜ (boxes). Solid line: Ψ3n 2 . . One Bloch period is shown in the plots except ξ = 0 where the period ˜ ˜ is inﬁnite. middle: condensate fraction (Exact: solid line. for which a population resonance occurs i. . . . . . . . . . . .2 75 . 59 59 5. 57 Upper plot: ξ = 2ξmax max 4. . Lower plot: ξ = 2ξ res /7. BdG(and HFBPopov): dotted line). ordered in decreasing magnitude.59 . . . . . . squares). . . . Inhomogeneous condensate (dotted curves). right: superﬂuid fraction fs (Exact: solid line. BdG(and HFBPopov): dotted line. . . 63 5. . . Inhomogeneous condensate: Ψ0 2 ˜ 1 2 (stars). .075ER . . Evolution of Ψn 2 for various interaction strengths. . . . where we have taken γ as the local mean ﬁeld energy. the values of ξ ˜ for which Ψ0 periodically disappears. . Comparisons of the exact solution and BdG (and HFBPopov) solutions as a function of Vef f = U/J. .15 Evolution of the normalized population for diﬀerent values of ξ. . . . Ψ ˜ ˜ ˜ Ψ3n 2 (dashdot line). . . The vertical line shown in the plots is an estimation crit of Vef f when the system is commensurate . . . . .e. b) Pair excitations . The parameters used were J = 0.79:red line ). fs is also shown in these plots.18. . q = −2π/3 (dashdot line. . . Ψ3n+1 2 (solid line). dotted line: Ψ3n+1 2 . dashdot line: 2 . It can also be seen in the plots that ξ = ξmax /2 res and ξ = ξmax are resonant values. . . . . .1 Scattering processes included in the quadratic hamiltonian: a) Direct and exchange excitations. xv 55 56 4. . The ”nonclassical” motion can be seen where the 3n+1˜ Ψ3n+2  well populations initially increase more rapidly that the populations res of the 3n+2wells. . .33. ξ = 2 and γeﬀ = 1. . . .17 Eﬀects of interactions on generalized Bloch oscillations for the pat˜ tern loaded system. . . 4. . . . res . . 4. . . for a system with M = 3 and ﬁlling factors n = 5 and 5. The exact second (2) order term (dashed line) of the superﬂuid fraction. . . . but replacing γ by an average value γave = 1. . . and Ψ3n+2 2 (dotted line).
Ω/J = 9. . . . . Lowlying quasiparticle amplitudes found by numerically solving the BdG equations. BdG(and HFBPopov): dotted line).5 × 10−4 and N = 100. . . .3 LIST OF FIGURES Comparisons of the exact solution and BdG(and HFBPopov) solutions as a function of Vef f = U/J. . .The site indices i are chosen such that i = 0 corresponds to the center of the trap. . . . . . . . . .79): red line ).5 × 10−4 and N = 100. . . . analytic Eq. . .5 × 10−4 and N = 100. crosses (Vef f = 3). . Although these quantities are deﬁned only at the discrete lattice sites we join them to help visualization. . . . . . . . . . The exact second order (2) term (dashed line) of the superﬂuid fraction. . . . BdG(and HFBPopov): dotted line. .8 s Comparisons between the product Fn Gs calculated from the Thomasn Fermi approximation and the exact numerical solution of the BdG equations. . . middle: condensate fraction (Exact: solid line.5 82 5. . . .09). . 86 5. Comparisons of the quasiparticle spectrum found by numerically solving the BdG equations(BdG) with the ThomasFermi solution (TF) and the noninteracting energies (Free). fs is also shown in these plots. ﬁlled diamonds (Vef f = 11). .33. . . . .2. . The parameters used were U/J = 0. .7 84 5. . . The empty boxes represent the exact solution for the case J=0. . . . . . In this case the agreement is much better. . Ω/J = 9. . .5 × 10−4 and N = 100. . . . The parameters used were U/J = 0. . . The depletion is also shown in the plot. .6 83 5. .xvi 5. . . . . . . . . . . Ω/J = 9. empty boxes (Vef f = 100) and pentagons (Vef f = 312). . . Comparisons of the condensate fraction given by the BdG solutions and the HFB equations as a function of Vef f = U/J. 85 5.01). . . The letter q labels the quasiparticle energies in increasing order.9 Condensate density (triangles). . . . . .2. . . . . . . . . . . . . . . . total density (ﬁlled boxes) and local depletion (empty diamonds) as a function of the lattice site i for diﬀerent values of Vef f. Comparisons of the condensate wave function found by numerically solving the BdG equations with the ThomasFermi solution. 76 5.2. . . stars (Vef f = 0. 87 . . . . . . . . . . . . .10 Quasiparticle spectrum predicted by the HFBPopov theory for diﬀerent values of Vef f : Empty diamonds (Vef f = 0. . for a system with M = 3 and ﬁlling factors n = 50 and 50. The parameters used were U/J = 0.2. for a system with M = 3 and ﬁlling factor n = 10. . . Left: number ﬂuctuations (Exact: solid line. . . . . . . . . . . . Ω/J = 9.4 78 5. . . . . . . . BdG(and HFBPopov): dotted line). The parameters used were U/J = 0. . . . .(5. . . . . right: superﬂuid fraction fs (Exact: solid line. The quasiparticle energies are in recoil units. . . . . .
xvii 88 89 90 92 97 98 98 6.33. . 5. . . 5. . . . . .1 Twoloop (upper row) and threeloop diagrams (lower row) contributing to the eﬀective action. .The site indices i are chosen such that i = 0 corresponds to the center of the trap. . for a system with M = 3 and ﬁlling factors n = 50 and 50.33. The maximum value reached by the proﬁle decreases as Vef f is increased.LIST OF FIGURES 5. .17 Comparisons between no /nTmb = fc Tmb and U as a function of U for a system with M = 3 and ﬁlling factors n = 5 and 50. = 11. = 3.16 Comparisons of the condensate fraction predicted from the exact diagonalization of the BHH (red).11 Number ﬂuctuations in the self consistent HFBPopov approach as a function of lattice site for Vef f = 0. for diﬀerent values of Vef f . . . . . the improved Popov approximation (yellow). b) the settingsun and c) the basketball. .01. . the BdG (and HFBPopov) equations (blue) and the test ﬁeld approximation (green) as a function of Vef f = U/J. In this ﬁgure kl designates the initial states. The empty boxes shown for each of the curves correspond to the number ﬂuctuations predicted by the homogeneous HFBPopov model using a local density approximation. 5. . = 312. . = 0. Explicitly. the BdG (and HFBPopov) equations (blue) and the test ﬁeld approximation (green) as a function of Vef f = U/J. .13 First order onsite superﬂuid fraction as a function of the lattice site i for diﬀerent values of Vef f . . . empty boxes: Vef f . Filled boxes: Vef f . . . stars: Vef f . . the diagram a) is what we call the doublebubble . . . 102 . . . . . .01 (boxes). . for a system with M = 3 and ﬁlling factors n = 5 and 5. . .09. =0. 5. . 5. . crosses: Vef f . . Vef f = 0. . . q the quasimomentum. . .15 Comparisons of the condensate fraction predicted from the exact diagonalization of the BHH (red). . . . . . Vef f = 100 (stars) and Vef f = 312 (diamonds). . . . . . . . . .09 (crosses). . .Vef f = 11 (triangles). . . . . .12 Quasimomentum distribution as a function of qa. a the lattice spacing. .14 Diagrammatic representation of the twobody T2b scattering matrix. empty diamonds: Vef f . ij the ﬁnal states and np a set of intermediate states . the improved Popov approximation (yellow). . . = 100 and triangles: Vef f . . . . . . 5. . . . Vef f = 3 (circles). . .
C2 for the fractional condensate population in the initially empty one and CT for the total condensate fraction. .2 ˆ the condensate population per well calculated from the exact solution:  a0 2 and  a1. . The time is given in units of /J. . . . . 141 7. . strongly correlated regime (γ = 12). . . intermediate regime (γ = 2). J = 1 and U/J = 1/3. The time is given in units of /J. . . . . . . . 138 Comparisons for the case M = 3. . . .5 . . the secondorder large N approximation (pentagons) and the full 2PI secondorder approximation(crosses) for the very weak interacting regime. . . In the momentum distribution plot the upper curve corresponds to the k = ±2π/3 intensities and the lower one to the k = 0 quasimomentum intensity. . . . . 140 Time evolution of the condensate population per well and the total condensate population. . . N = 6. . .1 LIST OF FIGURES Comparisons between the exact and the DNLSE solutions for six atoms and three wells. . . The time is given in units of /J. J = 1 and U/J = 1/30. .2 2 . . . J = 1/2. . . . 40 and 80. . . . a† a1. . . In the plots the abbreviation 1st is used for the initially occupied well and 2nd for the initially empty wells. . . middle panel. . . . . In the plots P1 stands for the fractional atomic population in the initially populated wells and P2 for the population in the initially empty wells. . . . . weakly interacting regime (γ = 0. . Time is in units of /J. In the plots C1 stands for the fractional condensate population in the initially populated well. the HFB approximation (boxes). Due to the symmetry of the initial ˆ ˆ periodic conditions the curves for the i = 1 and 2 wells are the same in all depicted curves . 139 Comparison between the evolution of the atomic population per well for M = 2. bottom panel. . condensate and depletion per well are depicted the top curves correspond to the initially populated well solutions and the lower to the initially empty wells. . . . The pentagons show ˆ0 ˆ ˆ1. Notice the diﬀerent scale used in the depletion plot. . . The parameters used were M = 3. . . .xviii 7. . . . . Time is in units of /J. . In the quasimomentum plots k = 2π/a is the reciprocal lattice vector with a the lattice spacing.2 . . .3 7. .4 7. The number of atoms N is explicitly shown in each panel. Top panel. . . In the plots where the population.4. N = 8. . 137 Comparisons between the exact solution(solid line). . . 7. .2 7. . . . . . . . .2 (t)2 . . . The solid line is the DNLSE prediction for the population per well: z0 (t)2 and z1. N U/J = 4 and N = 20. .2). . . . the triangles are used to represent the exact solution for the population per well calculated using the Bose Hubbard Hamiltonian: a† a0 .for the same parameters as Fig. . . .
. . . . . . . . . 142 Contour plot of the two dimensional band of the 1ph excitations to ﬁst order in perturbation theory. . . . . . . . Comparisons between the ﬁrst order corrections to the 1ph excitations calculated by diagonalizing the BoseHubbard Hamiltonian inside the 1ph subspace and the analytic solution Eq. . . . . . Bragg frequency curves calculated from the exact diagonalization of the manybody Hamiltonian (red) and the from the HFBPopov approximation(blue) for a trapped system with N = 9. Comparisons between the exact (red) and the perturbative (blue) spectra for a trapped system with N = 9. The horizontal axis is in units of . . . Eq. The lines depicted in the plot are located at the diﬀerent quasiparticle energies excited by the Bragg perturbation. . . . Time is in units of /J. . . 9. . . .4 166 9. 40 and 80. (10. . . . Comparisons between the energy eigenvalues calculated by the exact diagonalization of the Hamiltonian (red dots). . 183 . . . . . . . . . . . . .23). . . . The parameters used were M = 2. .2 164 9.875. . . . .875 deep in the Mott regime. . . . Quasimomentum distribution as a function of U/J for a unit ﬁlled lattice with six wells (M = 6). . . .22). . . . . .LIST OF FIGURES 7. . (9.3 Comparisons of the imparted energy vs. See text Eq. . The horizontal axis is in units of . . . . . The plots are scaled to set the integrated quasimomentum density to one for all N . . . . In the plots ko denotes the k = 0 quasimomentum component and k1 the k = π/a one (a the lattice spacing). . . . .2 Imparted energy: Comparisons between the the exact and Bogoliubov approximation for N = M = 9. . The lattice potential with atoms loaded into the ground band is perturbed by a shallow running wave perturbation (the Bragg potential). .3 165 9. . . . . . 174 10. . . . . . the number of wells M and the ratio U/J is indicated in each plot. . . . . . . . The lines indicate the energies of the states that are coupled directly to the ground state. M = 11 and Ω/J = 1. The number of sites used for the plot was M = 11. . . . . . . . .1 163 9. . . . . . . . . . The ﬁlling factor g.5 170 10. M = 11 and Ω/J = 1. . . . . . . . . 181 10. Energies are in units of J. For the plot we used Jτ / = 10. . . . The index n labels the eigenvalues in increasing order of energy. . . . (9. Energies are in units of J.22). . . . . . . . . . Due to the lattice symmetry n4π/3 = n2π/3 and n5π/3 = nπ/3 . J = 1/2. The labels are ky = 2π/M R and kx = 2π/M R. the diagonalization of the Hamiltonian in the restricted 1ph subspace (green triangles) and the analytic solution (blue crosses). . . In the plot the brighter the color the higher the energy. . . .1 Bragg spectroscopic scheme considered in this paper. . . . N U/J = 4 and N = 20. . . . . . .6 xix Dynamical evolution of the quasimomentum intensities. . The exact solutions are displayed with solid lines and the ﬁrst order perturbative results with dotted lines.
. .6 (blue). . . . . . .4 (red) and the analytic solution Eq. 191 10. . . M = 11 and Ω/J = 1. . .875. The perturbation time was set to Jτ / = 10. . . . . . . 186 10.2 Percentage ﬁdelity as a function of time calculated using the perturbative solution 11. . . 11. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. Note the change of the vertical scale between the two panels. . . . . The horizontal axis is in units of .10Transfer energy as a function of the temperature for diﬀerent U/J parameters.4 Density proﬁle calculated from the exact diagonalization of the manybody Hamiltonian (red) and from the HFBPopov approximation (blue) for a trapped system with N = 9.9 Transfer energy as a function of the temperature for diﬀerent U/J parameters. . Note the change of the vertical scale between the two panels. . The vertical scale is in units of . 198 . . .7 Transfer energy vs. . . . . . . . . . . . . the perturbative solution inside the one particle hole subspace (red) and the analytic solution Eq. . . . 192 11. . . . . .The number of sites used for the plot is M = 31 . . . . due to a magnetic ﬁeld. . 11. . . . . . . ω/J in the ThomasFermi (blue) and HFBPopov (red) approximations. . Bragg frequency for a zero temperature trapped system with parameters U/J=45. . e. . .6 (blue). 196 11. acts to ﬁll gaps in the central region of the trap. . 187 10. . . . . The perturbation time was set to Jτ / = 10. . . .857. . .6 Eigenvalues of the quasiparticle spectrum plotted in increasing energy order. . 186 10. . .5×10−4 and N = 100. . . . Ω/J = 9. . . . . . The center subspace R of the lattice deﬁnes the quantum computer register containing K < N qubits. . . . . The horizontal axis is in units of . . Ω/J = 1.875. . 185 10.1 Comparisons between the time evolution of the percentage ﬁdelity calculated by diagonalizing the BoseHubbard Hamiltonian (dots). . . . . . .3 Schematic of an inhomogeneous lattice ﬁlled with N qubits with an onsite interaction energy U . . . . . . . . . . . An externally applied trapping potential of strength Vj = Ωj 2 . . . . . The horizontal axis in in units of . . . . The horizontal axis is in units of . . The plots are for an trapped system with M = 11. . . . . . . . 184 10. . .g. Bragg frequency for a zero temperature homogeneous system with M = N = 9 and diﬀerent transfer momenta q. . . . . The horizontal axis in in units of . . The plots are for a homogeneous system with N = M = 9. . . .xx LIST OF FIGURES 10. . . 197 11. . . . . N = 9 and Ω/J = 1.5 Dynamical structure factor vs. n labels the lattice site.8 Transfer energy vs. The system parameters are U/J = 0. . . M = 11 and N = 9 and diﬀerent transfer momenta q.
. . . t) calculated by evolving the initial target state using the BoseHubbard Hamiltonian (dots). . . The bound states quickly decay and give the possibility of monitoring population in the “faulty” ± register states Sj . . . . . . 11. . . The population in T R for η = 1 is indistinguishable from one. . . . the restricted one particle hole basis (red) and the analytic solution Eq. . . . (b) Long time dynamics for K = 501. . 11. . (11. . . The unit ﬁlled ± state T describing a target quantum register and the states Sj having one doubly occupied lattice site and a neighboring hole are coupled to ﬁrst order in J. . .T (0). . . . Also shown is the oscillatory dynamics at fundamental frequency U described by Eq. The plots show dynamics appropriate to tunneling in one dimension with U/J = 500. .4 Comparisons between the time evolution of the percentage ﬁdelity Fcom (K = 5. . . . . .16) if the measurement is turned oﬀ after the target state is reached. . A catalysis laser resonantly couples the ± ± ground states Sj to the excited states Mj describing a bound molecule at the doubly occupied site. . . . .16 (blue). . . . . . . . . . N = 551 and ﬁnite detector eﬃciencies η. . . (a) Quantum trajectories corresponding to a null measurement result for three diﬀerent register sizes K. xxi 200 201 202 203 . . . . . In the plot we assumed a commensurate unit ﬁlled lattice with ﬁve sites and inﬁnitely high boundaries. . .6 Schematic of the relevant couplings in the problem. .LIST OF FIGURES 11. . .5 Percentage ﬁdelity as a function time calculated using the approximated solution 11. . The arrow indicates ρT. The time scale to saturate the target state is independent of the number of qubits: tsat ≈ κ−1 . We assumed inﬁnity high boundaries and K = 31. . . .7 Population in the unit ﬁlled register state T R during continuous measurement of the register beginning in the BoseHubbard ground state Ψg .16 (blue) and the ﬁdelity found numerically by restricting the dynamics to the one particle hole subspace. . . 11. . . . . 11. . . . . .
Theoretically. which scales like T −1/2 . Although Einstein’s prediction applied to a gas of noninteracting atoms. the weakly interacting regime has the advantage that all the atoms can be described by a . BEC could be the mechanism underlying the phenomenon of superﬂuidity in 4 He [3]. generally up to 106 . In such a condensate a macroscopic number of atoms. the relatively weak twoparticle interaction in dilute alkali atoms allows these systems to be used as a theoretical and experimental arena to study coherent matter wave properties. collectively occupy the lowest energy state. It was not until 1995. At this point the wave functions of the particles are suﬃciently smeared out so that there is always some overlap and a Bose Einstein condensate is formed. using the advances made in laser cooling techniques[9]. is decreased the deBroglie wave length. Further evidence for this point of view came from neutron scattering experiments [4]. The basic idea of BEC is that below a critical temperature a macroscopic number of particles occupy the lowest energy state: as the temperature. that BEC in dilute alkali atomic gases was achieved. London suggested that. increases and at the critical point it becomes comparable to the interparticle mean separation. The ﬁrst series of experiments were done with Rubidium [6]. They are then transferred to a magnetic trap where evaporative cooling allows the system to be cooled to nanoKelvin temperatures. In contrast to experiments with liquid helium. despite the strong interactions in this system. sodium [7] and lithium [8] vapors. At a critical phase space density BEC takes place.Chapter 1 Introduction Bose Einstein condensation (BEC) in bosonic gases was predicted by Einstein in 1925 [1] based on the quantum statistics ideas developed by Bose for photons [2]. In these experiments atoms are typically collected in a magnetooptical trap (MOT) and compressed and cooled to microKelvin temperatures using laser cooling techniques. The ﬁrst experiments concentrated on using atomic Hydrogen but mainly the large rates of inelastic collision prevented these experiments from succeeding. The experimental realization of BEC in alkali gases opened unique opportunities for exploring quantum phenomena on a macroscopic scale. where the strong interactions between particles wash out the eﬀects due to the BEC. T. Experimental eﬀorts to create a BEC in dilute gases date back to the 1980’s [5].
An entirely diﬀerent way to reach the strongly correlated regime is by using optical lattices. the ratio between these two energies is proportional to n1/3 as . However. In order to be in the weakly interacting regime. sound velocity and superﬂuid ﬂow phenomena [12][19]. 2 n is the mean particle density and m is the atomic mass) and Ekin ∼ 2m n2/3 . • Optical lattices Optical lattices are periodic Stark shift potentials created by the interference of two or more laser beams. the GPE together with Bogoliubov analysis have been in general very successful describing these experiments. The . 24] with applications to atom optics and atom interferometry [25]. weakly interacting dilute gases described by a mean ﬁeld picture are the simplest many body systems one can possible found. Ekin . 21]. The problem of this approach is. In dilute alkali vapors this ratio is generally of order 0. a description beyond the simplest mean ﬁeld theory can be made by treating the small quantum ﬂuctuations as perturbations. Eint . They have been widely used in atomic physics in the context of atom diﬀraction [23. Much theoretical and experiment work has been done studying condensate properties such as condensate collective excitations. an obvious way to proceed is either to raise the density or to raise the scattering length. By increasing the depth of the optical lattice the ratio between kinetic energy and potential energy can be changed without aﬀecting the density or the scattering length. This treatment was proposed by Bogoliubov in 1947 [11]. however. This allows a very intuitive understanding of the system based on the so called GrossPitaevskii equation (GPE) [10].02.2 Chapter 1 Introduction single macroscopic wave function. It is indeed possible to tune the scattering length to large values by using a Feshbach resonance. For three dimensional systems Eint ∼ n 4π m as (as is the scattering length. where by turning oﬀ the trapping ﬁelds the atoms are allowed to expand and the wave packets to interfere with each other. In the weakly interacting regime. the ratio between the interaction energy of uncorrelated atoms at a given density. must be small. The GPE assumes that all the atoms are in the condensate and neglects completely quantum correlations. that the lifetime of the condensate strongly decreases due to threebody losses [22]. and the quantum kinetic energy needed to correlate particles by localizing them within a distance of order of the mean interparticle 2 distance. which fully characterizes the low energy scattering processes. Experimentally. the macroscopic wave packet can be probed by interference experiments. Thus. This has recently been realized for example in 85 Rb where the scattering length was tuned over several orders of magnitude [20. The beauty of this approach is that the lattice depth can be used as an experimental knob to change the kinetic to interaction energy ratio allowing us to reach diﬀerent manybody regimes. They have also been used as a way to trap and cool atoms. phonon modes. To enter the strongly correlated regime. Since in most of the experiments with dilute gases the condensate depletion is at most 3%. The ﬂuctuations lead to a small depletion of the condensate mode since other excited states diﬀerent from the condensate get populated.
for example by measuring number squeezing [45] or by studying the collapse and revival of coherence in a matter wave ﬁeld [47]. the system is in the weakly interacting regime and most of the atoms are Bose condensed. the ultimate source of coherent atom . optical lattices provide a way of exploring a quantum system analogous to electrons in crystals but with complete control over the lattice and the atoms. 37]. Bragg scattering and Bloch oscillations [38]. Besides the possibility of high ﬁdelity initialization. the easy scalability. 48. in recent years there have been many impressive experiments which have demonstrated the loss of quantum coherence as the system approaches the strongly correlated regime.P. In fact. superﬂuidity [41] and quantum chaos [42]. 40]. Fisher et al. which describes bosons with local repulsive interactions in a periodic potential. It was ﬁrst realized by Jaksch et al. 50]. 32]. one of the most successful techniques to load ultracold atoms in the ground state. When ultracold bosonic atoms are loaded in shallow lattices. The negligible number ﬂuctuations makes it possible to prepare the ﬁducial state with exactly one atom per site needed to initialize a quantum computer register [43. that a BEC loaded in a lattice potential is a nearly perfect experimental realization of the BoseHubbard Hamiltonian. with almost no discernible thermal component. Deep in the Mott insulator regime the kinetic energy is very small with respect to the interaction energy and it is energetically favorable for the atoms to remain localized without tunneling. [44] predicted that a system modeled by the BoseHubbard Hamiltonian exhibits a quantum phase transition from a superﬂuid to an insulator state (superﬂuidMott insulator transition) as the interactions are increased. [26][28] followed by Grynberg et al. is by ﬁrst forming a Bose Einsten condensate in a weak magnetic trap and then adiabatically turning on the lattice by slowly ramping up the intensity of the laser beams. Combined with BEC. There has also been spectacular recent experimental progress in the strongly correlated regime. Moreover. [43]. [29].A. by which atoms have recently been partially cooled to the ground sate with ﬁlling factors of order one [30.3 ﬁrst experiment where atoms were cooled to the microkelvin regime in a multidimensional optical lattice was carried out by Hemmerich et al. 34]. There have been various attempts to cool atoms directly in an optical lattice. the Mott insulator transition in a 3D lattice starting from a BEC has experimentally observed by M. Atom dynamics in optical lattices is closely related to electron dynamics in solid state crystals. . One of the most important potential applications of the Mott insulator transition is to use it as a mean to initialize a quantum computer register. but optical lattice have favorable attributes such as the absence of defects and the high degree of experimental control [33. 31. 36. low noise and high experimental control make ultracold neutral atoms loaded in an optical lattice one of the most attractive candidates for implementations of quantum computation. Among them we can mention Raman cooling techniques. M. as well as coherent matter wave interferometry [39. However. Beautiful experiments have been done in this regime and have provided an elegant demonstration of band structure [35. Greiner and coworkers [46].
where a mean ﬁeld treatment is valid. there is a superimposed harmonic conﬁnement potential.4 Chapter 1 Introduction The overall goal of this thesis is to study equilibrium and non equilibrium properties of cold bosons loaded in optical lattices starting from the superﬂuid regime. I give a review of the generic issues that characterize the superﬂuid to Mott insulator transition. I go one step further and consider the manybody properties of the system by introducing the BoseHubbard Hamiltonian. the HartreeFockBogoliubov (HFB) approximation and the HFBPopov approximation. the small quantum corrections can be included by using the Bogoliubov approximation. In chapter 4. I also report on our idea of using the superﬂuid fraction to study . To test the validity of the diﬀerent approximations and their departure from the exact solution as the interactions are increased I compare them with the exact numerical diagonalization of the BoseHubbard Hamiltonian. and going into the rich and complex strongly correlated regime where the standard GPE and Bogoliubov treatments fail to describe the system and a more general framework is required. I study the mean ﬁeld Discrete Nonlinear Schr¨dinger equation (DNLSE). the exact solution (for a system with N atoms and M wells the number of states scales like (N + M − 1!)/(N !M !)). In chapter 3. In the second experiment.the atoms were loaded into every third site of an optical lattice. Using the Bogoliubov approximation. with the aim of having large enough spatial separation to address individual atoms. and a system closer to real experimental situations where. I discuss the Bogoliubov de Genes equations (BdG). Most of the work is done in the context of ongoing experimental eﬀorts. an optical lattice was moved and the average displacement of the atoms was used as a means to probe the band structure of the system. a translationally invariant one with periodic boundary conditions which in general allows analytic solutions. where mean ﬁeld techniques can be applied. if the system is weakly interacting.to which I refer as the patterned loading experiment [33]. and use it to model two expero iments done by the lasercooling and trapping group at NIST. In the Bogoliubov approximation the complicated manybody quartic Hamiltonian is reduced to a quadratic one. besides the lattice potential. • Overview In chapter 2. the single particle band structure and the tight binding approximation. This patterned loading method may be a useful technique for the implementation of lattice based quantum computing proposals. In the ﬁrst one. The DNLSE completely neglects quantum ﬂuctuations. In chapter 5. Because of the exponential scaling of the dimensionality of the Hilbert space with respect to system size. the numerical comparisons are restricted to systems with a moderate number of atoms and wells. However. in chapter 5 I study diﬀerent standard quadratic approximations in two diﬀerent lattice systems. especially the ones trying to achieve lattice based quantum information processing. I start by reviewing the theory of optical potentials. which can be diagonalized exactly. I describe the equilibrium properties of lattice systems in the superﬂuid regime.
In chapter 6 and 7. they are able to capture damping eﬀects exhibited in the exact solution. to the manybody scattering matrix. . we are properly taking into account the eﬀect of the surrounding atoms in the properties of binary collisions. The idea presented here is to upgrade the bare potential. which is the one that explicitly appears in the manybody Hamiltonian. 53. they fail at late times. By deriving an expression for the superﬂuid density based on the rigidity of the system under phase variations we were able to explore the connection between the quantum depletion of the condensate and the quasimomentum distribution on one hand and the superﬂuid fraction on the other. While the secondorder approximations show a clear improvement over the HFB approximation. motivated by the patterned loading experiment.5 the approach to the Mott insulator transition. We use this system to illustrate many basic issues in nonequilibrium quantum ﬁeld theory. we show in this chapter how the full secondorder 2PI equations are in agreement with current kinetic theories [15]. By upgrading the bare potential. The idea in chapter 8 is to simplify the 2PI equations and to obtain near equilibrium solutions where. We show that because the secondorder 2PI approximations include multiparticle scattering in a systematic way. We derive mathematical expressions for the equations of motion in chapter 6 and apply them to the particular case of the patterned loaded lattice in chapter 7. we adopt a closed time path (CTP) [56] functionalintegral formalism together with a twoparticle irreducible (2PI) [57] eﬀective action approach and derive equations of motion. However. In particular. Even though a description of the dynamics of the patterned loading system using the DNLSE was derived in chapter 4. when interaction eﬀects are signiﬁcant. which a collisionless approach fails to produce. it is shown by comparisons with the exact quantal solution calculated by time propagating the initial conﬁguration with the BoseHubbard Hamiltonian that a mean ﬁeld solution is valid only for short times and in the very weakly interacting regime. b) the nexttoleading order 1/N expansion and c) a full secondorder perturbative expansion in the interaction strength. kinetic theories that describe excitations in systems close to thermal equilibrium are valid. 55]. The 2PI eﬀective action formalism provides a useful framework where the mean ﬁeld and the correlation functions are treated on the same selfconsistent footing. We retain terms of up to secondorder in the interaction strength when solving these equations. I present my attempt to approach the strongly correlated regime by using the improved Popov approximation [52. At the end of the chapter. To deal with the dynamics far from equilibrium. we adopt a functional eﬀective action approach capable of dealing with non equilibrium situations that require a treatment beyond mean ﬁeld theory. it yields dynamical equation of motion that are non local in time and hard to estimate analytically. such as nonlocal and nonMarkovian eﬀects. Under the 2PICTP scheme we consider three diﬀerent approximations : a) the time dependent HartreeFockBogoliubov (HFB) approximation.[58][62] and reproduce in equilibrium the higher order perturbative corrections well known in the literature since Beliaev’s work [63]. pertaining to the dynamics of quantum correlation and ﬂuctuations.
One key piece of evidence for the Mott insulator phase transition is the loss of global phase coherence of the matter wavefunction when the lattice depth increases beyond a critical value [46]. . I derive expressions for the excitation spectrum of the Mott state for both homogeneous and trapped systems and compare them with the solutions obtained by exact diagonalization of the BoseHubbard Hamiltonian. 66]. so loss of coherence is not a proof that the system resides in the Mott insulator ground state. in the experiments by Greiner et al. Bragg spectroscopy [65. However. In chapter 10. where I report on our proposal to solve this problem by using the spatial inhomogeneity created by a quadratic magnetic trapping potential together with a continuous measurement procedure which projects out the components of the wave function with more than one atom in any well. We test the accuracy of the approximations and their deviation from the full quantal behavior as usual by comparing them with numerical solutions obtained by diagonalizing the BoseHubbard Hamiltonian for a moderate number of atoms and wells. Speciﬁcally. The ground state has a remaining coherence proportional to the tunneling matrix element. [46] a potential gradient was applied to the lattice to show the presence of a gap in the excitation spectrum. The main purpose of this chapter is to understand the manybody properties of the Mott insulator ground state and the nature of its manybody excitations which will be crucially important as more elaborate experiments with optical lattices in the strongly correlated regime are undertaken. for this reason. we study the total momentum and total energy deposited in the system by the Bragg perturbation calculated under a linear response analysis and obtain analytical solutions in the superﬂuid and deep Mott insulator regimes. I ﬁnish this thesis with chapter 11. there are many possible sources of phase decoherence in these systems.6 Chapter 1 Introduction In chapter 9. Such architecture requires a lattice commensurately ﬁlled with atoms. which does not correspond exactly to the ground Mott insulator state. All of the proposals for quantum computation which utilize a latticetype architecture have the Mott insulator transition as the initialization scheme to load exactly one atom per lattice site. Indeed. I jump into the Mott insulator phase and I study the physical properties of the system deep in the Mott insulator regime. It is known that substantial decoherence can be induced by quantum or thermal depletion of the condensate during the loading process. but also can be used to map out the excitation spectrum and to determine the temperature of the system when it is deep in the Mott regime. This degrades the initialization of the quantum computer register and can introduce errors during error correction. not only can identify the excitation gap that opens up in the Mott regime. which is the other regime where an analytic treatment based on perturbation theory is possible. A perturbative analysis should be applicable to study current experiments [64] that reach the strong Mott regime. we show that another common experimental technique.
Rev. C. Clark CW. M. Rev. J. Opt. Rey. E. Rey AM. 825 (2003). M. P. Mod. L. Rev A. condmat/0406552. A. 033610 (2004). K.. G. A. . (2004). Brennen.7 Publications in the PhD work: • Chapter 4 Dynamics of a periodthree pattern loaded BoseEinstein condensate in an optical lattice. Ana M. 1. M. A. M. Charles W. • Chapter 10 Bragg spectroscopy of ultracold atoms loaded in an optical lattice. Phys. Charles W. submitted to Phys. 69. L. Opt. Rey. 36 . condmat/0312069. Clark. Charles W. Phys. Ana Maria Rey. • Chapter 6 and 7 Nonequilibrium Dynamics of Optical Lattice . Phys. Las. Pupillo. Rey. Carl J. Phys.Loaded BEC Atoms: Beyond HFB Approximation. C. Clark (Proceedings of the Laser Physics Workshop 2003). Clark. J. A. Clark. Calzetta. Carl J. A. B. Lett. Williams CJ. Calzetta. Clark. B. Blakie. Mol. Roth R. B. Williams. Guido Pupillo. 14. submitted to Phys. Scalable quantum computation in systems with BoseHubbard dynamics. A. W. Hu. Rev. B: At. BEC with ﬂuctuations: beyond the HFB approximation.. P. • Chapter 11 Scalable register initialization for quantum computing in an optical lattice. Guido Pupillo. Roura. Charles W. Lett. 053610 (2003). Roura. A 67. Blair Blakie. quantph/0403052. Clark. Rey. Gavin Brennen. Hu. Phys. Rey. Charles W. Edwards M. E. Burnett K. G. • Chapter 5 Bogoliubov approach to superﬂuidity of atoms in an optical lattice. Williams. Williams. to appear in J.
8 Chapter 1 Introduction .
the resultant optical potential felt by the atoms will have diﬀerent potential wells separated by a distance of the order of the laser wavelength. The depths of the optical potential wells that can be obtained in an experiment are in the microKelvin range. The idea of this chapter is to introduce the basic theory of optical lattices and to review the single particle properties of atoms loaded in such periodic potentials. optical lattices have several advantages with respect to solid state systems. in contrast to solids. where the lattice spacings are generally of order of Angstrom units. a spatially dependent intensity induces a spatially dependent potential energy. However. Therefore.1 Basic theory of optical lattices Neutral atoms interact with light in both dissipative and conservative ways. If such a modulation is obtained by the interference of several laser beams. the lattice can be moved by changing the polarization of the light or chirping the laser frequency and the lattice geometry can be modiﬁed by changing the laser conﬁguration. Optical lattices also can be controlled very easily by changing the laser ﬁeld properties. atoms can be trapped in this potentials when cooled at low temperatures. Cold atoms interacting with a spatially modulated optical potential resemble in many respects electrons in ionlattice potential of a solid crystals . They can be made to be largely free from defects. the lattice constants in optical lattices are typically three order of magnitude larger. When an atom interacts with an electromagnetic ﬁeld. The conservative interaction comes from the interaction of the light ﬁeld with the induced dipole moment of the atom which causes a shift in the potential .Chapter 2 Optical lattices Optical lattices are periodic potentials created by lightmatter interactions. Moreover. by laser and evaporative cooling techniques. Nevertheless. the energy of its internal states depends on the light intensity. such defects for example prevented the observation of Bloch oscillations in crystalline solids. For example the lattice depth can be changed by modifying the laser intensity. 2.
2. ∆ = ω − ωo .1 AC Stark Shift Consider a two level atom. 2. N has similar energy to the state with the atom in an excited state and N − 1 photons 1 ≡ . a ˆ (2.1.10 Chapter 2 Optical lattices energy called acStark shift. Because of conservation of momentum. the dissipation comes due to the absorption of photons followed by spontaneous emission. interacting with a monochromatic electromagnetic ﬁeld with frequency ω = 2πν as schematically shown in Fig. Figure 2.1) where a is the photon annihilation operator. If the detuning of the laser from ˆ the atomic transition.1. Laser cooling techniques make use of this light forces. is small ∆ ωo . 0 ≡ g. the net eﬀect is a dissipative force on the atoms caused by the momentum transfer to the atom by the absorbed and spontaneously emitted photons. On the other hand.1: AC Stark shift induced by atomlight interaction. then the state with the atom in the ground state and N photons in the ﬁeld. This is the physics that describes optical lattices. with internal ground state g and excited state e and energy diﬀerence ωo in a lossless cavity of volume V . Assume also that the experiment is performed within a time smaller than the spontaneous emission rate so that spontaneous emission can be neglected. For large detunings spontaneous emission processes can be neglected and the energy shift can be used to create a conservative trapping potential. The laser frequency is ω = 2πν which is detuned from the atomic resonance by Λ The uncoupled Hamiltonian describing the atoms and the electromagnetic ﬁeld is given by ˆ Ho = ωo e e + ω(ˆ† a + 1/2).
In the previous line we also assume a large number of photons and neglect the variation in the coupling constant due to ∆N. for example u(x) = e−ik·x ). Since the atoms are practically always in the ground state. d the dipole moment of the atom and u(x) the ﬁeld mode evaluated at the atomic position x.1. 2. valid in the limit ∆ ωo . the atoms are illuminated with superimposed counterpropagating laser beams.(for plane waves. In the rotating wave approximation. This .1 Basic theory of optical lattices 11 e. the energy of the atoms is changed according to the stark shift Ω(x)2 4∆ . However. the resonant process correspond to either the excitation of the atom along with the emission of a photon or the relaxation of the atom with the absorbtion of a photon. The interaction Hamiltonian is then reduced to Ω∗ (x) Ω(x) e g aei∆t + ˆ e g a† e−i∆t . Under the dipole approximation which assumes that the spatial variation of the electromagnetic ﬁeld is small compared with the atomic wave function.1 is given by E0. the ˆ coupling Hamiltonian denoted as HI in the interaction picture is given by: ˆ HI = −d · E = e g eiωo t + g e e−iωo t Ω∗ (x) 2 a† eiωt + ˆ ω N 2∈V (2.2) Ω(x) −iωt ae ˆ . ˆ 2 2 ˆ HI ≈ (2. i. ∆ 4∆ (2. are neglected and only the near resonant frequency processes are considered.3) Physically. with ε the Here Ω(x) is the Rabi frequency given by Ω(x) = −2 unit polarization vector of the ﬁeld. This energy shift is the so called acStark shift.e N N − 1.2. 0 and 1 . exp(±i(ωo + ω)t). N − 1 . the eﬀect of the interactions on the states. can be determined with second order (2) perturbation theory.1 = ± (2) ˆ Ω(x)2  1 Hint 0 2 =± . 2 u(x)ε · d. which deﬁnes the optical potential. ∆ Ω.2 Dissipative interaction In the above discussion we implicitly assumed that the excited state has an inﬁnite lifetime. E1 − E0 = − ∆. Furthermore. If the detuning is large compared to the Rabi frequency. the beams interfere and the interference pattern results in a periodic landscape potential or optical lattice. if instead of interacting with a monochromatic electromagnetic ﬁeld. the energy shift E0. In this case. The eﬀect of the interactions is to couple these states. in reality it will decay by spontaneous emission of photons.4) with the plus and minus sign for the 0 and 1 states respectively. the type of processes with a rapidly oscillating phase.
2. Because the eﬀective spontaneous emission rate of the atoms increases with the light intensity. Vlat (x) = Vo sin2 (kx) = . On one hand. such that a standing wave interference pattern is created.6) (2. due to the constructive interference of the lasers. If the excited state has a life time 1/Γe corresponding to a efolding time for the occupation probability of the state.8) ∆ where k = 2π/λ is the absolute value of the wave vector of the laser light and Vo is four times times the depth of a single laser beam without retroreﬂection. ∆ > 0. 4∆2 − Γ2 4∆ e Ω(x)2 Γe Ω(x)2 Γe ≈ . 4 ∆ − iΓe /2 Ω(x)2 ∆ Ω(x)2 ≈ . in a red detuned light ﬁeld. the laser detuning should be chosen as large as possible within the available laser power in order to minimize inelastic scattering processes and create a conservative potential. ∆ < 0. The proper detuning for an optical lattice depends on the available laser power I (Ω2 ∝ I) and the maximum dissipative scattering rate that can be tolerated. and the potential minima correspond to the points with zero light intensity.7) The real part of the energy corresponds to the optical potential whereas the imaginary part represents the the rate of loss of atoms from the ground state. the potential is attractive and the minima correspond to the places with maximum light intensity. the spontaneous emission in a red detuned optical lattice will always be more signiﬁcant than in a blue detuned one. with small detuning it is possible to create larger trap depths for a given laser intensity since the optical potential scales as V ∼ I/∆.12 Chapter 2 Optical lattices eﬀect can be taken into account phenomenologically by attributing to the excited state an energy with both real an imaginary parts. 2 4∆2 − Γ2 8∆2 e (2. (2. It can be created by retroreﬂecting a laser beam. On the other. On the other hand. the inelastic scattering rate is inversely proportional to the detuning squared and scales like Γe /∆ V . For blue detuning . Therefore. This results in a Rabi frequency Ω(x) = 2Ωo sin(kx) which yields a periodic trapping potential given by Ω2 o sin2 (kx). the corresponding life time for the amplitude will be twice this time.5) (2.1. The sign of the optical potential seen by the atoms depends on the sign of the detuning.3 Lattice geometry The simplest possible lattice is a one dimensional(1D) lattice lattice. the sign is positive resulting in a repulsive potential. The energy of the perturbed ground state becomes a complex quantity which we can write as E0 (2) = V (x) = γsc (x) = Ω(x)2 = V (x) + iγsc (x).
In this case the interference term vanishes and the resulting potential is just the sum of two superimposed 1D lattice potentials. Generalization to higher dimensions can be done straightforwardly. The main purpose of this section is to review the basic aspects that describe the behavior of noninteracting particle subject to a periodic potential. Typically a negligible frequency diﬀerence compared with the optical frequency is required to achieve independence. To create a two dimensional lattice potential for example. ε1 and ε2 are polarization vectors of the counter propagating set and φ is relative phase between them. In this case the lattice potential has the form Vlat (x. Consider a one dimensional particle described by the Hamilp2 ˆ tonian H = 2m + Vlat (x). they can be made independent by detuning the common frequency of one pair of beams from that the other. two orthogonal sets of counter propagating laser beams can be used. 68] . 2. This leads to a variation of the geometry of the lattice in a chequerboard like pattern. they interfere and the potential is changed depending on the relative phase of the two beams.9) Here k is the magnitude of the wave vector of the lattice light. where Vlat x) = Vlat (x + a).2. A more general class of two 2D lattices can be created from the interference of three laser beams [34. 3 we show a variety of possible 2D optical lattice geometries that can be made by three and four interfering laser beams. y) = Vo cos2 (ky) + cos2 (kx) + 2ε1 · ε2 cos φ cos(ky) cos(kx) . thus even in this case to a good approximation the wave vectors can be considered equal. Such lattices can provide tighter onsite conﬁnement.2 Single particle physics 13 Periodic potentials in higher dimensions can be created by superimposing more laser beams. especially if the lattice geometry is separable. Even if the polarization of the two pair of beams is the same.2 Single particle physics In this section for simplicity we are going to restrict the analysis to a one dimensional lattice. (2. Bloch’s theorem [67. A simple square lattice with one atomic basis can be created by choosing orthogonal polarizations between the standing waves. In Fig. If the polarization vectors are not orthogonal and the laser frequencies are the same.2. 2. 33] which in general yields non separable lattices. better control over the number of nearest neighbors and signiﬁcantly reduced tunneling between sites compared with the counter propagating four beam square lattice.1 Bloch functions One of the most important characteristics of a periodic potential is the emergence of a band structure. Similarly a 3D lattice can be created by the interference of at least 6 orthogonal sets of counter propagating laser beams.
The quantity q should be viewed as a quantum number characteristic of the translational symmetry of the periodic potential.(a)(e) potentials for diﬀerent conﬁgurations of 3 beams. This ﬁgure is a courtesy of P.11) Using this ansatz into the Schr¨dinger equation.14 Chapter 2 a Optical lattices 1 y [λ] 0 −1 (a) a2 a1 (b) a2 (c) 2 a1 a1 1 y [λ] 0 −1 (d) a2 (e) a 2 (f) a1 a2 a1 a1 −1 0 x [λ] 1 −1 0 x [λ] 1 −1 0 x [λ] 1 VLatt(r) [ER] −10 −5 0 5 Figure 2. ER is the atomic recoil energy. yields o (n) an equation for uq (x) given by: (ˆ+ q)2 p (n) (n) + Vlat (x) uq (x) = Eq u(n) (x) q 2m (2. it turns out that q plays the same fundamental role in the dynamics in a periodic potential as the momentum does in the absence of the lattice. Blair Blakie [34] ) states that the eigenstates φq (x) can be chosen to have the form of a plane wave times a function with the periodicity of the potential: φ(n) (x) = eiqx u(n) (x). ER = 2 k 2 /2m.10) (2. To emphasize this similarity .2: Optical lattice potential.12) Bloch’s theorem introduces a wave vector q. q (n) (n) (n) (n) (2. Hφq (x) = Eq φq (x). Even though it is not the same. q q u(n) (x q + a) = u(n) (x). (f) potential for the 4 counterpropagating laser beam conﬁguration (The two pair of light ﬁelds are made independent by detuning the common frequency of one pair of beams from the other. just as the momentum is a quantum number characteristic of the full translational symmetry of the free space.
each of them. For very deep lattices the spectrum is almost degenerate in q and exhibits a dependence on n similar to the one of a particle in a ﬁxed ﬁnite interval(determined by the period of the potential). 0 < x < a. i. is referred to as the band structure. s = V0 /4ER and a = Eq /ER − V0 /2ER . dx2 (2. because the wave vector q appears only as a parameter in Eq. It can be observed how the amplitude of the wave function in between adjacent lattice sites decreases with increasinglattice depth. As the depth increases. As the potential is increased the band structure appears. In this case the Schr¨dinger Equation is given by o − d2 (n) Eq V0 (2 − 2 cos(2y)) φ(n) (y) = φ(n) (y) φ (y) + q 2 q dy 4ER ER q (n) (2. Eq. (2. In general the wave vector q is conﬁned to the ﬁrst Brilloiun zone. which is the one used in experiments. 2. On the other hand. lattice depths are generally speciﬁed in recoil units. −π/a < q ≤ π/a. the band gap increases and the band width decreases. for an inﬁnite lattice.3 shows the band structure of a sinusoidal potential for diﬀerent potential depths. The description of energy levels in a periodic potentials in terms of a family of (n) continuous functions Eq each with the periodicity of a reciprocal lattice vector. . In the absence of a lattice the eigenfunctions of the free system are plane waves. 2π/a.2 Single particle physics 15 q is called the quasimomentum or crystal momentum . In the simple case of a sinusoidal potential.14) .13) where ER = 2 k 2 /2m is the atomic recoil energy and y = kx. In Fig. the particles are free so the spectrum is quadratic in q. Therefore. The index n appears in Bloch’s theorem because for a given q there are many solutions to the Schr¨dinger equation. For convenience. the energy levels for a ﬁxed n has to vary continuously as q varies. 2. one eigenvalue problem for each q. Fig.13) is just the Mathieu equation ([69]) d2 y + (a + 2s cos(2x))y = 0.12) can be seen as a set of eigenvalue o problems in a ﬁxed interval. For small V0 the discontinuity occurs only at the edge of the ﬁrst Brillouin zone qa ± π and the gap is proportional to V0 /2. s) is the characteristic parameter which is a complicated function of s and ν.4 the probability density of (n=0) the Bloch wave function φq=0 is plotted for diﬀerent lattice depths. Solutions of the Mathieu equation are generally written in the Floquet form eiνx P (x) where ν is known as the characteristic exponent and a = a(ν. the band structure can be solved analytically. For V0 = 0. Vlat (x) = V0 sin2 (kx). As the lattice depth is increased the barrier height between adjacent lattices sites increases and the eigenstates of the system tend to get localized at each lattice site (regions around the potential minima). In the lattice language (n) ν correspond to the quasimomentum .2. on general grounds has an inﬁnite family of solutions with a discretely spaced spectrum of modes labelled by the band index n. Equation (2.e. (2.12).
The oscillations are needed to ensure orhogonality between Wannier functions.5 qa Π 1 Figure 2.5 0 0.5 qa Π Vo 20ER 1 E ER 1 0.2. the plots use a logscale for the density color map.5 we show the Wannier orbital centered at the origin site for each of the lattice conﬁgurations shown in Fig.2.16 Chapter 2 Optical lattices Vo 0ER 8 6 4 2 0 8 6 4 2 0 E ER E ER Vo 0. q q (2.5 0 0.5 qa Π Vo 4ER 1 1 8 6 4 2 0 15 E ER 10 5 0 1 0. Under this approximation the Wannier function centered at the origin can be shown to be wn (x) ≈ u(n) (x) sin(kx) .10 to be approximately the same for all Bloch states in a band.3: Band structure of an optical lattice 2.5 qa Π 1 0.16) This looks like u(n) (x) at the site center. They constitute a more appropriate representation as the lattice depth is increased and particles get localized at individual lattices sites. 2.2. kx (2.2 Wannier orbitals Wannier orbitals are a set of orthonormal wave functions that fully describe particles in a band and are localized at the lattice sites.15) where the sum is over the ﬁrst Brillouin zone. Wannier orbitals are thus a unitary transformation of the Bloch functions and are formally an equivalent representation to describe the periodic system. M is the total number of lattice sites and xi is the position of the ith lattice site.5 0 0. .5ER 1 0. The actual form of a Wannier function may be seen if we assume the (n) periodic function uq (x) in equation 2. To observe the small oscillatory tails at the neighboring sites. but spread out with oscillations of gradually decreasing amplitude. In Fig. They are deﬁned as: 1 wn (x − xi ) = √ M e−iqxi φ(n) (x).5 0 0. The lattice depth in this pictures in deep enough that the Wannier orbitals are well localized.
2 Single particle physics Φ01 x Density 2 17 Φ01 x Density 2 0 1 2 3 xa Φ01 x 2 4 0 1 2 3 xa Φ01 x 2 4 Density 0 1 2 3 xa 4 Density 0 1 2 3 xa 4 Figure 2. By expanding the wave function in Wannier orbitals.18) = − = − ∗ dxwn (x − xi )Ho wn (x − xj )dx. ∗ dxwn (x − xi )U(x)wn (x − xj )dx.2. (2. (2.2. we get the following equations of motion −i with Jij Vij (n) ∂ (n) z = ∂t i −Jij δnn zj jn (n ) (n ) (t) + Vij (n.n ) . (n=1) 2.4: Probability density of the Bloch wave function φq=0 for diﬀerent lattice depths. It can be observed the localization of the wave function increases with the lattice increases. ψ(x. Sometimes a very good approximation is only to take into account overlap between nearest neighbor orbitals.3 Tightbinding approximation The tightbinding approximation deals with the case in which the overlap between Wannier orbitals at diﬀerent sites is enough to require corrections to the picture of isolated particles but not too much as to render the picture of localized wave functions completely irrelevant. (n) (2.i zi (t)wn (x − xi ). This tightbinding model is commonly used to solve the problem of a particle in a periodic potential when also an external potential is applied and it is going to be fundamental when considering particle interactions.n ) (n ) zj (t). t) = n.17) and using it in the Schr¨dinger equation that describes a particle moving in the o potential of a 1D lattice plus a perturbative external potential V (x).19) (2.20) (n.
Eq. ∗ dxw0 (x)Ho w0 (x)dx ∂ zi (t) = −J (zi+1 (t) + zi−1 (t)) + V (xi )zi (t) + εo zi (t).2. Vij = V (xi )δji δn.n ) sumed constant inside each individual lattice site. if the lattice is deep enough such that tunneling to next to nearest neighbors can be ignored. 2. (2. the equations of motion reduce to −i with ∗ dxw0 (xi )Ho w0 (x − xi+1 )dx. Moreover.22) (2. and the external perturbation is a slowly varying function such that it can be as(n.21) J = − (2. o For deep lattices the localized Wannier orbitals can be approximated by a Gaussian function. i.18 Chapter 2 Optical lattices 1 y [λ] (a) 3 2 1 1 2 1 3 (d) 2 3 2 1 3 2 1 1 2 3 3 (b) 3 1 3 1 3 2 1 3 3 1 (c) 3 2 1 0 3 −1 1 2 3 2 2 3 3 2 1 1 2 3 1 2 3 0 1 x [λ] 1 y [λ] 0 −1 (e) 1 1 3 2 3 2 2 1 1 2 0 x [λ] 2 (f) 3 2 1 3 −1 0 x [λ] 1 −1 1 −1 log10w0(r)2 [λ−2] −15 −10 −5 0 Figure 2. However.23) εo = where J is the tunneling matrix element between nearest neighboring lattice sites. This ﬁgure is a courtesy of P.21) is known as the discrete Schr¨dinger equation (DSE) or o tightbinding Schr¨dinger equation.n . Blair Blakie [34] ) If we assume that the external perturbation is not strong enough or sharp enough to induce interband transitions. ∂t (2. εo is the unperturbed on site energy shift and and zi the ﬁrst band coeﬃcients (0) zi (t) = zi (t). and so on. we may represent the moving particle quite satisfactory by using Wannier functions of only the ﬁrst band. Dotted white lines indicate the potential energy contours.e. The numbers labels the near neighbor sites: 1 the nearest neighbors.5: Wannier state logdensity distribution corresponding to the lattice conﬁgurations shown in Fig. because the Gaussian ansatz neglects the small .
2.2 Single particle physics
19
oscillations characteristic in the tail of the Wannier function, the tunneling matrix element J is then underestimated by almost an order of magnitude. For the special case of a sinusoidal potential Vlat (x) = Vo sin2 (kx) for which Mathieu functions are the exact solutions an analytic expression for J can be obtained by using tabulated Mathieu functions in Eq. (2.22) and ﬁtting the data. The best ﬁt we get is: √ β ˜ ˜ J/ER = α Vo e−γ Vo (2.24) ˜ with α =1.39666, β =1.051 and γ =2.12104 and Vo = Vo /ER .
2.2.4
Semiclassical dynamics
Formalism Solving Eq. (2.21) is presumably not possible for arbitrary V (xi ). However, we get general insight to the nature of solutions by an application of the correspondence principle. It is well known that wavepacket solutions of the Schr¨dinger equation o behave like classical particles obeying the equations of motion derived from the same classical Hamiltonian. The classical Hamilton equations are x= ˙ ∂H , ∂p p=− ˙ ∂H . ∂x (2.25)
Using the correspondence principle and the resemblance of the crystal momentum or quasimomentum to real momentum, the semiclassical equations of a wave packet in the ﬁrst band of a lattice can be written as ([67], [68]) dV (x) 1 dEq , q=− ˙ . (2.26) x = v (q) = ˙ dq dx The semiclassical equations of motion describe how the position and wave vector of a particle evolve in the presence of an external potential entirely in terms of the band structure of the lattice. If we compare the acceleration predicted by the model with the conventional newtonian equation, m¨ = −dV (x)/dx, we can x associate an eﬀective mass induced by the presence of the lattice, m∗ . This is given by
(0) (0)
1 1 d2 (0) = 2 2 Eq . m∗ dq Translationally invariant lattice
(2.27)
If no external potential is applied, V (x) = 0, plane waves are the solutions of DSE: (q) (q) zj (t) = fj e−it(Eq +εo )/ , fj = √1 eiqaj with M is the total number of lattice M sites. If also periodic boundary conditions are assumed, the quasimomentum q 2π is restricted to be an integer multiple of M a . The lowest energy band dispersion relation in this case is given by Eq = −2J cos(qa). (2.28)
20
Chapter 2
Optical lattices
From the above equation it is possible to see the connection between J and the band width J = Eq= π − Eq=0 /4. (2.29) a Linear external potential and Bloch oscillations If V (x) is assumed to be a linear potential, V (x) = Jξ x, it can be shown (see a for example [70],[71],[72]), that in this case the energy spectrum of Eq. (2.21) is discrete and evenly spaced (Es − E0 ) = Jξs s = 0, 1, 2, 3 . . . . (2.30)
The corresponding eigenfunctions are spatially localized and the amplitudes zj (t) given by : zj (t) = fj e−iEs t/ ,
(s) fj (s)
(2.31) (2.32)
(s)
= Js−j (−2/ξ) s = 0, 1, 2, 3, . . . .
With Js the sth order Bessel function of the ﬁrst kind. Because the quantities fj are just displaced Bessel functions, the sth eigenstate of the system tends to be exponentially localized around the sth site as ξ is increased. The localization is almost complete when ξ = 4, i.e. when the potential energy drop over a lattice period is equal to the zero ﬁeld band width. The localization can be observed in (0) Fig.2.6 where the ground state amplitudes fi are plotted for diﬀerent values of ξ.
Ξ 0.1 0.2 0.1 fi 0 fi 0 0 0.1 0.2 30 20 10 0 10 i site Ξ 1 20 30 0.4 0.2 0 0.2 0.4 30 20 10 0 10 i site Ξ 4 20 30 Ξ 0.5
0.6 0.4 0.2 0 0.2 0.4 0.6
0.8 0.6 0.4 0.2 0 0.2 30 20 10 0 10 i site 20 30
fi 0
30
20
10 0 10 i site
20
30
Figure 2.6: Ground state amplitudes fi diﬀerent values of ξ
(0)
An insight on the dynamics of a particle in this linear potential is obtained by calculating the time dependent probability of ﬁnding it at the site j at time t ,
fi 0
as a function of the lattice site i for
2.2 Single particle physics
21
zj (t)2 , if it was in the state localized at the origin at time zero zj (0) = δ0j . After some algebra it can be shown to be given by
2 Pj = zj (t)2 = Jj
4 sin(ξτ /2) ξ
,
(2.33)
with τ = tJ/ . The dynamical evolution of Pj=0 , for a given j consists of an initial period of growth until a maximum is reached followed by an oscillatory decay. The sequence is repeated again backwards until a time τ = 2π/ξ at which Pj=0 vanishes and a new cycle starts. In Fig. 2.7, the diﬀusion of a square wave packet occupying at t = 0 the central 20 sites is plotted. It can be seen that the center of mass of the wave function accelerates and then returns to its initial position after a period τ = 2π/ξ.
Τ 2Π Ξ 0. Τ 2Π Ξ 0.2
Ψi 2
40
0 20 i site Τ 2Π Ξ 0.4
20
40
Ψi 2 40
0 20 i site Τ 2Π Ξ 0.6
20
40
Ψi 2
40
0 20 i site Τ 2Π Ξ 0.8
20
40
Ψi 2 40
20
0 20 i site Τ 2Π Ξ 1.
40
Ψi 2
40
20
0 20 i site
40
Ψi 2 40
20
0 20 i site
40
Figure 2.7: Diﬀusion of a square wave packet in the presence of a linear potential.The parameters for the plot are ξ = 0.1. A simpliﬁed picture of the dynamics of a wave packet in this linear potential can also be obtained if we use the semiclassical model. If we assume a packet initially centered at x(t = 0) = 0, with average initial quasimomentum q(t = 0) = qo , the semiclassical model predicts periodic oscillations of the center of mass with frequency ξ and amplitude 2/ξ. 2 x(t) = (1 − cos(ξτ + qo a)). ξ (2.34)
22
Chapter 2
Optical lattices
This periodic motion, known as Bloch oscillations, has been observed in systems of cold atoms [73] and BoseEinstein condensates in optical lattices [38, 36]. On the other hand, if the external potential is strong enough to cause interband transitions, the tightbinding model does not apply anymore. In this case the atoms may gain enough energy from the external ﬁeld to tunnel out through the energy gap into a second band when they reach the Brillioun zone boundary. This phenomenon is known as of LandauZener tunneling and it also has been observed in experiments [40]. Harmonic potential A case of experimental interest is when, besides the periodic potential, the system 2 is subject to a harmonic conﬁnement, V (x) = 1 mωT a2 (x/a)2 ≡ Ω(x/a)2 , with ωT 2 the trapping frequency (note that Ω is unrelated to the Rabi frequency discussed in section 2.1). The energy spectrum in this case is determined by the equation Es fi
(s)
= −J(fi+1 + fi−1 ) + Ωi2 fi .
(s)
(s)
(s)
(s)
(2.35)
It is possible to show that the solutions fi of Eq. 2.35 are the Fourier coeﬃcients of the periodic Mathiue function with period π and the eigenvalues are the characteristic value of these periodic Mathiue functions [69]. The symmetric and antisymmetric solutions and eigenvalues are given respectively by
(s=2r)
fi fi
= =
(s=2r+1)
Es=2r =
1 2π 4J cos(2ix)dx r = 0, 1, 2..., ce2r x, π 0 Ω 1 2π 4J sin(2ix)dx, se2r x, π 0 Ω 4J Ω 4J Ω a2r Es=2r+1 = b2r , 4 Ω 4 Ω
(2.36)
(2.37)
with ce2r (x, 4J/Ω) and se2r (x, 4J/Ω) the even and odd periodic solution (with period π) of the Mathieu equation with characteristic parameter a2r (4J/Ω) and b2r (4J/Ω) respectively. The eigenvalues Es are complicated functions of 4J and Ω therefore we are only going to consider expressions for certain limits: • Case J Ω: This is the case where most experiments have been developed, for example for a 5 ER lattice the parameter q = 4J varies between 1.7x104 Ω for a magnetic trap of 9 Hz to 35 for a trap of 200 Hz. In this regime the excitation spectrum is given by: √ 1 4ΩJ(s + ) − Ω 2 (2s + 1)2 + 1 32 + ...
Es = −2J + −Ω
(2.38)
Ω J
(2s + 1)3 + 3(2s + 1) 210
12 1 + . 2m∗ ∂x2 2 m∗ ω ∗ . . 2 (2.41) The average width of the wave function is just aho = • Case J Ω: The single particle spectrum in this case is given by Es=2n ≈ Es=2n−1 = Ω 4 (2n)2 + Ω 4 4J Ω 2 1 +O ((2n)2 − 1) 2 4J Ω 4 + .42) 4+ 4− − 5 + ... The diﬀerential equation obtained is just the harmonic oscillator equation.8 the spectrum is shown for diﬀerent values of 4J . 2 It can be appreciated that the spectrum is nearly twofold degenerate and it goes like Ωn2 to ﬁrst order in (J/Ω) . 30 (2. In this regime then. n ≥ 3..2 Single particle physics 23 which to a ﬁrst approximation is just the harmonic potential found if no lattice were present but with an eﬀective frequency ω ∗ and eﬀective mass m∗ given by ω∗ = m∗ = √ 4ΩJ = ωt 2 m . The harmonic character of the spectrum can be easily checked if we take the continuous limit of Eq.35). the trapping potential is what mostly determines the energy spacing. . 2. .39) (2.40) 2Ja2 .. (2. (Es + 2J)f (s) (x) = − ∂f (s) (x) 1 ∗ ∗2 2 (s) + m ω x f (x). m∗ (2..2. In Fig. It is worth it to emphasize that this result was derived under the tightbinding approximation and therefore it is not valid for very shallow lattices. E4 ≈ E3 = E2 = E1 = E0 = Ω 4 Ω 4 Ω 4 16 + 4J Ω 4J Ω 4J Ω 2 2 4J Ω 1 + .. 12 2 1 + . . Ω .
8: Harmonic conﬁnement plus lattice spectrum: The index n labels the eigenvalues in increasing energy order. .1 20 321 En J En J 5 10 nth 15 20 162 4 3208 1624 40 5 10 nth 15 20 Figure 2.24 Chapter 2 Optical lattices 4J En J 4 2 5 10 nth 4J 100 33 En J 17 1 5 4J 10 15 1 20 10 15 nth 4J 0.
critical behavior of 4 He and. It includes the main physics that describe strongly interacting bosons. cold atoms in optical lattices [43]. 75. The simplest non trivial model that describes interacting bosons in a periodic potential is the Bose Hubbard Hamiltonian. The Bose Hubbard Hamiltonian predicts a quantum phase transition from a superﬂuid to a Mott insulator state. recently. This transition has been observed experimentally in atoms conﬁned in a 3D optical lattice [46]. This model has been used to describe many diﬀerent systems in solid state physics. 3.1 BoseHubbard Hamiltonian We begin the analysis by considering the second quantized Hamiltonian that describes interacting bosonic atoms in an external trapping potential plus lattice. which is the competition between kinetic and interaction energy. like short correlation length superconductors.Chapter 3 The BoseHubbard Hamiltonian and the superﬂuid to Mott insulator transition In chapter 2 we studied the physics of non interacting particles in a periodic system. Josephson arrays. In this chapter we go one step ahead and describe the basic properties of a bosonic many body system with repulsive interactions in a periodic lattice potential. In this chapter we start by reviewing the basic properties of the Bose Hubbard Hamiltonian and outline the most important aspects that characterize the superﬂuid to Mott insulator transition. It has been studied analytically with many diﬀerent techniques such as mean ﬁeld approximations [74. In the grand canonical ensemble it is given by: . Numerically most of the studies are based on quantum Monte Carlo methods and density matrix renormalization techniques [79]. renormalization group theories [44] and strong coupling expansions [78]. 77].
4) . the only relevant scattering process is the swave scattering. and an is the annihilation operator at site n which obeys bosonic canonical ˆ commutation relations.3) is inserted in H and only tunneling between nearest neighbors is considered. where we used Wannier orbitals to expand the single particle wave function. ˆ where Φ† (x) is the bosonic ﬁeld operator which creates an atom at the position x. ˆ (3.m> ˆ HBH with = −J a† a† an an + ˆn ˆn n n (Vn − µ)ˆ† an .3) where w0 (x) is the Wannier orbital of the lowest vibrational band localized at the origin. an ˆ (3. (3. (3. Vlat (x) is the periodic lattice potential. only the ﬁrst band has to be considered and the ﬁeld operator can be written as: ˆ Φ(x) = n an w0 (x − xn ). µ is the chemical potential and acts as a lagrange multiplier to ﬁx the mean number of atoms in the grand canonical ensemble and Vat (x) is the interatomic scattering potential. since for cold gases the thermal de Broglie wavelength is much larger than the eﬀective extension of the interaction potential. The sum is taken over the total number of lattice sites. it is convenient to expand the ﬁeld operator in a Wannier orbital basis. we obtain the Bose Hubbard Hamiltonian 1 a† am + U ˆn ˆ 2 <n. to a good approximation. However.Chapter 3 26 The BoseHubbard Hamiltonian and the superﬂuid to Mott insulator transition ˆ H = + ˆ dxΦ† (x) − 1 2 2 2 2m ˆ ˆ ˆ + Vlat (x) Φ(x) + (V (x) − µ)Φ† (x)Φ(x) (3. In this thesis we are going to assume that the scattering length is positive so the interactions are repulsive. If the chemical potential of the system is less than the single particle excitation energy to the second band. Similar to the noninteracting situation. which is in general a complicated function. V (x) denotes any additional slowlyvarying external potential that might be present (such as a magnetic trap).1) ˆ ˆ ˆ ˆ dxdx Φ† (x)Φ† (x )Vat (x − x )Φ(x)Φ(x ). the interatomic potential can be replaced by an eﬀective contact interaction: Vat (x) ≈ 4πas m 2 δ(x). Therefore. ˆ If Eq.2) with as is the scattering length and m the mass of an atom. and the actual interparticle potential plays a minor role.
5) to the oscillatory tails characteristic of the Wannier orbitals. 3. N is the mean number of atoms.5) (3. The parameter U measures the strength of the repulsion of two atoms at lattice site n.6) (3. V0 .e. 2π aho (3. J is the hopping matrix element between nearest neighbors (see also 2. and a Gaussian approximation can be used to estimate it [43]. 2m (3. Under the Gaussian approximation. As the intensity of the lattice is increased the tunneling rate decreases exponentially and the onside d/4 interaction increases as a power law. The ﬁrst term in the Hamiltonian proportional to J is a measure of the kinetic energy of the system. Here the notation < n. (3. dx w0 (x)4 . (d is the dimensionality of the lattice).2 SuperﬂuidMott insulator transition 27 J U Vn = − = ∗ dxw0 (x − xn ) ˆ p2 + Vlat (x) w0 (x + xn+1 ).7) 4πas 2 m = V (xn ). with ahoj = mωhoj and .The integral (3.6) is not as sensitive as Eq.2 SuperﬂuidMott insulator transition At zero temperature the physics described by the BoseHubbard Hamiltonian can be divided into two diﬀerent regimes.8) with the bar indicating the geometric mean. Using this ansatz in the integral (3. This inequality is readily satisﬁed in practice.3. has the form w0 (0) ∼ ωhoj = given by 4Er Vj0 2 j exp − x √ j 2ahoj 2 . i. One is the interaction dominated regime . the mean interaction energy per particle U N must be smaller than the energy gap to the ﬁrst vibrational excitation. the ground state wave function centered at the origin of a separable sinusoidal lattice. m > restricts the sum to nearestneighbors sites. The third term in the Hamiltonian takes into account the energy oﬀset at site n due to a slowly varying external potential V (x).6) yields an onsite interaction U = a ω √ s ho . and to a good approximation they can be neglected. with lattice depth Vj0 in the j th direction. Nexttonearest neighbor tunneling amplitudes are typically two orders of magnitude smaller than nearestneighbor tunneling amplitudes. The second term determines the interaction energy of the system. U N < ω ho .24). The realization of the BoseHubbard Hamiltonian using optical lattices has the advantage that the interaction matrix element U and the tunneling matrix element J can be controlled by adjusting the intensity of the laser beams. To guarantee the validity of the one band approximation.
Because atoms are delocalized over the lattice with equal relative phases between adjacent sites. The onset of superﬂuidity is a consequence of the competition between the kinetic energy.[74][76] that describes the Mott insulator transition of a translational invariant system with Vn = 0 exhibits lobelike Mott insulating phases in the Jµ plane. quantum correlations can be neglected and the system can be described by a macroscopic wave function since the many body state is almost a product over identical single particle wave functions. Atoms tend to get localized at individual lattice sites and number ﬂuctuations are reduced. by decreasing µ from a point in the Mott phase. There is a macroscopic welldeﬁned phase and the system is a superﬂuid. The lowest lying excitations that conserve particle number are particlehole excitations (adding plus removing a particle from the system). In the Mott insulator phase the ground state of the system instead consists of localized atomic wave functions with a ﬁxed number of atoms per site. The gap is determined by the energy necessary to create one particlehole pair. The phase diagram [44]. as expected from an array of phase coherent matter wave sources. Similarly. Mott insulator phases occur only at integer densities. With an extra particle free to move around the lattice phase coherence is recovered and the system enters the superﬂuid regime. 3. at some point eventually it will be energetically favorable to remove one particle from the system. Since the kinetic energy of the system increases with J the width of the lobes decreases with J. and the interaction energy. ∂ρ/∂µ with ρ the average density of the system. which tries to delocalize the particles. The other is the kinetic energy dominated regime. and the system is in the Mott insulator phase. This phase is characterized by the existence of an energy gap. As interaction increases the average kinetic energy required for an atom to hop from one site to the next becomes insuﬃcient to overcome the potential energy cost. there is going to be a point at which the kinetic energy of adding an extra particle and letting it hop around will balance the interaction energy cost. vanishes. In the superﬂuid regime. Also at J = 0 the intersection points between the lobes with density ρ = n and the ones with density ρ = n + 1 are degenerate. see Fig.1. At J = 0 the gap is just equal to U . noninteger density contours lie entirely in the superﬂuid phase because there is always an extra particle that can hop without energy cost. Inside these lobes the compressibility. Because there is no energy barrier to the addition of an extra particle at these degenerated points the system remain superﬂuid. Each Mott lobe is characterized by having a ﬁxed integer density. The extra mobile hole will induce also phase coherence and the system will condense in a superﬂuid state. they exhibit an interference pattern when the lattice is turned oﬀ.Chapter 3 28 The BoseHubbard Hamiltonian and the superﬂuid to Mott insulator transition when J is much smaller than U . where tunneling overwhelms the repulsion and the system exhibits superﬂuid properties. The distance in the µ direction at ﬁxed J between the upper and lower part of the lobe is the energy gap. The physics behind this diagram can be understood as follows: If we start at some point in the Mott insulating phase and increase µ keeping J ﬁxed. . the kinetic energy term dominates the Hamiltonian. In this regime. which tries to localize them and make the number ﬂuctuations small.
(3. The parabolic phase boundary of the lobes in the plane µ vs (J/U ) vs is given by 1 (2n − 1) − Jt/U ± + (Jt/U )2 − 2(Jt/U )(2n + 1) + 1 . The solution of the mean ﬁeld model [74]. with d the dimensionality of the system. the compressibility has a singular part scaling near the transition as δ −α . where besides the lattice there is a harmonic trap that collects the atoms at the center. One type takes place at any generic point of the phase boundary. the system can be treated locally as an homogeneous system. This transition is driven at ﬁxed density by decreasing U/J and enabling the bosons to overcome the on site repulsion.9) where n is the integer density of the lobe. it is possible to deﬁne an eﬀective local chemical potential. and it is driven by the energy cost to add or subtract small numbers of particles to the incompressible Mott state as explained above. as the ratio U/J is changed the system can locally cross the boundary between the superﬂuid and Mott insulator phases. µn = µ−Vn . The two kinds of phase transition belong to diﬀerent universality classes. t the number of nearest neighbors.10) 2 µ/J = Up to this point we have discussed the basic aspects of the superﬂuidMott insulator transition in a translationally invariant system. the parameter equivalent to the reduced temperature δ = T − Tc for ﬁnite temperature transitions that measures the distance from the transition is δ ∼ µ − µc . Therefore even for the situation when the local density was not commensurate at . This scaling does not apply at the special ﬁxed density MottSuperﬂuid transition for which δ ∼ (J/U ) − (J/U )c and diﬀerentiation with respect to delta becomes inequivalent to diﬀerentiation with respect to the chemical potential. In this case the density of atoms is not ﬁxed since the atoms can redistribute over the lattice and change the local ﬁlling factor. (3. To deal with the inhomogeneous case.2 SuperﬂuidMott insulator transition 29 The phase diagram includes two diﬀerent types of phase transition. the local chemical potential is ﬁxed by the density. with µc the chemical potential at the phase boundary. at each lattice site n [43]. This is the case realized in experiments. On the other hand. α > 0. Because in the inhomogeneous case. As shown in ([44]) provided δ ∼ µ − µc . In the generic one. For the special ﬁxed density on the other hand one must take δ ∼ (J/U ) − (J/U )c . The situation is fundamentally diﬀerent for a inhomogeneous system with a ﬁxed total number of atoms and external conﬁnement. t = 2d. If the change in the mean number of atoms between neighboring sites is small.3. the other type only occurs at ﬁxed integer density and takes place at the tip of the lobes.[76] predicts that the critical value at the tip of the Mott lobe depends on the density and dimensionality of the lattice as (U/J)c = t 2n + 1 + (2n + 1)2 − 1 . Most of the features of the phase diagram discussed above can be veriﬁed by simple calculations done using a mean ﬁeld approximation.
They found that Mott phases exist in extended domains above a threshold interaction strength.1: Zerotemperature phase diagram. the authors used quantum Monte Carlo simulations to study the ground state of the one dimensional Bose Hubbard model in a trap. . The vertical lines indicate critical values for diﬀerent ﬁlling factors the beginning. For the inhomogeneous case. the gradient in the local chemical potential leads to a shell structure with Mott insulator regions and superﬂuid regions in between.Chapter 3 30 The BoseHubbard Hamiltonian and the superﬂuid to Mott insulator transition Phase Diagram n 4 µU n 3 n 2 n 1 JUc JU Figure 3. as the ratio U/J is changed a local phase transition can take place. even without the commensurate ﬁlling required in the translationally invariant case. In [79].
2) x −x→∞ Strictly speaking. In this regime most of the atoms occupy the same condensate wave function. The function ϕ(x) is a classical ﬁeld having the meaning of an order parameter and is often called the ”condensate wave function”. It characterizes the oﬀdiagonal long range order present in the one particle density matrix [81].1 Discrete nonlinear Schr¨dinger equation (DNLSE) o In the weakly interacting regime. 80]. 4. ˆ Φ(x) = ϕ(x).1) This procedure can be interpreted as giving to the ﬁeld operator a non zero average. neither the concept of broken symmetry. lim ˆ ˆ Φ† (x )Φ(x) = ϕ∗ (x )ϕ(x). to a good approximation the ﬁeld operator can be replaced by a cnumber. . quantum ﬂuctuations can be neglected to a good approximation. ˆ Φ(x) → ϕ(x). However the condensate wave function can still be determined by diagonalization of the one particle density matrix.Chapter 4 Mean ﬁeld theory In this chapter we introduce the mean ﬁeld Discrete Nonlinear Schr¨dinger equao tions and use it ﬁrst to investigate the behavior of a BEC adiabatically loaded into an optical lattice moving at constant velocity. this deﬁnition of BEC corresponds to an spontaneous symmetry breaking [11. The eigenstate with larger eigenvalue corresponds to ϕ(x). (4. The macroscopic occupation of a single mode implies that the commutator of the annihilation and creation operators that create and destroy a particle in the condensate mode can be neglected with respect to the total number of atoms in the mode. (4. Because the original Hamiltonian is invariant under global phase transformation. and thus a well deﬁned phase to system. nor the one of oﬀdiagonal long range order can be applied to ﬁnite size systems with well deﬁned number of particles. Therefore. and second to model the dynamical evolution of a BEC initially loaded into every third site of an optical lattice. 12.
2 (4. the parameters J and U are the ones deﬁned in Eq.1) in the manybody Hamiltonian (3.i.3) dxϕ† (x)ϕ† (x)ϕ(x)ϕ(x).32 Chapter 4 Mean ﬁeld theory If we use the approximation (4.1) we get a classical Hamiltonian given by: 2 H = + dxϕ† (x) − 1 4πas 2 m 2 2 2m + Vlat (x) ϕ(x) + (V (x) − µ)ϕ† (x)ϕ(x) (4.4). the condensate order parameter can be expanded in a Wannier basis keeping only the lowest band orbitals: ϕ(x. The normalization condition for the amplitudes is given by: zn 2 = 1.5) and (3. m > restricts the sum over nearest neighbors.m> ∗ zn zm + n Vn zn 2 + U zn 4 .7) Notice the DNLSE can also be derived by assuming the ﬁeld operators present in the BoseHubbard Hamiltonian to be cnumbers.6) and N is the total number of particles. which can be minimized using the variational principle. (4.e. (3. The time dependent equations of motion for the condensate wave function can be ∂ ˆ obtained from the variational principle by minimizing the quantity i ∂t − H. H = −J <n. the GrossPitaevskii equation reduces to the discrete nonlinear Schroedinger equation (DNLSE).6) where < n. It was derived by independently by Gross and Pitaevskii [10].m> (4. If the chemical potential is small compared to the vibrational level spacing and the lattice depth is deep enough that a tight binding picture is valid.5) Using this ansatz in Eq. an = zn (t). = −J ∂t <n. t) = n zn (t)w0 (x − xn ). n (4. where the zn amplitudes satisfy: i ∂zn zm + (Vn + U N zn 2 )zn . This assumption ˆ leads to a tightbinding classical Hamiltonian.8) .4) This equation is known as the GrossPitaevskii equation (GPE). The minimization yields ∂ϕ(x) ∂t 2 i = − 2 2m + Vlat (x) + V (x) + 4πas m 2 ϕ(x)2 ϕ(x) (4. (4.
21). The only diﬀerence with respect to the single particle case is an energy shift in the dispersion relation which takes into account the average mean ﬁeld energy U no . In the translational invariant case with no other external potential present. This loop reﬂects the existence of new solutions. but if interactions are strong enough the superﬂuid ﬂow can no longer be stopped. Then we discuss the eﬀects of the trapping potential and interaction eﬀects on the dynamics. which takes into o account mean ﬁeld interaction eﬀects. that outside the range of validity of a tightbinding model. but with an extra cubic term.9) In the tightbinding approximation. zn = no exp(−inq) exp (−iEq t/ ). mean ﬁeld interactions do not modify the shape of the lowest band: it has still a cosine dependence with bandwith equal to 4J. especially at the edge of the Brillouin zone. We ﬁrst study the dynamics in a translationally invariant lattice using a simple single particle picture. To see the modiﬁcation that the nonlinear term introduces in the dispersion relation.2 Dragging experiment 33 The DNLSE is similar to its linear counterpart. For free particles the ﬂow is stopped completely by Bragg scattering from the periodic potential. self focusing eﬀects and dynamical stabilities might be observed.2 Dragging experiment In this section we investigate an experiment done at NIST where a BoseEinstein condensate was prepared in a harmonic trap and adiabatically loaded into a one dimensional optical lattice moving at constant velocity Ref. 4. the lowest band becomes triple valued near the Brilloiun zone edge and a loop appears at this point.4. as opposed to the linear case. characteristic nonlinear eﬀects such as solitons. . It has been shown in the literature [82. the DSE or tightbinding Schr¨dinger equation (see 2. inﬂuencing the condensate dynamics in the lattice. they cause the eﬀective mass of the Bloch state to depend on time. nonlinear eﬀects does not necessarily change the dispersion relation in a simple way. Using this plane wave ansatz in the DNLSE we get a dispersion relation given by: Eq = −2J cos(qa) + U no (4. These new solutions have a non zero velocity at the edge of the Brillouin zone. with no = N/M the condensate density. It is important to mention though. [37]. 83] that if the chemical potential is bigger than the gap vibrational excitation energy to higher bands. which only exist when the tightbinding approximation is not valid. and show how the centerofmass motion of the condensate can be used as a probe of the lattice band structure. As a consequence. we consider a one dimensional optical lattice with M sites and periodic boundary conditions. When mean ﬁeld eﬀects are included. the solutions of the DNLSE that are similar to those of the noninteracting √ case are. Vn = 0. The nonzero velocity carried by the Bloch waves is a manifestation of the superﬂuidity of the system.
up to some maximum height. The dashed line indicates the ﬁnal location of a comoving point with the optical lattice. linearly in time. The temporal dependence of the optical lattice depth is shown in Fig. Figure 4. a condensate prepared in the ground state of a harmonic trap is loaded in an optical lattice. The general observation is that the deeper the lattice. which is moved at constant velocity. at t = tr .1 Experiment In the experiment. In the ﬁgure.2. the greater the response of the condensate to the lattice motion.1: Condensate density in the xdirection at the observation time tf . In this experiment.34 Chapter 4 Mean ﬁeld theory 4. the lattice depth is held constant for a ﬁxed period of time and the system is allowed to expand in the lattice.1. After the maximum is reached. The loading of the condensate in the optical lattice is done by turning on the lattice. experimental results obtained for the ﬁnal condensate density for diﬀerent lattice depths are also shown. Light colors areas indicate regions of high condensate density. Finally at t = tf the condensate density distribution along the lattice axis is imaged. the condensate has a very narrow momentum spread compared to the lattice recoil momentum. . The important part of this turning on is that it is done suﬃciently slow to be adiabatic with respect to band excitations. 4.
During this period of time the average velocity of the atom. the average velocity of the atoms is given by: vatom = 1 dEq ko + m dq (n) q=−ko .2 we plot the atom’s average velocity in the lab frame vs. has the form ϕ(x ) = e−iko x . In Fig. but due to the fact that the Hamiltonian is invariant under translation. in the lattice frame. (4.2 Dragging experiment 35 4.2. during tr to tf . How much the atoms are dragged depends on the derivative of the dispersion relation evaluated at the loaded quasimonentum. the optical potential is ramped up. with vlatt = ko /m (we used primes to denote the coordinates and wave functions in the moving frame). the initial wave function can be approximated by a plane wave with zero momentum. (4. the quasimomentum is conserved (at least in this linear model with no other potential than the lattice) during the ramping time and the average quasimomentum at the end of the loading should remain −ko .11) is calculated using the Bloch dispersion relation for a three recoil lattice. Even though the tightbinding velocities do not exactly match the velocities calculated using the Bloch dispersion relation.10) implies that in the lab frame. .11) Because x(tf ) is the observable of the experiment. Transforming back to the lab frame. 4. (4. From tr to tf the lattice depth is held constant. The velocity in Eq. (4.12) Therefore. the wave function evolves. vlat = ko /m. to move the lattice at diﬀerent speeds corresponds to loading the condensate at a diﬀerent Bloch state. the lattice velocity. which is moving with velocity vlatt with respect to the lab frame. We emphasize the following points: • Because the initial quasimomentum is determined by the lattice velocity.10) with Eq the nth band Blochlike atomic dispersion relation evaluated at the ﬁnal lattice depth. This wave function. by measuring the average position of the atoms for a ﬁxed tf − ti the experiment measures the atom’s velocity. is: v= (n) 1 dEq dq (n) q=−ko . Eq. which is a determined by the band structure of the lattice. Because the momentum spread of the condensate is very narrow compared with the lattice recoil momentum at t = 0. we can calculate it as δx = x(tf ) − x(tr ) ≈ vatom (tf − tr ).4.(4. we can say that for a three recoil lattice the tightbinding approximation is fair. as seen in the lattice frame. The tightbinding results (dotted) are also shown.2 Linear free particle model In this section to a ﬁrst approximation we are going to neglect interactions and apply a single particle picture to understand the dynamics. As t increases.
5 0 0 0.2: Atom’s average velocity in the lab frame vs. On the other hand. • For small lattice velocities (ko a 1. ko = −qB . a the lattice constant) the atoms are almost stationary because then the dispersion relation is more freeparticle like.5 1 vlatt vB 1. as opposed to the (n) free free system. 1 dEq dq (n) ko /m and therefore Eq. in the lattice frame. Following this model. as ko approaches the k of the Brillouin zone edge.5 Figure 4.2. vB = π/am. q=−ko = 4.11) predicts no dragging. dots: tightbinding approximation results.e dEq /dqq=−qB = 0. the time evolution of the average position and wave vector of . the strength of the harmonic trap is less than one percent of the potential depth and we expect the semiclassical model to be valid. lattice velocity.36 Chapter 4 Mean ﬁeld theory Vo 3 Er 1 TBA Bloch vatom vB 0. the atoms are dragged the most. Line: results from the Bloch dispersion relation. As the lattice depth increases the dispersion relation desviates more from freeparticle behavior and dragging ensues. This is consistent with the experimental results shown in Fig.3 Eﬀect of the magnetic conﬁnement In the experiment a harmonic conﬁnement was present in addition to the lattice potential. Bloch waves carry zero velocity. To model the eﬀect of the magnetic trap on the atoms dynamics we used a semiclassical model (see chapter 2). • At zero lattice depth the dispersion relation is quadratic. 4. i.1. At this point. (4. The semiclassical model is a very good approximation whenever it is unnecessary to specify the position of the atom on a scale comparable with the spread of the wave packet. The validity of the semiclassical model requires that the external potential varies slowly over the dimension of the wave packet and ignores the possibility of inter band transitions. In the experiment.
then at tf the ﬁnal quasimomentum is q(tf ) = −ko + ∆(ko .13) with ω the magnetic trapping frequency. tf ) is a negative quantity because the force is a restoring force. trapped atoms move faster than the untrapped ones.26): x = v (n) (q) = ˙ 1 dEq . tf ). trapped atoms slow down compared with the ones in the homogeneous lattice. is less or greater than the velocity of the atoms without the external trap depending on the . the atoms feel a force opposite to the direction of motion. vB = π/am. dq (n) 37 q = −mω 2 (x + vlat t). vB < vlatt 3vB /2.000025. Here ∆(ko . However.4. For larger velocities. The force not only increases with time but also with the lattice velocity (see Eq. 4. the ﬁnal velocity of the atoms in the trap . (4.5 Figure 4. (2. the opposite behavior is observed and for an interval of velocities. This behavior can be understood as follows: Because of the trap. if the lattice velocities exceeds the Brillouin velocity. We observe in the plot that when atoms are loaded in the ﬁrst band.13)). Therefore. the motion of the atoms with and without trap is almost the same. The initial conditions are k(0) = −ko and x(0) = 0. We solved these equations of motion. We used the parameters V0 = 3ER and 1/2mω 2 /ER = 0.3. lattice velocity in the presence of a magnetic trap. Semiclassical 1 trap no trap model Vo 3 Er vatom vB 0.5 1 vlatt vB 1. vB . as long as the atoms remain in the ﬁrst band. The results are shown in Fig. The average quasimomentum of the system evolves in the presence of the force.3: Atom’s average velocity in the lab frame vs. ˙ (4.5 0 0 0.2 Dragging experiment the atoms are given by Eq. using the analytic dispersion relation for a sinusoidal lattice in terms of Mathieu functions. for a range of lattice velocities below half the Brillouin velocity (vlatt vB /2 with vB = π/am). and if at time t = 0 the atoms are loaded with quasimomentum q(0) = −ko.
k = − ∂ξ ˙ ξ = 2J sin(k)e−η . Besides the sign diﬀerence in the curvature the second band has a larger band width than the ﬁrst one.).17) (4. We numerically integrated the equations of motion assuming no other external trapping potential besides the lattice.. was found using Mathieu functions assuming a 3ER lattice. ˙ ˙ δ = 2J cos(k)e−η 2 2 (4. 4.2. The curve found is plotted in When mean ﬁeld eﬀects are included. The bandwidth J. (0 < q < π/2a. In these reference the authors assume a Gaussian proﬁle wave packet zn (t) = 4 2 [−(n−ξ)2 /γ 2 +ik(n−ξ)+iδ(n−ξ)2 /2] e . 4 − δ2 γ4 γ = 2J cos(k)e−η γδ. Depending on the the initial quasimomentum and width of the packet.15) (4. when the eﬀective mass is negative. V = 2/γ 2 π dxV (x) exp[−2(x − ξ)2 /γ 2 ] and the time scaled as t → tEr / . If the eﬀective mass is positive. k. We plot Fig.16) 4 ∂V 2N U . δ as variational parameters. the dynamics of the system becomes very rich. In this case the Hamiltonian is invariant under translations and we expect conservation of the quasimomentum. The parameters were chosen according to the experiment: γ(0) = 10. 85]). Notice that the variable k is just the average quasimomentum of the wave packet.4 vatom vs vlatt evaluated at t = 50. because if V = 0 then dk/dt = 0. .38 Chapter 4 Mean ﬁeld theory sign of the eﬀective mass (sign of the curvature of the dispersion relation) evaluated d2 E at the ﬁnal momentum. π/a < q < 3π/2a. We can also associate an inverse eﬀective mass given by m∗−1 = 2J cos(k)e−η .4 Interaction eﬀects To model the eﬀect of the interatomic interactions on the dynamics we used the tightbinding variational method proposed in Ref. On the other hand. 4. γ2π (4. δ(0) = 0 and N U = 2ER .).14) and used ξ.18) γ 1 with η = 2γ 2 − δ 8 . The more pronounced diﬀerence between trapped and untrapped atom velocities observed in the second band with respect to the ﬁrst one...(π/2a < q < π/a. ([84. γ. the group velocity is a decreasing function of q and the trapped atoms speed up with respect to the untrapped ones. +√ 3− γ ∂γ πγ (4.. is not only because ∆ increases with the lattice velocity but also because of the discontinuous change of the band structure when crossing the Brillouin zone edge. dq2q > 0. V = 0.. the group velocity is an increasing function of q and the velocity of the trapped atoms is less than the untrapped atoms’ velocity.. The equations of motion they found are: ∂V ˙ . This is seen in the above equations.
The ﬁnal value for γf . Physically. again assuming γo 1.8 vlatt vB 1 1.8 vatom vB 0.19) With γo = γ(0) the initial width of the packet. m∗ = eη /(2J cos k). the diﬀerent regimes can be understood by realizing that the nonlinear term makes the eﬀective mass time dependent. and for special cases solitons or breathers [84]. is characterized by the unbounded growth of the width of the packet. U N − 4J πcos(k)γo (4.2 0 0 0.20) . diﬀusion. The model predicts that the transition between the two regimes occurs. γf = U N γo √ . for a packet in the selffocusing regime predicted by the variational model is.6 0.4 Figure 4.6 0. The eﬀective mass. a variety of dynamical regimes can be observed. at √ (U N )c = 4J πcos(k)γo k < π/2 √ 2J πcos(k) (U N )c = k > π/2 γo (4.4 0. not only depends on the initial quasimomentum but also on the width of the wave packet. on the other hand. In particular the selftrapped wave packet cannot translate along the array. The selffocusing is a genuine nonlinear eﬀect. This is the behavior expected in regular linear quantum mechanics. δ(0) = 0 and N U = 2ER .4: Velocity of the atoms predicted by the variational model in the case when interactions are taken into account (dotted line).4. The single particle case is shown in red.4 non linear linear vatom vlatt 0. The diﬀusive regime. characterized by a diverging eﬀective mass. which is assumed big compared to one. The diﬀerent dynamical behaviors include eﬀects such as selftrapping. The parameters used were γ(0) = 10. and when the nonlinearity is present η is no longer constant.2 0.2 Dragging experiment 39 1 0.2 1.
3 Dynamics of a periodthree patternloaded BoseEinstein condensate in an optical lattice In this section we discuss the dynamics of a BoseEinstein condensate initially loaded into every third site of a one dimensional optical lattice.6 vlat vB 0. 4. Outside the strongly . Consistently. the predicted dynamical phase diagram as a function of U N vs vlat is shown in Fig. To appreciate the diﬀerent behaviors exhibited by the system below and above vB /2 we plot in Figs. motivated by the recent experimental realization of this system by the NIST group [33]. It can be seen that for U N = 2ER . 4.4vB the dynamics of the system is similar to the non interacting case. On the other hand. for two diﬀerent initial velocities. 4. the atoms stop moving in the lattice frame.4 0. For vlatt = −0.7. the velocity of the interacting system becomes greater than the noninteracting one.5vB the noninteracting eﬀective mass becomes negative. A condensate loaded in this way is not an eigenstate of the ﬁnal perioda lattice.4 we see that for lattice velocities approximately greater than vlat = 0. This regime is a selftrapping regime and nonlinear eﬀects tend to localize the atoms in the lattice frame. and the condensate will continue to evolve in the ﬁnal system.8 1 Self Focusing Figure 4.15vB the wave packet diﬀuses.2 0.40 Chapter 4 Mean ﬁeld theory Using a potential depth of 3ER and γo = 10. 4.4vB .7vB the wave packet localizes.4vB . Phase Diagram Vo 3 Er 8 6 UN 4 Diffusion 2 0 0 0.5.6 and 4. for vlatt 0. whereas for vlatt = −0.7vB the velocity of the atoms is just the velocity of the lattice so. and approximately at vlat = 0. in Fig.5: Phase diagram U N vs vlatt for Vo = 3Er and γo = 10. At vlat = 0. the evolution of the wave packet. the selffocusing regime is expected for lattice velocities greater than vlat = 0.
4. We show that for large ratios of the interatomic interaction strength to tunneling energy the condensate evolves with selfmaintained population imbalance.5 0 30 0 n t 50 30 zn t 0.5 0 2 2 60 30 0 30 n Figure 4. symmetry arguments can be used to reduce the wavefunction evolution to a two mode problem (analogous to a double well system with an energy oﬀset) for which an analytic solution of the dynamics can be given. and would be a suitable experimental observable.3 Dynamics of a periodthree patternloaded BoseEinstein condensate in an optical lattice t 0 2 2 41 t 10 zn t zn t 0.5 0 30 0 n t 40 30 2 t 20 0. We show that the momentum distribution of an interacting condensate changes in time in a manner which can be related to the spatial tunneling of condensate. 88. .5 60 30 0 30 n 0. In our system we assume the lattice is suﬃciently deep that a tightbinding description is applicable.6: Localization of the wave packet in the selffocusing regime vlatt = −0. For the periodic initial condition of equal wavefunction amplitude at every third site and zero at all others. the GPE equation is expected to give a good description of the condensate dynamics. 90.5 0.15vB .7: Evolution of the wave packet in the diﬀusive regime vlatt = −0.5 2 1 1.5 0 60 30 0 30 n t 40 0.7vB .5 0 60 30 0 30 n zn t zn t 0 zn t 0.5 1 0. A similar phenomenon has been studied in doublewell systems [87. 89. 91].5 0 60 30 0 30 n t 50 zn t 0 zn t 60 30 0 30 n t 30 2 0.5 0 30 0 n 30 Figure 4.5 0 2 1.5 1 0. whereby the condensate population tends to remain localized in the initially occupied lattice sites. Notice the change in the vertical scale in the last row t 0 2 2 t 10 2 t 20 zn t 0. correlated regime. and the DNLSE is valid.5 0 30 0 n t 30 30 2 2 zn t zn t 0 30 0 n 30 30 0 n 30 zn t 0.
and the addition of the second lattice will not shift the locations of the . of loading a condensate into the ground band of lattice with periodicity 3a.1 Experiment Figure 4. A second lattice of periodicity a which is parallel to the ﬁrst lattice.3.42 Chapter 4 Mean ﬁeld theory Under the inﬂuence of an external force a Bloch state will exhibit Bloch oscillations. as shown in Fig. as described in chapter 2. 4.8. To illustrate how a linear potential aﬀects the motion of a pattern loaded condensate we ﬁnd analytic solutions for the noninteracting case with periodic initial conditions and show how the dynamics for this system can be interpreted in terms of the interference of three Bloch states of the lowest band undergoing Bloch in unison. For the ideal case both lattice potentials are inphase. In the experiment. We present numerical results for more general (nonperiodic) initial conditions and consider characteristic properties of the momentum distribution. a combination of two independently controlled lattices was used to load the condensate into every third site of a single lattice. is then ramped up so that the superimposed light potentials form a superlattice of period 3a.8: Patterned loading experiment. Brieﬂy the procedure consists. so that the condensate is well localized in the potential minima of this lattice. 4.
∂τ (4. This initial condition is homogeneous in the sense that each occupied site has the same amplitude and phase along the length of the lattice.2 Case of no external potential The major interest of this section is in the tunneling properties of the condensate in the lattice.3 Dynamics of a periodthree patternloaded BoseEinstein condensate in an optical lattice 43 potential minima from those of the ﬁrst lattice alone. the amplitudes zn (τ ) are given by z3n (0) = √ ρ.24) (4. For an inﬁnite lattice. This allows us to reduce the full set of equations (4. and the amplitudes for the initially unoccupied sites satisfy z3n+1 (τ ) = z3n+2 (τ ) for all τ and all n.6) takes the form i ∂zn = −(zn−1 + zn+1 ) + (En + Λzn 2 )zn .21) to a set of two coupled equations i ∂z0 ∂τ ∂z1 i ∂τ = −2z1 + Λz0 2 z0 .21) • The case of periodic initial conditions: reduction to a two mode system We treat ﬁrst a model case in which no external potential is present. the amplitudes for all initially occupied sites (z3n ) evolve identically in time. (4. N ρ represents the initial number of atoms per occupied site. by removing the ﬁrst period3a lattice on a time scale long compared to band excitations. and dimensionless energy En = Vn /J and coupling constant Λ = N U/J.4. At τ = 0. 4. The normalization condition is z0 2 + 2z1 2 = ρ. or one with periodic boundary conditions. and in which the condensate initially has a uniform phase. = −(z1 + z0 ) + Λz1 2 z1 .23) z3n+1 (0) = z3n+2 (0) = 0. where ρ = 3/M with M the total number of lattice sites. Finally. (4.22) (4. and the condensate will remain localized at these positions. the condensate will be left in every third site of the perioda lattice. (4. (En = 0). For this reason it is convenient to deﬁne a new dimensionless scale of time τ = Jt/ .26) . but short compared to the characteristic time of transport within the lattice. and in which the initial occupancies of each third site are the same.25) where z3n ≡ z0 and z3n+1 = z3n+2 ≡ z1 for all n. and in the noninteracting case the time scale for tunneling is determined by the hopping matrix element J.3. In terms of these new variables the tightbinding evolution equation (4.
(4. . to a system of [m/2] + 1 equations. we can transform Eqs. We note in passing that a similar reduction.25) and (4. With this factor incorporated. found using a procedure similar to that presented by Raghavan et al.24)(4.31) The analytic solutions of Eqs.29)(4.31).27) √ By writing z0 =ψ0 ρ and z1 =ψ1 to the form ∂ i ∂τ ψ0 ψ1 = . where f.28) where γ = Λρ is the ratio of onsite repulsion to tunneling energies.26) is now ψ0 2 + ψ1 2 = 1. The factor of 2 diﬀerence in the deﬁnition of ψ0 and ψ1 arises because ψ1 represents the amplitude of the two initially unoccupied sites. can be expressed in terms of Weierstrassian elliptic functions ℘(τ . The analytic solutions are f (τ ) = g(τ ) = 1− 24 . g = −f sin φ.28) is explicitly Hermitian. We note that this equation of motion is identical to that for a condensate in a double well trap in the two mode approximation [92. (4. 4.25) √ γψ0 2 − 2 √ 2 √ − 2 γ ψ12 − 1 ψ0 ψ1 (4.32) (4. g3 ) [69]. 2 2 .24) √ √ and (4. [87]. g3 ) + (9 + 2γ + γ 2 ) (4.24).33) 1 − f (t)2 . exists for lattice systems that are loaded such that only every mth site is initially occupied. γ γ 4 = −4f g cos φ + f + 2g 4 − 2g 2 . and the √ normalization condition (4. (4. 93]. the matrix appearing in Eq.30) (4. of this system is (from Eq. • Solution of the equations of motion It is convenient to write the lattice site amplitudes appearing in Eqs. g2 .8) H = J Λ ∗ ∗ − 2(z0 z1 + z0 z1 ) − 2z1 2 + z0 4 ρ 2 +Λz1 4 = JΛρ . g2 . H. By introducing the phase diﬀerence φ = θ0 − θ1 . 2 ρ/2. (4. g. (4. 2 .27) can be recast as f = 2g sin φ. and θ are real. (4.25) as z0 =f eiθ0 ρ and z1 = geiθ1 ρ. Eqs.29) (4. (4. 12℘(τ .44 Chapter 4 Mean ﬁeld theory The Hamiltonian function. (4.
(4.28) is constant in time. we ﬁnd that the oscillation period is essentially constant (see Fig. Analysis of Eqs. separated by γ = 2. in which case the matrix of Eq. where the two levels are coupled √ by a Rabi frequency of strength 2. f (τ ) > g(τ ) for all τ . 4. T (γ). Interaction dominated regime . (4. 729 + 243γ − 46γ − 15γ + γ 2 3 4 6 45 (4. are determined by the parameter γ (see Figs. The Rabi model predicts that the cycling frequency between the levels is equal to the diﬀerence between the eigenvalues of the matrix of matrix of Eq. which is detuned from resonance by −1.9 and 4. (4. and the behavior can be approximately understood by taking γ = 0.35) The solutions f (τ ) and g(τ ) are oscillatory functions whose amplitudes and common period. (4.34) /216.9). Because the coupling is detuned from resonance the transfer of populations between wells is incomplete.29) .(4. It is useful to qualitatively divide this behavior into two regimes. which gives the period of oscillation as TRabi = 2π/3 in units of /J (see Fig. T (h/J) γ (J) Figure 4.31) shows that f (τ0 ) = g(τ0 ) for some value of τ0 when γ ≤ 2. 4.10). The equations of motion in this case are equivalent to those of a twostate Rabi problem [94]. with ψ1 2 attaining a maximum value of 8/9. 2.9: Oscillation period (in units of /J) as a function of the interaction strength γ.28). In this case the role of interactions is relatively small. and for γ > 2. Tunneling dominated regime For γ (4.4.9). This system will undergo Rabi oscillations whereby atoms periodically tunnel from the initially occupied site into the two neighboring sites.3 Dynamics of a periodthree patternloaded BoseEinstein condensate in an optical lattice where the parameters g2 and g3 are given by g2 = g3 = 81 − 14γ 2 + 4γ 3 + γ 4 /12.
4. (4.10 where we plot the minimum value of f 2 occurring during the oscillation as a function of γ.36) .32) and (4.4) is expressed as a sum over the lattice sites .33). τ ) (4. ˜ is expressible as the Fourier series ϕ(k.10: Minimum value of f 2 during an oscillation period as a function of γ. and show how this relates to the evolution of the spatial amplitudes given in Eqs. but with reduced amplitude. τ ) = ˜ m zm (τ )e−ikam χ(k). the tunneling between sites occurs at a higher frequency. τ ). The population of the initially occupied sites becomes self trapped by the purely repulsive pair interaction. (4. In contrast to the tunneling dominated regime. • Momentum space dynamics Typically the spacing between individual wells in an optical lattice is too small to resolve the localized density distributions of atoms in neighboring sites using standard imaging techniques.46 Chapter 4 Mean ﬁeld theory The eﬀect of interactions on the meanﬁeld dynamics is to cause the energies of the initially occupied sites to shift relative to those of the unoccupied sites. Because of the periodicity of the system. the condensate tends to be localized on the initially occupied sites in the interactiondominated regime. the momentum space wavefunction. In the tightbinding approximation the condensate order parameter ϕ(x. As γ increases the population imbalance between wells increases (see text). which we denote as ϕ(k. Here we calculate the momentum dynamics of the condensate loaded into every third site of an optical lattice. where tunneling periodically populates all sites equally. This is demonstrated quantitatively in Fig. ( ) γ (J) Figure 4. The momentum distribution is a more convenient observable which approximately corresponds to the expanded spatial distribution of the released condensate. As γ increases and this energy shift increases relative to the strength of coupling between sites. which in the context of a double well system has been called “macroscopic quantum self trapping” [93].
40)).37) e−ikx w0 (x)dx.4.39) (4. q/2]. where the selftrapping eﬀect causes the tunneling between lattice sites to stop (i.e. 3 4 1 γ = 1 − (3f 2 + 1)(f 2 − 1) 3 8 = c3n+2 (τ )2 .40) (4. we invoke the identity M −1 eikna = n=0 M δk.41) In the reduced zone scheme. (4. where the value is maximized by evaluating the cn 2 at the time when f takes its minimum value (we note that f (τ ) is given by Eq. The magnitude of the time variation of the cn 2 is proportional to γ.3 Dynamics of a periodthree patternloaded BoseEinstein condensate in an optical lattice where 1 χ(k) = √ 2π ∞ −∞ 47 (4. f 2 ≈ 1 at all times). 4. arising from the 3lattice site spatial periodicity of the condensate wavefunction. and m an integer.38) χm = χ(qm/3). The momentum distribution of the system has very sharp peaks of relative amplitude cm 2 at momentum k = qm/3.qm/3 .e. even though tunneling occurs between the lattice sites.32)). (4. the momentum wavefunction then consists of three peaks corresponding to Bloch states of quasimomenta 0. In addition. the momentum components are constant in time (see Eqs. Using the analytic solutions for f and g. and Eq. ±q/3.2πm/a . deﬁned as ∆max ≡ (c1 2 − c0 2 )max .39) and (4. Since there are only two independent amplitudes in the set {zm }. the momentum intensities explicitly depend on the occupations of each site and will vary in time when tunneling occurs. 1 cm (τ ) = z0 + z1 e−iqm/3 + z2 e−i2qm/3 . However. (4. We note that the maximum contrast occurs for γ ≈ 3. In Fig.11 we show the maximum contrast between the intensity of the Fourier peaks. in this case. 3ρ where q is the reciprocal lattice vector q = 2π/a. where M is the number of lattice sites. γ = 0). τ ) = ˜ 3/ρcm (τ )χm δk. but will reduce for suﬃciently large values of γ. (4. If interactions between the atoms are ignored (i.31). The identical behavior of c3n+1 2 and c3n+2 2 means that the ±q/3 peaks always have the same intensity. where we only consider momenta in the range k ∈ [−q/2.9. we ﬁnd that ϕ(k. (4. we obtain only two independent Fourier amplitudes c3n (τ )2 = c3n+1 (τ )2 1 γ 1 + (3f 2 + 1)(f 2 − 1) . when interactions are considered. To compute the momentum distribution. the zero quasimomentum component c0 (τ ) . χm describes a slowly varying envelope determined by the localization of the Wannier states at each lattice site.
3 ∆max2 0.42) (4.1 0 0 5 10 15 20 γ (J) Figure 4. applicable to a condensate initially prepared in a harmonic trap.11: Maximum contrast of the Fourier components as a function of γ.2 0. but also lie in a range in which the homogeneous model is expected to give a fair description. these correspond to a condensate of 105 atoms of 87 Rb produced in a magnetic trap with axial and radial frequencies of 9 and 12 Hz respectively. so that the initial state is z3n+1 (0) = z3n+2 (0) = 0. The maximum contrast is deﬁned as ∆max ≡ (c1 2 − c0 2 )max with the maximum value occurring when f 2 is at its minima. we use inhomogeneous to refer to the overall spatial envelope of the period three initial condition. U = 2. • Application to an inhomogeneous condensate Here we wish to consider the dynamics for an inhomogeneous condensate. For values of γ greater than 3.5ER .075 ER .2kHz.induced selftrapping. For the simulations shown here.11 × 10−5 ER and J = 0.9 the contrast between the intensities starts to decrease due to the reduction in tunneling caused by the nonlinearity. z3n (0) = 3 2 πM 2 1/4 (4. The previous homogeneous theory we have presented is expected to accurately describe inhomogeneous cases when the initial pattern of population in every third site extends over many lattice sites i.43) exp(−(9n/M )2 ).e. M = 76.5 0. with ER = 2. and loaded into a lattice with a depth of 4. the parameters used were N = 105 . i=1 2 Taking a particular example we choose a Gaussian envelope to the periodic arrangement of atoms into every third site. The total number of occupied wells and the strength of the on site interatomic interaction were calculated by preserving the value . These parameters are typical of the experimental regime.48 Chapter 4 Mean ﬁeld theory vanishes once during each period of oscillation T (γ). 0. M 1 so that meanﬁeld energy associated with each triplet of sites U (n) = Λ 3 z3n+i 4 varies slowly across the system. For the pattern loaded condensate.4 0.
64. In Fig.13 the numerical Fourier spectrum for the inhomogeneous case evaluated at several diﬀerent times. The variation of the intensities of the peaks. is related to the spatial tunneling between lattice sites in the presence of the mean ﬁeld. Figure 4. can be seen in the plot. The parameters used for the simulation were J = 0. 4. We see that the numerical and analytical results agree well at short times. we show in Fig.075ER and γeﬀ = 2.4. We used γeﬀ as the local mean ﬁeld energy (see text). as shown below. as time progresses the mean ﬁeld variation across the lattice causes the tunneling rates to vary with position and leads to momentum peak broadening. This eﬀect eventually causes the homogeneous model to become an inaccurate description of the inhomogeneous system. Initially all occupied sites are in phase and the three distinctive momentum peaks have a narrow width determined by the intrinsic momentum uncertainty of the condensate envelope.3 Dynamics of a periodthree patternloaded BoseEinstein condensate in an optical lattice 49 of the chemical potential of the system upon reduction to one spatial dimension and by assuming that each of the localized orbitals in the tightbinding description are Gaussian. which. and the solid line represented the initially unpopulated wells. Inhomogeneous condensate: stars for the initially populated well and boxes for the initially empty wells. That is the reason why the homogeneous model ﬁts very well. but diﬀer more as time progresses due to the diﬀerent meanﬁeld seen by diﬀerent wells. 4.12: Comparison between the population evolution of the central three wells for the inhomogeneous condensate and the homogeneous model. Fig.12 shows the results of numerical integration of the equations of motion and the approximate analytical solution given by the quasihomogeneous model described above. We note that momentum space signature for spatially tunneling in the in . 4. To understand the disagreement as time evolves. The hopping rate J was estimated by using Mathieu functions. However. Homogeneous model: dashed line represents the initially populated wells.12 we show the evolution of the population of the central wells (normalized to one) compared with the homogeneous model with γ taken to be the local mean ﬁeld energy γeﬀ = Λ 2 π (9/M ).
4. For longer times. units) 0 q/3 2q/3 momentum 0 q/3 2q/3 momentum Figure 4. the width of the Fourier peaks increases.14. c3n+1 (τ )2 .7ms 0 q/3 2q/3 momentum t = 2. each centered around the respective peak and adding the square of the norm of the Fourier components within each section) vs. and as shown in the plots the model predictions are fair at least for one period of oscillation. c3n+1 (τ )2 . c3n+2 (τ )2 (calculated from the numerical simulation by partitioning the numerical Fourier spectrum in three equal non overlapping sections. we conclude that the Fourier distribution can be used as a signature of the meanﬁeld quantum tunneling inhibition. c3n+2 (τ )2 are not meaningful anymore. the ones calculated with the homogeneous model. 4.8ms 0 q/3 2q/3 momentum t = 1.12. teracting system is still present in the inhomogeneous case. units) t = 0. This is shown in Fig. Because the parameters used for the numerical calculations were chosen to be experimentally achievable. where we plot the evolution of the quantities c3n (τ )2 .6ms density (arb. At this point the quantities c3n (τ )2 . but using an averaged value γave ≡ Λ n zn 4 instead of γ = Λρ. It can be observed that the predictions of the simple model are in very good agreement with the numerical results when the three peaks of the spectrum are well deﬁned.50 Chapter 4 Mean ﬁeld theory t = 0ms density (arb. .13: Momentum distribution of the inhomogeneous condensate evaluated at various times for the same parameters as used in Fig. until a point when they split.
Inhomogeneous condensate (dotted).85. 4. (4.4. we obtain the evolution equations i ˜ ∂ Ψ3n ∂τ ˜ ˜ = −(Ψ3n−1 ei3ξτ + Ψ3n+1 ) ˜ ˜ +ΛΨ3n 2 Ψ3n . corresponding to a uniform force parallel to the lattice.46) (4.44) i ˜ ∂ Ψ3n+1 ∂τ ˜ ∂ Ψ3n+2 i ∂τ Assuming periodic boundary conditions. ˜ ˜ ˜ = −(Ψ3n+1 + Ψ3n+3 e−i3ξτ ) + 2ξ Ψ3n+1 ˜ ˜ +ΛΨ3n+1 2 Ψ3n+1 .3. Taking the initial conditions (4. In what follows we assume that the force is suﬃciently weak that band excitations due to LandauZener tunneling are negligible. In this case the evolution equation diﬀers from what we considered in the previous section by the term En in Eq. ˜ ˜ ˜ = −(Ψ3n+2 + Ψ3n ) + ξ Ψ3n+1 ˜ ˜ +ΛΨ3n+1 2 Ψ3n+1 . Parameters are the same as in Fig. so that a tightbinding picture of the lowest band is suﬃcient to describe the dynamics. 1.23).14: Evolution of momentum peak populations. where ξ is the potential diﬀerence between lattice sites (in units of the hopping matrix element J).3 Dynamics of a periodthree patternloaded BoseEinstein condensate in an optical lattice 51 Figure 4. homogeneous result (solid line).3 Dynamics with a constant external force • Homogeneous three state model In this section we consider the dynamics of a periodically loaded condensate in the presence of a linear external potential. and ˜ transforming the Wannier amplitudes as z3n+j (τ ) = Ψ3n+j (τ )e−i3nξτ (j = 0. 2).21) taking the form En = nξ. the periodicity of the initial conditions and equations of evolution allow considerable simpliﬁcation from the .12. where the comparison is made by replacing γ by an average mean ﬁeld energy γave ≡ Λ n zn 4 = 1. Upper curves: population of the q = ±2π/3 momentum states. (4.45) (4. Lower curves: population of the q = 0 momentum state. 4.22)(4.
e. Ψ2 (τ )). Ψn = Ψn+3 ) and so the evolution of the system can hence be reduced to the three independent equations ˜ ∂ Ψ0 ∂τ ˜ ∂ Ψ1 i ∂τ ˜ ∂ Ψ2 i ∂τ i ˜ ˜ ˜ ˜ = −(Ψ2 ei3ξτ + Ψ1 ) + ΛΨ0 2 Ψ0 .47) (4. ˜ ˜ ˜ ˜ ˜ = −(Ψ2 + Ψ0 ) + ξ Ψ1 + ΛΨ1 2 Ψ1 . which provides valuable insight into the complicated tunneling dynamics the system exhibits in the absence of nonlinearity.49) are more diﬃcult to treat analytically than the case considered in the last section due to the presence of a linear potential. • Analytic solution for linear dynamics ˜ ˜ ˜ Deﬁning the vector x(t) = (Ψ0 (τ ). (4. In this section we derive an analytic solution for a noninteracting condensate (i.47)(4. and Ψ2 ↔ {Ψ3n+2 }.e. In particular.49) can be decoupled. these assumptions mean that ˜ ˜ every third Wannier amplitude evolves identically (i. (4.51) the linear version of Eqs. and using the transformation y(τ ) = P (τ )x(τ ). Ψ1 (τ ). yet should furnish a good description for dilute condensates satisfying γ 1. Ψ1 ↔ {Ψ3n+1 }. directly yielding the solutions ˜ Ψn (τ ) = 2nπ e−inξτ −i∆0 (τ ) √ e + ei( 3 −∆1 (τ )) 3M 2nπ +e−i( 3 +∆2 (τ )) .52) .48) ˜ where the new amplitudes map onto the original set according to Ψ0 ↔ ˜ 3n }.50) M j=0 The equations of motion (4. Λ=0). For the nonlinear regime we present numerical results to illustrate the typical behavior. (4. (4. ˜ ˜ ˜ = −(Ψ1 + Ψ0 e−i3ξτ ) + 2ξ Ψ2 ˜ ˜ +ΛΨ2 2 Ψ2 . where P (τ ) is the unitary matrix P (τ ) = e−iξτ e−i2ξτ 1 2π 2π e−i(ξτ − 3 ) e−i(ξτ + 3 ) .47)(4. 2π 2π e−i(2ξτ + 3 ) e−i(2ξτ − 3 ) 1 1 (4.49) (4. and obey the normalization ˜ ˜ ˜ ˜ {Ψ condition 2 3 ˜ Ψj 2 = .52 Chapter 4 Mean ﬁeld theory full set of M coupled equations.
4. a 3 (4. During this evolution each state accumulates phase at a rate determined by the instantaneous Bloch energy. the system consists of three Bloch states whose quasimomenta will translate in unison under the action of the external force. ξ 3 3 53 (4. (4. i. (4. τ ) = α0 φ− ξ t (x)e−i∆0 (τ ) + α+ φ− ξ t+ q (x)e−i∆1 (τ ) a a 3 +α− φ− ξ t− q (x)e−i∆2 (τ ) .3 Dynamics of a periodthree patternloaded BoseEinstein condensate in an optical lattice where we have deﬁned the phase terms ∆n (τ ) = − for n = 0. the wavefunction can be expressed as a superposition of three Bloch waves (of the lowest band) which are symmetrically spaced in quasimomentum.56) where kn (τ ) = −ξτ /a+kn (0) is the quasimomentum of Bloch state n at time t.53) ˜ These solutions for the spatial amplitudes Ψi (τ ) can be most easily understood by considering a Bloch state decomposition of the condensate wavefunction. at this time the wavefunction must be of the form ϕ(x. 1. with α+  = α− . 2. Assuming the condensate initially has a total crystal momentum of zero. 0) = α0 φ0 (x) + α+ φq/3 (x) + α− φ−q/3 (x). 2nπ 2nπ 2 sin(ξτ − ) + sin( ) . The nature of our system allows us to construct an analytic form for the initial wavefunction. and the α are complex constants determined by the lattice depth.58) . and hence of the group velocity of the Bloch wave. The action of an external force on a Bloch state causes it to linearly change its quasimomentum in time according to ξ k(τ ) = − τ + k(0). τ ∆n (τ ) = 0 E(kn (s))ds.57) for which ∆n (τ ) can be evaluated. a (4. In the tightbinding approximation the dispersion relation for the Bloch states has the analytic form E(k) = −2 cos(ka).53)). The wavefunction evolution in the Bloch basis is ϕ(x.54) where φk (x) is a Bloch state with quasimomentum k. Because the system has a three lattice site period and is assumed to be in the lowest band. and yields the results given in Eqs.e. For the case we are considering here. (4. gives rise to the wellknown phenomenon of Bloch oscillations (see chapter 2).55) The periodicity of the Bloch dispersion relation in k. (4.
Eqs. 2 (4. For ˜ ˜ ˜ ξ > 0. . Note: we take all α = ρ as determined by the 3 initial conditions. 4.(4. making the system closer to resonance.we refer to these as resonances. For this case. we ﬁrst start by considering the case when ξ is small. in the sense of Eq. It is interesting to note that the system exhibits ”nonclassical” dynamics whereby the atoms start to tunnel in the direction opposite the direction of the force (this statement applies even when the external ﬁeld is not weak). (4. where the resonance condition results in the initially populated wells becoming empty at some later time.60) (4.15 we show the temporal evolution of the spatial amplitudes Ψi for a range of values of ξ.28).52). where ˜ ˜ Ψ1 and Ψ2 behave identically.54 Chapter 4 Mean ﬁeld theory From this solution we can obtain solutions for the evolution of the spatial amplitudes Eqs.61) τ3 (6τ + 3τ cos(3τ ) − 4 sin(3τ )) .22) and (4.23). φk (x) = n eikna w0 (x − na) (see chapter 2) and make use of Eq. Small ξ solution .Non classical transport To understand the dynamics. This is the normal Bloch oscillation period. the population in the wells is given by ˜ Ψ0 (τ )2 = ˜ Ψ1 (τ )2 = ˜ Ψ2 (τ )2 = where h(τ ) = 1 5 + 4 cos(3τ ) + ξ 2 h(τ ) 3M 1 (2 + 3ξ)[1 − cos(3τ ) − 3M 1 (2 − 3ξ)[1 − cos(3τ ) − 3M (4. and in particular the population in each well is then determined by the interference of the Bloch phases ∆n . To do this we expand the Bloch states in terms of Wannier functions according to.59) ξ2 h(τ )] 2 ξ2 h(τ )] 2 (4. Bloch Oscillations The evolution of the spatial amplitudes. For certain choices of ξ the initially occupied ˜ Ψ0 amplitude periodically disappears . These functions are all periodic in time with period τB = 2π/ξ (in units of τ = tJ/ ).5). (4. and gives the time scale over which the quasimomenta of the Bloch states increases by exactly one reciprocal lattice vector. atoms in Ψ0 start to tunnel to Ψ1 more rapidly than to Ψ2 . the eﬀect of a linear potential is to enhance the tunneling from the initial populated 3n wells to their 3n + 1 neighbors ones. in this weak limit. (4. Thus. the degeneracy ˜ ˜ in the population of the wells represented by Ψ1 and Ψ2 is lifted. Resonances ˜ In Fig.62) The above solution shows that when the force is applied. This should be compared with the results in the absence of the force.
dashdot line: Ψ3n+2 2 . The ”nonclassical” motion can be seen where the 3n+1well populations initially increase more rapidly that the populations of the 3n + 2wells.52) we obtain the following conditions on the phases for these resonances π + 2πs .15: Evolution of the normalized population for diﬀerent values of ξ. (4. ∆2 (τ ) − ∆1 (τ ) = ± 3 ∆1 (τ ) − ∆0 (τ ) = ± (4. 3 π + 2πm . ˜ By requiring Ψ0 = 0 in Eq. When these conditions are satisﬁed the population is not shared between the adjacent sites. (4. the +sign choice in Eqs.63) and (4. One Bloch period is shown in the plots except ξ = 0 where the period is inﬁnite. It can also be seen in the plots that res res ξ = ξmax /2 and ξ = ξmax are resonant values.3 Dynamics of a periodthree patternloaded BoseEinstein condensate in an optical lattice ξ =0 ξ= 1 ξ 2 55 0 0 τ (h/J) ξ= 2 ξ 7 τ (h/J) ξ= ξ τ (h/J) ξ= 5 ξ 14 τ (h/J) ξ =3ξ 0 τ (h/J) τ (h/J) Figure 4.63) (4.63) and . Similarly the −sign case gives the condition for complete ˜ tunneling into Ψ2 . dotted line: Ψ3n+1 2 . (4. Solid ˜ ˜ ˜ line: Ψ3n 2 .4. Solving for the values of ξ for which Eqs. In particular. but preferentially tunnels to one of the neighbors.64) where s and m are integers and the same sign choice on the right hand side must be made for both equations.64) gives the phase condition for complete tunnel˜ ing into Ψ1 .
2 0. Large Force Limit res For ξ > ξmax the system exhibits a population imbalance as the Bloch oscillation suppresses the ability of the system to tunnel between sites. one obtains the spectrum √ 6 3 (1 + 3s) res ξn=(s.e.16.5 Chapter 4 Mean ﬁeld theory 0.66) (4. ξτ 2 .1 0 20 40 60 80 100 n Figure 4.67) (4.68) .56 0. For large s or m the resonant values of ξ are close together. and become spaced further apart as the magnitude of s decreases. the values of ξ for which Ψ0 periodically disappears.65) which is shown in Fig. (4.64) hold. since for values of ξ greater than this the amplitudes of the phase oscillations ∆i are so small that they cannot satisfy the condition for population resonance. . In the regime res ˜ ξ > ξmax .e. In the absence of an applied force (i. ordered in decreasing mag˜ nitude.16: The spectrum of values of external force. 2 + 3(s + 2m + 1)2 π (1 + 3s) (4. ˜ Ψ0 (τ )2 −→ ˜ Ψ1 (τ )2 −→ ˜ Ψ2 (τ )2 −→ 3 8 1− 2 M ξ 12 1 sin2 M ξ2 12 1 sin2 M ξ2 sin2 ξτ 2 ξτ 2 . The largest value of ξ for which a res√ res onance can be found is ξmax ≡ 6 3/2π.4 0. When the external force is applied the phases oscillate at a rate that increases with ξ and an amplitude that decreases with ξ.m) = ± . ∆1 = −1 and ∆2 = −1 so that the resonance conditions can never be achieved. for which a population resonance occurs i. Ψ0  exhibits only one (nonzero) minima per Bloch period and the dynamics of the system is dominated by the Bloch oscillations. ξ = 0) the phase terms are timeindependent with ∆0 = −2. 4. (4.3 0. small amplitude oscillations around its initial value. res In the limit ξ ξmax the population in the wells is described by rapid.
• Numerical results for the nonlinear case (h/J) (h/J) Figure 4. the introduction of interactions brings the . Furthermore.17: Eﬀects of interactions on generalized Bloch oscillations for the pattern ˜ loaded system.17. the resonance picture still gives us useful guidance concerning the expected behavior. In general. Although the dynamics in this situation is considerably more complicated than in the linear case. res 4.3 Dynamics of a periodthree patternloaded BoseEinstein condensate in an optical lattice 57 res The case of ξ = 3ξmax depicted in Fig.4. Lower plot: ξ = 2ξ res /7.15 shows an approach to this behavior. we see that for ξ = 2ξmax . ξ = 2ξmax max Because of the diﬃculty of solving analytically the equations of motion the nonlinear equations were solved numerically. For example. there exist critical values of ξ and Λ above which no resonances occur.4. nonlinearity tends to destroy the periodicity of the Bloch oscillations that is present in the noninteracting case. Evolution of Ψn 2 for various interaction strengths. Upper plot: res . in Fig.
analogous to what is seen in Fig. For longer times.14) (which extends over M = 76 sites). if. and for the more inhomogeneous initial state used in Figs. In Figs. interatomic interactions induce time variation of the momentum intensities. then eventually decreases towards zero when macroscopic imbalance is achieved.8ER . as in the untilted case. the model predictions start to disagree with numerical simulation due to the spread of the three Fourier peaks.12 4. we use an eﬀective mean ﬁeld energy. the inhomogeneous result much more rapidly departs from the homogeneous prediction than the example presented here. (4.075 ER and ξ = 2. similarly to the untilted case. the introduction of interactions eventually draws the system out of resonance. Because the external ﬁeld together with the nonlinearity breaks down the periodicity of time evolution. we again solved numerically the discrete nonlinear Schr¨dinger equation for a condensate loaded into o every third lattice site but. Concerning the momentum distribution. We observe that the dynamics are much more sensitive to inhomogeneous eﬀects in the presence of an external force. M = 290. initially with a Gaussian proﬁle. To compare the predictions of the homogeneous model in the presence of a linear external potential to a more realistic case. on the other hand. J = 0. the dynamics of the momentum distribution is quite complex.18 and 4. nonresonant res behavior is restored. 4.19 we present a comparison between the evolution of the normalized population at the central wells found numerically and the prediction of the model for the parameters: N U = 4. instead of being homogeneous. As Λ increases.11 for the untilted case. For ξ = 2ξmax /7. Very good agreement between the model and the numerical results was found for short times and modest meanﬁeld energies. . 4. the contrast increases to a maximum value. The contrast between momentum components vanishes at Λ = 0.58 Chapter 4 Mean ﬁeld theory system into resonance. but that for large interaction strength.
59 Figure 4.19: Evolution of the momentum peak populations. stars).11 for the homogeneous model. diamonds). Homogeneous case: Ψ3n 2 (dashdot line).18: Comparison between the evolution of a inhomogeneous condensate ˜ ˜ with the homogeneous result. and Ψ ˜ (solid line). ξ = 2 and γeﬀ = 1. Inhomogeneous condensate (dotted curves). homogeneous model (lines) using the same parameters as those in Fig. q = 0 (dashed line. . q = 2π/3 (solid line. The parameters used were J = 0.075ER . but replacing γ by an average value γave = 1. 4. q = −2π/3 (dashdot line.3 Dynamics of a periodthree patternloaded BoseEinstein condensate in an optical lattice 59 Figure 4. Ψ3n+1 2 ˜ ˜ (stars). and Ψ3n+2 2 (dotted line). squares). Inhomogeneous condensate: Ψ0 2 (boxes).4. Ψ1 2 ˜ 2 2 (diamonds).18. where we have taken γ as the local mean ﬁeld energy.
60 Chapter 4 Mean ﬁeld theory .
is to treat the ﬁeld operator as a cnumber plus a ﬂuctuating term. the HFBPopov approximation and the improved Popov approximation where the bare potential is upgraded to the manybody scattering matrix. They will fail to give a good description as interactions become important and the system approaches the Mott insulator transition. 95]. The nonquadratic terms should be of higher order and can be treated perturbatively. with a chemical potential smaller than the energy gap to the ﬁrst vibrationally excited state. The cnumber describes the condensate or the coherent part of the matter ﬁeld. which was ﬁrst proposed by Bogoliubov in 1947 [11]. we apply the Bogoliubov approximation to the Bose. [77. In this chapter. when a condensate is present. and therefore the dominant terms in the Hamiltonian are quadratic in the ﬂuctuating ﬁeld. the analysis in the present chapter is going to be restricted to the zero temperature .Chapter 5 Quadratic approximations As discussed in chapter 3. We also deal with the higher order terms by considering diﬀerent approximations. quantum ﬂuctuations are small.Hubbard Hamiltonian and derive the correspondent Bogoliubovde Genes(BdG) equations as developed in Refs. All of these approximate approaches are valid only in the weakly interacting regime when quantum ﬂuctuations are small. The aim of this chapter is to rewrite it so that a systematic approach can be developed in the weakly interacting regime. Among them we discuss the HartreeFockBogoliubov (HFB) approximation. Because we are interested in the quantum eﬀects caused by interactions that drive the quantum phase transition instead of thermal quantum ﬂuctuations. when the loading process produces atoms in the lowest vibrational state of each well. We show how the superﬂuid fraction can be thought of as a natural order parameter to describe the superﬂuid to Mott insulator transition. In the weakly interacting regime. the BoseHubbard Hamiltonian is an appropriate model. To test the validity of the approximations we compared them with numerical solutions obtained by diagonalizing the BoseHubbard Hamiltonian. The quartic form of the Hamiltonian makes it very diﬃcult to deal with it in all the diﬀerent regimes. In this chapter we also derive an explicit expression for the superﬂuid density based on the rigidity of the system under phase variations. The basic idea.
the creation and annihilation operators on each site can be replaced by a cnumber(see chapter 4). 2 (Vn − µ)zn 2 + (5.m> ϕn zm ˆ ∗ + n ∗ Vn − µ + U zn 2 zn ϕn + h.3) (5.1 The characteristic Hamiltonian In the very weakly interacting regime.2) U zn 4 . with H0 = −J <n. Chapter 5 Quadratic approximations 5.7) U 2 ˆ H3 = U ϕ2† zn + ϕ2 zn + (ϕ† ϕn + ϕn ϕ† )zn 2 . ˆn ˆn ˆ n ˆ H4 = U 2 ϕ† ϕ† ϕn ϕn .6).62 case. The ﬁrst step in the diagonalization is the minimization of the energy functional H0 .m> ϕ† ϕm + ˆn ˆ n (Vn − µ)ϕ† ϕn + ˆn ˆ (5.. The terms of the Hamiltonian have been grouped in equations according to the number of non condensate operators which they contain.1) Replacing this expression for an in the BoseHubbard Hamiltonian Eq.n> zm + (Vn + U zn 2 )zn (5.c.5) (5. For the time independent situation the DNLSE can be written as µzn = −J <m.8) . to a good approximation.4) ˆ leads to : ˆ ˆ ˆ ˆ ˆ H = H0 + H1 + H2 + H3 + H4 . (3. To include the small quantum ﬂuctuations in the description of the system we assume that the ﬁeld operator can be written in terms of a cnumber plus a ﬂuctuation operator: an = zn + ϕn .m> ∗ zn zm + n (5.4) ˆ H1 = −J <n. 5. ˆn ˆn ˆ ˆ n Where < n.6) (5. stands for the hermitian conjugate. m > restricts the sum over m to nearest neighbors and h.c. ˆ ˆ H2 = −J <n.2 Bogoliubovde Gennes (BdG) equations In this section we neglect higher order terms and focus only on the quadratic Hamiltonian. which in turn means that the linear Hamiltonian H1 vanishes. ˆn 2 ˆn 2∗ ˆn ˆ ˆ ˆn n ϕ† ϕ† ϕn zn + h.. This requires the condensate amplitudes zn to be a stationary ˆ solution of the DNLSE (4.c. ˆ ˆ (5.
The zero mode is essentially non perturbative and it introduces an artiﬁcial infrared divergence in low dimensional models. The transformation ˆn ˆ ˆ† ˆ is known as Bogoliubov transformation and is given by ϕn = ˆ s=0 ∗s † us αs − vn αs .5. A diﬀerent way to deal with the zero mode has been proposed by Castin and Dum [98]. This mode is the Goldstone boson associated with the breaking of global phase invariance by the condensate. b) Pair excitations The quadratic Hamiltonian can be diagonalized by ﬁnding a basis of socalled quasiparticle states which do not interact which each other. The quadratic Hamiltonian H2 allows the leading order eﬀects of the non condensate to be taken into account. Schematically they are shown in ﬁg.2 Bogoliubovde Gennes (BdG) equations 63 The Hamiltonian H0 gives a description which includes only the contribution from the condensate.5. By including second order terms in the Hamiltonian two classes of interactions are included besides the condensatecondensate ones: a) Interactions between one excited atom and one condensate atom leading to transitions of the form 0j → 0i (direct and exchange excitations)and b) interactions between two condensate atoms which cause the atoms to be excited to non condensate states 00 → ij (pair excitations). Figure 5. For this reason quadratic approximations are actually improved if the contribution from this mode is neglected all together [96]. ˆ nˆ (5.1: Scattering processes included in the quadratic hamiltonian: a) Direct and exchange excitations.1. αs . Gardiner [97] and Morgan [16].9) In general the spectrum of ﬂuctuations includes a zero mode. Mathematically this involves a linear canonical transformation of the singleparticle creation and annihilation operators a† and an into quasiparticle operators αs . Here the theory is written in terms of operators which exchange particles between zero and .
10) (5.. i. = 0. To obtain the desired orthogonal modes one n zn vn way to proceed is to solve the BdG equations setting the cs to zero and then to remove from the solutions their projection onto the condensate.64 Chapter 5 Quadratic approximations nonzero modes.13) (5. and one further operator which changes total particle number.) and z = (z1 .).12) s s with us = (us . These parameters are given by 2 ∗ s n zn  (zn un s − zn vn ) cs = U 2 n zn  (5. n zn un and n (zn un − zn vn ) = 0. (5... z2 . vs = (v1 . vn } are constrained by the conditions: n ∗s s u∗s us − vn vn n n n s ∗s us vn − vn u∗s n n n = δss . The matrices L and 1 2 M are given by Lnm = −J Mnm = δnl δlm + δnm (2U zn 2 + Vn − µ) (5. Each of the terms.) .9) is required to be canonical.(5. nevertheless the solutions are orthogonal to the condensate only in ∗ s s ∗ s the general sense.16) . As shown by [16] both the latter and the former approaches lead to the same physical predictions. which means that it preserves the commutation relations and leads to bosonic quasipars ticles. so for simplicity we are going to ignore zero mode ﬂuctuations and restrict the ﬂuctuation operators to act only on the excited states. the quadratic Hamiltonian takes the form ˆ H2 = s=0 B † ωs αs αs + ˆ ˆ 1 B ω − Lss 2 s ... If the BdG equations are satisﬁed.12).14) <n. with δnl the Kronecker delta which is one if n = l and zero otherwise. . (5.e.l> 2 −U zi δnm .11) The necessary and suﬃcient conditions that the quasiparticle amplitudes have to fulﬁll to diagonalize the Hamiltonian are provided by the so called Bogoliubovde Gennes (BdG) equations L M M∗ L us vs B = ωs us −vs + cs z −z∗ .. (5. To satisfy this. . v2 . The Bogoliubov transformation Eq. . the amplitudes {us . us . s are not necessarily zero.15) The BdG equations without the cs parameters have the same quasiparticle energies as Eq. (5. conserving the total particle number. The paramB eters cs ensure that the solutions with ωs = 0 are orthogonal to the condensate [16].
The solutions ωs come in pairs with positive and negative energies. ˆn ˆ ˆn ˆ ˆn ˆ an am − an am = ϕn ϕm . (5. if B is a solution for the amplitudes (us . z). ρnm and the anomalous average matrix. There is always a solution with ωs = 0 and in this case the 0 . s∗ us vm .2. u0 ) ∝ (z. the depletion and condensate atoms are related by the equation: N = n a† an = ˆn ˆ n zn 2 + ρnn . the quantity ρnn represents the non condensate population (or depletion) at position n or depletion at position n. the average is over the ground quasiparticle state given by αs 0 = 0 and the ˆ ρnm and mnm can be written in terms of quasiparticle amplitudes as: ρnm = s=0 s∗ s vn vm . mnm as: ρnm = mnm = a† am − a† am = ϕ† ϕm . Taking into account higher order terms we are mainly introducing manybody eﬀects on the scattering. 5. At zero temperature.17) (5. we deﬁne the one body density ﬂuctuation matrix.21) The interpretation to the physical meaning of the anomalous term is postponed to section 5.3 Higher order terms In the preceding section we described the diagonalization of the Hamiltonian containing only terms up to quadratic order in the ﬁeld operators.18) The averages denotes an ordinary quantum expectation value.3 Higher order terms 65 ˆ If the Hamiltonian H2 . the eigenvalues of the system B are real.7.20) mnm = − Physically.5. ˆ ˆ ˆ ˆ ˆ ˆ (5. We explicitly amplitudes must be proportional to the condensate (v exclude the zero mode solution to guarantee that the excitations are orthogonal to the condensate. If the average number of atoms is N . vs ) then −ω B is also a solution for the ωs s B amplitudes (vs∗ . We now want to go beyond this approximation and consider the corrections introduced by the cubic and quartic terms.19) (5. n s=0 (5. us∗ ). is positive deﬁnite. . To have a complete description of the quadratic Hamiltonian. which is the case when the condensate amplitudes correspond to a stable state of the system.
. (5.23) introduces also a energy shift in H0 : ˆ ∆H0 = ˆ H3 U (2ρ2 + mnn 2 ) nn 2 (2ρnn ϕn + mnn ϕ† )zn + h.(5. ˆn ˆ ˆ ˆ n The corrections from higher order terms yield a modiﬁed quadratic Hamiltonian which can be also diagonalized. 5.7)) one can reduce them to quadratic forms in terms of the functions ρnm and mnm . (5. n n ϕ† ϕ† ϕn ϕn n n → 4ρnn ϕ† ϕn n + mnn ϕ† ϕ† n n + m∗ ϕn ϕn nn − (2ρ2 nn 2 (5.27) LHF B MHF B M∗HF B L∗HF B LHF B = −J nm <n.c.31) The parameters cHF B are there again to enforce the orthogonality of the quasis particle modes with the condensate [16].c. s ∗ U n zn 2 (zn us − zn vn ) n = U n zn s ∗ n (mnn zn un 2 − ∗ s mnn zn vn ) . which gives ϕ† ϕ† ϕn ϕn = 2ρ2 + n n nn 2 .(5. The diagonalization leads to the following equations µHF B zn = −J <m. ˆ ˆn = U n (5.29) (5.26) ˆ H4 = U 2 (2ρnn ϕ† ϕn + mnn ϕn ϕn ) + h. which relies upon the factorization approximation: products of many operators are approximated by paring the operators in all possible ways and replacing these pairs by their expectation value [15].25) (5.6) and (5.28) δnl δlm + δnm (2U zn 2 + Vn − µHF B + 2U ρnn ). (5.l> us vs = HF ωs B us −vs + cHF B z s + −cHF B z∗ s − .23) ( Eq. using Eq..22) in H3 and H4 ( Eqs. Using the factorization approximation.30) MHF B nm cHF B s 2 = −(U zi + U mnn )δnm ..(5.1 HartreeFockBogoliubov equations Third and quartic terms in the Hamiltonian can be included by treating them in a selfconsistent mean ﬁeld approximation. mnn  ˆ ˆ Therefore. (5.(5.66 Chapter 5 Quadratic approximations 5. while Eq.(5.24) (5. one can reduce third and quartic products of ﬂuctuation ﬁeld operators to: ϕ† ϕn ϕn → 2ρnn ϕn + 2mnn ϕ† . The cnumber term in Eq.23) is justiﬁed by Wick’s theorem [51].(5.23) and Eq.22) + mnn  ).3.n> zm + (Vn + U zn 2 + 2U ρnn + U mnn )zn .22) is justiﬁed by analogy [16].
This procedure is known as HFBPopov approximation. is associated with a velocity ﬁeld for the condensate by v0 (x) = m Θ (x) .32) LP M∗P MP L∗P P = ωs + cs . 2 n zn  5. in the e1 –direction and a ground–state energy. ∗ s zn 2 (zn us − zn vn ) n cs = U n . The HFBPopov equations have the form: µP zn = −J <m.30) are known as HartreeFockBogoliubov (HFB) equations. vs (x) ≡ v0 (x) ([100].e. and they are considered to be a better approximation than the HFB equations because they yield a gapless spectrum. x ) has to have exactly one macroscopic eigenvalue which deﬁnes the number of particles in the condensate.36) 2 MP nm = −(U zi )δnm .l> δnl δlm + δnm (2U zn 2 + Vn − µP + 2U ρnn ). Θ (x) = θx1 /L with a total twist angle θ over .27) to (5.34) (5. the onebody density matrix ρ(1) (x. Formally.4 The BoseHubbard model and superﬂuidity The concept of superﬂuidity is closely related to the existence of a condensate in the interacting manybody system. i. Even though they take into account higher order corrections. us vs us −vs z −z∗ (5. calculated with periodic boundary conditions.4 The BoseHubbard model and superﬂuidity 67 The above set of equations (5. Θ (x). 5.33) LP nm = −J <n.2 HFBPopov approximation One way to solve the gap problem is to set the anomalous term mnn to zero in HFB equations. (5. the corresponding eigenvector describes the condensate wave function φ0 (x) = eiΘ(x) φ0 (x). (5. A spatially varying condensate phase. Now we impose a linear phase variation.n> zm + (Vn + U zn 2 + 2U ρnn )zn .[101]) and enables us to derive an expression for the superﬂuid fraction. they have the problem that violate the Hugenholtz and Pines theorem [99] which states that the energy spectrum of a Bose gas is gapless. Consider a system with a ﬁnite linear dimension.35) (5. E0 .37) This irrotational velocity ﬁeld is identiﬁed with the velocity of the superﬂuid ﬂow. L. (5.5. The HFBPopov equations were ﬁrst introduced by Popov [52]. fs . there exists an excitation with an energy which tends to zero as the momentum tends to zero.3.
37) leads to a fundamental relation for the superﬂuid fraction. more convenient to map the phase variation by means of a unitary transformation onto the Hamiltonian [103].68 Chapter 5 Quadratic approximations the length of the system in the e1 –direction. The current operator J (Note that 1 the physical current is given by this expression multiplied by ) and the hopping . 105]. Technically the phase variation can be imposed through socalled twisted boundary conditions [102]. In this case.39) where the second equality applies to a one dimensional lattice system on which a linear phase variation has been imposed. Ts . and the number of sites is M . the phase variation over this distance is ∆θ. 1 2 Eθ − E0 = Ts = mN fs vs . (5.41) Using this expansion the twisted Hamiltonian (5. 2 (5. Replacing the superﬂuid velocity. These phase factors show that the twist is equivalent to the imposition of an acceleration on the lattice for a ﬁnite time. vs with the phase gradient according to Eq. 2 (5.42) ˆ where we retain terms up to second order in ∆θ. θ π. Here the distance between sites is a. Eθ − E0 . of the superﬂow generated by the phase gradient. We calculate the change in energy Eθ −E0 under the assumption that the phase change ∆θ is small so that we can write: e−i∆θ 1 1 − i∆θ − (∆θ)2 . Eθ will depend on the phase twist. fs = 2m L2 Eθ − E0 1 Eθ − E0 = . For very small twist angles. (5. 2 (5. It is interesting to note that the present experiments enable us to make a speciﬁc connection between the formal and operational aspects of the system. however. 2 N θ2 N J (∆θ)2 (5. In the context of the discrete BoseHubbard model it is.40) takes the following form: ˆ Hθ ˆ ˆ ˆ ˆ ˆ 1 H0 + ∆θJ − (∆θ)2 T = H0 + Hpert . the energy diﬀerence. The resulting “twisted” Hamiltonian in this case M M ˆ Hθ = n=1 nn Vn − J ˆ (e−i∆θ a† an + ei∆θ a† an+1 ) + ˆn+1 ˆ ˆn ˆ n=1 U 2 M nn (ˆ n − 1) . The resulting ground–state energy. For simplicity we are going to focus in a one dimensional lattice. J ≡ 2 /(2ma2 ). Thus. can be attributed to the kinetic energy.38) where m is the mass of a single particle and N is the total number of particles so that mN fs is the total mass of the superﬂuid component.40) ˆ n n=1 exhibits additional phase factors e±i∆θ — the socalled Peierls phase factors — in the hopping term [104.
4 The BoseHubbard model and superﬂuidity ˆ operator T are given by: M 69 ˆ J = iJ ˆ T = −J (ˆ† an − a† an+1 ) an+1 ˆ ˆn ˆ (ˆ† an + a† an+1 ) . The superﬂuid fraction is then given by the contribution of both the ﬁrst and second order term: fs = fs(1) − fs(2) .47) Thus we obtain for the energy change up to second order in ∆θ Eθ − E0 = (∆θ)2 − D ≡ 1 M − 1 ˆ Ψ0 T Ψ0 − 2 ˆ  Ψν JΨ0 2 Eν − E0 = M (∆θ)2 D. an+1 ˆ ˆn ˆ (5. Eν − E0 (5. 1 ˆ Ψ0 T Ψ0 . NJ Eν − E0 ν=0 (5. 2.27).5. (5. deﬁned above.44) n=1 M n=1 The change in the energy Eθ − E0 due to the imposed phase twist can now be evaluated in second order perturbation theory Eθ − E0 = ∆E (1) + ∆E (2) . The second order term is related to the matrix elements of the current operator involving the excited states Ψν (ν = 1. In general both the ﬁrst and the second order term contribute. Eν − E0 The quantity D. fs(1) ≡ − 2N J ˆ 1  Ψν JΨ0 2 fs(2) ≡ .) of the original Hamiltonian ∆E (2) = − ν=0 ˆ  Ψν Hpert Ψ0 2 = −(∆θ)2 Eν − E0 ν=0 ˆ  Ψν JΨ0 2 . For a translationally invariant lattice the second .45) The ﬁrst order contribution to the energy change is proportional to the expectation value of the hopping operator 1 ˆ ˆ ∆E (1) = Ψ0 Hpert Ψ0 = − (∆θ)2 Ψ0 T Ψ0 .46) Here Ψ0 is the ground state of the original BoseHubbard Hamiltonian (4. is formally equivalent to the Drude weight used to specify the DC conductivity of charged fermionic systems [106].43) (5.. (5. .49) Here N is the number of atoms in the lattice. 2 (5.48) ν=0 1 ˆ Ψ0 T Ψ0 − 2 ν=0 ˆ  Ψν JΨ0 2 ..
i.51) n=1 We use ﬁrst order perturbation theory on the wave function to obtain the following expression: ˆ Ψ(∆θ)Jθ Ψ(∆θ) = 2∆θ − 1 ˆ Ψ0 T Ψ0 − 2 ˆ  Ψν JΨ0 2 Eν − E0 (5. However.5 Expectation values Using the quadratic approximations we can evaluate expectation values of meaningful physical quantities. Useful quantities as the system approach the Mott insulator state are the number ﬂuctuations and the quasimomentum distribution.: M ˆ Jθ ˆ J + J∆θ ˆ ˆ (ˆ† an + a† an+1 ) = J − T ∆θ .50) n=1 We expand this to ﬁnd the lowest order contributions. the physical current.53) multiplied by 1 .e.54) m∗ a2 This is the total ﬂux. Thus. js = 1 ˆ Ψ(∆θ) Jθ Ψ(∆θ) M ∆θ m∗ a = vs ns .53) ν=0 = 2N Jfs ∆θ.55) Js = 1 ˆ Ψ(∆θ)Jθ Ψ(∆θ) = N fs ∆θ So we see that the Drude formulation of the superﬂuid fraction (5.e. i. an+1 ˆ ˆm ˆ (5. Eq. We can further understand this approach to the superﬂuid density by calculating the ﬂow that is produced by the application of the phase twist.56) . (5. in exact calculations the second order term plays a role. js . can be expressed as: . (5. The number ﬂuctuations at the site xi are given in the Bogoliubov approximation by ∆nn = a† an a† an − a† an ˆn ˆ ˆn ˆ ˆn ˆ s∗ us − vn 2 .52) (5. 5. n s 2 = zn 2 (5. Js . ˆn+1 ˆ ˆn ˆ (5.70 Chapter 5 Quadratic approximations term vanishes (as is going to be shown latter) if one uses a quadratic approximation.49) gives an intuitively satisfying expression for the amount of ﬂowing superﬂuid. and we need to divide by M to get the ﬂux density. = N fs aM (5. To do this we work out the expectation value of the current operator expressed in terms of the twisted variables: M ˆ Jθ = iJ (e−i∆θ a† an − ei∆θ a† an+1 ) .
when it is released from the lattice the gas shows a nice interference pattern. zn = no . and instead of interference peaks a incoherent background is observed. M (1) fs = n=1 (2) fs = (1) fsn = (5. the quasiparticle modes have a plane wave character and therefore can be related to quasimomentum modes : 1 1 q uq = √ eiq·xn uq . The quasimomentum distribution function nq is deﬁned as [103] nq = 1 M eiq·(xn −xm ) a† am n = 1 M n. M is the total number of wells. In the superﬂuid regime because of the phase coherence of the system.57) where the quasimomentum q can assume discrete values which are integer multiples 2π of ai Mi .60) − 2ωs 2 s us vn+1 ) n . Due to the translational symmetry of the system the condensate amplitudes √ are constant over the lattice. Vn = 0. Also.5.58) M ∗ ∗ (zn+1 zn + zn+1 zn ) + n=1 s s vn s s∗ s∗ s (vn vn+1 + vn vn+1 ) .6. the superﬂuid fraction in terms of quasiparticles amplitudes and energies can be written as: (1) (2) fs = fs − fs . vn = √ eiq·xn uq . J N − ωs + ωs + δss 5.m 2 + n. (5. As the interactions are increased phase coherence is lost.6 5.61) n M M . (5. Neglecting interaction eﬀects during the expansion. We assume a d dimensional separable square optical lattice with equal tunneling matrix element J in all directions and periodic boundary conditions.m zn eiq·xn n ρnm eiq·(xn −xm ) .59) 1 2N J N s n un + s s s n un+1 vn s.6 Applications 71 The quasimomentum distribution of the atoms released from the lattice is important because it is one of the most easily accessible quantities in the experiments. where ai and Mi are the lattice spacing and number of lattice sites in the i direction and M = i Mi is the total number of lattice sites.s (zn+1 − zn−1 ) ωs s us vn+1 n 2 2 + s s n (un+1 vn (5. (5. For a one dimensional optical lattice. the quasimomentum distribution represents the Fourier transform of the original spatial distribution in the lattice.1 Applications Translationally invariant lattice To understand manybody eﬀects included in the quadratic approximations we start by studying the case where no external conﬁnement is present.
vq = v−q ).64): 2 ωq = L2 − M2 = qq q−q q 2 + 2U no q . 2ωq + no U − ωq . where t is the number of nearest neighbors t = 2d. 2ωq (5.66) (5. • Bogoliubovde Gennes (BdG) equations and HFB. k (5. n = N/M .72 Chapter 5 Quadratic approximations Here the vector q denotes the quasimomentum whose components assume discrete 2π values which are integer multiples of √ with a is the lattice spacing.68) n = no + (5. (5.65) u2 = q 2 vq = Lqq + ωq = 2ωq Lqq − ωq = 2ωq q + no U + ωq .67) q uq vq = − and Mq−q no U = 2ωq 2ωq 1 M 2 vq . Mq−q = no U. The BdG equations in the translationally invariant case become the following 2 × 2 eigenvalue problem Lqq −Mq−q Mq−q −Lqq with Lqq = q a M (5.Popov approximation In the simplest quadratic approximation the DNLSE reduces to µ = −tJ + no U.62) uq vq = ωq uq vq . q=0 (5.(5.64) Here we have introduced the deﬁnition = 4J d 2 ki a i=1 sin ( 2 ). The constrain that ﬁxes the number of particles can be written as .63) + no U.69) with n the total density. The quasiparticle energies ωq and modes are found by diagonalizing Eq. The d amplitudes uq and uq must satisfy the condition uq 2 − vq 2 = 1 and can all be chosen to be real and to depend only on the modulus of the wave vector (uq = u−q . The translationally invariant case has the advantage that the quasiparticle modes are always orthogonal to the condensate and therefore the parameters cs are always zero. (5.
(5. due to the cancellation of the ρnn . 1 (5. Thus. ∞) = = 1 √ ln(cot(qo )) 2 2π 0. (5. ∞) α(3. Therefore. the most important contribution to the depletion comes from the lowlying modes.5. In the thermodynamics limit. terms in the quasiparticle equations (5. 0. M → ∞ the sum can be replaced by an integral and we ﬁnd α(1. α is a dimensionless quantity which depends only on the dimensionality of the system.6 Applications 73 no = n − 1 M q q=0 + U no 1 − 2ωq 2 .77) (5.73) (5. in the regime U no /J > 1 the interaction term dominates for all quasimomentum and so ωq ∼ 2U no q . where the single particle energy dominates at high momenta (it grows as q 2 ).72) 1 α = α(d.76) (5. For these modes is also true that ωq ∼ 2U no q . (5. in the weakly interacting regime. M ) ≡ M g = n+ n M −1 2M √ J . to a good approximation the condensate fraction can be written as: U no α. On the other hand. (5. 2 2 q q=0 U n/J U n/J 1 .74) In Eq. the HFBPopov approximation leads to the same quasiparticle energy spectrum and amplitudes than the BdG equations. the single particle excitations in the presence of the lattice are always bounded by 4Jd. The only diﬀerence is in the chemical potential which has an extra term 2U ρnn µ(P ) = −tJ + U no + 2U ρnn . ∞) → α(2. J no ≈ g − with (5.227293.74) the term M −1 is a ﬁnite size eﬀect term which is important to 2M keep for low density systems.78) qo →0 .71) As opposed to the free particle system.70) In the homogeneous case.34). .75) (5.160287.
the momentum distribution and superﬂuid fraction. Eq.k . M 1 1 q + U no = no δq.60)]. In Eq.80) (5. q δq.79) tells us that for very weak interactions almost all the atoms are in the condensate.k .83) (5.80) is consistent with those of other approaches previously reported in the literature [107. There the authors show that Eq.(5. J 4J 2 (5. M ) has a ﬁnite value. It is important to emphasize that due to the translational invariance of the lattice [see Eq. 109.0 + − M 2ωk 2 v q 2 cos(qa) . The expression for the superﬂuid fraction gives a direct insight into the behavior of the system as atoms are pushed out of the condensate due to interactions.65)(5. but under the BdG (and HFBPopov) approximation it only vanishes when U/J → ∞.82) (5. the second order term vanishes in the Bogoliubov limit.66) are completely determined if no is known. If we restrict our attention to the one dimensional system we have 2 2J q − − → no arctan( −− ).81) nq = no δq. M 1 (1) fs = no + N M (2) fs = 0. (5. (5. 110].72) we obtain all necessary information.79) Eq. The calculated quasiparticle amplitudes can be used to get analytic expressions for the number ﬂuctuations. Because Eqs.(5.74 Chapter 5 Quadratic approximations The infrared divergence in the one dimensional thermodynamic limit is a consequence of the importance of long wave length correlations in low dimensional systems. (5. ωq M →∞ π no U ∆nn = no q (5. which is given . As interaction increases the condensate fraction decreases. For ﬁnite one dimensional systems however α(1.84) (5. 108. (5. The BdG (and HFBPopov) equations therefore do not predict any superﬂuid to Mott insulator phase transition. by solving Eq.(5.0 + 1 q2 v  δq. (5. The solution of the algebraic equation is: no ≈ g + α2 U − 2J g α2 U α4 U 2 + .84) the sum involving the Bogoliubov amplitudes v q characterizes the diﬀerence between the condensate fraction. [95] to produce results for squeezing.80) has been studied in Ref.85) (1) (2) fs = fs − fs .
6 0.4 0.33. The exact second order (2) term (dashed line) of the superﬂuid fraction.2 fs. for a system with M = 3 and ﬁlling factors n = 5 and 5.4 0. This happens in the Mott insulator state.33 fc 0.3 we compare the number ﬂuctuations per lattice site. 5. 2 1.8 0. As interactions are increased the depleted population spreads into the central part of the band.5.6 0. This.8 0. middle: condensate fraction (Exact: solid line. crit The vertical line shown in the plots is an estimation of Vef f when the system is commensurate In Figs.6 Applications 75 by the ﬁrst term.2 n 5. N . BdG(and HFBPopov): dotted line).8 0.5 0.5 n 5.33 5 10 15 20 25 30 Veff 0. For weak interactions and small depletion.2 n 5 10 20 30 40 Veff fs.4 0. (where the cos(qa) term has a negative sign) and the superﬂuid fraction is reduced.2: Comparisons of the exact solution and BdG (and HFBPopov) solutions as a function of Vef f = U/J. not a perfect switching oﬀ of the superﬂuid. the population in the upper quarter of the band again produces a positive contribution to the superﬂow and the contribution from the sum decreases. The size of the matrix +M −1)! needed in the exact solution for N atoms and M wells scales as (N!(M −1)! .2 and 5. BdG(and HFBPopov): dotted line). We were restricted to consider only three wells due to computational limitations.4 0. however can lead to perfect ﬁlling and cancellation of the ﬂow. fs2 n n 5. which cannot be described by the Bogoliubov approximation.33 5 10 15 20 25 30 Veff 5 10 15 20 25 30 Veff Figure 5. fs2 n fc n 5 n 5 10 20 30 40 Veff 10 20 30 40 Veff 2 1. In the case of our Bogoliubov description we can only see reduction of the ﬂow. and the superﬂuid fraction. the condensate fraction and the total and second order superﬂuid fraction determined from the exact solution of the BoseHubbard Hamiltonian to the solution obtained from the BdG (and HFBPopov) equations as a function of the ratio Vef f = U/J. M = 3.33 and n = 50 and 50. fs is also shown in these plots.79:red line ). right: superﬂuid fraction fs (Exact: solid line.6 0.5 1 0. and commensurate ﬁlling factors n = 5 and 5. Left: number ﬂuctuations (Exact: solid line.analytic (5.5 1 0. BdG(and HFBPopov): dotted line.33. The systems used for the comparisons are one dimensional lattices with three wells.6 0. Finally for even larger interaction strengths. the depletion of the condensate has initially little eﬀect on superﬂuidity. In a sense the interactions are playing a role akin to Fermi exclusion ”pressure” in the case of electron ﬂow in a band.8 0.2 0.
Left: number ﬂuctuations (Exact: solid line. analytic Eq.5Vef f crit ( with Vef f = (U/J)c .2 n 50. It can be observed that the analytic solution agrees well with the numerical solution of the BdG( and HFBPopov) equations.8 0. if the approximate approach works well for these small systems we expect it to provide a good description of the larger systems prepared in the laboratory.4 0.8 0.4 0.9)). fs2 n 50.60)). middle: condensate fraction (Exact: solid line. However.(5. BdG(and HFBPopov): dotted line). provided it predicts accurately the ﬁrst order term.79).6 0.6 0.3: Comparisons of the exact solution and BdG(and HFBPopov) solutions as a function of Vef f = U/J. we only expect them to give a good description of the superﬂuid fraction in the region where the second order term is extremely small.33.2 n 50 50 100150200 Veff 0. and the region in which the BdG (and HFBPopov) equations are accurate gets larger. the BdG (and HFBPopov) equations crit starts to fail.33 50 100150200 Veff Figure 5. BdG(and HFBPopov): dotted line.2 fs. BdG(and HFBPopov): dotted line). fs2 n 50 50 100150200 Veff 5 n n 50.49) vanishes in the quadratic approximations (see Eq. In the plots for the condensate fraction we also show the analytic approximation given by Eq.79): red line ). Its predictions of squeezing agree very well with the exact solutions right up to the point where the number ﬂuctuations become less than unity.2 fs. When the second order term starts to grow. An estimate of Vef f is shown by a vertical line in some of the ﬁgures. This is exactly what is observed in the crit plots. right: superﬂuid fraction fs (Exact: solid line. With increasing ﬁlling factor the critical value is shifted towards larger values of the interaction strength. Because the second order term of the superﬂuid fraction (second term of Eq.8 0.(3. For the cases with noncommensurate ﬁllings the agreement is signiﬁcantly .76 n 50 fc Chapter 5 Quadratic approximations 5 n 3 1 50 100150200 Veff 0.4 0. for a system with M = 3 and ﬁlling factors n = 50 and 50.33 fc 3 1 50 100150200 Veff 0.(5. In this case the agreement is much better.8 0.33 50 100150200 Veff 0. It is interesting to note that the number ﬂuctuations predicted by the theory are accurate in a greater range than the other physical quantities shown.6 0.(5.4 0. fs is also shown in these plots. The exact second order (2) term (dashed line) of the superﬂuid fraction. (5. typically above 0. see Eq.6 0.
It can be seen that the condensate fraction is always higher in the HFB solutions and remains constant at higher values of U/J which is not physically correct. 5.90) (5. As discussed in the previous section.92) (5. The diagonalization of them yields µ = −tJ + U no + 2U n + U m. the most important contribution to the depletion of the condensate in the BdG equations comes from the low lying modes. In Fig. for a system with M = 3 and ﬁlling factor n = 10. because of the ωq −1 dependence. (5.91) + 2U (no + m) q − 4U 2 no m. (5. As can be seen in the plots for these cases the second order term is always very small. q=0 q 2 (5.5. 2ωq 2ωq ωq 2 = with n ≡ 1 M 2 vq . 2ωq 2ωq Mq−q no U + U m uq vq = = .4 we compare the condensate fraction predicted by the HFB equations with the condensate fraction obtained from the BdG equations. This is not surprising because when the ﬁlling is not commensurate there is always a superﬂuid fraction present and the Mott transition doesn’t occur.86) (5.87) + no U − U m.93) m ≡ − 1 M vq u q .88) (5. Lqq − 2ωq q + no U − U m − ωq vq 2 = uq 2 − 1 = = .89) (5. q=0 From the above equation it can be seen that the quasiparticle energies predicted under the HFB approximation don’t approach zero as q goes to zero. . The ﬁnite value of ωq has in the HFB approximation as q goes to zero explains why the HFB equations are always going to predict smaller condensate depletion than the gapless approximation. Mq−q = no U + U m.6 Applications 77 better for all quantities. • HFB equantions In the translationally invariant case the HFB Hamiltonian leads to the following equations: Lqq −Mq−q Mq−q −Lqq with Lqq = q uq vq = ωq uq vq .
8 0 10 20 Veff BdG HFB 30 40 Figure 5.78 Chapter 5 Quadratic approximations n 10. • BdG equations In the presence of the harmonic trap the BdG equations take the form: .2 Onedimensional harmonic trap plus lattice In this section we consider the experimentally relevant case when there is an external harmonic magnetic conﬁnement in addition to the lattice potential.85 0. For simplicity we focus our attention on the one dimensional case. for a system with M = 3 and ﬁlling factor n = 10. M 3 1 0. Secondly we study the HFBPopov approximation and use it to explore the limits of validity of the quadratic approximations as the interactions are increased.9 0.95 fc 0.6.4: Comparisons of the condensate fraction given by the BdG solutions and the HFB equations as a function of Vef f = U/J. Firstly we study the BdG equations in the weakly interacting regime and get some insight on the solutions by deriving analytic results using the so called ThomasFermi approximation [12]. We consider the two most relevant approximations which are the BdG equations and the HFBPopov approximation. 5.
takes a signiﬁcant contribution only near the edges of the wavefunction. The quantum pressure. s 2 ∗ s n zn  (zn un s − z n vn ) nn = ˜ cs = U 2 n zn  .97) (5. To ﬁrst order it is possible to neglect the depletion of the condensate and assume that zn 2 = N . (2.101) .39)) the Bogoliubov approximation takes a rather simple analytic form.94) + (2U zn  − µ + Ωn 2 2 2 2 s )vn − ∗ cs zn = s s −J(vn+1 + vn−1 ) ∗2 −U zn us . N = n (zn  + nn ).96) (5. B s −ωs vn (5. 18. and to a good approximation it can be neglected.6 Applications 79 B ωs us + cs zn = −J(us + us ) + (2U zn 2 − µ + Ωn2 )us n n+1 n−1 n 2 s −U zn vn . RT F ≈ 3 NU 4 Ω 1/3 . This approximation has been very useful to derive analytic expressions in magnetic conﬁned condensates without the lattice and proved to be in agreement with experimental measurements [12. Changing the sum to an integral it can be solved for µ to get 2 µ = ΩRT F . (5. which is proportional to the kinetic energy. The eﬀect of increasing the ratio N U/Jaho is to push the atoms outwards. ﬂattening the central density and increasing the width of the condensate wavefunction. ˜ s vn 2 . 19]. n 2 (5.5. Under the ThomasFermi approximation one gets a condensate density proﬁle of the form of an inverted parabola which vanishes at the turning points RT F = µ/Ω: zn 2 = µ−Ωn2 U 0 n < RT F n > RT F (5.98) µzn = −J(zn+1 + zn−1 ) + (U zn  + Ωn )zn . (5. This is the socalled ThomasFermi approximation (TF). but the number of atoms is large enough such that N U/Jaho 1.99) In the parameter regime where U/J 1.95) (5.100) The chemical potential is determined by ﬁxing the total number of particles. where aho = m∗ ω∗ (See Eq.
(5. (n − 1)2 − 1− 2 RT F s s with Fn = Fn / 1 − n2 2 RT F n2 1− 2 RT F s Fn−1 − s Fn . ∂ 2 ∂ ∂x (1 − x ) ∂x Pn + n(n + 1)Pn = 0. 111] and use the parameter J as an expansion parameter in these equations.100) into Eqs. .(5. . We correct this at the end by removing from the modes their projection onto the condensate. to the exact solutions of Eqs.80 Chapter 5 Quadratic approximations To solve for the excitation spectrum we use the ThomasFermi condensate amplitude. We deﬁne the variables q q s Fn = uq + vn and Gs = us − vn .(5. Eq. (5.(5. Therefore the solutions of Eq.107) ΩRT F (2s + 1) B ωs B ωs (2s + 1) 3 4ΩRT F Gs = n n RT F The larger the size of the condensate the closer are the approximate solutions.105) n2 2 Ps RT F Ps n RT F . Eq.107). (5. To ﬁrst order in J we get the following set of equations: n2 2 RT F 2 2RT F Ω 1 − B s Gs = ωs Fn .104) are approximately given by : B ωs = s Fn = ω∗ s(s + 1) . 2 1− 1 1− n2 2 RT F (5.104) ωB =2 s ω∗ 2 s Fn . For simplicity we are going to set the conn n n stants cs to zero and ignore the orthogonality constraint at the beginning.95) [18.106) (5. n These two equations can be combined to get a simple equation of the form 1− (n + 1)2 2 RT F 1− n2 2 RT F s s Fn+1 − Fn (5.104).103) B = ωs Gs . n (5. The above equation is just the discretized form of the Legendre equation.94)and (5.102) (n+1)2 2 RT F −J 1− s s s Fn+1 + Fn−1 − Fn + 1− n2 2 RT F (n−1)2 2 RT F 1− (5.
5 we plot the condensate density proﬁle found by numerically solving the BdG equations and compare it with the TF condensate solution. (2.110) it follows that the desired F and G amplitudes orthogonal to the condensate are thus given by s s B Fn → Fn − 2cs zn /ωs . (5. s = 2: cs = 4 ω ∗ R5 Ω2 δs. A small condensate depletion guarantees the validity of the quadratic approximation to describe the manybody system.101). In Fig. n n s vn (5. To perform the sum.37)). If we calculate the constants cs using Eq. to make the solutions orthogonal to the condensate we have to remove their projection onto the condensate mode: B us → us − cs zn /ωs . In Fig. The TF approximation reproduces the numerical solution very accurately except at the edges where the kinetic energy can not be ignored with respect to the potential energy.(5. and orthogonality properties of the Legendre polynomial we ﬁnd that the only mode that needs to be corrected is the quadrupole mode.5. the TF quasiparticle and excitation spectra of the noninteracting system (see Eq. 5. In the same plot we also show the condensate depletion which is small for the chosen parameters. Finally. 5.6 Applications 81 Finally.6 we show comparisons between the quasiparticle energies found numerically.s vn 2 . N = n. to ﬁx the number of particles to N it is necessary to renormalize the condensate wave function and the chemical potential.110) From Eqs. (5. 15U (5. 2 TF The divergence however is nonphysical and only indicates the failure of the TF approximation near RT F . It can be seen in the plot that only the lowest a4 1/3 2 −1/2 . It is important to point out that the term 1 − Rn in the G amplitudes makes the sum divergent. This is done by renormalizing the Thomas Fermi radius and replacing N by N − N in Eq.108) (5.111) (5. (5.2 .109) → s vn − ∗ B cs zn /ωs .112) Gs n → Gs . we have to exclude the where the contribution from the boundary layer of thickness d = 2RhoF T TF approximation starts failing [111]. it is possible to calculate the s depletion of the condensate. n Once solved for the quasiparticle amplitudes.99).
in Fig.5: Comparisons of the condensate wave function found by numerically solving the BdG equations with the ThomasFermi solution. 5.5 × 10−4 and N = 100.5 1 0. the ﬁrst quasiparticle excitation energy coincide with the ﬁrst noninteracting excitation energy .5 2 # atoms 1. (5.105). The depletion is also shown in the plot. In the regime where Ω < J and the system’s size is large enough that discretization eﬀects. In general. We observe good agreement between the two solutions except near the TF . quasiparticle excitations other than the dipole are aﬀected by interactions and they have diﬀerent energy than the noninteracting excitations. (5. In the same graphic the quasiparticle amplitudes for the s = 2. To test the accuracy with which the TF approximation describes the quasis particle amplitudes. Ω/J = 9. This ﬁrst excitation is known as the dipole mode and describes the oscillatory motion of the center of mass when the system is displaced. 4 excitation modes are also depicted.2. 5.111)) to the product found by numerically solving the BdG equations. Higher order modes describe more complicated collective excitations.5 0 30 20 10 0 n 10 20 30 Condensate depletion TF Figure 5.8 we compare the product Fn Gs as a function n of the lattice site n using the TF approximation results (Eqs.7.112) and (5. Higher excitation energies are closer to the noninteracting ones. The parameters used were U/J = 0. 3. calculated by numerically solving the BdG equations n are shown in Fig.82 Chapter 5 Quadratic approximations 2. The dipole quasiparticle ams=1 plitudes (us=1 . As shown in Fig. 5. vn ). are not important. neglected in Eqs. lying modes are well described by the TF approximation. The s = 2 mode is known as the quadrupole mode and is related to the breathing of the condensate when shaken.6 TF quasiparticle energies lie below their noninteracting conterparts.
15 0.6 Applications 4 83 U J 0.05 BdG TF Free J 9. This behavior is consistent with the fact that near the edge of the atomic cloud. • HFBPopov approximation In the translationally invariant system.5. radius. However. The parameters used were U/J = 0.1 0. when the harmonic conﬁnement is present this is no longer the case.2.6: Comparisons of the quasiparticle spectrum found by numerically solving the BdG equations(BdG) with the ThomasFermi solution (TF) and the noninteracting energies (Free).5 x 10 N 100 2 4 6 8 10 12 sth eigenvalue 14 16 18 20 Figure 5.2 Ωs ER 0. the kinetic energy becomes comparable to the potential energy and therefore the TF approximation breaks down.25 0. Ω/J = 9. both the BdG and the HFBPopov approximations give the same quasiparticle amplitudes and energies. The HFBPopov approximation shifts the quasiparticle energies depending on the spatial variation of the non condensate density in the region of the condensate.5 × 10−4 and N = 100. We use the HFBPopov approximation instead of the BdG approximation because of the relevant role of the noncondensate atoms into the condensate as the system is driven to the Mott insulator transition. In this section we investigate the limits of validity of the HFBPopov approximation as the parameter U/J is increased. In the presence of an harmonic conﬁnement the HFBPopov equations take .2 0.
(5.114) zn . The {un n . In the second s ˜ step we solve for {us .115) (5.117) by using the nn obtained ˜ in the previous cycle.113) using the nn from the previous n s . s (5.118) The HFBPopov are nonlinear equations and therefore it is more complicated to get analytic approximations. The parameters used were U/J = 0. ˜ 2 ∗ s n zn  (zn un s − z n vn ) cs = U 2 n zn  .115) subject to the constraint Eq. s −ω s vn 30 20 10 0 10 20 30 n s 4 0 0 30 20 10 0 10 20 30 n Figure 5.2. we solved the HFBPopov equations numerically by an iterative procedure. [19]. Ω/J = 9. (5. (5. vn } in Eqs. (5. This generates new values for the zn . the form: ωs us + cs zn = −J(us + us ) + (2U (zn 2 + nn ) − µ + Ωn2 )us ˜ n n+1 n−1 n (5. In the ﬁrst step we solve Eq. n zn−1 ) + U (zn 2 + 2˜ n ) + Ωn2 n s vn 2 .113) s nn ) − µ + Ωn2 )vn ˜ − ∗ cs zn = µz n = −J(zn+1 + nn = ˜ s s −J(vn+1 + vn−1 ) + (2U (zn 2 + ∗2 −U zn us . v s } are used then to update n . similar to the one followed in Ref.116) (5.84 Chapter 5 Quadratic approximations s 1 us vs n n us vs n n vs n us n 0 s 2 0 30 20 10 0 10 20 30 n s 3 us vs n n us vs n n 30 20 10 0 10 20 30 n 2 s −U zn vn .5 × 10−4 and N = 100. ˜n cycle and the newly generated zn . Each cycle of the iteration consists of two steps. (5.7: Lowlying quasiparticle amplitudes found by numerically solving the BdG equations. Instead.117) N = n (zn 2 + nn ).
J was varied to achieve a range of Vef f = U/J between 0.5 × 10−4 and N = 100.13).8: Comparisons between the product Fn Gs calculated from the Thomasn Fermi approximation and the exact numerical solution of the BdG equations. Ω/J = 9. The range was chosen based on a local mean ﬁeld approach [77]. Because the HFBPopov is gapless.0ER .01 and 312. In general we observe the reduction of the condensate population and thus the increment . 5. the condensate population (triangles) and the onsite depletion (empty diamonds) as Vef f is increased.95.5. with ER the one photon recoil energy. The parameters used were U/J = 0. for comparison purposes.113) with the cs set s to zero but removing in each cycle the projection of the calculated {us . N = 1000. The results of the numerical calculations are summarized in Figs. and set U N = 1. We used a total number of 1000 atoms. (5. (5. which for the case of a rubidium condensate corresponds to a trap frequency of approximately 90 Hz.0015ER . The parameters chosen for the numerical calculations were Ω = 0. Convergence is reached when the change in n 2 n ˜ n  from one cycle to the next is smaller than a speciﬁed tolerance. the ground state density proﬁle for J = 0 (empty boxes). This has the advantage that it can be calculated exactly from the Hamiltonian.9 we plot the evolution of the density proﬁle (black boxes). In Fig.2. In the plots we also show.6 Applications 85 s 1 BdG TF s 2 BdG TF Fs Gs n n Fs Gs n n 0 0 20 0 n s 3 BdG TF 10 20 20 0 n s 4 10 20 Fs Gs n n Fs Gs n n BdG TF 0 20 0 n 10 20 0 20 0 n 10 20 s Figure 5. which for our parameters estimates the transition region between Vef f ≈ 640 (at the center where the local ﬁlling factor is approximately 80) and Vef f ≈ 12 (at the wings). it is possible to keep the orthogonality of the excitations to the condensate by solving Eqs. vn } n amplitudes onto the condensate.
For the chosen parameters. total density (ﬁlled boxes) and local depletion (empty diamonds) as a function of the lattice site i for diﬀerent values of Vef f. the density proﬁle has a parabolic shape reﬂecting the conﬁning potential. of the depletion with increasing interaction strength. The empty boxes represent the exact solution for the case J=0. For lower values of J the HFBPopov density starts to diﬀer from the J = 0 limit.01 Veff =0. By comparing the evolution of the density as J is decreased with the exact solution at J = 0. but as J is decreased the depletion of the condensate becomes very important. We can appreciate that around Vef f = 3 both proﬁles are almost equal.9: Condensate density (triangles). even though the system is closer to the J = 0 limit. The density evolves from a Gaussian type (see plots for Vef f = 0. begin to be important.The site indices i are chosen such that i = 0 corresponds to the center of the trap. We can say that beyond this point higher order correlations.01 and 0. we can crudely estimate the validity of the HFBPopov calculations. The departure of the HFBPopov density proﬁle from the J = 0 one as J is decreased begins at the edges (see the panels corresponding to Vef f = 11 and 100). neglected by the theory. This is expected if we consider the onsite depletion. most of the atoms are in the condensate. Although these quantities are deﬁned only at the discrete lattice sites we join them to help visualization. When the system is in the superﬂuid regime.09) with smooth edges towards a ThomasFermi proﬁle with sharp edges adjusting its shape to the J = 0 proﬁle.86 Chapter 5 Quadratic approximations 80 60 40 20 Veff =0. .09 80 60 40 20 80 60 40 20 Veff =3 Veff =11 80 60 40 20 80 60 40 20 10 5 0 i Veff =100 Veff =312 80 60 40 20 5 1010 5 0 i 5 10 Figure 5. For such values of Vef f the local depletion in the wings .
09).5. For higher values of Vef f .01). For the smallest ﬁlling factor (see Fig. At this point the failure of the method is clear and a fully quantal method is required.4 ωq 0. 5. As the optical lattice depth is ramped up the adiabaticity criteria is harder to . crosses (Vef f = 3). 0. ﬁlled diamonds (Vef f = 11).2 0 0 2 q 4 6 Figure 5.6 0. The HFBPopov quasiparticle spectrum is shown in Fig.10: Quasiparticle spectrum predicted by the HFBPopov theory for different values of Vef f : Empty diamonds (Vef f = 0. and thus the validity of the HFBPopov assumptions starts to be dubious. the HFBPopov density predictions diﬀers from the J = 0 solution even at the central wells.6 Applications 87 corresponds to a considerable percentage of the condensate populations.2) the diﬀerences between the homogeneous HFBPopov calculations and the exact solutions become important for values of Vef f greater than 20. see plot for Vef f = 312. 5. It is worth mentioning that the small energy diﬀerence between the ground and ﬁrst excited states for high values of Vef f makes the numerical solution very unstable in the sense that it is very easy to jump to an excited state when solving for the condensate wave function. empty boxes (Vef f = 100) and pentagons (Vef f = 312). The letter q labels the quasiparticle energies in increasing order. The homogeneous results shown in the previous sections corroborate our present statements for the conﬁned system. stars (Vef f = 0.10. The quasiparticle energies are in recoil units. The decrement in the energy spacing predicted by the HFBPopov theory as the system approaches the transition is very useful to keep in mind for the experimental realization of the Mott transition. It can be observed how the lower energy eigenvalues evolve from a linear non degenerated spectrum to an almost degenerated one as J is decreased.
(5.67)). 5.12 we present the quasimomentum distribution for the same parameters used in the previous plots. Vef f = 3 (circles).66) and (5. but replacing the condensate density in each lattice site by the one found numerically for the trapped system (see Fig. we expect that the inhomogeneous HFBPopov results for squeezing also agree with the exact solutions right up to the transition. (5.88 fulﬁll. v q } amplitudes found for the homogeneous system (Eqs. Based on this agreement and the results for the homogeneous system shown in the previous section. In Fig. We also show the number ﬂuctuations evaluated by using a local density approximation (empty boxes).Vef f = 11 (triangles). The number ﬂuctuations proﬁle reﬂects the condensate proﬁle. The latter was calculated by substituting in the number ﬂuctuations expression (Eq.11: Number ﬂuctuations in the self consistent HFBPopov approach as a function of lattice site for Vef f = 0. The complete agreement between the two approaches justiﬁes the validity of the local density approximation for the estimations of local quantities in conﬁned systems.9). As the hopping rate is decreased we observe that the sharpness of the central peak decreases and .01 (boxes). 5.11 we plot the results for the number ﬂuctuations found numerically using the inhomogeneous HFBPopov approach. The empty boxes shown for each of the curves correspond to the number ﬂuctuations predicted by the homogeneous HFBPopov model using a local density approximation. The maximum value reached by the proﬁle decreases as Vef f is increased. Vef f = 0.09 (crosses). 5. Chapter 5 Quadratic approximations 6 ∆ni 4 2 0 10 5 0 i 5 10 Figure 5. Vef f = 100 (stars) and Vef f = 312 (diamonds). In Fig.56)) the {uq . The distribution for the two lowest values of Vef f corresponds to the one that characterizes an uncorrelated superﬂuid phase with a narrow peak at small quasimomenta.
a the lattice spacing.5. The step function like shape of the condensate proﬁle causes an oscillatory  sin(x)/x shape of the quasimomentum distribution.01 Veff =0. This agrees with numerical solutions of the BoseHubbard Hamiltonian using Monte Carlo simulations [112]. The evolution of the onsite superﬂuidity as the interaction strength is in(1) .09 nq 200 100 0 200 Veff =3 Veff =11 Veff =100 Veff =312 nq 100 0 0 1 aq 2 30 1 aq 2 3 Figure 5. We attribute the origin of the small peak to the depletion of the condensate at the wings. for diﬀerent values of Vef f . 5. q the quasimomentum.6 Applications 89 400 300 nq 200 100 0 300 Veff =0.(5. It is interesting to note the appearance of a small peak between q = 0. The curves corresponding to Vef f = 0. the most important contribution to the quasimomentum distribution still comes from the condensate atoms.12: Quasimomentum distribution as a function of qa.5 and 1 which is most noticeable for the Vef f = 3 case.01 − 11. the longrange order starts to decrease and the Fourier spectrum becomes broader. depict how as Vef f is increased the superﬂuid proﬁle decreases faster at the wings and at the center but no major change is observed in the middle section. the distribution extends towards large quasimomenta. As the lattice depth is increased the hopping becomes energetically costly. For the parameters when the small peak is present.60).13 we plot the ﬁrst order on site superﬂuid fraction fsn which was deﬁned in Eq. which are in the regime where the HFBPopov is expected to be valid. In Fig.
01.06 1) fsc( 0. creased. crosses: Vef f . We expect the HFBPopov method to give a better description of the transition as the dimensionality of the system is increased and therefore to be a good model in an experimental situation.02 0 10 5 0 i 5 10 Figure 5. empty boxes: Vef f . = 312. is in agreement with the development of uncompressible regions surrounded by superﬂuid rings predicted for trapped systems [79] as the transition is approached. for systems with ﬁxed integer density as the interaction strength is changed.09. and one mean ﬁeldlike exhibited when the transition is induced by changing . = 0. its behavior depends strongly on the dimensionality of the system.like. empty diamonds: Vef f . we considered one dimensional systems. [79] the Mott transition in a ddimensional homogeneous system has two diﬀerent critical behaviors: one (d+1) XY. = 100 and triangles: Vef f .The site indices i are chosen such that i = 0 corresponds to the center of the trap. the Mott transition has been achieved [46] in a 3 dimensional lattice with ﬁlling factors between 1 and 3. stars: Vef f .1 0. In the present analysis.08 0. due to computational limitations.04 0. Experimentally. we showed how the method is incredibly powerful in describing most of its characteristic features as they are driven from the superﬂuid regime towards the transition. = 3. As shown in previous studies [44]. =0. exhibiting a domain localized decrement instead of a global one.13: First order onsite superﬂuid fraction as a function of the lattice site i for diﬀerent values of Vef f .90 Chapter 5 Quadratic approximations 0. Even though the HFBPopov approach fails to describe the strong coupling regime for the one dimensional systems we considered. Filled boxes: Vef f . The Mott transition is a quantum phase transition and as for all critical phenomena. = 11. .
they take into account the eﬀect of the surrounding atoms on the nature of the interatomic collisions. commensuration is only well deﬁned locally. together with the fact that the ﬁnite length scale introduced by the trap suppresses the long wave ﬂuctuations which are responsible for destroying the mean ﬁeld [16]1 . we expect that for trapped systems in d = 3. 1 One obvious consequence of this is that BEC is possible in one and two dimensions in a trap whereas in the homogeneous. The problem with the HFB equations is that not all the interactions are upgraded and the diagonal elements still contain bare interaction terms. thus the total superﬂuid fraction doesn’t vanishes in the Mott regime. (5. In this section.5.7 Improved HFBPopov approximation The corrections to the quadratic Hamiltonian from higher order terms have essentially two major eﬀects. the range of validity of the HFBPopov extends closer to the transition. Because the critical dimension for the latter type of transition is two [44]. Secondly. Firstly. The factorization is appropriate for the quartic term in the Hamiltonian (as it is justiﬁed by Wick’s theorem [16]) but not for the cubic terms. Diﬀerent from the homogeneous case where the Mott transition is characterized by the global oﬀset of the superﬂuidity. The inhomogeneity introduced by the conﬁning potential allows the existence of extended Mott domains (above a critical interaction strength) surrounded by superﬂuid ones [79]. 5. make us believe the critical behavior in conﬁned systems to be more meanﬁeld like. thermodynamic limit it can not occur in fewer than three dimensions . they include the eﬀect of the interactions between condensate and noncondensate atoms on the behavior of the condensate. To solve the problem the ﬁrst approximation that one might think of is to upgrade by hand the bare interactions in the diagonal terms and replace them by the manybody scattering matrix. This issue. we discuss the improved Popov approximation and apply it to a translationally invariant system. 16]. In the ﬁrst part of this section we show. This procedure is known as the improved Popov approximation [53. for conﬁned systems. We explicitly show that the improved Popov approximation gives a better description of the manybody physics as the Mott insulator transition is approached in comparison with the regular HFBPopov approximation.7 Improved HFBPopov approximation 91 the density.18)) we actually upgrade the bare interaction potential U to the many body scattering matrix which gives a better description of interparticle collisions. by taking into account the anomalous average m (Eq. We mentioned previously that although the HFB treatment should represent an advance over the BdG theory it is not useful in practice because it is not gapless. how. [79]. The diﬀerent treatment of the oﬀ diagonal and diagonal terms is caused by the factorization approximation on which the HFB approximation is based.
ij the ﬁnal states and np a set of intermediate states T2b describes the interaction between two particles in vacuum. In this ﬁgure kl designates the initial states. Figure 5. plays the same role as its twobody counterpart. (5. To describe their inﬂuence on the scattering the concept of the manybody scattering matrix is introduced. Inserting a complete set of eigenstates in Eq. but describes scattering occurring in manybody systems. they are usually referred as ladder diagrams. Tmb . Because of the ladder shape of the diagrams included in the scattering matrix.92 Chapter 5 Quadratic approximations 5.1 The twobody and manybody scattering matrices The twobody scattering matrix.119) ˆ z − H sp ˆ Here V is the interatomic potential and H sp is the single particle Hamiltonian. The manybody scattering matrix. In the interacting Bose system. For a bosonic gas there are two major eﬀects . A diagrammatic representation of T2b is shown in Fig. and the kets pq are the corresponding two particle eigenstates and describe the intermediate states in the collision of two atoms.7. this is physically interpreted as the energy of the scattering process. T2b . describes the scattering of two particles in vacuum. (5. binary collisions do not occur in vacuum but in the presence of other atoms.14: Diagrammatic representation of the twobody T2b scattering matrix.120) ˆ where εp and εq are single particle energies (eigenvalues of H sp ).14. 5. Although T2b is deﬁned for a general complex parameter z.119) one obtains: T2b (z) = V + pq T2b (z) = V + V 1 V pq 1 z − (εp + εq ) pqT2b (z). (5. It is deﬁned as a function of a complex parameter z by the LippmannSchwinger equation:[51] T2b (z).
53. However this simpler Tmb matrix is the one that naturally appears in the manybody theory. let us take the inner product of Eq. see Eq. Tmb should account for Bose enhancement of a scattering process. 5. At this point it is important to highlight that the zero energy of the manybody scattering matrix is not the same as the zero energy in its twobody counterpart. in this case the z is measured relative to the energy of a stationary particle. The anomalous average m is deﬁned in terms of the quasiparticle amplitudes uq and vq according to Eq. on the other hand. z − (ωp + ωq ) (5. Second. is related to the oﬀ diagonal matrix elements.122) p To make the connection with the manybody scattering matrix.5. In the manybody case Tmb is deﬁned in terms of quasiparticle energies and therefore z is measured with respect to the condensate chemical potential.121). with the states q − q and 00 and evaluate it at . at least at lowest order.7 Improved HFBPopov approximation 93 which need to be accounted for as compared to scattering in vacuum. (5. the relevant states which enter in the matrix elements should be manybody states rather than single particle ones. A more general manybody scattering matrix which includes the full quasiparticle wave functions has been discussed by Bijlsma and Stoof [53].93). −(ωp + ω−p ) no (5. The full quasiparticle character of the intermediate states is not taken into account in this simple deﬁnition. and pq correspond to singleparticle wave functions.2 The anomalous average and the manybody scattering matrix In this section we are going to follow [16. For interpretation purposes we are going to restrict the analysis to the translationally invariant system. 113] depending upon the approximations made. ωp and ωq are quasiparticle energies. nq and np are thermal quasiparticle occupation factors which vanish at T = 0.90). the ladder diagrams included in the Tmb give the largest contribution to the modiﬁcation of the bare potential due to interactions.7. The one that we are going to use is a simplest generalization of the twobody matrix given by [16. 113]: Tmb (z) = V + pq V pq 1 + np + nq pqTmb (z). The product uq vq . 53] to show how by including the anomalous average the manybody scattering matrix is introduced into the theory. Because the twobody matrix is written in terms of single particle energies. If we combine the two equations we get an expression for the oﬀ diagonal matrix element Mq−q given by Mq−q U =U+ no M Mq−q 1 . where the interparticle distance is large compared with the swave scattering length. Various forms for the Tmb have been been proposed in the literature [16.121) where z is again a complex parameter. 54.(5. First of all. In the dilute gas limit.(5.
94
Chapter 5
Quadratic approximations
z = 0. At zero temperature, the thermal occupation factors vanish, nq = 0 and we get: 1 pkTmb (0)00 . −(ωp + ωk )
q − qTmb (0)00 = q − qV 00 +
pk
q − qV pk
(5.123) The quantity k1 k2  V k3 k4 is the momentum representation of the twobody interaction potential V . In the BoseHubbard Hamiltonian we assume a contact potential with amplitude U which yields a potential in the momentum representation given by k1 k2  V k3 k4 = U δk +k ,k +k . M 1 2 3 4 (5.124)
Using Eq. (5.124) into (Eq. 5.123) we obtain 1 M 1 Tmb , −(ωp + ω−p )
Tmb = U + with
U
p
(5.125)
p − pTmb (0)00
≡
1 Tmb . M
(5.126)
Eqs. (5.125) and (5.122) imply that Mq−q = no Tmb , and therefore that by including the anomalous average in the equations, the oﬀ diagonal matrix element given by U no in the BdG equations is upgraded to Tmb no in the HFB equations. The manybody scattering matrix also appears in the DNLSE since Eq. (5.89) contains m: µ = −tJ + U no + 2U n + U m = −tJ + Tmb no + 2U n. (5.127)
Finally, to conclude this section we just rewrite the HFB equations in terms of Tmb obtaining: + no (2U − Tmb ) no Tmb − −no Tmb q − no (2U − Tmb ) uq vq uq vq
q
= ωq
.
(5.128)
This allow us to explicitly see the diﬀerent treatment that the theory gives to the diagonal and oﬀ diagonal terms: The bare potential is completely upgraded to Tmb in the oﬀ diagonal terms but is not in the diagonal ones, and in the chemical potential.
5.7 Improved HFBPopov approximation
95
5.7.3
Improved Popov approximation
The appearance of a gap in the HFB equations is because of the diﬀerent treatment of the oﬀ diagonal and diagonal terms. We will show in chapter 8 that as we go beyond the quadratic approximation of the Hamiltonian and include higher order corrections we start to incorporate the manybody scattering in the diagonal terms. However, a naive way to correct for the gap problem in the HFB equations is simple to upgrade by hand the interactions in the diagonal terms and replace them by the manybody scattering matrix. This way to proceed is known as the improved Popov approximation [16, 53]. The matrix we have to diagonalize under the improved Popov approximation is:
q
+ no Tmb −no Tmb no Tmb −( q + no Tmb )
uq vq
= ωq
uq vq
(5.129) (5.130)
µ = −tJ + Tmb no + 2Tmb n
Eqs. (5.129) and (5.130) together with Eq. (5.125) and Eq. (5.69) form a closed set of equations. They are exactly the same than the HFBPopov equations if U is replaced by Tmb . The reason why there are two diﬀerent coupling constant in the equation for the chemical potential Tmb and Tmb , is because they describe two diﬀerent scattering processes. The second term in the right hand side of Eq. (5.130) describes the scattering between two atoms in the condensate colliding at zero momentum, q − qTmb (0)00 , and that is why the coupling constant is Tmb (see Eq. 5.126). On the other hand, the third term describes the collision between one condensate atom and one atom out of the condensate and the coupling constant in this case should be evaluated at diﬀerent energy Tmb ∝ p p0Tmb (ωp )0p . The extra diﬃculty that the improved Popov equations have compare with the HFBPopov equations is that Tmb depends on the quasiparticle energies and all equations must be solved in a self consistent way. To get simple analytic expressions we start by analyzing limiting regimes. • Case no Tmb > J In this regime the quasiparticle energy can be approximated by ωq 2no Tmb q . Using this approximation in the expression for the manybody scattering matrix we obtain Tmb = U 1+
U 2 α2 no JTmb
,
(5.131)
with α deﬁned in Eq. (5.73). Solving the algebraic equation we get two possible solutions Tmb = U ± U2 2Jno α2 + α4 + 4Jno α2 U . (5.132)
96
Chapter 5
Quadratic approximations
Because we want the root which increases as U increases we choose the solution with positive sign. • Case no Tmb < J In general one can show that the condition no Tmb < J is only satisﬁed if the quantity Tmb /J is small. This also implies that the parameter U/J must be small. In this regime, therefore, we can perturbativelly expand Eq.( 5.131 ) in powers of U/J to get: Tmb U 1+O U J . (5.133)
The fact that Eq. (5.134) has the correct asymptotic behavior in the limit no Tmb < J (it approaches to U as U goes to zero) even though it was derived under the assumption no Tmb > J, allow us to extend the validity of Eq. (5.132) to all regimes. Therefore, to a good approximation, we get a general formula for Tmb given by: Tmb = U + U2 2Jno α2 + α4 + 4Jno α2 U . (5.134)
We have checked numerically that in fact, Eq. (5.134) is a very good approximation of Tmb in all regimes. Having an analytic expression for Tmb , we can solve for the condensate density to get: no = n − 1 M
q q=0
+ U no 1 + , 2ωq 2 (5.135)
≈g−α
Tmb no . J
In the above equation we used the same approximations that lead to Eq.(5.72), which was explicitly showed to be good in Figs. 5.2 and 5.3 and the deﬁnition of g given in Eq. (5.74). If we solve Eq. (5.135), we ﬁnally obtain an analytic expression for no Ug . (5.136) J Surprisingly, the expression for the condensate density that we get after including the manybody scattering matrix is exactly the same, as the one we get if in the BdG equations for the condensate, Eq. (5.72), we replace no by g in the right hand side. Except for the term M −1 , this corresponds to the lowest order solution 2M we can get of Eq. (5.72), where instead of solving self consistently the algebraic equation we replace no by n. The term M −1 is very small at high densities when 2M n + 1/2 ≈ n, regime where in fact we only expect a mean ﬁeld treatment to be valid and it might be interpreted as a ﬁnite size eﬀect. Hereafter we are going no ≈ g − α
5.7 Improved HFBPopov approximation
97
to refer to the non self consistent solution of the BdG equations as the test ﬁeld approximation. Finally, if we use the expression for the condensate density Eq. ( 5.136) in Eq. (5.134), and assume again that n + 1/2 ≈ n, after some algebraic manipulations the ﬁnal result we get is no Tmb = U n. (5.137)
For us this a striking result. The net eﬀect of including the manybody scattering matrix in the theory reduces to replacing U no in the BdG equations with U n. This result is hard to understand because what it implies is that the lowest order approximation beyond the simple DNLSE, is the one that reproduces the best the exact solution as the interactions in the system are increased.
n 5 M 3 1 0.9 fc 0.8 0.7 0 5 10 15 UJ 20 25 30
Exact ManyB Popov test
1 0.9 fc 0.8 0.7 0 5
n 5.33
M 3
Exact ManyB Popov test
10
15 UJ
20
25
30
Figure 5.15: Comparisons of the condensate fraction predicted from the exact diagonalization of the BHH (red), the improved Popov approximation (yellow), the BdG (and HFBPopov) equations (blue) and the test ﬁeld approximation (green) as a function of Vef f = U/J, for a system with M = 3 and ﬁlling factors n = 5 and 5.33. To check the improvement that we get by including the manybody scattering matrix in the theory, in Figs 5.15 and 5.16 we show comparisons of the condensate fraction, fc = no /n, as a function of U/J calculated from the exact diagonalization of the manybody Hamiltonian (red) with the improved Popov(yellow) and the regular BdG (and HFBPopov)(blue) approximations. We also plot the test ﬁeld approximation results (green). In the plots we used a one dimensional lattice with M = 3 and densities n = 5 and n = 5.33 in Fig. 5.15 and n = 50 and n = 50.33 in Fig. 5.16. It is clear in the plots that the improved Popov is a better approximation than the regular BdG (and HFBPopov) approximation. In particular in the non commensurate, high density case the agreement between the improved Popov and the exact solution is very good. It also can be seen in the plots how the improved Popov reduces to the simple test ﬁeld approximation. The small diﬀerences seen in the low density case disappear as the density is increased. To corroborate the
3 M 3 Exact ManyB Popov test 100 150 UJ 200 250 0 50 100 150 UJ 200 250 Figure 5.17: Comparisons between no /nTmb = fc Tmb and U as a function of U for a system with M = 3 and ﬁlling factors n = 5 and 50. (5.8 0. Phase Transition One of the signatures of the entrance to the Mott insulator phase is a zero condensate density. the improved Popov approximation (yellow). for a system with M = 3 and ﬁlling factors n = 50 and 50.8 0.9 fc 0. n 5 30 25 20 15 10 5 0 0 5 250 200 150 100 50 0 0 50 n 50 Tmb fc Tmb fc Tmb fc U Tmb fc U 10 15 20 25 30 U 100 150 200 250 U Figure 5.17 Tmb fc . calculated from Eqs. validity of Eq. As we mentioned below. we plot in Fig. at which the condensate density vanishes: . vs U . the BdG (and HFBPopov) equations (blue) and the test ﬁeld approximation (green) as a function of Vef f = U/J. On the other hand.98 Chapter 5 Quadratic approximations 1 0.134). The curves just overlap in the high density regime n = 50 and in the low density case n = 5 the disagreement is very small. M 3 Exact ManyB Popov test 1 0. 5. the BdG (and HFBPopov) approximation does not predict the Mott insulator transition because the condensate density only vanishes in the limit when U/J → ∞.16: Comparisons of the condensate fraction predicted from the exact diagonalization of the BHH (red). Vef f = (U/J)c .33.7 0 50 n 50.7 n 50.137). does predict the existence of a critical value . (5.136) and (5. the improved Popov c approximation.9 fc 0.
α → ∞ and the critical point approaches zero. Furthermore. the quadratic approximations are developed assuming small ﬂuctuations and therefore as soon as quantum correlations become important it is clear that a fully quantal method is required. both equations describe a critical point which scales linear with n and with a proportionality constant which depends only on the dimensionality of the system. the phase transition predicted by the improved Popov approximation does not have a clear connection to the real Mott insulator phase transition.8 Conclusions 99 n + M −1 n 2M −→ − . This distinction is crucial to reproduce the characteristic superﬂuid to Mott insulator phase diagram. However.5.138) n 1 2 α α2 If we compare the above expression with the equation of the critical point calculated from a mean ﬁeld model starting in the Mott insulator phase (see chapter 3) c given by. We have shown that this method can be used to predict the relevant physical quantities over a useful range. . c Vef f = 5. there is not clear signature that the system becomes incompressible in the Mott phase. and no gap in the excitation spectrum opens up as the transition is reached. Vef f = t (2n + 1) + (2n + 1)2 − 1 .8 Conclusions In summary. However. Nevertheless. this is consistent with the calculations done in Ref. For a one dimensional lattice in the thermodynamic limit the improved Popov approximation has the infrared divergency problem. the only regime where mean ﬁeld theories are expected to be valid. (5. In the improved Popov approximation the proportionality constant is 1/α2 . A speciﬁc problem is the fact that the improved Popov approximation does not distinguish between commensurate and incommensurate ﬁllings. in the mean ﬁeld theory starting from the Mott phase it is 8t (remember t is the number of nearest neighbors). in this chapter we have developed quadratic approximations for describing the approach of a superﬂuid system towards the Mott insulator transition. we see that in the high density limit. [44] where the authors showed using renormalization group techniques that the upper critical dimension for the transition at constant integer density is dc = 2.
100 Chapter 5 Quadratic approximations .
127. 123]. 125] and condensed matter systems [124] as well as addressing the issues of thermalization and quantum phase transitions [126. the nexttoleading . In this chapter extend our analysis to non equilibrium systems. 115. In this chapter we consider under the 2PICTP scheme diﬀerent approximations: the HartreeFockBogoliubov (HFB) approximation. BEC [54.Chapter 6 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism In the last section we focused our attention on quadratic approximations of the many body Hamiltonian and we used them to describe equilibrium properties of ultra cold atoms loaded in optical lattices. 118.[58][62]. To date most theoretical descriptions of nonequilibrium dynamics of BEC’s have been based on the time dependent GrossPitaevskii equation. in this chapter we adopt a closed time path (CTP) [56] functionalintegral formalism together with a twoparticle irreducible (2PI) [57] eﬀective action approach. collapses and revivals of the coherent matter ﬁeld [47] and the NIST patterned loading experiment described in chapter 4. experiments such as those mentioned above have been able to achieve regimes where the standard mean ﬁeld description is inapplicable. However. particles and ﬁelds [122. We retain terms of up to secondorder in the interaction strength when solving these equations. 128]. To treat farfromequilibrium dynamics. motivated by recent experimental achievements such as the colliding and collapsing condensates [20. so new methods are required. 114]. This method has been generalized for and applied to the establishment of a quantum kinetic ﬁeld theory [117.[116]). coupled with extended kinetic theories that describe excitations in dilute weakly interacting systems close to thermal equilibrium ([15]. 119] with applications to problems in gravitation and cosmology [120. 21. 121]. The description of the evolution of condensates far from equilibrium has gained considerable importance in matterwave physics.
By requiring the generalized (master) CTP eﬀective action [118] to be stationary with respect to variations of the correlation functions we obtain an inﬁnite set of coupled (SchwingerDyson) equations for the correlation functions which is a quantum analog of the BBGKY hierarchy. serves this purpose.Chapter 6 102 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism order 1/N expansion of the 2PI eﬀective action up to secondorder in the interaction strength and a secondorder perturbative expansion in the interaction strength.1) . The two particle irreducible (2PI) formalism [57] where the correlation functions appear also as independent variables. 1 (ˆ† ai+1 + a† ai ) + U ai ˆ ˆi+1 ˆ 2 ˆ H = −J i a† a† ai ai + ˆi ˆi ˆ ˆ i i Vi a† ai ˆi ˆ (6.1 2PI eﬀective action Γ(z. The CTP or SchwingerKeldysh eﬀective action formalism [56] serves this purpose.1: Twoloop (upper row) and threeloop diagrams (lower row) contributing to the eﬀective action. In this section we shall focus on the 2PI formalism. and then upgrade it to the CTP version in the next section. b) the settingsun and c) the basketball. In chapter 7 we apply the 2PICTP approximations derived here to describe the the patterned loading experiment [33] previously discussed in chapter 4. The second requirement is to describe the evolution of the correlation functions and the mean ﬁeld on an equal footing. The ﬁrst requirement for the study of nonequilibrium processes is a general initialvalue formulation depicting the dynamics of interacting quantum ﬁelds. The 2PI eﬀective action produces two such functions in this hierarchy. G) Figure 6. Our starting point is the one dimensional BoseHubbard Hamiltonian (see chapter 3). Explicitly. 6. the diagram a) is what we call the doublebubble .
ai ] = i + − dt i dt i i a∗ (t)∂t ai (t) i + ai (t)a∗ (t) i+1 (6. it is straightforward to generalize them to higher dimensions or to include additional external potentials.5) In terms of the familiar Pauli matrices. with a chemical potential smaller than the distance of the ﬁrst vibrationally excited state. Here. i i 4N (6. b = (1. (6. the parameter U denotes the strength of the onsite repulsion of two atoms on the site i.varying external potential that might be present (such as a magnetic trap). we denote the total number of atoms by N and the number of lattice sites by M . 2) deﬁned by i ai = a1 . σab = σx and hab = −σy . As usual.2) J a∗ (t)ai+1 (t) i dt i U ∗ a (t)a∗ (t)ai (t)ai (t). The classical action associated with the BoseHubbard Hamiltonian (6. and J denotes the hopping rate between adjacent sites.1 2PI eﬀective action Γ(z. The Bose Hubbard Hamiltonian should be an appropriate model when the loading process produces atoms in the lowest vibrational state of each well.which is two in this case. tunneling to nexttonearest neighbor sites can be neglected. i i i In terms of these ﬁelds the classical action takes the form 1 hab aa (t) ∂t ab (t i i 2 U (σab aa (t)ab (t))2 . the parameter Vi denotes the energy oﬀset of each lattice site due to an additional slowly. Because the nexttonearest neighbor amplitudes are typically two orders of magnitude smaller. Hereafter we consider for simplicity only a homogeneous lattice with periodic boundary conditions and no other external potential (Vi = 0).3) S[a] = + dt i dt i (6. 2) is implied and hab and σab are matrices deﬁned as hab = i 0 −1 1 0 .6. . i 2 i To simplify our notation we introduce aa (a = 1. Once the equations of motion are derived.4) Jσab aa (t)ab (t) − i+1 i where N is the number of ﬁelds. σab = 0 1 1 0 .1). G) 103 where ai and a† are the bosonic operators that annihilate and create an atom ˆ ˆi on the site i. a∗ = a2 . is given in terms of the complex ﬁelds ai and a∗ by i S[a∗ . and summation over repeated ﬁeld indices a.
9) The functional integral (6. i TC aa (t)ab (t ) − aa (t) i i i ab (t ) . the mean ﬁeld. K] such that δW [J. t )].12) . K] [57] as Z[J. t ).10) (6. (6.Chapter 6 104 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism After second quantization the ﬁelds aa are promoted to operators. K] = eiW [J. t ) Gab (t.K]/ i S[a] + Daa exp = a (6.e. To get an ij expression for the eﬀective action we ﬁrst deﬁne the functional Z[J. K] δKijab (t. i. i (6. (6. We deﬁne Γ[z.8) is a sum over classical histories of the ﬁeld aa in the i presence of the local source Jia and the non local source Kijab . ij 2 i Expressing J and K in terms of z and G yields a Jia (t)zi (t) − i Γ[z.6) (6. t ) = ij The brackets denote the expectation value with respect to the density matrix and TC denotes time ordering along a contour C in the complex plane. G] = W [J.7) Gab (t. i j dt i Jia (t)aa (t) + i 1 2 dtdt ij where we have introduced the following index lowering convention Xa = σab X b . G]. t ) a = zi (t). K] − 1 − 2 dtdt dt 1 2 dtdt ij a b zi (t)zj (t )Kijab (t. the two particle irreducible generating functional for Green’s a functions parameterized by zi (t) and the composite ﬁeld Gab (t. ij ij (6. t ).8) aa (t)ab (t )Kijab (t. Physically. t ). t )Kijab (t. The coherent state measure is included in Da. z a (t)2 is the expectation value of the composite ﬁeld by Gij i condensate population and the composite ﬁelds determine the ﬂuctuations around the mean ﬁeld: a zi (t) = aa (t) . We denote i a the expectation value of the ﬁeld operator. K] δJia (t) δW [J. G] as the double Legendre transform of W [J. All correlation functions of the quantum theory can be obtained from the eﬀective action Γ[z.11) = 1 a b [z (t)zi (t ) + Gab (t. by zi (t) and the ab (t. t ) . The addition of the twoparticle source term is what characterizes the 2PI formalism.
where T r means taking the trace. 2 In order to get an expression for Γ[z. Γ[z. (6. N the solution of the functional integrodiﬀerential equation (6. t ) −1 = ) (6. it can be written as i i exp Γ[z.1 2PI eﬀective action Γ(z.13) (6.(6. t ) δG δGij dt i . G) From this equation the following identity can be derived: δΓ[z.16) can be expressed as . G] δGab (t. ϕ] = S[z + ϕ] − δΓ[z. t ))zj (t ). we have i i δΓ[z.17) By introducing the classical inverse propagator iD−1 (z) given by δS[z] a b δzi (t)δzj (t iDijab (t.(6. G] notice that by using Eq. t ) i dti (6. G] a ϕ (t) − a δzi (t) i δΓ[z. i δGij (t. G. G] = a Daa exp + 1 2 S[a] + a dti Jia (t) [aa (t) − zi (t)] i (6. t ).12) for Γ[z. (6. G. G] a δzi (t) δΓ[z. ϕ] .18) = (δij hab ∂t + J(δi+1j + δi−1j )σab ) δ(t − t ) U c − (2zia (t)zib (t) + σab zi (t)zic (t)) δij δ(t − t ). t )zj (t ) − T rGK i j 2 δΓ[z.16) S[z.8) for W [J. t )ab (t ) − zi (t)Kijab (t. For simplicity we have denoted a Deﬁning the ﬂuctuation ﬁeld. G] δG Dϕa exp a by dti . G] − T rG = −i ln i S[z. K] and placing it in Eq.15) = a 1 a b dti dtj aa (t)Kijab (t. G].6. G] b dti dtj ϕa (t) ab ϕ (t ).14) 1 = − Kijab (t. G] b ab (t ) − zi (t ) + T rG i ab (t. G] a i a S[a] − dti [a (t) − zi (t)] Daa exp a δzi (t) i a dti dtj [aa (t) − zi (t)] i − δΓ[z. G] δΓ[z. t ) ij 105 = −Jia (t) − dt j d (Kijad (t. ϕa = aa − zi .
13) and (6. (6.20) i i N It consists of all twoparticle irreducible vacuum graphs (the diagrams representing these interactions do not become disconnected by cutting two propagator lines) in the theory with propagators set equal to G and vertices determined by the interaction terms in S[z + ϕ] . (6. (6. t) + Giida (t. G] . a δzi (t) (6. t ) = −J(Gab (t. t ) ≡ 2i δΓ2 [z. G] = S[z] + T r ln G−1 + T rD−1 (z)G + Γ2 [z.24) a dtk Σikc (t. . t )zic (t) + i ij N i U d (zid (t)zi (t))Gab (t.22) (6.19) in equations (6. t))zia (t) N U d + ((Giiad (t. G] − .23) Equation (6. This procedure leads to the following equations: Sint [z + ϕ] = − b hab ∂t zi (t) = −J(zi+1a (t) + zi−1a (t)) U d c + (zid (t)zi (t) + Gii c (t.24) once Γ2 [z.14) . G] is speciﬁed.19) 2 2 The quantity Γ2 [z.22) can be rewritten as a partial diﬀerential equation suitable for initial value problems by convolution with G. kj The evolution of z a and Gab is determined by Eqs. t ) ij (6. t )Gcb (t . t ) + Gab (t. G] is conveniently described in terms of the diagrams generated by the interaction terms in S[z + ϕ] which are of cubic and higher orders in ϕ U dti (ϕib (t)ϕb (t))2 i 4N U − dti ϕa (t)zia (t)ϕb (t)ϕib (t). the dynamical equations of motion for the mean ﬁeld and the propagators are found by using the expression (6.22) and (6. t )) + c i+1j i−1j 2U a z (t)Gcb (t. δGab (t. and setting the right hand side equal to zero. t ) + iδ ab δij δ(t − t ). This diﬀerential equation reads explicitly cb ha ∂t Gij (t.21) and G−1 (t.Chapter 6 106 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism i i Γ[z. ijab Σijab (t. t))zi (t) N δΓ2 [z. t ) + ij N (6. t ). Since physical processes correspond to vanishing sources J and K. t )−1 − Σijab (t. t ) = Dijab (t. G] + const.
Fig. 6. we recover the DNLSE. The HFB approximation neglects multiple scattering. Timedependent HartreeFockBogoliubov (HFB) approximation: A truncation of Γ2 retaining only the ﬁrst order diagram in U. The unequal treatment of the condensate and ﬂuctuations present in this approximation introduces limitations and makes this approach not very attractive for studying the non equilibrium dynamics. in which the system remains a pure condensate. This gives us the usual meanﬁeld description.. but conserves energy and particle number [132. 15. Various approximations for the full 2PI eﬀective action can be obtained by truncating the diagrammatic expansion for Γ2 . 6. It can be . 3. Meanﬁeld approximation: If. The oneloop approximation has an equation for the ﬂuctuations identical to the BdG equations (See chapter 5). Oneloop Approximation: The next approximation consists of discarding Γ2 altogether. 133]. (6.e. i.2 Perturbative expansion of Γ2 (z. This yields the socalled oneloop approximation.1.2 Perturbative expansion of Γ2 (z. G) and approximation schemes The diagrammatic expansion of Γ2 is illustrated in Fig. The presence of this terms is necessary for the time dependent evolution because they guarantee the conservation of particle number and energy. we discard all terms to the right of S[z]. The dots where four lines meet represent interaction vertices.19).1 The standard approaches 1. It is important to point out that between all the quadratic approximations discussed in chapter 5 only the HFB approximation is suitable for describing non equilibrium dynamics because it is the only one that obeys conservation laws. The expression corresponding to each vacuum diagram should be multiplied by a factor (−i)l (i)s−2 where l is the number of solid lines and s the number of loops the diagram contains. Which approximation is most appropriate depends on the physical problem under consideration. This approximation violates Goldstone’s theorem. The action Γ including the full diagrammatic series for Γ2 gives the full dynamics. G) and approximation schemes 107 6. It is of course not feasible to obtain an exact expression for Γ2 in a closed form. 2. 6.1.6. The HFB equations can also be obtained by using cumulant expansions up to the secondorder [136] in which all cumulants containing three or four ﬁeld operators are neglected. where two and threeloop vacuum diagrams are shown. however the equation of motion for the condensate does include the depletion and anomalous terms. in Eq. yields equations of motion for z and G which correspond to the time dependent HartreeFockBogoliubov (HFB) approximation. keeping only the doublebubble diagram.2.
the bare potential in the diagonal terms is also upgraded to the many body scattering matrix up to secondorder in the ladder expansion.4. the Tmatrix can be approximated by a constant proportional to the scattering length (the pseudopotential approximation). However this approximation is only valid in the weak interaction limit and neglects all momentum dependencies which appears in the problem as higher order terms. the ladder diagrams give the largest contribution to the fourpoint vertex. Though strict dissipation can never be observed in an energy conserving closed system. but is a necessary price to pay for a fuller account of the quantum dynamics. In chapter 8 we will also show that by taking into account the twoloop contribution of the 2PI eﬀective action. The twoloop contribution includes diagrams topologically identical to those found by Beliaev. • Nonlocal Dissipation and NonMarkovian Dynamics Higher order terms lead to nonlocal equations and dissipation. 54]). In the dilute gas limit. in chapter 5 we explicitly showed how by including the anomalous average in the equations the bare potential was upgraded in the oﬀ diagonal elements to the many body scattering matrix. thus accounting for the repeated scattering of the bosons. It will be described in more detail in section 6. Many well acknowledged approaches to the quantum .2 Higher order expansions We make a few remarks on the general properties of higher order expansions and then specialize to two approximations. The presence of nonlocal terms in the equations of motion is a consequence of the fact that the 2PI eﬀective action really corresponds to a further approximation of the master eﬀective equation [118]. This makes numerical solution diﬃcult. In that sense the 2PI eﬀective action approach allows us to go beyond the weakly interacting limit in a systematic way and to treat collisions more accurately.Chapter 6 108 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism interpreted as an expansion in terms of U t/J. (where t is the time of evolution) and is good for the description of short time dynamics or weak interaction strengths. To lowest order in the diluteness parameter.2. In particular. NonMarkovian dynamics is a generic feature of the 2PI formalism which yields integrodiﬀerential equations of motion. 6. where the interparticle distance is large compared with the swave scattering length. • 2PI and Ladder Diagrams Since the work of Beliaev [63] and Popov [52] it is well known in the literature that including higher order terms in a diagrammatic expansion corresponds to renormalizing the bare interaction potential to the fourpoint vertex (see for example Ref. characteristic features of dissipative systems like exponential damping of correlations can be exhibited once interactions are properly taken into account. but with the exact propagator instead of the oneloop propagator. [53.
Now. 2.2 Perturbative expansion of Γ2 (z. The Markovian approximation assumes that only the current conﬁguration of the system. determines its future evolution. In the low kinetic energy. from a physical point of view the zero mode exists and is quantum in nature. in which the lattice which conﬁnes the atoms to the bottom of the wells with enhanced interaction eﬀects. However.diﬀerential equations which depend on the time history of the system. which one can use to justify the Markovian approximation.6. 119].nonMarkovian dynamics needs to be confronted squarely in systems such as the patterned loaded lattice. In this sense the method is a controlled expansion in a small parameter. There . weakly interacting regime the time between collisions (or mean free path) is long compared to the reaction time (or scattering length): τc >> τ0 . or in the kinetic theory context that the time scales between the duration of binary collisions τ0 and the inverse collision rate τc are wellseparated. That is the rationale for our adoption of the CTP 2PI scheme. but not its history. the settingsun and the basketball (see Fig. which corresponds to an expansion around the vacuum. Secondorder terms lead to integro.1). In the large N approach the ﬁeld is modelled by N ﬁelds and the quantum ﬁeld generating functional is expanded in powers of 1/N . The long separation between collisions and the presence of intermediate weak ﬂuctuations. By including the settingsun and the basketball in the approximations we are taking into account two particle scattering processes [121. In the linearized theory. G) and approximation schemes 109 kinetics of such systems adopt either explicitly or implicitly a Markovian approximation [62]. accompanied by the low dimensionality of the system and farfromequilibrium initial conditions. but unlike perturbation theory in the coupling constant.6. Markovian approximations are made if one assumes instantaneous interactions. 6. LargeN approximation The 1/N expansion is a controlled nonperturbative approximation scheme which can be used to study nonequilibrium quantum ﬁeld dynamics in the regime of strong interactions[127. for the speciﬁcs: 1. Secondorder expansion: A truncation retaining diagrams of secondorder in U consists of the doublebubble. However. 126]. the large N expansion corresponds to an expansion of the theory about a strong quasiclassical ﬁeld. allow for a rapid decay of the temporal and spatial correlations created between collision partners.3 Zero mode ﬂuctuactions In chapter 5 we neglected zero mode ﬂuctuations due to the non perturbative character of the zero mode. this approach introduces an artiﬁcial infrared divergence in low dimensional models thus the theory is actually improved if the contribution from this mode is neglected all together [96].2.
t ) = θctp (t. because the 2PI formalism goes beyond the linearized approximation. to get the generating functional we seek. the standard path integral representation for each one can be introduced. the ﬁelds and integration contours suggest introducing a 2 x 2 matrix notation. This way is cleaner notationally and has the advantage that all the functional formalism of the previous section may be taken with the only diﬀerence of path ordering according to the complex time contour Cctp in the time integrations. the zero mode does not have the impact it has in the linearized formalism and it is not clear that subtracting it necessarily leads to a better approximation. 6. the initial state that we are going to assume for our analysis is a coherent state rather than a proper state of the total particle number as will be described in chapter 7.Chapter 6 110 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism are both fundamental and practical reasons why isolating and subtracting the zero mode is not as compelling in the dynamical evolution we want to describe in this chapter. 119]. otherwise all the dependence on the source drops out.22) and (6. The basic idea of the CTP formalism relies on the fact that a diagonal matrix element of a system at a given time. The twopoint functions are decomposed as Gab (t. [126] and [127] and introduce the CTP formalism in our equation of motion by using the composition rule for transition amplitudes along the time contour in the complex plane. Moreover. However. However we are going to follow Refs. ij ij ij where (6. (6. in this chapter we shall not attempt to isolate the contributions from the zero mode. Discarding these ﬂuctuations would spoil the integrity of the formalism. t)Gab< (t. Therefore. A full nonperturbative treatment in the future is certainly desirable.25) . and nonnegligible. Since each term in the product is a transition matrix element of the usual or time reversed kind. Secondly. The doubling of sources. This notation has been discussed extensively in the literature [117. t )Gab> (t. t ). Firstly. t = 0. t ) + θctp (t .3 CTP formalism In order to describe the nonequilibrium dynamics we will now specify the contour of integration in Eqs. we have to require that the forward time evolution takes place in the presence of a source J + but the reversed time evolution takes place in the presence of a diﬀerent source J − . can be expressed as a product of transition matrix elements from t = 0 to t and the timereverse (complex conjugate) matrix element from t to 0 by inserting a complete set of states into this matrix element at the later time t . The SchwingerKeldysh formalism is a powerful method for deriving real and causal evolution equations for the expectation values of quantum operators for nonequilibrium ﬁelds. quantum ﬂuctuations in the total particle number are real. as the total particle number is not very high.24) to be the SchwingerKeldysh contour [56] along the realtime axis or closed time path (CTP) contour.
t )Σ< (t. t ) + θ(t . t )Σ> (t.26) (6. t ) = 0 t dt (Σ> (t. we have to multiply by a negative sign to take into account the opposite direction of integration.30) + t ∞ − 0 dt Σ< (t. t )Σ (t. t )G< (t . t )G(t . t ) dt Σ> (t. t )(G> (t . t )Σ< (t. where we have omitted the indices because they are not relevant for the discussion. t ) = > > > < (6. t )G> (t . t) for t and θctp (t. t)θctp (t . t )θctp (t .29) θctp (t.6. t )G> (t . t ))G> (t . t )Σ (t. t ) = ij ϕa (t)ϕb (t ) . If t > t . When integrating over the second half C − . t )G> (t .27) with ϕi being the ﬂuctuation ﬁeld deﬁned prior Eq. t )Σ< (t. t ) for t and θ(t . t )G (t . i j (6. t )Σ> (t. t ) = − 1 for t on C 0 for t on C + t both on t both on and t on and t on C+ C− . Using the rule for CTP contour integration we get t dt Σ(t. t )G> (t .24).28) With these deﬁnitions the matrix indices are not required. t ) dt θ(t . C+ C− (6. t ). (6. t )θctp (t . t ) = 0 dt ∞ θ(t . (17) and θctp (t − t ) being the CTP complex contour ordered theta function deﬁned by θ(t. t )Σ< (t.31) − 0 . (6. t ) − Σ< (t. we have t dt Σ(t. t )G(t . t )G (t . t ) θctp (t . t )). t ) + θ(t . t ) − G< (t . t ). t ) (6.3 CTP formalism 111 Gab> (t. i j ϕb (t )ϕa (t) . t )G< (t . To show explicitly that the prescription for the CTP integration explained above does lead to a wellposed initial value problem with causal equations. t )G(t . The integrand has the CTP ordered form Σ(t. t ) + θctp (t . t)θctp (t . t )G< (t . t ) = ij Gab< (t. t ) + θctp (t. let us explicitly consider the integral in Eq.
It is convenient to express the evolution equations in terms of two independent twopoint functions which can be associated to the expectation values of the commutator and the anticommutator of the ﬁelds. t ) − G< (t . i i (t. 2 11(ρ) (ρ) Gij (t. [127] the functions (F )ab (ρ)ab Gij (t. t ))G< (t . t ) + dtk Σik (t.32) − 0 The above equations are explicitly causal.35) (F ) b − 0 dtk Σik (ρ)ab (F )ac (t. t ) − Σ< (t. t )(G> (t . t )zic (t) + N t t 0 U N c zic (t)zi (t))Gij (ρ)ac (F )ab (t.(6. t ) + Gi−1j (t. ρ. We deﬁne. m. t )).40) Gij Gij (t. t ) = i ϕi (t)ϕj (t ) − ϕj (t )ϕi (t) . i i 2 (6. we deﬁne the normal. t ) = (F ) 1 ϕi (t)ϕj (t ) + ϕj (t )ϕi (t) . t ) + 2U a (F )cb zi (t)Gij (t. t ) = 0 t dt (Σ> (t.34) Gij (t. With these deﬁnitions Eq. t ) = 1 Gab> (t. t )Gkjc (t . t ) + Gab< (t. (6. t ) + Gi−1j (t. t ) ≡ ρij (t.38) (6. following Ref. t ) = i ϕ† (t)ϕj (t ) − ϕj (t )ϕ† (t) . spectral and statistical functions as 21(F ) 21(ρ) (F ) Gij (t.Chapter 6 112 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism On the other hand. t ). t + dtk Σik (t. t ) = (ρ) 1 ϕ† (t)ϕj (t ) + ϕj (t )ϕ† (t) . t )G(t . t )zic (t) + N t t U N c zic (t)zi (t))Gij (ρ)ac (ρ)ab (t. if t < t t dt Σ(t.39) (6. and anomalous. t ) = −J Gi+1j (t. t ) ≡ mij (t.24) can be rewritten as: (F )cb (F )ab (F )ab ha ∂t Gij c (t. t ) − Gab< (t. t )Gkjc (t . ij ij where the (F ) functions are usually called statistical propagators and the (ρ) are called spectral functions [137]. t ) ≡ mij (t. t ). t )Gkj c (t . t ) = i Gab> (t.33) (6. t ) dt Σ< (t. t ) ij ij 2 (6. 11(F ) . t ) ≡ ρij (t.(6. t ) . t ) (6. (ρ)ab (ρ)b ha ∂t Gij c (ρ)cb (t. t ) = −J Gi+1j (t.37) (6. t ) + 2U a (ρ)cb zi (t)Gij (t.36) (ρ)b In particular.
these kinds of damping 1 found in the HFB approximation (due to phase mixing. 6. (6.42) U N d d a zid (t)zi (t) + Gii d (t. the latter with theoretical meaning. Using the ﬁrst order expression for Γ2 in Eqs. . in the Boltzmann sense.41) a b Giia (t.1 HFB approximation Equations of motion The HFB equations include the leading order contribution of Γ2 .24) yields the following equations of motion. ii N (6. 134].4 HFB approximation 113 With these relations in place. Multiple scattering processes are neglected in this approximation. They describe the coupled dynamics of condensate and noncondensate atoms which arise from the most important scattering processes which are direct. the former referring simply to the phenomenological decay of some function. as in the Vlasov equation citeBalescu) are diﬀerent in nature from the collisional dissipation (as in the Boltzmann equation) responsible for thermalization processes. t)Gab (t.43) we make a distinction between the meaning of the words ‘damping’ and ‘dissipation’. b z a a ζHF B ≡ −J zi+1 (t) + zi−1 (t) + (6.22) and (6.4. t) + 2Giiab (t.. However. we now proceed to derive the time evolution equations for the mean ﬁeld and the twopoint functions from the 2PICTP eﬀective action for the BoseHubbard model under the three approximations described before. t) . e. t) . exchange and pair excitations. We expect the HFB equations to give a good description of the dynamics in the collisionless regime when interparticle collisions play a minor role. b z ha ∂t zi (t) = ζHF B .6. ii the factor of two arises because the direct and exchange terms are identical. The leading order contribution of Γ2 is represented by the doublebubble diagram. The basic damping mechanisms present in the HFB approximation are Landau and Beliaev damping associated with the decay of an elementary excitation into a pair of excitations in the presence of condensate atoms[133. t) zi (t) + 1 2U zib (t)Gab (t.4 6.g. t)Giib (t. Its contribution to Γ2 is z independent and has an analytic expression of the form Γ2 [G] = U − dti 4N (1) (6.
t ).46) (6.40).49) with Lij (t) = −J(δi+1j + δi−1j ) + 2U δij zi (t)2 + ρii (t. t))Gab (t.2 Mode expansion of the HFB equations To decouple the HFB equations we apply the well known Bogoliubov transformation to the ﬂuctuation ﬁeld (see chapter 5) ϕj (t) = q ∗q uq (t)ˆ q − vi (t)ˆ q . 6. Because they are local in time they can be decoupled by a mode decomposition. and setting N = 2. The conservation properties of the HFB equations can also be understood by the fact that these equations can also be derived using Gaussian variational methods [15].50) (6. Eqs. t ) ij N + 2U a a (z (t)zic (t)Gcb (t. t ) i ∂t ij = Lik (t)ρkj (t. t ) ∂t ij ∂ (ρ) −i ρ (t. t ) + Gab (t. t ).4. t ) + M∗ (t)mkj (t. ik = Lik (t)ρkj (t. t)Gdb (t. = Lik (t)mkj (t.47) (6. t ). Mij (t) = U δij zi (t)2 + mii (t. t) . t))zi (t) ∗ +U mii (t. t ) + Mik (t)ρkj (t. These methods always yield a classical Hamiltonian dynamics which guarantees probability conservation. (6. t ) ∂t ij ∂ (F ) i m (t. t )) + i+1j i−1j (6. t )) + iδ ab δij δC (t − t ). (F ) (F ) (6.Chapter 6 114 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism G ha ∂t Gcb (t. t ). ij ij N i In terms of the spectral and statistical functions. t ) + Mik (t)ρkj (t. t) . (F ) (F ) (6.48) (6. t ) = ζHF B . the above equations take the form i ∂t zi (t) = −J(zi+1 (t) + zi−1 (t)) + U (zi (t)2 + 2ρii (t.44) U d d (zid (t)zi (t) + Gii d (t. It can be checked that they preserve important conservation laws such as the number of particles and energy.52) .45) (F ) −i ∂ (F ) ρ (t.37) to (6.51) The time dependent HFB equations are a closed set of selfconsistent equations that describe the coupled dynamics of the condensate and noncondensate components of a Bose gas. ik = Lik (t)mkj (t. t ) + Gii d (t. c ij G ζHF B ≡ −J(Gab (t. t ) + M∗ (t)mkj (t. α α† i (6. t ) ∂t ij ∂ (ρ) m (t. t)zi (t). (ρ) (ρ) (F ) (F ) (ρ) (ρ) (F ) (6.
4 HFB approximation 115 where (ˆ q .59) (6.53) and (6. t ) = − (ρ) (F ) 1 2 ∗q ∗q uq (t)vj (t ) + uq (t )vi (t) . i j q ∗q ∗q uq (t)vj (t ) − uq (t )vi (t) .60) mii (t.58) ρij (t. t) satisfy 1 2 (F ) ρii (t.49) and using the constraints (6.54) In the zero temperature limit. t ) = i mij (t. αq αk = 0. vi (t)}. t) (F ) and mii (t. αq ) are time independent creation and annihilation quasiparticle operα ˆ† ∗q ators and all the time dependence is absorbed in the amplitudes {uq (t). t)zi (t).53) (6. i .55) (6.56) (6. where the quasiparticle occupation number vanishes.54) we recover the standard time dependent equations for the quasiparticle amplitudes ∂ zi (t) (6. j i q q ∗q vi (t)vj (t ) − uq (t )u∗q (t) . the amplitudes must fulﬁll the relations q ∗k uq (t)u∗k (t) − vi (t)vi (t) i i i q k uq (t)vi (t) − vi (t)uk (t) i i i = δqk . (6. i To ensure that the quasiparticle transformation is canonical.57) (6.60) into Eqs.54). i Replacing Eqs. the statistical and spectral functions take the form ˆ† ˆ 1 2 ρij (t. (6. (6. t))zi (t) + U mii (t.(6.45)(6. t) = q (F ) q vi (t)2 + (6.59)(6.53) (6. t ) = −i Notice that at equal time and position due to Eqs. i j q mij (t. t ) = (ρ) (F ) q ∗q vi (t)vj (t ) + uq (t )u∗q (t) .61) ∂ (F ) (F ) ∗ = −J(zi+1 (t) + zi−1 (t)) + U (zi (t)2 + 2ρii (t. j i q (6. ρii (t. t) = − q (F ) ∗q uq (t)vi (t).6. = 0.
(5. we notice that the equations are identical as they should be if we replace q ∂ ∂ q i ∂t uq → ω HF B uq . t) + zi (t)∗2 )uq (t). t )Gbd (t. i i If we compare Eqs. t )Gdd (t.87).Chapter 6 116 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism ∂ q u (t) (6. t )Gijdd (t.1b and the basketball Fig. The settingsun diagram depends on z and contains only threepoint vertices. Once the time dependent quasiparticle amplitudes are calculated we can derive the dynamics of physical observables constructed from them as a function of time. (5.87). t )Gdd (t. The factor of 1/2 i i diﬀerence between ρii and ρii (see Eq.5 6. They can be solved using standard time propagation algorithms. t )Gijdd (t. (6. which were neglected in Eqs. t )Gijdd (t. t) + zi (t)2 )vi (t).61) (6.19)) just leads to a trivial renormalization of the chemical potential. t )+ ij ij 2Gijbb (t. (F ) 6. etc. t ) + ij ij ij ij i U 2N 2 dti dtj Gijbb (t. t))vi (t) − U (mii (t. Hence ± we do not need the cHF B variables. t ) . where M is the total number of lattice sites.64) dti dtj zib (t)zjb (t ) × Gbb (t. G] = i U N 2 (2) (6. t )Gbb (t. Eqs. t )Gdb (t. The basketball diagram is independent of the meanﬁeld and is constructed with only quartic vertices.27) to (5.1 Secondorder expansion Equations of motion Full secondorder The secondorder contribution to Γ2 is described in terms of the settingsun Fig. 6. which simply keep the orthogonality between s of the condensate and the excitations. (5.62) ∂t i (F ) (F ) q = −J(uq (t) + uq (t)) + 2U (zi (t)2 + ρii (t. The secondorder (2) Γ2 eﬀective action can be written as Γ2 [z. 6. such as the average number of particles in a well ni (t). (6.63) with their time independent version. t))uq (t) − U (mii (t. t )Gdb (t. Eqs.61)(6.27) to (5. i+1 i−1 i ∂ q −i v (t) = (6. ij ij .5. t ) + 2Gbd (t. i ∂t vi → ω HF B vi and i ∂t zi → µHF B zi .1c diagrams.63) ∂t i (F ) ∗(F ) q q q −J(vi+1 (t) + vi−1 (t))) + 2U (zi (t)2 + ρii (t.63) correspond to a set of M (2M + 1) coupled ordinary diﬀerential equations. t )Gijdd (t. In the time dependent equations we have included the zero mode ﬂuctuations.
68) (6. tj ) + fij (t. t With the above deﬁnitions we ﬁnd from Eqs. t )Gijac (t.66) (6. t ). (F ) (F ) (F ) (6. Λij a (t. tj )gij (ti . t ij ) = −(zid (t)zjb (t ) + Gijdb (t. t )zic (t)zkd (t ) Gdb (t .71) 2 i 2U N 2U N 2U N dtj zjb (t ) Πij (t. t ). t ) = −D(t. ji ij (6. t )gij (ti . t )Gcb (t . t ) + Λik d (t. t ) = −Gij (t. t )Gac (t.70) )Gba ij (t.45). t ).ρ) and we introduce the functions Ωij [f . t ) = zib (t)zja (t b (6. (6. the equations of motion derived under the full secondorder approximation can be written as: . t )Gba (t . D(t. ij 2 Ξijab (t.43) and (6.73) (6. g] = fij (ti . (6. t )Gdb (t .22) and (6. t ij )+ Ξac (t. t )Gab (t.5 Secondorder expansion To simplify the notation. (6. t ) = ζHF B + c ij 2 i +i a zi (t) 2 dtk zkc (t ) Πik (t. t ) kj kj a Θik d (t. let us introduce the following deﬁnitions [127]: 117 1 Πij (t.ρ) To get explicit expressions of the equations of motion in terms of ρ(F. g] = fij (ti .72) b G ha ∂t Gcb (t. tj )gij (ti . t ) = −Gcb (t. tj )gij (ti . t ij )Gdc (t.67) (6. t) + Λij c (t. t ). Θac (t. t ) − Πij (t.65) (6. 4 (ρ) (F ) (ρ) (ρ) (F ) Ωij [f . t )Gijca (t.74) Using the spectral and statistical functions and setting the number of ﬁelds N equal to 2.69) ))Gab (t. t ) = − Gijab (t. tj ) . t ). t )Gijab (t.6. t ij ). m(F. t ) . tj ). t ) + Λca (t.24) the following equations of motion: b z ha ∂t zi (t) = ζHF B + b (6. t ). tj ) − 1 (ρ) (ρ) fij (ti . ij b bc Λij a (t. ik kj c dt z G where ζHF B and ζHF B are deﬁned in Eqs.
ρ] + 2zi zk Ωik [ρ. ρ∗ ] + Ωik [m. . m∗ ] (ρ) (ρ) ρkj (ρ) ∗ ∗ dtk 2zi zk Ωik [m∗ . m∗ ] (F ) (F ) (F ) (F ) (F ) (ρ) (ρ) ρkj (F ) ∗ ∗ dtk 2zi zk Ωik [m∗ . m∗ ] + zk Ωik [m. m∗ ] (F ) (F ) (F ) (F ) (ρ) mkj . m∗ ] (F ) (F ) (F ) (ρ) (ρ) (ρ) (F ) (F ) mkj ∗ ∗ ∗ dtk zi zk Ωik [ρ. m∗ ] + zk Ωik [m∗ . ρ∗ ] mki .75) (F ) −2U 2 k 0 ti ∗ dtk zk Ωik [ρ. ρ] ρkj ∗ dtk Ωik [ρ. m∗ ] mkj ∗ ∗ dtk Ωik [m∗ . ρ] ρkj (ρ) (ρ) (ρ) (F ) (ρ) −2U 2 −2U 2 −2U 2 +2U 2 +2U 2 +2U 2 +2U 2 ∗ dtk Ωik [ρ. ρ∗ ] mki (F ) (F ) (F ) (ρ) (ρ) (ρ) (ρ) (ρ) −2U 2 k 0 ti (F ) +2U 2 k 0 ti ∗ dtk zk Ωik [ρ. m∗ ] + zk Ωik [m∗ . ρ∗ ] + Ωik [m. ρ∗ ] + zk Ωik [m. ρ∗ ] + zk Ωik [m. m∗ ] + zk Ωik [m. Υ] + 2zi zk Ωik [ρ. (ρ) −i ∂ti ρij = (F ) −J(ρi+1j (ti . ρ] + zi zk Ωik [m∗ . ρ∗ ] + zk Ωik [m. ρ∗ ] + zk Ωik [m. ρ] + 2zi zk Ωik [ρ. m∗ ] + 2zi zk Ωik [m.76) + (F ) ρi−1j (ti . ∆] + 2zi zk Ωik [ρ. ρ∗ ] + Ωik [m. m∗ ] + 2zi zk Ωik [ρ∗ . tj )) (ρ) + 2U (zi 2 + (F ) (F ) ρii )ρij + ∗(F ) U (mii (ρ) + (F ) ∗2 zi )mij (F ) −2U ∗ ∗ ∗ dtk zi zk Ωik [ρ. m∗ ] + 2zi zk Ωik [m. ∆] + 2zi zk Ωik [ρ. ρ∗ ] + 2Ωik [m. Υ] + zi zk 2Ωik [ρ. ρ] + zi zk Ωik [m∗ . ρ] ρki (F ) (F ) (F ) (ρ) +2U 2 k 0 ti ∗ ∗ zk Ωik [ρ. ρ] ρki ∗ ∗ zk Ωik [ρ. m∗ ] mkj ∗ ∗ dtk Ωik [m∗ . tj ) ti 2 k 0 ti k 0 ti k 0 ti k 0 tj k 0 tj k 0 tj k 0 tj k 0 (F ) (6.Chapter 6 118 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism ∗ i ∂ti zi = −J(zi+1 (ti ) + zi−1 (ti )) + U (zi 2 + 2ρii )zi + U mii zi (F ) (ρ) (F ) (6. m∗ ] + 2zi zk Ωik [ρ∗ .
m] ρkj (ρ) +2U 2 k 0 tj dtk Ωik [m. m] + 2zi zk Ωik [ρ. m∗ ] (F ) (F ) (F ) (ρ) (ρ) +2U 2 k 0 tj mkj . m∗ ] (F ) (F ) (F ) (ρ) (ρ) (ρ) (F ) (F ) −2U 2 k 0 ti mkj +2U 2 k 0 tj ∗ ∗ ∗ ∗ dtk 2zi zk Ωik [m. ρ∗ ] + Ωik [m. ρ∗ ] + Ωik [m. ∆] + 2zi zk Ωik [ρ. ρ] + zi zk Ωik [m∗ . ρ∗ ] + 2zi zk Ωik [ρ∗ . m∗ ] + 2zi zk Ωik [ρ∗ .77) ∗ ∗ ∗ dtk zi zk Ωik [ρ. ∆] + 2zi zk Ωik [ρ. Υ] + 2zi zk Ωik [ρ. Υ] + 2zi zk Ωik [ρ. tj )) + 2U (zi 2 + ρii )ρij + U (mii −2U 2 k ti tj ti k tj ti k tj ti k tj (ρ) (ρ) (ρ) (ρ) (F ) (ρ) ∗(F ) (ρ) ∗2 + zi )mij (ρ) (ρ) (6. ρ∗ ] mkj ∗ dtk Ωik [ρ∗ . m∗ ] (ρ) (ρ) (ρ) (ρ) (ρ) (ρ) (ρ) −2U 2 −2U 2 −2U 2 ρkj (ρ) ∗ ∗ dtk 2zi zk Ωik [m∗ .5 Secondorder expansion 119 −i ∂ti ρij = −J(ρi+1j (ti . i ∂ti mij = (F ) (6. m] + 2zi zk Ωik [m∗ . m] + 2zi zk Ωik [m∗ . m∗ ] mkj ∗ ∗ dtk Ωik [m∗ . m∗ ] (ρ) (ρ) (ρ) (ρ) (ρ) (ρ) ρkj (F ) −2U 2 k 0 ti ∗ ∗ ∗ dtk zi zk Ωik [ρ∗ . m] ρkj 0 k −2U 2 k 0 ti dtk Ωik [m. ∆] + 2zi zk Ωik [ρ. m] + 2zi zk Ωik [ρ. m∗ ] (ρ) (ρ) (ρ) (ρ) (ρ) mkj . .6. ρ∗ ] + 2zi zk Ωik [ρ∗ . ρ∗ ] + Ωik [m. m∗ ] + 2zi zk Ωik [m. ρ∗ ] + Ωik [m.78) (F ) (F ) (F ) (F ) (F ) 2 (F ) −J(mi+1j (ti . ρ] ρkj ∗ dtk Ωik [ρ. ρ∗ ] + Ωik [m. ρ∗ ] + zi zk Ωik [m. ρ∗ ] + Ωik [m. tj ) + mi−1j (ti . m∗ ] (F ) (F ) (F ) (F ) (F ) (F ) ρkj (ρ) +2U 2 k 0 tj ∗ ∗ ∗ dtk zi zk Ωik [ρ∗ . Υ] + 2zi zk Ωik [ρ. ρ] + 2zi zk Ωik [ρ. ρ∗ ] mkj ∗ dtk Ωik [ρ∗ . tj ) + ρi−1j (ti . tj )) + 2U (zi 2 + ρii )mij + U (mii + zi )ρij ti (ρ) (F ) ∗ ∗ ∗ (ρ) ∗ (ρ) −2U 2 dtk 2zi zk Ωik [m. ρ∗ ] + zi zk Ωik [m.
ρ) Υij (F. At the next to leading order level. ρ∗ ] + 2zi zk Ωik [ρ∗ . the next to leading order (NLO) contributions are of order 1 and so on. m∗ ] (ρ) (ρ) (ρ) (ρ) (ρ) mkj .ρ) 2Ωij [ρ. At leading order only the ﬁrst term of Eq. ∆] + 2zi zk Ωik [ρ.ρ) [ρ. (6. m∗ ] . tj ) respectively for the two point functions. It can be shown that all graphs contained in an O(N ) expansion can be built from the irreducible invariants[127]: z 2 . t )Gijdd (t. ρ∗ ] + 2Ωik [m. ρ∗ ] + 2Ωij + (F. These notation is also going to be used in the equations for the 1/N expansion. t ) + ij ij (6. m∗ ] (ρ) (ρ) (ρ) (ρ) (ρ) (ρ) ρkj (ρ) +2U 2 +2U 2 with ∗ ∗ ∗ dtk zi zk Ωik [ρ∗ . ρ∗ ] + Ωik [m. T r(Gn ) and T r(zzGn ) . The leading order large N approximation scales proportional to N .ρ) Ωij [m.Chapter 6 120 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism i ∂ti mij = (ρ) (6.ρ) = Ωij = (F. U N 2 Γ2 i (2)1/N [z. (6. m] ρkj +2U 2 tj k +2U 2 k ti tj ti k tj ti k tj dtk Ωik [m. with n < N . In the above equations we adopted the notation zk meaning zk (tk ) for the condensate and mkj and ρkj meaning mkj (tk . ρ∗ ] (F.81) (F. (6. the doublebubble is totally included but only certain parts of the settingsun and basketball diagrams are: the ﬁrst term in both of the integrals of Eq. t )Gbb (t. ij ij . G] = i 2 dti dtj zib (t)zjb (t ) Gbb (t.42) contributes. m∗ ] . tj )) + 2U (zi 2 + ρii )mij + U (mii + zi )ρij ti (ρ) (ρ) ∗ ∗ ∗ (ρ) ∗ (ρ) dtk 2zi zk Ωik [m. m] + 2zi zk Ωik [m∗ . ∆ij (F. t ) . t )Gdd (t. if we truncate up to secondorder in the coupling strength.80) (6. ρ∗ ] mkj ∗ dtk Ωik [ρ∗ . The factors of N in a single graph contributing to the same 1/N expansion then have two origins: a factor of N from each irreducible invariant and a factor of 1/N from each vertex. tj ) and ρkj (tk . t )Gdd (t. 2PI1/N expansion The 2PI eﬀective action is a singlet under O(N ) rotations. m] + 2zi zk Ωik [ρ. Υ] + zi zk 2Ωik [ρ. t )Gijdd (t.79) (ρ) (ρ) (F ) (ρ) (F ) 2 (ρ) −J(mi+1j (ti .82) U 2N dti dtj Gijbb (t. tj ) + mi−1j (ti . ρ∗ ] + zi zk Ωik [m.ρ) [m.64).
m∗ ] (F ) (F ) (F ) (ρ) (ρ) (ρ) (F ) (F ) −U 2 k 0 ti mkj +U 2 k 0 tj ∗ ∗ ∗ dtk zi zk Ωik [ρ. m∗ ] (F ) (F ) (F ) (F ) (F ) (F ) (ρ) (ρ) +U 2 k 0 tj ρkj +U 2 k 0 tj ∗ ∗ dtk zi zk Ωik [m∗ . Π] + zi zk 2Ωik [ρ. and Θ = Ξ. m∗ ] mkj ∗ ∗ dtk Ωik [m∗ .6. m∗ ] + zi zk Ωik [m. ρ] + zi zk Ωik [m∗ . Π] + zi zk 2Ωik [ρ. m∗ ] + zi zk Ωik [m.83) (ρ) (ρ) −U 2 k 0 ti dtk Πik (ρ) ∗ zk ρki + zk mki (F ) (F ) + U2 k 0 ti dtk Πik (F ) ∗ zk ρki + zk mki . ρ∗ ] + Ωik [m. In terms of the spectral and statistical functions the equation of motion can be written as: ∗ i ∂ti zi = −J(zi+1 (ti ) + zi−1 (ti )) + U (zi 2 + 2ρii )zi + U mii zi (F ) (F ) (6. ρ] + zi zk Ωik [ρ. m∗ ] (ρ) (ρ) (ρ) (ρ) (ρ) (ρ) ρkj (F ) −U 2 k 0 ti ∗ ∗ dtk zi zk Ωik [m∗ . . −i ∂ti ρij = (F ) (6. ρ] + zi zk Ωik [m∗ . ρ] ρkj ∗ dtk Ωik [ρ. m∗ ] + zi zk Ωik [ρ∗ . Π] + zi zk Ωik [ρ. ρ] + zi zk Ωik [ρ. ρ∗ ] + 2Ωik [m. m∗ ] mkj ∗ ∗ dtk Ωik [m∗ . Π] + zi zk Ωik [ρ. m∗ ] (F ) (F ) (F ) (F ) (ρ) +U 2 k 0 tj mkj . ρ∗ ] + 2Ωik [m.84) (F ) (F ) (F ) (F ) ∗(F ) (F ) ∗2 −J(ρi+1j (ti .5 Secondorder expansion 121 The equations of motion under this approximation are the ones obtained for the full secondorder expansion but with Λ = Λ = 0. tj )) + 2U (zi 2 + ρii )ρij + U (mii + zi )mij ti (ρ) (F ) ∗ (ρ) ∗ ∗ (ρ) −U 2 dtk zi zk Ωik [ρ. tj ) + ρi−1j (ti . m∗ ] + zi zk Ωik [ρ∗ . ρ] ρkj 0 k −U 2 k 0 ti ∗ dtk Ωik [ρ. ρ∗ ] + Ωik [m.
m] ρkj (ρ) +U 2 k 0 tj dtk Ωik [m. m] ρkj (ρ) (ρ) (ρ) (F ) −U 2 k 0 ti dtk Ωik [m. Π] + zi zk Ωik [ρ. Π] + zi zk 2Ωik [ρ. m∗ ] (ρ) (ρ) (ρ) (ρ) (ρ) (ρ) (ρ) −U 2 −U 2 −U 2 ρkj ∗ ∗ dtk zi zk Ωik [m∗ . tj ) + ρi−1j (ti . m∗ ] + zi zk Ωik [ρ∗ .85) ∗2 + zi )mij (ρ) (ρ) ∗ ∗ ∗ dtk zi zk Ωik [ρ. m∗ ] + zi zk Ωik [m. m∗ ] mkj ∗ ∗ dtk Ωik [m∗ . ρ∗ ] + Ωik [m. ρ] + zi zk Ωik [ρ. m∗ ] (F ) (F ) (F ) (ρ) (ρ) +U 2 k 0 tj mkj . Π] + zi zk 2Ωik [ρ. ρ∗ ] + Ωik [m. tj )) + 2U (zi 2 + ρii )ρij + U (mii −U 2 k ti tj ti k tj ti k tj ti k tj (ρ) (ρ) (ρ) (ρ) (ρ) (F ) (ρ) ∗(F ) (ρ) (6. m] + zi zk Ωik [m∗ . . m] + zi zk Ωik [ρ. ρ∗ ] + 2Ωik [m. ρ∗ ] mkj ∗ dtk Ωik [ρ∗ . m∗ ] (ρ) (ρ) (ρ) (ρ) (ρ) mkj . m∗ ] (F ) (F ) (F ) (F ) (F ) (F ) ρkj (ρ) +U 2 k 0 tj ∗ ∗ ∗ dtk zi zk Ωik [ρ∗ . Π] + zi zk 2Ωik [ρ. ρ∗ ] + zi zk Ωik [ρ∗ . ρ] + zi zk Ωik [m∗ . tj ) ti −U 2 dtk 0 k (F ) (6.86) + (F ) mi−1j (ti . ρ] ρkj ∗ dtk Ωik [ρ. m] + zi zk Ωik [m∗ . ρ∗ ] + Ωik [m. m] + zi zk Ωik [ρ. m∗ ] (ρ) (ρ) (ρ) ρkj −U 2 k 0 ti ∗ ∗ ∗ dtk zi zk Ωik [ρ∗ . ρ∗ ] + zi zk Ωik [ρ∗ . ρ∗ ] + Ωik [m. i ∂ti mij = (F ) −J(mi+1j (ti . Π] + zi zk Ωik [ρ. ρ∗ ] + 2Ωik [m. Π] + zi zk 2Ωik [ρ. m∗ ] (F ) (F ) (F ) (ρ) (ρ) (ρ) (F ) (F ) −U 2 k 0 ti mkj +U 2 k 0 tj ∗ ∗ ∗ ∗ dtk zi zk Ωik [m. ρ∗ ] + zi zk Ωik [m.Chapter 6 122 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism −i ∂ti ρij = −J(ρi+1j (ti . tj )) (ρ) + 2U (zi 2 + (ρ) (F ) (F ) ρii )mij + (F ) U (mii (ρ) 2 + zi )ρij (F ) (F ) ∗ ∗ ∗ ∗ zi zk Ωik [m. ρ∗ ] mkj ∗ dtk Ωik [ρ∗ . ρ∗ ] + zi zk Ωik [m.
z and anomalous ﬂuctuations. In Ref.2 Conservation laws For a closed (isolated) system the mean total number of particles N and energy are conserved quantities as they are the constants of motion for the dynamical equations. (6. are neglected. ρ∗ ] + zi zk Ωik [ρ∗ . m. These equations depict the nonlinear and nonMarkovian quantum dynamics.89) = N. 6. tj )) + 2U (zi 2 + ρii )mij + U (mii + zi )ρij +U 2 +U 2 +U 2 +U 2 with ∗ ∗ ∗ ∗ dtk zi zk Ωik [m. which we consider as the primary distinguishing features of this work. Π] + zi zk 2Ωik [ρ. [61] the authors gave a nonMarkovian generalization to the quantum kinetic theory derived by Walser et. It supersedes what the secondorder kinetic theories currently presented can do. m] + zi zk Ωik [m∗ . Π] + zi zk Ωik [ρ.5. m] + zi zk Ωik [ρ.ρ) [m. However in that work symmetry breaking ﬁelds. m] ρkj (ρ) (ρ) (ρ) (ρ) dtk Ωik [m. .87) (ρ) 2 −J(mi+1j (ti . The mean total number of particles is given by (F ) ˆ N = i a† ai = ˆi ˆ i zi 2 + ρii − 1 2 (6. their going beyond the HFB approximation notwithstanding. ρ∗ ] + Ωij (F. m∗ ] (ρ) (ρ) (ρ) (ρ) (ρ) (ρ) ρkj (ρ) ∗ ∗ ∗ dtk zi zk Ωik [ρ∗ . Πij (F.ρ) [ρ. ρ∗ ] + Ωik [m. m∗ ] . m∗ ] (ρ) (ρ) (ρ) (ρ) (ρ) mkj . ρ∗ ] + zi zk Ωik [m.5 Secondorder expansion 123 i ∂ti mij = (ρ) (ρ) (F ) (ρ) (F ) ti k tj ti k tj ti k tj ti k tj (ρ) (6.[58] by including memory eﬀects. [62] presents a kinetic theory approach that includes binary interactions to secondorder in the interaction potential but uses the Markovian approximation. For example Ref. Particle number conservation is a consequence of the invariance of the Hamiltonian under a global phase change. tj ) + mi−1j (ti .ρ) = Ωij (F.88) We end this section by emphasizing that the only approximation introduced in the derivation of the equations of motion presented here is the truncation up to secondorder in the interaction strength. ρ∗ ] + 2Ωik [m. ρ∗ ] mkj ∗ dtk Ωik [ρ∗ . al.6.
See Eqs. While number conservation can be proved explicitly.Chapter 6 124 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism The kinetic equation for N is then d ˆ N (t) dt ∂ ∗ z (t) ∂t i = i 2Re zi (t) + lim t→t (6. energy and momentum. (45). conserve particle number. and cancelling terms. t) ∂t ∂t ii = 0.42). In their . Φ = Γ2 .76). for the full secondorder expansion. conserve the particle number. we have presented a new approach for the description of the nonequilibrium dynamics of a BoseEinstein condensate and ﬂuctuations in a closed quantum ﬁeld system. All the approximations we consider in this paper are phi derivable and thus they obey energy. where the mean ﬁeld and the correlation functions are treated on the same footing selfconsistently and which respects conservations of particle number and energy. HFB. namely. This can be shown ˆ by plugging in the kinetic equation of N (t) (Eq.86)).(6. with Φ a functional of G. 6.6 Conclusions In summary.(46).90) ∂ (F ) ∂ ∗(F ) ρii (t.77) and Eqs (6.(6. 127] and references therein. 6. to prove total energy conservation is not obvious as the Hamiltonian cannot be represented as a linear combination of the relevant operators. It is important to note that even though total population is always conserved there is always a transfer of population between condensate and non condensate atoms.82). To discuss energy conservation we can use the phiderivable criteria [138] which states that nonequilibrium approximations in which the self energy Σ is of the form δΦ/δG. (6. (6. (6. For a detailed discussion of the complete next to leading order 1/N expansion see Refs. The formalism allows one to go beyond the well known HFB approximation and to incorporate the nonlinear and nonMarkovian aspects of the quantum dynamics as manifest in the dissipation and ﬂuctuations phenomena. [126. However.85). The CTP formalism ensures that the dynamical equations of motion are also causal. Φ = Γ2 + Γ2 and for (1) (2)1/N the secondorder next to leading order 1/N expansion. t ) + ρ (t .90) the equation of motion for the mean ﬁeld and the normal statistical propagator Eqs. The 2PI eﬀective action formalism provides a useful framework. particle number and momentum conservation laws. the introduction of approximation schemes that truncate the inﬁnite hierarchy of correlation functions at some ﬁnite order with causal boundary conditions may introduce dissipation [118].(6. Φ = Γ2 + Γ2 . (1) (1) (2) For HFB. Eqs.23).64) and (6. 1/N expansion and full secondorder expansion. It is clear that the exact solution of a closed system is unitary in time and hence disallows any dissipation. All three approximations we have considered.
.6 Conclusions 125 current form the scattering terms are nonlocal in time and thus hard to estimate analytically. this systematic approach can be used as a quantitative means to obtain solutions in diﬀerent regimes and make comparisons with kinetic theory results where a Markovian approximation is assumed. We postpone these discussions to chapter 8.6. and their calculation is numerically demanding. However.
Chapter 6 126 The two particle irreducible eﬀective action (2PI) and the closed time path (CTP) formalism .
and therefore a mean ﬁeld approach is only accurate for short times. However.Chapter 7 Nonequilibrium dynamics of a patterned loaded optical lattice: Beyond the mean ﬁeld approximation In this chapter the twoparticle irreducible (2PI) closedtimepath (CTP) eﬀective action formalism derived in chapter 6 is used to describe the nonequilibrium dynamics of a Bose Einstein condensate (BEC) selectively loaded into every third site of a onedimensional optical lattice. our numerical results also show that all of the approximations fail at late times. interatomic collisions play a crucial role in determining the quantum dynamics of the system. which the mean ﬁeld and the HFB approximations fail to reproduce. 7. We show that because the secondorder 2PI approximations include multiparticle scattering in a systematic way. when interaction eﬀects become signiﬁcant. The exact numerical solution is also used to test the validity of the various methods derived under the 2PICTP approximation. However. we show here that even in the case when the kinetic energy is comparable to the interaction energy. In chapter 4 we used a mean ﬁeld approach to describe the dynamics of this system. This result is demonstrated by comparison between the mean ﬁeld solutions and the exact numerical time evolution of the initial state using the Bose Hubbard Hamiltonian for systems with small numbers of atoms (N ∼ 10) and lattice sites (M = 2 or 3). in order to provide the context for the subsequent discussion of the . they are able to capture damping eﬀects exhibited in the exact solution.1 Mean ﬁeld dynamics In this section we review the basic features of the mean ﬁeld results obtained in chapter 4.
2. due to the fact that the Hilbert space needed for the calculations increases rapidly with the number of atoms and wells. As γ increases the tunneling between sites occurs at a higher frequency. and in which the condensate initially has a uniform phase. . 7. but with reduced amplitude. For γ = 0 the period of oscillation is 2π . The population of the initially occupied sites becomes eﬀectively selftrapped by the purely repulsive pair interaction.1.1 Dynamical evolution By making the mean ﬁeld ansatz in the BoseHubbard Hamiltonian. 3 Interaction dominated regime: The eﬀect of interactions on the mean ﬁeld dynamics is to cause the energies of the initially occupied sites to shift relative to those of the unoccupied sites. (τ ≡ tJ ). the amplitudes for all initially occupied sites z3i (τ ) = z0 (τ ) evolve identically in time.1. The J mean ﬁeld dynamical behavior can be qualitatively divided into two regimes: The tunneling dominated regime (γ < 1): In this regime the oscillation period is essentially constant. T (γ).Chapter 7 128 Nonequilibrium dynamics of a patterned loaded optical lattice: Beyond the mean ﬁeld approximation dynamics in the 2PICTP formalism. We use a modest number of atoms and lattice sites for the comparisons. the role of interactions is relatively small. z3i+1 (0) = z3i+2 (0) = 0. and the amplitudes for the initially unoccupied sites satisfy z3i+1 (τ ) = z3i+2 (τ ) = for all τ .2 Comparisons with the exact solution To check the validity of the mean ﬁeld approximation. or one with periodic boundary conditions. are determined by the parameter γ ≡ 3 MU = 3λ. Thus at τ = 0. (4. In section 4. where M is the total number of lattice sites.3. This allows us to reduce the full set of equations to a set of two coupled equations for z0 (τ ) and z1 (τ ). we replace the ﬁeld operator ai by a cnumber zi (t). The amplitudes zi (t) satisfy the DNLSE ˆ (Eq. For an inﬁnite lattice. This system will undergo Rabi oscillations whereby atoms periodically tunnel from the initially occupied site into the two neighboring sites. and the equations of motion are equivalent to those of a twostate Rabi problem. The solutions z0 (τ ) and z1 (τ ) are oscillatory functions whose amplitudes N and common period. we made comparisons with the exact many body solution for N = 6 atoms and M = 3 wells. we treated a model case in which the initial occupancies of each third site are the same.6)). the amplitudes zi (τ ) are given by z3i (0) = 3/M . 7.
2): In this regime the DNLSE gives a good description of the early time dynamics. [129] the authors show the validity of N the semiclassical approach when τ < τcl ∼ M γ in the case of two lattice sites. We expect the semiclassical approach to be valid for time scales less than the inverse energy level spacing. The fully quantal solution was found by evolving the initial state in time with i ˆ the BoseHubbard Hamiltonian. To do the numerical calculations we partitioned the Hilbert space in subspaces with a ﬁxed number of atoms and propagated independently the projections of the initial state on the respective subspaces. Weakly interacting regime (γ = 0. We chose this state because in the experiment the loading of the atoms was done slow enough with respect to band excitations but fast with respect to many body excitations that at time t = 0 most of the atoms in the initially populated wells were condensed. and therefore during the dynamics the diﬀerent subspaces never get mixed. After τcl quantum eﬀects become important. i. 2.1. (7. for three diﬀerent values of γ. 7. In Ref. The salient features observed in these comparisons are: 1. so that ϕ(t) = e− Ht ϕ(0) . The number of subspaces used for the numerical evolution were such that no change in plots of the dynamical observables was detected by adding another subspace. We observe in Fig.7. this √ condition was achieved by including the subspaces between N − 4 N and N + 4 N atoms.1 that the total population per well predicted by the mean ﬁeld solution agrees with the exact solution and also that the condensate population remains large for the time under consideration.1 Mean ﬁeld dynamics Exact solution For the exact solution we used an initial state given by ϕ(0) = (e−N/2 e √ N a† ˆo 129 0 ) ⊗ 0 ⊗ 0 . 7. In this regime the ratio between interaction and kinetic energy is . Intermediate regime (γ = 2): Quantum ﬂuctuations lead to a nontrivial modulation of the classical oscillations. This time scale is in good agreement with the numerical results shown in in Fig.e. zi (t)2 .1 we plot the average population per well a† (t)ˆi (t) and the condenˆi a sate population per well  ai (t) 2 and compare them with the mean ﬁeld predicˆ tions.7. This procedure could be done because the Hamiltonian N ˆ ˆ commutes with the number operator i ai † ai . Numerical comparisons In Fig. A subspace with Nn number of atoms and M wells is spanned n +M −1)! by (Nn !(M −1)! states. Generally for N atoms in the initial √ state.1) The initial state represents a coherent state with an average of N atoms in the initially populated well and zero atoms in the others.
In Ref. leading i=j . even for rather early times. the phase factors add to an integer value of 2π. in the sense that both predict selftrapping of atoms in the initially populated wells. For a uniform loaded lattice. At time t = 0 we have a condensate fraction of order one. However. and nonzero correlations are developed in the dynamics. To understand the dynamics in the weak and intermediate regimes. the authors show. In this regime. it can not remain zero forever because we are dealing with a closed quantum system. how the phase coherence is restored dynamically. 3. this is no longer the case for t > 0 . However. However. The collapses and revivals of the condensate population in the strong interacting regime can be easily estimated by considering the energy spectrum. after some time. the collapse and revivals of the condensate in this strong interacting regime and the importance of quantum eﬀects have been experimentally observed[47]. with ﬁnite recurrence time. However. Mean ﬁeld results are accurate only for a short time.Chapter 7 130 Nonequilibrium dynamics of a patterned loaded optical lattice: Beyond the mean ﬁeld approximation small enough to allow the atoms to tunnel but not too small to make interaction eﬀects negligible. After the collapse. [129]. The dynamics of the system is described by the interference of the diﬀerent nparticle Fock states that span the coherent state of the initially populated well. it can be seen to coherence of the system. Therefore at some time trev we expect the condensate to revive again. If we look at the initial . The dynamical restoration of the phase coherence which tends to distribute atoms uniformly among the lattice sites and to damp the oscillations characterizes the dynamics in the superﬂuid regime. the exact solution exhibits damped oscillations of the atomic population. the phase is going to randomize and this will lead to the collapse of the condensate population. Strongly correlated regime (γ = 12): The system exhibits macroscopic quantum selftrapping of the population. Qualitatively. It is expected then that. not for a patterned loaded initial state but for an initial Mott state also with zero initial phase coherence. the fast decrease of the condensate population and its subsequent revivals (as found in the exact solutions) give us an idea of the importance of correlation eﬀects beyond mean ﬁeld. Even though there is minimal initial coherence between adjacent sites due to the patterned loading procedure we are still preparing the system in a superﬂuid state in the initially populated well. Quantum scattering eﬀects are crucial. both the mean ﬁeld and the exact solutions agree. At integer values of trev = (U/h)−1 . determined by a† (0)ˆj (0) ˆi a be zero due to the patterned loading. the system will remain for a while with zero condensate population. the ground state of the system is not superﬂuid. In this regime the energy eigenstates of the system are almost number (Fock) states and the energy spectrum is almost quadratic. we have to focus on the coherent properties of the system. Even though interactions can be strong. the ground state is indeed superﬂuid. En ≈ n(n−1)U/2. due to interactions.
the only relevant parameter for describing the dynamics of the system is the ratio U N/J. tj ) and mij (ti .1. tcoll .2. 0) = 1 δij . The initial conditions correspond to an initial coherent state with N atoms in the initially populated well. J = 1 and U/J = 1/30. it is given by: tcoll ∼ trev /(2πσ). also shows how the collapse time. the full secondorder approximation and the 1/N expansion up to secondorder in the interaction strength. For the parameters used in the strongly correlated regime. ρij (ti . tj ). γ = 12 and N = 6. where the ratio U/J is small enough that truncation up to secondorder makes sense but not too small so that interaction eﬀects still have to be taken into account. explicitly. (F ) (ρ) (ρ) (F ) (ρ) . 7. the √ initial distribution is Poissonian and tcoll is given by tcoll ∼ /( N U ). With this choice we wanted to show the validity of a mean ﬁeld approach to describe this regime and the corrections introduced by the higher order approximations. To study the kinetic energy dominated regime we chose for the simulations three diﬀerent sets of parameters:. At the mean ﬁeld level (using the DNLSE) for a given number of wells. Because in all these approximations we use U/J as an expansion parameter we will focus our calculations on the intermediate regime. ρij (0. Here zi (t) are the condensate amplitudes and ρij (ti .[130] for a more general situation. depends on the variance of the initial atomic distribution.7. In this section we proceed to test the validity of the diﬀerent 2PICTP approximations derived in chapter 6.2 2PICTP approximations 131 to a revival of the initial state. The second set of parameters is M = 3. In this regime the kinetic energy is big enough to allow tunneling but the eﬀect of the interactions is crucial in the dynamics. M = 3. 7. That reference. We start by describing the initial conditions chosen for the numerical calculations. J = 1 and U/J = 1/3. This time scale agrees with the one estimated in Ref. and ﬁnally we discuss the results. 7. 0)) = 2 mij (0. N = 6.2 2PICTP approximations It was shown in the previous section that to describe the dynamics of the patterned loaded optical lattice. then we outline the numerical algorithms used. If the initial state is a coherent state. 0) = 0. N = 8.1 Initial conditions and parameters To model the patterned loading. the initial conditions assumed for the numerical (F ) (ρ) (F ) solutions were zi (0) = N δi0 . tj ) are the statistical and spectral normal and anomalous propagators as deﬁned in chapter 6. mij (ti . we observe that the estimated collapses and revival times are in agreement with what is shown in Fig. 0) = −iδij and mij (0. the time dependent HFB approximation. approximations beyond the standard mean ﬁeld theory are required. This is not the case in the exact solution where both U N/J and N are important. tj ). ρij (0. The ﬁrst set is chosen to be in the very weak interacting regime.
The main numerical limitation of .. (ρ) s. and therefore we expect better agreement of the truncated theories with the exact solution with larger N .ρ) (F. Due to the symmetries of the matrices only half of the (n. To increase the number of atoms in the exact calculations we had to reduce the number of wells to two due to the fact that the dimension of the Hilbert space scales exponentially with N and M .ρ) (F. To study the dependence of the dynamics on the total number of atoms. that is.ρ) (F.ρ) one speciﬁes ρij00 . the spatial dimension is discrete (indices i and j). and negative sign for the spectral functions. with n and m integers and took the advantage that. further decreasing it did not change the results. To ensure that the discretized equations retain the conservation properties present in the continuous ones one has to be very careful in the evolution of the (F ) diagonal terms of ρiinn and take the limit m → n in a proper way: ρijn+1n+1 − ρijnn = ρijn+1n − ρijnn (F. (7. For the numerical solution we employed a time discretization t = nat .2. mρ = 0 are ijnn ijnn ﬁxed for all time due to the bosonic commutation relations.ρ) (F. mij00 and zi0. (F.ρ) (F.ρ) . The important point to note is that all equations are causal in time. The greater the parameter U N/J.ρ) ± ρjin+1n − ρjinn (F.2) ∗(F.Chapter 7 132 Nonequilibrium dynamics of a patterned loaded optical lattice: Beyond the mean ﬁeld approximation The larger the initial population N . but the number of atoms was changed from 20 to 80. the larger the initial population in the coherent matter ﬁeld. We used the fourth order RungeKutta algorithm to propagate the local part of the equations and a regular one step Euler method to iterate the non local parts. due to the presence of the lattice.ρ) for the time evolution of the matrices ρijnm .. and all quantities at some later time tf can be obtained by integration over the explicitly known functions for times t ≤ tf . As initial conditions (F. Though the equations are very complicated.ρ) (F. m) matrices have to be computed and the values ρρ = −i. for the time step n + 1 one computes successively all entries with m = 0. n.ρ) ± mjin+1n − mjinn . The discretized equations (F. mijnm and zin advance time stepwise in the ndirection for ﬁxed m. n + 1 from known functions evaluated at previous times. the third set of parameters in our solutions is chosen to be M = 2.ρ) (7.ρ) (F. (F ) s. the smaller is the time step required..3) mijn+1n+1 − mijnn = mijn+1n − mijnn (F..ρ) ∗(F.ρ) (F. they can be solved on a computer... For the integrals we used the standard trapezoidal rule.2 Numerical algorithm for the approximate solution The time evolution equations obtained in chapter 6 are nonlinear integrodiﬀerential equations. The time step at was chosen small enough so that convergence was observed. J = 1/2 and N U/J was ﬁxed to 4. 7.ρ) with the positive sign for the statistical propagators. Starting with n = 1. t = mat .
A characteristic time. This time is described by the exponential damping of the twopoint correlator at time t with the initial time[127].5.2 the dynamics of the atomic population per well resembles the Rabi oscillation phenomenon. In this regime damping eﬀects remain very small for the time depicted in the plots. periodic boundary conditions enforce equal evolution of the initially empty ones. within a typical numerical precision it was usually not necessary to keep all past values of the two point functions in the memory.6 we show our numerical results. with M the total number of lattice sites and a the lattice spacing. In general. (7. Notice that even though there are three wells. i zi (t)2 . The quasimomentum distribution of the atoms released from the lattice is important because it is one of the most easily accessible quantities from an experiment.3 Results and discussions In Figs. This shows up in the plots of the condensate population and depletion. t) − 1 . In the duration depicted in the plots of Fig. The numerical simulations show a general agreement between the diﬀerent approaches and the exact solution.j (F ) where the quasimomentum k can assume discrete values which are integral mul2π tiples of M a . We focus our attention on the evolution of the condensate population per well. zi (t)2 . 7. The basic features of the plots can be summarized as follows: The very weakly interacting regime: In Fig. We used for the calculations a single PII 400 MHz workstation with 260 Mb of memory.2.2 the total condensate constitutes an important fraction of the total population. ρii (t.2 to 7. 7.2 2PICTP approximations 133 the 2PI approximation is set by the time integrals. 7. The eﬀect of including higher order terms in the equations of motion is to introduce small corrections which improve the agreement with the exact dynamics. which make the numerical calculations time and memory consuming. Regarding the quasimomentum distribution. (see sec. zi 2 + ρii (t.7. after which the inﬂuence of the early time in the late time behavior is negligible. the total atomic popu(F ) lation per well. the depletion per well or atoms out of the 2 1 condensate. and the total condensate population. In our numerics we extended the length of the employed time interval until the results did not depend on it. For a typical run 12 days of computational time were required. it was less than the inverse damping rate. The secondorder 1/N expansion gives an improvement over the HFB and the complete secondorder perturbative expansion almost matches the exact solution perfectly.4) i M i. The total population is also explicitly shown in the ﬁgures to emphasize number conservation.5) The quasimomentum distribution function nk is deﬁned as 1 nk (t) = eik(i−j) a† (t)aj (t) . t) − 2 . 7. is given by the inverse damping rate. we observe that similarly to the spatial . where the small diﬀerences can be better appreciated. However.
4 to 7. In contrast. We observe as a general issue in this regime that. Even though none of them are satisfactory after the ﬁrst oscillation the HFB approximation is the only one that fails to capture the exponential decrease of the condensate population. In Figs. The k = 0 and ± 2π intensities oscillate with 3 3 the same frequency as the atomic population per well. regardless of the fact that the secondorder solutions capture the damping eﬀects. As soon as the third order terms start to be important the large 1/N expansion gives a better qualitative description. even though we observe the collapse of the condensate population. they enforce equal evolution of the ± 2π quasimomentum intensities.6 as the number of atoms is increased (see discussion below). The failure can be understood under the following lines of reasoning. they depart from the exact dynamics: the secondorder approaches predict faster damping rates. At zero temperature condensate atoms represent the most ”classical” form of a matter wave. This behavior is better appreciated in ﬁgs. They modulate the oscillations in the population per well and scatter the atoms out of the condensate. 7. J = 1 and U/J = 1/3.Chapter 7 134 Nonequilibrium dynamics of a patterned loaded optical lattice: Beyond the mean ﬁeld approximation distribution where the initial conﬁguration and periodic boundary conditions reduce the three well system to a double well one. the exponential decay of the condensate is present in the secondorder approximations. it includes the eﬀects of collisions only indirectly through energy shifts.3 we plot the numerical solution for the parameters M = 3. as soon as the condensate population decreases to a small percentage of the total population. the noncondensate population increases. Thus. Comparing the two secondorder approaches we observe that the full secondorder expansion gives a better description of the dynamics than the 1/N solution only in the regime when the perturbative solutions are close to the exact dynamics. the total population is always conserved: as the condensate population decreases. and breaks down outside the collisionless regime where multiplescattering eﬀects are important. to have a more . In contrast to the case of the very weak interacting regime. The overdamping is more severe in the dynamics of the population per well than in the condensate dynamics. It is important to point out that. This is expected. The intermediate regime: We can see the eﬀect of the interactions in the dynamics. When they decay. because even though this approximation goes beyond mean ﬁeld theory and takes into account the most important scattering eﬀects. 2. 7. Non local parts of the selfenergy included in them encode scattering eﬀects responsible for damping. At this point the higher order terms neglected in the secondorder approximations are the ones that lead the dynamical behavior. both are also well described by the approximations in consideration. 1. the role of quantum correlations becomes more important. it is only at early times that any of these approximations is close to the exact solution. N = 8.
4 to 7. 7.5 and the quasimomentum intensities in Fig. as the number of atoms is increased. In the plots we show the numerical solutions found for a double well system with ﬁx ratio U N/J = 4 and three diﬀerent values of N : N = 20. thus tcoll ∝ N increases with N as observed in the numerical J calculations. In this way for all the cases we start with an atomic population of magnitude one in the initially populated well. The N revival time is proportional to U −1 and varies with N for ﬁxed U N/J as √ trev ∝ N . as a result of the increase in the initial number of coherent atoms is in fact observed in the plots. The decrease of the damping rates as the number of atoms is increased is not surprising because by changing the number of atoms we aﬀect the quantum coherence properties of the system.6. 7. In the exact dynamics we see that as the number of atoms is increased the damping eﬀects occur at slower rates.6. 7. the qualitative behavior of the exact quantum dynamics is not much aﬀected as the number of atoms is increased for a ﬁxed U N/J.7. 7. Even though the problem of underdamping in the HFB approximation and the overdamping in the secondorder approaches are not cured. 3. 7. This feature can be noticed in the quantum dynamics of all of the observables depicted in the plots 7. Similarly to what happens to the spatial observables. The better agreement of the 1/N expansion relies on the fact that even though the number of ﬁelds is only two in our calculations the 1/N expansion is an expansion about a strong quasiclassical ﬁeld conﬁguration. . Damping eﬀects are also quite noticeable in the quantum evolution of the quasimomentum intensities.4. 40 and 80.4 to 7.2 2PICTP approximations 135 accurate description of the dynamics after the coherent matter ﬁeld has decayed one needs a better treatment of correlations.6. Perhaps this issue can be more easily observed in the quasimomentum distribution plots. As discussed above the collapse time t√ of the condensate population is approximately given by tcoll ∼ 2πrev .6 we explore the eﬀect of the total number of atoms on the dynamics. the condensate population per well and total condensate population in Fig. The improvement of the 2PI approximations as N is increased. We present the results obtained for the evolution of the atomic population per well in Fig. Fig. To make the comparisons easier we scaled the numerical results obtained for the three diﬀerent values of N by dividing them by the total number of atoms. In Figs. The 1/N expansion shows the fastest convergence. Besides damping rates. the HFB approximation fails completely to capture the damping eﬀects present in the evolution of the Fourier intensities whereas the secondorder approaches overestimate them. we do observe a better matching with the full quantal solution.
one may try to adopt a stochastic approach. We have carried out the analysis up to secondorder in the interaction strength. Alternatively. [127] but it is not obvious that this will lead to the required improvement. Even though. but it is suﬃcient to account for dissipative eﬀects due to multiparticle scattering that are crucial even at early times. as is clear that the secondorder 2PI approximations fail to capture the fully correlated dynamics in the system. Because of their success in describing moderately correlated regimes. Our formulation is capable of capturing the salient features of the system dynamics in the regime under consideration. but the challenge will be shifted to the derivation of a noise term (which is likely to be both colored and multiplicative) which contains the eﬀects of these higher correlations and the solution of the stochastic equations.Chapter 7 136 Nonequilibrium dynamics of a patterned loaded optical lattice: Beyond the mean ﬁeld approximation 7. . However.This approximation is introduced so as to make the numerical solution manageable. the secondorder approximations could become a useful tool for describing other experimental situations. To improve on this a better treatment of higher correlations is required. the secondorder approximations used here predict an overdamped dynamics. at the point where an important fraction of the condensate population has been scattered out. such as the decay of the condensate population and the damping of the oscillations of the quasimomentum and population per well unaccounted for in the HFB approximation. One might try to include the full next to leading order large N expansion without the truncation to secondorder as done in Ref. it has been proved to work at intermediate times when correlations are not negligible and standard mean ﬁeld techniques fail poorly.3 Conclusions In this work we have used the CTP functional formalism for 2PI Green’s functions to describe the nonequilibrium dynamics of a condensate loaded in an optical lattice on every third lattice site.
Due to the ˆ ˆ symmetry of the initial periodic conditions the curves for the i = 1 and 2 wells are the same in all depicted curves .2 ˆ per well calculated from the exact solution:  a0 2 and  a1.2 2 . 2 0 6 γ=2 # of atoms 4 2 0 6 γ = 0. Top panel. a† a1. # of atoms 4 DNLSE Conden. middle panel.1: Comparisons between the exact and the DNLSE solutions for six atoms and three wells. The solid line is the DNLSE prediction for the population per well: z0 (t)2 and z1.2). the triangles are used to represent the exact solution for the population per well calculated using the Bose Hubbard Hamiltonian: a† a0 .2 # of atoms 4 2 0 0 2 4 t 6 8 10 Figure 7.2 (t)2 . The time is given in units of /J. The pentagons show the condensate population ˆ0 ˆ ˆ1. bottom panel.2 .3 Conclusions 137 6 γ = 12 Popul. strongly correlated regime (γ = 12). weakly interacting regime (γ = 0.7. . intermediate regime (γ = 2).
25 5 6 4 2 0 2.05 2 1. N = 6.8 Depletion/well 0.6 condensate/well Population/well Total Number 6 .5 5 6 Exact 2nd 1 /N HFB 5.2: Comparisons between the exact solution(solid line).2 0 0 2 4 t 6 8 10 0 2 4 t 6 8 10 Figure 7. In the plots where the population. The time is given in units of /J.4 0. momentum 0.95 1. the HFB approximation (boxes). Notice the diﬀerent scale used in the depletion plot.Chapter 7 138 Nonequilibrium dynamics of a patterned loaded optical lattice: Beyond the mean ﬁeld approximation Total Condensate 6 5.9 4 2 0 0. the secondorder large N approximation (pentagons) and the full 2PI secondorder approximation(crosses) for the very weak interacting regime. J = 1 and U/J = 1/30. The parameters used were M = 3.75 5. In the momentum distribution plot the upper curve corresponds to the k = ±2π/3 intensities and the lower one to the k = 0 quasimomentum intensity. condensate and depletion per well are depicted the top curves correspond to the initially populated well solutions and the lower to the initially empty wells.5 5.
J = 1 and U/J = 1/3.7. The time is given in units of /J. N = 8. In the quasimomentum plots k = 2π/a is the reciprocal lattice vector with a the lattice spacing. . In the plots the abbreviation 1st is used for the initially occupied well and 2nd for the initially empty wells.5 t 6 8 10 0 2 4 t 6 8 10 0 2 4 Figure 7.3: Comparisons for the case M = 3.5 3 k = q/ 3 2.3 Conclusions 139 8 6 4 2 0 8 6 4 2 0 3 2 1 0 # of atoms Total Number Total Condensate Exact 2nd 1N HFB # of atoms 1 st Condensate 1 st Depletion 1 st Population 2 nd Population 2 nd Condensate 2 nd Depletion 8 6 4 2 0 3 2 1 # of atoms qmomentum k=0 0 2 4 3 2 1 0 t 6 8 10 0 3.
40 and 80.4: Comparison between the evolution of the atomic population per well for M = 2. The number of atoms N is explicitly shown in each panel.4 0. N U/J = 4 and N = 20.2 0 1 ( atoms) /N # 0. In the plots P1 stands for the fractional atomic population in the initially populated wells and P2 for the population in the initially empty wells.4 0. N20 P1.6 0.8 0. N80 Figure 7. .Chapter 7 140 Nonequilibrium dynamics of a patterned loaded optical lattice: Beyond the mean ﬁeld approximation Exact 2nd 1 /N HFB 1 ( atoms) /N # 0. J = 1/2.2 0 0 2 4 t P2. Time is in units of /J. N20 6 8 10 0 2 P2.6 0.8 0. N80 6 8 10 P1. N40 4 t 6 8 10 0 2 4 t P2. N40 P1.
N40 C1.5: Time evolution of the condensate population per well and the total condensate population. In the plots C1 stands for the fractional condensate population in the initially populated well. N20 C2. Time is in units of /J.6 0.8 0.4 0.6 0.6 0.8 0.2 0 1 ( atoms) /N # 0.7. N80 C2.2 0 0 2 4 t 6 CT. N80 C1. . N80 8 10 0 2 4 t 6 8 10 0 2 4 t 6 8 10 Figure 7.2 0 1 ( atoms) /N # 0.for the same parameters as Fig.3 Conclusions 141 Exact 2nd 1 /N HFB 1 ( atoms) /N # 0. N20 C1. 7. N20 CT. N40 C2. C2 for the fractional condensate population in the initially empty one and CT for the total condensate fraction.4 0.4.4 0. N40 CT.8 0.
Time is in units of /J.3 0.Chapter 7 142 Nonequilibrium dynamics of a patterned loaded optical lattice: Beyond the mean ﬁeld approximation Exact 2nd 1N HFB 0. The plots are scaled to set the integrated quasimomentum density to one for all N .5 0.6 0.7 0. N40 ko. 40 and 80.4 0.4 0.6: Dynamical evolution of the quasimomentum intensities.7 qmomentum 0.8 qmomentum 0. N20 ko.6 0. N80 Figure 7.2 0 2 4 t 6 8 10 0 2 4 t 6 8 10 0 2 4 t 6 8 10 ko. . N80 k1. J = 1/2. In the plots ko denotes the k = 0 quasimomentum component and k1 the k = π/a one (a the lattice spacing).3 0. N20 k1. N40 k1.5 0. N U/J = 4 and N = 20. The parameters used were M = 2.
We have carried out the analysis up to secondorder in the interaction strength. The purpose of this chapter is then to simplify the complicated 2PICTP equations and show that they in fact reproduce. However. standard kinetic theories and equilibrium solutions for weakly interacting gases well known in the literature since the late 1950’s.Chapter 8 From the 2PICTP approximations to kinetic theories and local equilibrium solutions In chapter 6 we used the CTP functional formalism for 2PI Green’s functions to derive dynamical equations of motion. when the perturbation induces disturbances in the system of wave length longer than thermal wave lengths and frequencies much smaller than characteristic particle energies then the system is in a regime where standard kinetic theories give a good description of the dynamics. In chapter 7 we used the formalism to investigate the nonequilibrium dynamics of a condensate loaded on every third site of an optical lattice. Moreover. We showed that the formalism allowed us to go beyond the HFB approximation and to incorporate the nonlinear and nonMarkovian aspects of the quantum dynamics as manifested in the dissipation and ﬂuctuation phenomena. the numerical solutions are complicated enough that application to real system with many lattice sites are presently beyond the scope of standard computation capabilities. Nevertheless. in the slowly varying regime. in the present form the equations of motion are complicated nonlocal nonlinear equations from which we can hardly get any physical information regarding the system behavior except by numerical solution. .
H(ti . tj ) ≡ −iGij b (t. −1 where Do (ti . tk ) − S HF B (ti .2) (8. σx = . tk ) z (tk ) = k dt S(ti . tj ) N U T r H(ti .8) with σ being the Pauli matrices: 1 0 0 −1 0 1 1 0 σz = . (6. tj ) is the inverse free particle propagator given by: −1 Do (ti . (8. tj ) ≡ G< (ti . (8. a −iG> c (t.22) and Eq. We deﬁne the quantities a z(ti ) ≡ zi (t). t ). .45) they can then be rewritten as: ΣHF B (ti .26) and (6. (6. tj ) ≡ (iδij σz ∂t + J(δi+1j + δi−1j ) − δij Vi ) δ(t − t ). tj ) ≡ > (8.1 Rewriting the 2PICTP secondorder equations In order to make the comparisons with standard approaches we will start by getting rid of the matrix indices that were useful to derive the equations of motion but which complicate the notation.4) a −iG> c (t ji a −izi (t)zj b (t ). ij a −iG< c (t. dt k −1 Do (ti .24)).9) −1 In Do (ti . (8. t ) = ij . tk )z(tk ). tk ) G(tk . tj ) ≡ G (ti . tj ) = −δij δC (t − t ). (6. t ).7) (8. tk )G(tk . tj ) we have allowed the presence of an external potential Vi . (6.1) a G(ti .6) dt Σ(ti . (8. tk ) − ΣHF B (ti .(8.3) (8.11) . With these deﬁnitions we now explicitly separate the single particle.27). tj ) δ(t − t )δij . tj ) ≡ U T r H(ti .5) in Eqs.5) Here we used the notation introduced in Eqs.1)(8. tj ) + 2G(ti . tj ) + G(ti . tj ) k dt k −1 Do (ti . (8. tj ) + G(ti .10) δ(t − t )δij. The label HFB stands for the HFB contribution.Chapter 8 144 From the 2PICTP approximations to kinetic theories and local equilibrium solutions 8. the HFB and the secondorder contributions in the equations of motion for the condensate and the propagators (Eq. tj ) I + 2 H(ti . S HF B (ti . Using the deﬁnition (8.43) and (6. t). tj )I + G(ti . tj ) ≡ i N i (8.
72). tj ) ≡ − 2 + 2U N G( ) (ti . tj ). tj ) = t k −∞ (8. tj ) ≡ (Σ (ti .(6. ∞ k −∞ −1 dt H(ti . tj ) . tj ) − S < (ti .17) (8. tj ) + t −∞ dt H(ti .19) 2G( ) (ti . If we use the full secondorder expansion.8.12) and we have introduced the spectral functions γ(ti . tk ) G( ) (tk . tj )T r G( ) (ti . tj ) − G (ti . tj ) = t k −∞ dt A(ti . tj )G( ) (tj . tj )G( ) (tj . (8. tj ) − dt G( ) (ti .15) is the hermitian conjugate of Eq. tj ) ≡ (G (ti . ti ) (8. tj ). Eqs.14). Γ(ti .31) and (6. Eq. Γ and A to be consistent with standard Kadanoﬀ and Baym notation [137]. tj ) − dt Σ( ) (ti . tk ) − ΣHF B (ti . tk )Γ(tk . ∞ dt k −∞ t k −∞ ∞ k −∞ t k −∞ −1 Do (ti . (6. tj ) − Σ (ti . tk ) Do (tk . To evaluate the secondorder contribution. In the above equations. tj ) − S HF B (tk . tj ) is just the spectral function deﬁned in Eq. Eq.71 and 6. tj ) − ΣHF B (tk . tk ) − S HF B (ti . tj )).16) (8. tj )). tj )).12) = 0. tk )H(tk . tk )γ(tk . tj ) ≡ (S > (ti . tj ) (8. A(ti . it can be shown that the matrices S ( ) and Σ( ) are given by 2 S ( ) 1 (ti .32) but instead of setting the initial time to zero we choose it to be −∞ and in this way we recover the standard Kadanoﬀ and Baym integration contours [137]: ∞ t −∞ dt k −∞ −1 Do (ti . however. tk ) H(tk . . tj ) − dt γ(ti .34) multiplied by a minus sign. we adopted the notation γ.14) dt Γ(ti .18) Notice that A(ti .13) is the hermitian conjugate of Eq.(6.15) −1 dt G( ) (ti . (8. tk )G( ) (tk .13) = 0. (8. ti )G( ) (ti . (8. tk ) Do (tk . (8.1 Rewriting the 2PICTP secondorder equations 145 where I is the identity matrix. tk )A(tk . tj ) (8. tk )Σ( ) (tk . we use Eqs. > < > < (8.
tj ) = 2U N 2U N 2 2zi 2 + ρii + ρii zi + mii ∗2 + m∗ 2+ρ +ρ zi 2zi  ii ii ii δ(t − t )δij(8.26) (8. ti )G( ) (ti . tj ) +2G( ) (ti . δ(t − t )δij . ti )H(ti .27) (8. tj ) + G( ) (tj . ti )G( ) (ti . ti )G( ) (ti . tj ) H(tj .25) (8.Chapter 8 146 From the 2PICTP approximations to kinetic theories and local equilibrium solutions Σ ( ) 1 (ti . tj )T r G( ) (ti . . tj ) = < S11 (ti . tj ) ≡ − 2 2U N 2 × ( ) (8.20) H(ti .28) .24) zi 2 + ρii + ρii mii ∗ 2+ρ +ρ mii zi  ii ii > S11 (ti . tj )G( ) (tj . tj )Gik (tj . tj ) . ji N2 2 −i8U ρij (2mji m∗ + ρji ρij ). ti )H(ti . tj ) G( ) (ti . (8. ij N2 2 −i8U mji (mij m∗ + 2ρji ρij ). (8. ti )G( ) (ti . tj ) = < S12 (ti . tj ) = −i ρij mij m∗ ρji ji ρij m∗ ij mji ρji .23) . tj ) + G( ) (tj .21) (8. ji N2 2 −i8U mij (mij m∗ + 2ρij ρji ). tj ) = > S12 (ti .22) into the selfenergy equations we get ΣHF B (ti . It is convenient to decompose the above equations in their matrix components. (8.21) and (8. tj ) = S HF B (ti . tj ) = −i8U 2 ρij (2mij m∗ + ρij ρji ). tj ) + G( ) (tj .22) Using Eqs. To do that we introduce the deﬁnitions G> (ti . ti ) + 2H(ti . ti )G( ) (ti . ji N2 (8. tj )T r H(tj . tj ) + G( ) (tj . tj ) = −i G< (ti .
Σ22 (ti . .33) (8.36) The above expressions for the selfenergy exactly agree with the ones used in Refs. [59. tj ) {mji → mij ∗ } . ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) (8.29) ij N2 ρ2 zj zi ∗ + 2ρji ρij zi zj ∗ + 2mij mji ∗ zj ∗ zi + 2mij ρij zi ∗ zj ∗ ). tj ) = ij 11 N2 ∗ ∗ ∗ ∗ ∗ 2ρij ρji zi zj + 2mji zi mij zj + 2ρij mji zi zj ). In a classical point of view the two diﬀerent scales in the system can be understood as the time (or length) scale separation between the duration of a collision event (or scattering length) and the inverse collision rate (or the mean free path). not readily solvable in closed form. one related to quantum processes (microscopic) which determines the degree of quantummechanical entanglement of the system and one related to the statistical and kinetic behavior (macroscopic) determined by the range of the interactions among particles. ij −i8U 2 (2ρji mij ρij + 2ρji ρij zi zj + mij 2 mji ∗ + (8. In Refs. tj ) {zi → zj ∗ . tj ) = 11 147 −i8U 2 (ρji ρ2 + 2mij ρij mji ∗ + 2ρij mji ∗ zi zj + (8. tj ) {ρij → ρij } . To progress further we need to introduce approximations based on physical considerations. −i8U 2 (ρij 2 ρji + 2ρij mji mij ∗ + 2ρij zi zj mij ∗ + ρ2 zj zi ∗ + Σ< (ti . Σ> (ti . tj ) = Σ11 (ti .31) 2 −i8U (2ρij mji ρji + 2ρij ρji zi zj + mji 2 mij ∗ + Σ< (ti . tj ) = (8. S21 (ti .8. 8. tj ) {zi → zj .2 Boltzmann equations From previous sections it can be observed that the equations of motion at secondorder are quite involved: they are nonlinear and nonlocal integro diﬀerential equations. (8.30) N2 2mij zi zj mji ∗ + 2mij ρij zj zi ∗ + 2ρji mij zi zj ∗ + mij 2 zi ∗ zj ∗ ). mji → mij ∗ } . ρij → ρij } .32) 12 N2 2mji zi zj mij ∗ + 2ρij mji zj zi ∗ + 2mji ρji zi zj ∗ + mji 2 zi ∗ zj ∗ ). tj ) = Σ12 (ti . tj ) = 12 and S22 (ti . 62] the authors used these equations as their starting point before applying the Markovian approximation.34) (8. tj ) = S11 (ti . Close to equilibrium (when the external potential induces disturbances much longer than the lattice spacing and frequencies much smaller than the characteristic particle energies) and in the weakly interacting regime.35) (8. Σ21 (ti . which in a classical picture corresponds to assume that be the time between collisions (or mean free path) is long compared to the reaction time (or scattering length). 62]. tj ) = S12 (ti . we expect that a reasonable assumption would be to consider the kinetic scale larger than the quantum one.2 Boltzmann equations Σ> (ti . [59. 60. One of them is to recognize two time scales in the system.
˜ G( ) (ti .37) (8. (8. Since hereafter we will use the gaugetransformed functions exclusively. (8. ω). tj ) = H T+ τ 2 q R+r/2 (8. 4Ja2 s vi 2 − Vi . (8. tj )e−iθj (t )σz . (8. G . T− τ 2 R−r/2 1 ≡ −i 2πM dωe(iqr−iωτ ) H(Rq. [139] we introduce the gauge transformation zi (t) = eiθi (t) G ( ) (ti . tj ) = G( ) T+ 2 R+r/2 2 R−r/2 1 ≡ −i dωe(iqr−iωτ ) G( ) (Rq. v s and µ are expected to be slowly varying functions of R = (i + j)/2. and peaked about the zeros of r = (i − j). (8.39) (8. T. In contrast to normal systems were there is not a condensate. T− G( ) (ti . tj ) = e iθi (t)σz no (ti ).45) . To obtain the kinetic equations we Fourier transform the functions G and no with respect to r and τ . tj ) = ( δij (iσz ∂t − ∂t θi ) − δij Vi + J(eiσz ∆θi+1/2 δi+1j + e−iσz ∆θi−1/2 δi−1j ) δ(t − t ).Chapter 8 148 From the 2PICTP approximations to kinetic theories and local equilibrium solutions Using this approximation it is possible to recast the quantum dynamics into the much simpler forms of kinetic theory. The Fourier transform reads τ τ . Following Ref. in condensed systems making the scale separation requires ﬁrst to do a gauge transformation which makes it easier to identify (and coarsegrain away) the fast variations induced by the rapid change of the condensate phase.43) .40) ˜ After the gauge transformation the variables no . (8. The equations of motion are invariant under the phase transformation if we −1 ˜ −1 replace Do by Do : ˜ −1 Do (ti .42) T = (t + t )/2. Extending these identiﬁcations to the nonequilibrium system we deﬁne the chemical potential and superﬂuid velocity as 2 µi = − ∂t θi − s vi = 2J∆θi a2 .44) 2πM q H(ti .38) where no (ti ) and θi (t) are real. in the following the tildes will be dropped to simplify the notation. In equilibrium. T. ω). τ = (t − t ). ∆θi ≡ (θi+1/2 − θi−1/2 )/a is related to the superﬂuid velocity and the time derivative of the phase to the chemical potential.41) necessary conditions to derive Boltzmanntype kinetic equations.
A . the presence of the condensate gives non zero values to the oﬀ diagonal terms of ( ) the functions G12 (Rq. (8. in Eqs (8. B] = ∂A ∂B ∂A ∂B − + ∂R A∂q B − ∂q A∂R B.46). we get the following equations: i i −1 −1 Do − S + γ H = − Do .(8. T ) with quasimomentum q and energy ω. ω). Σ( ) 2 with −1 Do (Rq.48) 2 4 i i i −1 A = − Do . T ) is just related to the condensate density of atoms at the space time point (Ra. T. H 2 2 i i −1 −1 H Do − S − γ = − H. which is not explicitly written for simplicity.(8.46) In Eq.478. T. H] .51) Because all the quantities are slowly varying functions of R and T . G( ) + Σ. S] − [H. T. G11 (Rq. ∂ω ∂T ∂T ∂ω (8. the quantity no (R. G( ) − Σ .2 Boltzmann equations Neglecting secondorder variations of no (Rq. the twopoint function (<) G11 (Rq. G 2 i i −1 G( ) Do − Σ − Γ − G− A 2 2 i G. Γ + A.52) . ω) = 2πM (I + σx ) no (R. T. Σ + 2 2 1 1 − G( ) .[139] and assuming a (q. It can be interpreted as the density of noncondensed particles with quasimomentum q (>) and energy ω at the lattice site R and time T . In particular when the inverse propagator Do and the self energies vary very little as Ra is changed by a characteristic excitation wavelength or T is changed by an inverse excitation energy. 4 4 i i −1 Σ( ) = − G( ) .44). The generalized Boltzmann equations can be obtained as the Fourier transform of the equations of motion for the case in which the variations in R and T are very −1 slow.8. The brackets denote the generalized Poisson brackets deﬁned as: [A. T ) dependence of the variables. G( ) + 2 2 2 1 1 ( ) (8.47) 2 4 i 1 + [H. H] + [γ. T ). R. As opposed to a normal system. On the other hand. T.49) + Γ. T. T )δ(ω)δq0 . Σ( ) . i 1 [ S. 149 (8. ω. Do + G( ) . Following Ref. γ] . ω) ≡ (σz ( ω − vs 2J sin(qa)) + (2J cos(qa) + µR (T )) I) . We refer to them as the anomalous contributions to the respective two point functions. (8. ω) is essentially the density of states available to a particle that is added to the system at (Ra.50) we approximated the discretized equation to continuous diﬀerential equations.50) 4 4 + (8. Do 2 2 i −1 G+ Do − Σ + Γ G( ) − Σ( ) 2 i Σ( ) . ω) corresponds to the well known Wigner distribution function [140]. ω) we can approximate it as H(Rq. In Eq. (8.
50). T. T. A − [Γ. dω γ(Rq.54) (8.Chapter 8 150 From the 2PICTP approximations to kinetic theories and local equilibrium solutions In the equations we have also introduced the following functions: S(Rq. 2π ω − ω dω Γ(qR.55) (8. T.60). T. ω) + S B (Rq. they include collisional integrals to describe binary interactions.63) are our passage to the Boltzmann equations.53) (8. tj ). 2 2 i − G− A Π= (8. ω) = Σ S B (Rq. S HF (Rq. (8. T.57) (Rq. T. T. can be rewritten in term of statistical and spectral functions as: i −1 Do − Σ + Γ F 2 i G+ A 2 i 2 − Π − = (8.58) (8. Do − Σ − [ G. T. 2 Σ> + Σ< . ω).48) and (8. ω). F − Π. ω) + Σ (Rq. They describe the state of the gas at a given time. (8.59) Π = Eq. Γ(ti . T.(8. Π . 2π ω−ω B with P denoting the Cauchy principal value and γ(Rq. S (ti . Do − Σ − Γ − G − A. ω). 2 (8. T. ω). T. ω) = S HF (Rq. T.49)and Eq. The ordinary Boltzmann equation emerges from the approximation in which the self energies that appear on the left side of Eqs.61) 2 i i i −1 − F. T. ω) and A(Rq. If we deﬁne the statistical variables as: F = G> + G< .47).(8. ω) = P G = P HF (8. −1 A. ΣHF (Rq. ω ) . ω).48) and .62) (8. To progress further we can introduce even more simpliﬁcations based on physical assumptions. T. tj ) and A(ti . 2π ω − ω dω A(Rq. (8. tj ) respectively. ω) .63) Eqs. ω) understood as Fourier transforms of the functions HF γ(ti . 2 2 2 −1 Do − Σ.47). ω) = P ΣB (Rq. tj ). ΣHF (ti . (8. G + A . In contrast to the HFB equations. G] . T.60) F i −1 Do − Σ − Γ 2 i i −1 Do − Σ + Γ. tj ). Γ(Rq.56) (8. (8. ω) . Σ(Rq. Γ] . i 2 i −1 A Do − Σ − GΓ = − 2 −1 Do − Σ A − Γ G = − (8.
= 0. In this way.70) (8.66) . F 2 2 2 i −1 −1 Do − Σ A − Γ G = − Do − ΣHF .71) (8. A . In the derivation of the ordinary Boltzmann equations. we can treat the left and the right hand sides in diﬀerent ways. These two appearances of the selfenergy play a diﬀerent physical role in the description of the dynamics [137].47). how the collisions transfer particles from one energymomenta conﬁguration to another.75) Σ F −Π G Σ A−Γ G −1 Do −1 Do −1 T r Do − S HF . This is a reasonable assumption in dilute weakly interacting gases in which the duration of a collision is very short compared to the essentially interactionfree dynamics between isolated collisions.60). Because these two eﬀects are physically so diﬀerent.73) (8.72) (8.64) .F − ΣHF . = −T r(γH).(8.2 Boltzmann equations 151 (8.65) . the self energies on the left describe the quantum kinetic eﬀects due to interactions. i. (8. (8. = −T r(ΓF − ΠA).68) . H 2 2 i i −1 −1 H Do − S − γ = − H. On the other hand.74) (8. how interaction eﬀects change the energy momentum dispersion relations from that of free particles to a more complicated spectrum.63) are handled diﬀerently from those which appear on the right. they can be simpliﬁed to : Tr Tr Tr −1 Do − −1 Do − −1 Do − S H = 0. Do − ΣHF 2 2 2 i −1 −1 A Do − Σ − GΓ = − A.48) and (8. one completely neglects all the kinetic eﬀects in the secondorder self energies (the dependence on T and R in the secondorder self energy terms on the right hand side) and retain dynamical eﬀects (T and R dependence on the left hand side).67) .8. (8.e. (8. i. Do − S HF 2 2 i i i −1 −1 Do − Σ + Γ F − Π G+ A = − Do − ΣHF .60)(8. (8. A 2 i i i −1 −1 F Do − Σ − Γ − G − A Π = − F.69) If we take the trace of the sum and the diﬀerence of each one of the above equations with its hermitian conjugate. (8. The self energies on the right hand side describe the dynamical eﬀects of collisions. H Tr Tr −Σ HF . Do − ΣHF 2 . (8. Eqs. = 0 (8.(8..e. = 0. we get the familiar Boltzmann equations which describe the particles as free particles in between collisions. Neglecting kinetic eﬀects in the secondorder self energies.63) can be approximated to i i −1 −1 Do − S + γ H = − Do − S HF .
A .72.63) with its transpose we also get: 1 Im r D−1 − SHF . To close this set of equations we need an equation of motion for the superﬂuid velocity which can be found from the deﬁnitions Eq.Chapter 8 152 From the 2PICTP approximations to kinetic theories and local equilibrium solutions ∗ Moreover.(8. when the system has never been disturbed it remains in its equilibrium state. 8. to the linear equilibrium solutions discussed in chapter 5 upgraded with secondorder corrections in U not included in the quadratic approximations. when the system has had suﬃcient time to relax after an applied perturbation. They describe the quantum properties of the gas which is evolving according to the Boltmaznn equations (8. the functions G and H are completely independent of R and T . The equations have to be solved self consistently for any analysis. Equations (8.(8.78) o o 2 1 Im r D−1 − S H = − Re r D−1 − SHF . (8.75) and (8. in the absence of any external potential.81).82) Eqs.60) to (8. Firstly. (8.81) o o 2 with Re and Im denoting the real and imaginary parts.81) together with Eq. (8.75) and (8. H + (γH) .80) o o 2 1 Im r D−1 − Σ A − Γ G = − Re r D−1 − ΣHF .78) are usually called gap equations. if we deﬁne the operator rM = M12 + M21 and apply it again to the sum and the diﬀerence of each one of the equations (8. In this case the generalized Poisson brakets terms are zero and Eqs.3 Equilibrium properties for a homogeneous system In this section we will show how the nonequilibrium Boltzmann equations lead.708. A . F + r(ΓF − ΠA) . (8. 8.768.(8.798. F + r(ΓF − ΠA) .708. (8.72) and (8. in a special case. Secondly.76) o 2 1 Re r D−1 − Σ F − Π G = Im r D−1 − ΣHF . H + (γH) . Under the derived formalism the Boltzmann and gap equations form a coupled set of equations which replaces the original dynamics.738.82) form a closed set of equations that describe the state of the gas at a given time.(8. At equilibrium.39) and Eq. (8. (8.77) o o 2 1 Re r D−1 − Σ A − Γ G = Im r D−1 − ΣHF . There are two situations in which we expect an equilibrium solution to come from the Boltzmann equations.79) o o 2 1 Im r D−1 − Σ F − Π G = − Re r D−1 − ΣHF . (8.83) . 8.40) to be: Re r D−1 − S H o = 2Ja2 ∂ ∂vs =− ∂T ∂R (µ + V ) + 4Ja2 vs 2 .78.(8.75) imply that: −1 A Do − Σ − ( G)Γ = 0.
(8. ω) as indicated in Eq.57).84) G(q. ω) given by dω A(q. Since it is always possible to write [137] 1 2 F (q. (8.85) From Eqs. nq (ω) = eβ 1 −1 with β interpreted as ω the local inverse temperature in energy units [137. 139]. ΓF − ΠA = 0. we get from Eq. ω) is determined by A(q.80)) we also get that at equilibrium γ = 0.(8. ω). Eq. (8.86) and (8. (8. In the . (8.88) Π(q. ω) = i and the function i −1 − Do − S − Γ 2 −1 .89) is satisﬁed. ω ) 2π ω − ω i −1 Do − S + Γ 2 −1 G(q. (8. 0).87) can only be satisﬁed if nq (ω) + A(q.8. (8. (8.3.3 Equilibrium properties for a homogeneous system 153 Because G(q. (8. ω) are just delta functions with peaks at values of ω that matches the possible energy diﬀerence which results from adding a single particle with quasimomentum q to the system.74) and (8. (8.86) (8.87) Eqs.73).1 Quasiparticle formalism In the non interacting case the diagonal terms of A(q. ω).90) 8. ω) = P = 1 2 i −1 + Do − S − Γ 2 −1 . ω) is given by i −1 Do − S + Γ 2 −1 A(q. ω) = nq (ω) + 1 2 Γ(q.(8. ω) = then Eq. Since H contains delta functions in momentum and energy at equilibrium.70): µ = −2J + S11 (0. Detailed study of the structure of the self energy indicates that nq (ω) is related to the BoseEinstein distribution.87) are just the mathematical statement of detailed balance. 0) + S12 (0.(8. They just represent the physical condition that at equilibrium the net rate of change of the density of particles with momentum q and energy ω is zero. (8.83) is satisﬁed when A(q.79).
ρq (ω) and mq (ω) deﬁned as the Fourier transform of ρij . However. there are always sharp peaks in A. In the absence of vortices it is always possible to ﬁnd an ensemble in which the amplitudes (uq . ω). −ωq ) q−q qq uq vq = ωq 1 0 0 −1 uq vq . A∗ (q. ω) + 11 HF Mq−q (ω) = Σ12 B (q.84). −ω) q−q qq .96) (8. ω) are not delta functions but instead continuous functions of ω. ω) = ω with Lqq (ω) = −2J cos qa − µ + ΣHF B (q. it is possible to write the matrix components of D−1 as: D−1 (q. The quasiparticle approximation is obtained by considering Γ very small for small values of ω.94) and satisfy the normalization condition uq 2 − vq 2 = 1. using the deﬁnitions of F and A. ω ) . Eq.91) dω Γ11 (q. vq = v−q . −ω). are given by: 2 u2 vq q − ω − ωq + i0+ ω − ωq + i0− A11 (q. ρij and mij respectively (see Eqs. (8. ωq ) M∗ (q.Chapter 8 154 From the 2PICTP approximations to kinetic theories and local equilibrium solutions many body system the energy spectrum is suﬃciently complex so that the diagonal elements of A(q. vq ) are purely real and uq = u−q . ω) L∗ (−q. (8.93) dω Γ12 (q. 1 1 Using the assumption of a very small Γ. The zeros of D peaked are identiﬁed with the quasiparticle energies ωq .95) A12 (q.22)) in terms of quasiparticle amplitudes: . This i −1 assumption implies that D−1 ≡ Do − Σ − 2 Γ is essentially real with only an −1 about which A is very sharply inﬁnitesimal imaginary part.84) it is possible to see that the quasiparticle decay rate is determined by Γ. 2π ω − ω + i The quasiparticle amplitudes uq and vq are the solutions to the eigenvalue problem Lqq (q. These sharp peaks represent the coherent and long lived excitations. q (8. ω) = Finally. ω) Mq−q (q. ω) = 2Im vq uq uq vq − ω − ωq + i0+ ω − ωq + i0− = −2πuq vq [δ(ω − ωq ) − δ(ω + ωq )] . ω) + 1 0 0 −1 − Lqq (q.92) (8.21) and (8. These excitations are called quasiparticles. the matrix elements of the spectral function A. ω) M∗ (q. From Eq. ω) = −A11 (−q.98) A22 (q. ωq ) L∗ (−q. In terms of the quasiparticle amplitudes. A21 (q. ωq ) Mq−q (q. 2π ω − ω + i (8. we can express the matrix components: ρq (ω). ω) = −2Im 2 = 2π u2 δ(ω − ωq ) − vq δ(ω + ωq ) . (8.97) (8. (8. and the identity lim →0 ω−ω +i = P ω−ω − iπδ(ω − ω ). ω ) . which behave like weakly interacting particles. 12 (8.(8.
As a ﬁnal step. . Eq. they reduce to the HFB equations derived in chapter 5 using the quadratic approximation. the constraint n = no + n. to ﬁx the total density to n. 8.(8.105) uq vq (2nq (ωq ) + 1) . q (8. At zero temperature. mq (ω) = 2πuq vq [nq (ω)δ(ω − ωq ) − (1 + nq (ω))δ(ω + ωq )] .3 Equilibrium properties for a homogeneous system 155 2 ρq (ω) = 2π u2 nq (ω)δ(ω − ωq ) + vq (1 + nq (ω))δ(ω + ωq ) .(8. In terms of the quasiparticle amplitudes and setting N = 2.3.108) form a closed set of equations.101) ρq (ω) = 2π u2 (1 q + nq )(ω)δ(ω − ωq ) + 2 vq nq (ω)δ(ω + ωq ) .106) uq −vq . µ = −2J + U no + 2U n + U m.99) (8. For a given density and temperature Eqs. (8.90) then yield: −2J cos(qa) − µ + 2U (no + n) U (no + m) U (no + m) −2J cos(qa) − µ + 2U (no + n) uq vq = ωq (8. q q (8.103) 2 (1 + nq (ωq ))vq + u2 nq .102) (8. . ( 8.100) (8.104) (8.106).2 HFB approximation Under the HFB approximation the matrix Σ and S are just given by ΣHF B and S HF B . q In the HFB approximation.94) and Eq.108) has to be satisﬁed. (8. (8. they can be written as: ΣHF B = U S HF B = U with n = = m = 1 M 1 M 2 (no + n) no + m no + m 2 (no + n) no + 2n m m no + 2n .8.107) (8.
111) − 2U 2 M2 k. (8.110) Bk Bp Aq−k−p + 2Bk Cp Cq−k−p ω + ωk + ωp + ωq−k−p − i . whereas the solution of the HFB equations yields the whole quasiparticle spectrum. where the quantities A. Mq−q (q. ω) = U no + U m + k 2Bk Aq−k + 2Ck Ap + 2Ck Bq−k + 3Ck Cq−k − ω + ωk + ωq−k − i 2 2Ak Bp Cq−k−p + Ck Cp Cq−k−p 2Bk Ap Cq−k−p + Ck Cp Cq−k−p 2U + 2 − M ω − ωk − ωp − ωq−k−p + i ω + ωk + ωp + ωq−k−p − i k.p Lqq (q. The matrix which is to be diagonalized also become intrinsically nonlocal and to solve for a quasiparticle state with quasimomentum q we have to sum over all diﬀerent .112) The inclusion of secondorder terms modiﬁes the structure of the HFB equations. B and C are deﬁned as Ak = u2 .p 2Ak Cp Ck+p + Ak Ap Bk+p + 2Bk Cp Ck+p + Bk Bp Ak+p ωk + ωp + ωk+p .109) . For simplicity we restrict the calculations to the zero temperature case when nq = 0. In terms of the quasiparticle amplitudes the contributions to the selfenergy at secondorder are given by 2U 2 no M 2Ak Bq−k + 2Ck Ap + 2Ck Bq−k + 3Ck Cq−k ω − ωk − ωq−k + i (8. The matrix that we need to diagonalize to ﬁnd the quasiparticle energies depends now on the quasiparticle mode in consideration.3. Ck = −uk vk .Chapter 8 156 From the 2PICTP approximations to kinetic theories and local equilibrium solutions 8. This means that a separate nonlinear problem must be solved for every quasiparticle state.p (8. µ = −2J + U no + 2U n + U m 2Ak Bp Ck+p + 2Bk Ap Ck+p + 2Ck Cp Ck+p 2U 2 − 2 M ωk + ωp + ωk+p k. k 2 Bk = vk . ω) = −2J cos qa − µ + 2U no + 2U n + Ak Aq−k + 2Ak Bq−k + 4Ck Aq−k + 2Ck Cq−k 2U 2 no M ω − ωk − ωq−k + i k Bk Bq−k + 2(Bk Aq−k ) + 4Ck Bq−k + 2Ck Cq−k − ω + ωk + ωq−k − i 2 Ak Ap Bq−k−p + 2Ak Cp Cq−k−p 2U + − 2 M ω − ωk − ωp − ωq−k−p + i k.3 Secondorder and Beliaev approximations When secondorder terms are taking into account the matrixes Lqq and Mq−q become energy dependent.p (8.
113) Mq−q (q. . the equations that we get are the tightbinding version of the ones originally derived by Beliaev [63]: 2U 2 no M 2Ak Bq−k + 2Ck Ap + 2Ck Bq−k + 3Ck Cq−k ω − ωk − ωq−k + i .(8.115) become L(2) (q. ω − ωk − ωq−k − i (8.115) (8. secondorder corrections introduce the manybody scattering matrix in the diagonal terms. uq → 1. the diagonal elements are no longer equal as was always the case in all the quadratic approximations considered in chapter 5. When both HFB and secondorder corrections are included the quasiparticle energies not only are shifted with respect to the BdG quasiparticle energies but they also become complex. ω) = 4J sin2 (qa/2) + U no − U m + (8. (8. The physical meaning is that when the energy denominator in the secondorder terms vanishes a real decay process is energetically allowed. The imaginary part that the quasiparticle energies acquire comes from the poles of the secondorder terms and it is associated with a damping rate.121) shows that Lqq (q.117) k (2) Comparison with Eq. If we omit the secondorder terms containing no condensate amplitudes. (5.8. This can be seen by considering the secondorder Beliaev corrections in the particle approximation. In this case the secondorder terms in Eq. the HFB approximation has the problem that it is not gapless. It was shown by Beliaev that when secondorder Beliaev contributions are included the gap problem disappears. Finally. The reason is that as m introduces the manybody scattering matrix in the oﬀdiagonal terms. ω) = − qq 2U 2 no M 1 . .116) As discussed in chapter 5. The damping mechanism in which a quasiparticle decay into two of lower energy is known as Beliaev damping and was calculated by Beliaev in the case of a uniform Bose superﬂuid [63]. ω) is the second order contribution to the many body scattering matrix.114) 2n Ak Aq−k + 2Ak Bq−k + 4Ck Aq−k + 2Ck Cq−k 2U o M ω − ωk − ωq−k + i k Bk Bq−k + 2(Bk Aq−k ) + 4Ck Bq−k + 2Ck Cq−k − ω + ωk + ωq−k − i µ = −2J + U no + 2U n + U m. (8. ω) = U no + U m + − k 2Bk Aq−k + 2Ck Ap + 2Ck Bq−k + 3Ck Cq−k ω + ωk + ωq−k − i Lqq (q.3 Equilibrium properties for a homogeneous system 157 quasimomenta. vq → 0.
. We also showed that at equilibrium the full secondorder 2PI equations reproduce the secondorder corrections to the self energy well known since Beliaev. actually reduce to the standard kinetic theory solutions when the scale separation assumption is made.Chapter 8 158 From the 2PICTP approximations to kinetic theories and local equilibrium solutions 8. we showed in this chapter that the complicated non local non Markovian secondorder solutions derived under the 2PICTP formalism.4 Conclusions In summary.
1) (9. 9. In this chapter we use perturbation theory to describe the basic properties of the Mott phase.2) (9.3) (ˆ† an+1 + a† an ). the trapped system is studied ass