ABSTRACT
Title of dissertation: CONSTRUCTION OF APPARATUS
AND FIRST EXPERIMENTS INVESTIGATING
DYNAMICS OF BOSE-EINSTEIN CONDENSATES
IN DISORDERED OPTICAL LATTICES
Emily Elisa Edwards, Doctor of Philosophy, 2009
Dissertation directed by: Professor Steven L. Rolston
Department of Physics
Since the experimental achievement of Bose-Einstein condensation (BEC) in
dilute gases, ultra-cold atom systems have proven to be an unparalleled test bed for
condensed matter phenomena. With this in mind, our laboratory set out to build
an apparatus for the study of the effects of disorder on condensed matter phase
transitions using a BEC loaded into one, two, and three-dimensional lattices. My
thesis is divided into two main sections. In the first section I describe in detail
the design and construction of our apparatus. Our system is designed to form
87
Rb condensates of approximately 10
5
atoms. We have three possible experimental
science chambers.
1. One can perform 1D lattice experiments in the chamber where the con-
densate is formed. There is a mirror located in vacuum, which is suitable for this
purpose.
2. In a glass chamber one can do 1D, 2D, or 3D lattice experiments. Atoms
are loaded into a dipole trap (optical tweezer) prior to condensation and transported
approximately 20 cm to the glass cell where optical evaporation is performed to form
a BEC.
3. One can transfer, using the same techniques as (2), to a chamber with a
multi-channel plate detector for Rydberg atom experiments.
The experimental results described in my thesis pertain to situation (1). How-
ever, I describe some details of (2) in the construction section. The first experiment
presented in my thesis demonstrates the effect of disorder on the time-dependent dy-
namics of lattice systems. We observe that a small perturbation produces a dramatic
change in the adiabaticity criteria for loading a BEC into one-dimensional optical
lattice. I conclude with experiments that we expect to perform on this apparatus.
CONSTRUCTION OF APPARATUS AND FIRST EXPERIMENTS
INVESTIGATING DYNAMICS OF BOSE-EINSTEIN
CONDENSATES
IN DISORDERED OPTICAL LATTICES
by
Emily Elisa Edwards
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
2009
Advisory Committee:
Professor Steven L. Rolston, Chair/Advisor
Professor Kristian Helmerson
Professor Victor Galitski
Professor Christopher Monroe
Professor Amy Mullin
c? Copyright by
Emily Elisa Edwards
2009
Dedication
To my labmates.
ii
Acknowledgments
I would like to thank my advisor, Prof. Steve Rolston for the opportunity to
build an experiment and for mentoring me as a graduate student. Also, I would like
to acknowledge and thank my lab mate Matthew Beeler, who has been my partner
in this adventure and who is equally responsible for the apparatus. Additionally, I
would like to acknowledge to the following previous and current colleagues who have
contributed to this intellectual endeavor: Brendan Wyker, Dr. Kevin Teo, Jennifer
Robinson, Ilya Arakelyan, Dr. Ian Spielman, Dr. Trey Porto, and Dr. Zhao-Yuan
Ma. Additionally, I thank the group members, past and present, of the Rolston,
Orozco, Monroe, and NIST laser cooling research groups for consultation and for
lending equipment to us. Also, I would like to thank my friends and family for their
support.
This work was funded by NASA, ARO, and NSF.
iii
Table of Contents
List of Figures vi
List of Abbreviations vii
1 Introduction 1
1.1 Bose-Einstein Condensates . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Overview of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Construction 5
2.1 Vacuum System: UHV side . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Experimental Sequence for Creating a BEC . . . . . . . . . . . . . . 10
2.3 A Source of
87
Rb atoms . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Getters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.2 Rubidium Oven . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.3 Zeeman Slower . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.4 Trapping atoms: Lasers and Magnetic Fields . . . . . . . . . . . . . 20
2.4.1 Magneto-optical trap (MOT) and optical molasses . . . . . . . 20
2.4.2 Optical pumping . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.3 Magnetic confinement in a Ioffe-Pritchard type trap: Z-trap . 26
2.4.4 Coils: Bias and trimming fields . . . . . . . . . . . . . . . . . 28
2.4.5 Circuitry and power supplies . . . . . . . . . . . . . . . . . . . 29
2.4.6 Lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.5 Radio-frequency (rf) evaporation . . . . . . . . . . . . . . . . . . . . 40
2.6 Moving atoms: optical tweezer . . . . . . . . . . . . . . . . . . . . . . 43
2.7 Optical lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.8 Imaging atomic clouds and time-of-flight . . . . . . . . . . . . . . . . 47
2.9 Computer Control and Data Acquisition . . . . . . . . . . . . . . . . 48
3 Adiabaticity and Localization in Incommensurate lattices 51
3.1 Introduction and Motivation . . . . . . . . . . . . . . . . . . . . . . . 51
3.2 Disorder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3 Band Structure for Single Lattices . . . . . . . . . . . . . . . . . . . . 54
3.4 Experimental details and results . . . . . . . . . . . . . . . . . . . . . 57
3.4.1 Optical Lattices . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4.2 Lattice calibration . . . . . . . . . . . . . . . . . . . . . . . . 59
3.4.3 Sudden release of atoms from a lattice and incommensurate
potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.4.4 Adiabatic turn-off . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.5 Modeling results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.5.1 Band structure to approximate incommensurate lattices . . . . 64
3.6 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.6.1 Adding interactions: 1D Gross-Pitaevski equation . . . . . . . 73
3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
iv
3.8 Comments and Future directions . . . . . . . . . . . . . . . . . . . . 77
3.8.1 Incommensurate lattices as a model for disorder and finite-size
effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.8.2 Excited state populations . . . . . . . . . . . . . . . . . . . . 78
3.8.3 Band-mapping . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4 Final remarks 81
4.1 Non-equilibrium dynamics and the effect of lattices on the formation
of BECs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.2 Bloch Oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.3 Delta-kicked harmonic oscillator . . . . . . . . . . . . . . . . . . . . . 84
A Inventory 87
A.1 Vacuum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
A.2 Lasers, Fibers, AOMs, EOMs, Optoisolators . . . . . . . . . . . . . . 88
A.3 Power Supplies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
A.4 Miscellaneous parts and notable electronic bits . . . . . . . . . . . . . 89
Bibliography 91
v
List of Figures
2.1 Vacuum system: UHV side . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Vacuum System: Source and slower schematic . . . . . . . . . . . . . 16
2.3 Zeeman Slower circuitry . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Wire traps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 View from the bottom of the UHV chamber . . . . . . . . . . . . . . 30
2.6 IGBT circuitry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.7 Atomic transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.8 Layout for Repumper Laser set-up . . . . . . . . . . . . . . . . . . . 37
2.9 Layout for Trapping Laser set-up . . . . . . . . . . . . . . . . . . . . 38
2.10 Layout for optics in Zeeman slower, Probe beam, and Slower repumper 39
2.11 Evaporation sequence and images of BEC formation . . . . . . . . . . 42
2.12 Computer control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.1 Energy bands for a single lattice . . . . . . . . . . . . . . . . . . . . . 55
3.2 Lattice orientation in chamber . . . . . . . . . . . . . . . . . . . . . . 60
3.3 Absorption images of a BEC loaded into an incommensurate lattice . 63
3.4 Absorption images of results of adiabaticity test . . . . . . . . . . . . 65
3.5 Single particle band structure and calculated wavefunction of single
and combined lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.6 Using single particle solutions to calculate population and depletion
of ground band population . . . . . . . . . . . . . . . . . . . . . . . . 71
3.7 Comparison of 1D GPE calculations with experimental images . . . . 76
3.8 Growth of excited band populations . . . . . . . . . . . . . . . . . . . 79
vi
List of Abbreviations
Rb Rubidium
BEC Bose-Einstein Condensate or Bose-Einstein Condensation
UHV Ultra-high vacuum
MOT Magneto-optical trap
psd phase space density
IGBT insulated gate bipolar transistor
NASA National Aeronautics and Space Administration
ARO Army Research Office
NSF National Science Foundation
NIST National Institutes of Standards and Technology
JQI Joint Quantum Institute
vii
Chapter 1
Introduction
1.1 Bose-Einstein Condensates
A comprehensive review of the history and theory of Bose-Einstein conden-
sation (BEC) is given in [48, 12]. BEC was predicted in 1925 by Albert Einstein
using the statistics of photons developed by Satyendra Bose and extending it to
atoms. In the decades to follow, London suggested the existence of BEC in super-
fluid helium (He) which has been observed in neutron scattering experiments (see
Refs. in [48, 12]). Additionally, superconductors are also a system exhibiting BEC
behavior [33]. However, BEC could not be well-studied in these systems. BEC was
first achieved experimentally at near zero temperature in dilute gases of alkali atoms
in 1995 [1, 14]. Because of the robust nature of cold-atom experiments, BEC has
proven to be an invaluable source of research connecting the atomic physics realm
to condensed matter, fluid dynamics, quantum information, and even cosmological
theories. Here, I briefly describe the essential background theory necessary to the
understanding of our experiment.
The phase transition to BEC from a thermal state is a result of the statistics
of bosons (integer spin particles). The most succinct description is ?a macroscopic
occupation of the ground state.? For a system of non-interacting bosons, the occu-
1
pation number of a state with energy epsilon1
?
is
n(epsilon1
?
)=
1
e
(epsilon1
?
??)/k
b
T
?1
(1.1)
In Eq. 1.1 , T is temperature, k
b
is Boltzmann?s constant, and ? is the chemical
potential. Unlike fermions, bosons can occupy the same state and it was predicted
that for sufficiently low temperatures and high densities, a thermal ensemble will
begin to have a large number of atoms occupying the lowest energy level of the
system. One can calculate a critical temperature for this transition, T = T
c
, by
calculating the number of atoms in excited states of a uniform gas confined to a
cube of volume, V, using
N
ex
=
integraldisplay
?
0
depsilon1g(epsilon1)n(epsilon1
?
) (1.2)
where g(epsilon1) is the density of states. For the cube, the density of states is
g(epsilon1)=
VM
3/2
2
1/2
pi
2
?h
3
epsilon1
1/2
(1.3)
where M is mass and ?h is Planck?s constant [48]. By substituting Eq. 1.1 and Eq.
1.3 into Eq. 1.2 and solving for T
c
when ? = 0, one obtains
T
c
?
3.31?h
2
n
2/3
M
(1.4)
Here, n is the density of particles per unit volume , N/V. The phase space density
(psd) is the number of particles in a volume whose length scale is the thermal
deBroglie wavelength, ?
dB
= (2pi?h
2
/Mk
b
T)
1/2
, and is the experimental parameter
we optimize. One derives that the phase space density (psd) condition for BEC
(uniform gas) is n?
3
dB
? 2.612.
2
The critical parameters are slightly modified if we use a harmonic confining po-
tential, having trapping frequencies ?
x
,?
y
,?
z
, which is our experimental situation.
The density of states in this case is
g(epsilon1)=
1
2?h
3
?
x
?
y
?
z
epsilon1
2
(1.5)
Solving for T
c
one finds that the critical temperature depends on the harmonic
trapping frequencies and the number of atoms, N.
T
c
?
0.94?h
k
b
(?
x
?
y
?
z
)
1/3
N
1/3
(1.6)
Typical transition temperatures are sub-?K (in our system down to around 100 nK)
for dilute gas systems. Densities for these cold atom systems are around 10
13
?10
15
cm
?3
. For more details see Refs. [48, 12].
Although dilute gas BECs were first thought to be an achievement in atomic
physics with little relevance to other fields, they have emerged as tools for the
study of a wide variety of interesting physical problems. For example, there is
phase coherence across the condensate, which allows for diagnostic techniques and
experiments in matter-wave interference, and continued research in developing an
atom laser, a coherent source of matter waves. Additionally, a BEC in a dilute gas
can be repeatably created using current technologies in under a minute which makes
studying them very robust.
It was realized very quickly that BECs as sources of coherent atoms, coupled
with periodic optical potentials (lattices), allows for the study of many condensed
matter problems. The first clear demonstration of a condensed matter, many-body
problem studied with BEC was the observation of the superfluid to Mott insulator
3
transition [26]. Much of the current BEC research is aimed at the study of vari-
ous condensed matter phenomena, which are otherwise difficult to approach with
solid state experiments[45, 8]. However, they are also used for studying quantum
computing with neutral atoms [39] and look promising for the study of many-body
dynamics in non-equilibrium systems [64, 20, 23] .
1.2 Overview of Thesis
The work described in this thesis is divided into two main parts. The first
section is an overview of the construction of the BEC apparatus. I include the most
details on the parts that I was personally involved in building, characterizing, and
optimizing. In Chapter 3, I present an experiment studying the time-dependent
dynamics of loading a BEC into a pseudo-disordered potential. Chapter 4 gives an
overview of experiments that I have been involved in, but are in the early stages of
development. The Appendices are helpful references for the user of the apparatus.
4
Chapter 2
Construction
In the beginning...There was an empty lab. The following is a description of what
happened next.
As described in Chapter 1, theoretically there are only a few requirements for
achieving a BEC in a harmonic trap: a bosonic sample with phase space density
approximately equal to 1. In practice, there are many technological considerations
that accompany this experiment. Here, I describe the main details of the design of
our experiment. In each section, I explain the relevance to condensate formation.
All of this work is performed under ultra-high vacuum (UHV) conditions, so creating
this environment is the experimentalist?s first task when embarking on BEC research.
Then I will describe the theory and implementation of each stage of BEC formation.
Appendix A gives a table with vendors and part numbers.
2.1 Vacuum System: UHV side
In order to create a suitable environment for the production of a BEC we
needed to design a system with pressures on the order of 10
?11
torr. This stringent
requirement is a consequence of the time it takes to achieve BEC. In a magnetic
trap, forced radio-frequency (rf) evaporation typically takes 20-30 seconds, which
translates into needing a vacuum-limited lifetime at least comparable and preferably
5
longer than this time. Background collisions severely limit the lifetime of trapped
atoms, thus the lower the pressure the better. A quick calculation, assuming no
other loss mechanisms, shows that the vacuum limited lifetime of a magnetic trap
is of order 200 s at 5 ? 10
?11
torr (lifetime? 10
?8
/P
torr
) [42]. The following is a
description of our vacuum system in its current form.
The system, although originally a single chamber, was modified to include
two chambers: the source side and the ultra-high vacuum (UHV)/science side. A
schematic depicting the setup of the UHV/science-side with photos is shown in Fig.
2.1.
UHV is maintained in the science side of the chamber by a combination of a
150 L/s Varian Starcell (diode) double-ended ion pump and a titanium sublimation
pump (Ti-sub), which is situated inside of the ion pump. We have two types of
pumps because the ion pump is efficient at lowering the pressure due to most gases
through ionization, whereas the Ti-sub pump releases titanium which is effective at
pumping molecules like H
2
and H
2
O. The Starcell type pump is an option which
increases the efficiency of the pump for removing noble gases. The pressure is
monitored by a dual filament nude Bayard-type ion gauge (Granville-Phillips). It
should be noted that the ion pump is a poor pump for rubidium and will eventually
fail after years of use. Also, the ion gauge will pump rubidium but also act as a
diffusive source via outgassing, so it is not a very reliable measure of exact pressure.
The best test of vacuum quality is the lifetime of the atoms in the magnetic trap
measured after some evaporation.
The main chamber (where the condensate is formed) is a stainless steel Kim-
6
UHV/science side 
of vacuum system
150 L/s ion pump
ion gauge
science/glass cell
gold mirror/wire traps
Attached to slower 
(to source side)
AlN
block
Figure 2.1: Layout of UHV side of the vacuum system with photos of
the custom design feedthrough. The wire traps are shown with a false
translucent gold mirror overlaid to clarify the physical location of the
actual gold mirror.
7
ball Physics spherical octagon. The windows are anti-reflection coated (Spectrum
Thin Films) for either 780 nm or 1550 nm light depending on their intended purpose.
Most of the vacuum parts in both the source side and the science side are standard
issue, with the notable exception of two custom feedthroughs (Kurt Lesker, Inc.),
one in the main chamber and one in the source side. The special feedthrough in the
UHV side is a 6 inch con-flat flange that has both copper and water feedthroughs.
There are six-1/4 inch diameter copper feedthroughs for current to the wire traps
as well as a multi-pin connector, which was included originally for transmitting ra-
diofrequencies through an in-vacuum antenna. This antenna broadcasted to the
laser locking electronics, and so we opted for an external rf-coil instead. This un-
usual behavior was not understood and may not be present anymore because the
original Toptica DLX-110 laser has been refurbished (upgraded to a ?rock-solid?
configuration), which dramatically increased its stability. In addition, we replaced
the homemade locking circuitry with a commercial lock box (Precision Photonics).
There is an input and return stainless steel water feedthrough, which is at-
tached to a steel can inside the vacuum. This is intended to help cool the copper
block on which the aluminum nitride (AlN) bed for the wire traps (described in
Section 2.4) is situated. Good thermal contact between each piece of the assembly
is crucial for the traps to be efficiently cooled. We have recently observed that when
the cooling and heating of the traps during the experimental duty cycle is extreme,
the thermal stress can weaken the mechanical connections and potentially lead to
damage (the gold mirror breaking). The structure with the traps extends vertically
through a 6 inch flange, 4-way cross into the spherical octagon chamber (photos
8
shown in Fig. 2.1).
Achieving very low pressures requires care, and is made more difficult by intro-
ducing structures into the chamber. All parts of the attachments to the feedthrough,
with the exception of the gold-coated wafer and an AlN piece, were cleaned by hand
with vacuum grade acetone and Kimwipes R?, and using an ultrasonic bath, repeated
multiple times. The cleaning of the copper parts included an acid dip made of 3:1
mixture of water/12 M sulfuric acid. Acid type dips are suitable for cleaning OFHC
(oxygen-free high conductivity) copper, but caution should be taken for cleaning
other materials using this method due to the risk of etching. Vacuum equipment
ordered new was not cleaned, since typically these pieces are baked at the factory.
All gaskets were wiped with acetone. In an assembly such as ours, where many
materials are put under vacuum, gaskets and factory baked equipment are usually
not the limiting factors in achieving low pressures. All assembly screws are vented
to avoid trapped pockets of gases. Once the system was assembled (in its original
configuration without the slower/oven) we roughed out using an adsorption pump
until the ion pump could be turned on (typically around 10
?5
torr). After the pres-
sure lowered to 10
?8
torr , we wrapped the chamber in heater tape and aluminum
foil and baked at just under 150
?
C for 1-2 weeks. The limiting factor on the baking
temperature is the anti-reflection (AR) coatings on the windows. We take special
care not to create any large temperature gradients in the baking process due to the
fragile ceramic feedthrough, the coated windows, and the glass science cell (because
of the seal, see below). Once we added the Zeeman slower, the baking procedure
became more complex. The stainless steel 14 inch long nipple (inside tube that the
9
slower coils are wound on) is wrapped in heater tape and terminates at the gate
valve/differential pumping region. Very recently, upon breaking vacuum it became
critical for maintaining low pressures to bake the gate valve area. The maximum
temperature for this is 150
?
C when the valve is open and 120
?
C when it is closed
(limitation is the O-ring).
The science chamber is a commercially available 4.5 ? 2.2 ? 1.2 cm
3
quartz
cell. The cell is fused (by the glass blower at NIST, Gaithersburg) to a glass cup that
is sealed to a stainless steel 2.75 inch con-flat flange (Bomco). The glass to metal seal
on the flange is sensitive to both applied pressure changes and temperature gradients.
(When it was first attached to the vacuum, the stress from screw tightening cracked
the seal.) Being meticulous paid off, as we had no real difficulty in getting down to
4 ?10
?11
. After the addition of the source side, the UHV routinely is at 7-8?10
?11
torr. In later instances, we used a turbomolecular pump (Pfeiffer Vacuum)/roughing
pump combination for getting to pressures of 10
?6
Torr, before turning on the ion
pump(s).
2.2 Experimental Sequence for Creating a BEC
A typical duty cycle for creating BECs in our experiment is:
? Atomic source
? Collect 10
9
atoms from beam in a MOT
? Optical molasses stage
10
? Optically pump to trapped state
? Collect atoms in weak magnetic trap
? Compress magnetic trap
? RF evaporation to critical point
? Release atoms and take absorption image
These steps will be discussed in detail in the sections to follow.
2.3 A Source of
87
Rb atoms
This section describes the source of atoms we use for forming a condensate.
We have had two sources in our experiment: Rb dispensers (getters) and a Zeeman
slower. I will first discuss the getters, as that was our original source for rubidium
atoms and give some reasoning for why we later implemented a slower.
2.3.1 Getters
The boson we choose to work with is the very popular,
87
Rb. It is the less
abundant isotope of Rb (27 %). As compared with
85
Rb, it has a positive s-wave
scattering length, which means that its interactions are repulsive at very low tem-
peratures. It has been shown that condensates with attractive interactions and
sufficiently large numbers are unstable and collapse [18]. Rb has a single valence
electron so it is easily laser-cooled. Rb is also a good choice because of the com-
11
mercially available lasers with the required wavelengths. It is also inexpensive to
obtain, less than $100/gram.
We originally designed our system to have only a single vacuum chamber.
In this configuration, we used Rb alkali dispensers (SAES getters) as the atomic
source. Rb dispensers are essentially a tungsten envelope containing Rb bound
up in an inorganic compound. Rb is released diffusively from the envelope when
current flowing through the getters reaches a value corresponding to a threshold
temperature (at around 2.7 amps) required to free the Rb from the compound. The
dispensers have been shown to be effective in MOT loading and their turn-on/turn-
off properties have been studied [24, 51] . Our system was equipped with two sets
of three getters, one arranged in series and one in parallel.
In our set-up, using the dispensers proved troublesome to maintaining the
required vacuum pressure. When the getters release Rb into the vacuum, only a
fraction is captured by the MOT. The remaining Rb is pumped out by the walls,
the ion pump, and other surfaces inside the vacuum (i.e. gold mirror). There is
an inherent time constant (on the order of 10 s in our apparatus) associated with
this pumping. In addition, one must gently warm the getters just below threshold
current in order to maintain their cleanliness.
The release of Rb from the getters as a function of current is exponential. After
initially achieving BEC, during the optimization process, we accidentally released
too much Rb into the system which subsequently formed an alloy with the gold
mirror, requiring its replacement. In order to prevent destroying the mirror, we
built an electronic interlock system to shut-off current to the getters. We placed a
12
resistor in series with the output of the power supply and measured the temperature
of the resistor with an AD590 (Analog Devices) sensor. An AD590 is a chip that has
an output of 1 ?A/
?
K. After a few experimental duty cycles, the resistor temperature
reaches a steady state value. If the temperature of the resistor exceeds this value, the
supply is shut-off. After achieving BEC, we spent approximately 6 months trying
to optimize the duty cycle for the getters and eventually concluded that this type
of source did not lead to stable operating conditions conducive to doing reliable
experiments. The use of the single-chamber getter-based system was based on [67].
We believe that our difficulties were probably due to differences in chamber/pump
geometries. At their best operation, we ran the getters in pulsed mode, with a
cooling/pumping stage prior to loading the magnetic trap, leading to a total MOT
loading phase of 20 s. During this process, we began to design a substitute source.
Currently there are getters in the vacuum, which serve as a tool to implement an
electric field. However, they are not connected to operate in dispensing mode.
2.3.2 Rubidium Oven
Rubidium is a solid at room temperature with a boiling point of 688
?
C and
a melting point of 39.3
?
C. Our atomic beam is designed to be a crude version of
the candlestick source [63]. After breaking an argon-packed Rb ampoule (Strem
Chemical) under vacuum by bending a bellows (Kurt Lesker Inc.), we heat it to
105
?
C. The temperature is achieved by wrapping heater tape around the bellows
and the stainless steel ?T? housing the collimator. The oven region is wrapped in
13
sheet copper, insulating fiberglass material, and aluminum foil. The temperature is
measured by a thermocouple and maintained by a servo-controller (Omega).
The diffusive beam is approximately collimated by custom machined copper
apertures. In practice, this acts more as a nozzle, and un-collimated Rb mostly sticks
to a cold-plate. The atoms that do not enter the source chamber (spherical square,
Kimball Physics Inc.) are ideally recycled by a gold mesh via capillary action and
returned to the Rb reservoir. The cold plate is cooled far below the melting point
by an external TEC cooler clamped to a 3/4 inch copper feedthrough. The TEC is
maintaining a large temperature gradient and has one side water cooled for efficient
heat dissipation. When the water cooling stops the TEC cannot maintain constant
cooling of the copper block and the heat from the back of the TEC will heat the
copper cold plate. Since Rb tends to drastically shorten the lifetime of ion pumps,
this cold plate also serves as a preventative measure. Even with these measures, the
failure of the cold plate TEC cooler will lead to catastrophic poisoning of the source
side ion pump. As another measure, we heat the ion pump above the melting point
of Rb in attempt to prevent it from sticking to the walls. A Rb contaminated pump
exhibits high leakage current and eventually a high base pressure. (We keep a spare
30 L/s ion pump and when this damage occurs, we send the contaminated pump
for refurbishing). In short, it is critical to keep the Rb localized and away from the
ion pump or it will contaminate everything.
A schematic of the source side plus the Zeeman slower is shown in Fig. 2.2.
The pressure on the source side is maintained at two to three orders of magnitude
above the UHV side by a 30 L/s Varian-type pump (option diode or triode, Duniway
14
Stockroom). This is sufficient to maintain a 10
?8
Torr vacuum. The pressure on this
side is naturally higher due to the rubidium (Rb) oven. Separating the two sides is
a differential pumping section which is simply an aluminum tube of 8 cm in length
and 3/8 inches diameter. The desired dimensions of the tube were determined by
calculating the conductance. (Conductance is the inverse resistance of flow from one
location to another.) Other than preserving vacuum with a cold plate, separating the
two chambers is a manual gate valve (Kurt Lesker, Inc.). Adjacent to the differential
pumping stage we use a rotating ball-valve, which opens every experimental duty
cycle to allow the atomic beam to pass into the slower and UHV chamber. This ball
valve serves to quickly reduce the conductance of Rb between the source and UHV
chambers. It is affixed to an aluminum rod which screws into a custom feedthrough
(Kurt Lesker Inc.) next to the 3/4 inch copper feedthrough. The ball valve is rotated
via a magnetic rotary feedthrough (magidrive, Kurt Lesker Inc.). Externally, the
rotary feedthrough is attached to a stepper motor which is given ? 5 V to ?open?
and ?close,? respectively.
2.3.3 Zeeman Slower
There are two main alternative sources to getters: a double-magneto-optical-
trap (double-MOT) system and a Zeeman slower. Our associations at the ?birth-
place of the slower,? NIST, led us to choose the latter. This choice also proved
relatively convenient to implement. A brief description of the theory of a slower is
given here.
15
Source side of 
vacuum system
30 L/s 
ion pump
Rubidium ampoule 
inside bellows
Zeeman slower
Gate valve
cold plate
ball valve
dif_ferential 
pumping
To Main Chamber
OVEN
 
ion pump
Gate 
valve
Figure 2.2: Layout of source side of the vacuum system with photos of
the apparatus.
16
A slower uses the combination of two techniques to create a low temperature,
low velocity atomic beam: laser-cooling and the Zeeman shift. For a detailed de-
scription of these ideas, see [42]. If an atom is traveling in a given direction and
absorbs a resonant photon from the direction opposite to its propagation it acquires
a momentum kick, ??hk, where k is the wavevector of the photon. Upon relaxation,
the atom releases a photon, but as the spontaneous emission can statistically be
in any direction, the momentum transfer to the atoms from this process averages
to zero. The net momentum transfer is then ??hk. Following many cycles of the
absorption-emission process, one can imagine that the atomic velocity decreases. If
one applies this technique to a beam of atoms, one can essentially stop the sample
motion in the direction of propagation. The maximum deceleration of an atomic
beam is a
max
=?hk?/2M, where M is the mass and ? is the natural linewidth of
the transition which is 6 MHz for the D2 line of Rb. In the case of
87
Rb, a
max
is
0.113?10
6
m/s
2
(line data information from [62]).
Due to the distribution of the velocities of atoms in a collimated beam,
f(v
z
)=
parenleftbigg
v
3
z
2(k
b
T/M)
2
parenrightbigg
e
?v
2
z
2(k
b
T/M)
, (2.1)
compared with the narrow linewidth of the laser, only a fraction are in resonance
with the photons. Here, T is temperature and v
z
is the velocity along the atomic
beam direction. As the atoms are slowed, the Doppler shift, ?
D
= ?k ? v
z
, pushes
them out of resonance, which makes repeated absorption-emission cycles unlikely.
For
87
Rb at 100
?
C, v
rms
is approximately 360 m/s and ?
D
? -480 MHz. Each
absorption-emission cycle slows the atom by ?hk/m = 0.6 cm/s, which is an ?
D
? -
17
7 kHz. After about 800 cycles, the atoms are shifted by a linewidth. Fortunately,
there are ways to compensate for the Doppler shift, such as continuously changing
the laser frequency in time or shifting the atoms into resonance using a magnetic
field. These methods allow a significant portion of the atomic beam to be cooled
and slowed. We choose the latter of these techniques because chirping the laser
frequency produces a pulsed source of slow, cool atoms and we desire a continuous
beam.
The hyperfine splitting due to the coupling of the nuclear angular momentum
(I =3/2 for
87
Rb) with the total angular momentum of the electron, J, has de-
generate sub-levels. These levels (m
F
) are split in the presence of a magnetic field.
In a weak field, this phenomena is the Zeeman effect and the shift of the levels
is ?
B
m
F
g
F
B, where ?
B
is the Bohr magneton and g
F
is the g-factor. A Zeeman
slower uses the magnetic field dependence of the the levels to maintain resonance
between the slowing atoms and the laser beam. To keep a constant deceleration,
the parameters must satisfy the condition ?
D
+ ?B/?h = ?, where ? = ?
laser
??
0
is the detuning of the laser from atomic resonance (?
0
). Using the relationship
v
2
f
= v
2
i
?2az, one finds that the optimal field profile for slowing an atomic beam
is B(z)=B
0
radicalbig
1?z/z
0
, where z is along the axis of the atomic beam. Decreasing
the rms spread in velocity corresponds to lowering the temperature (i.e. deflection
or deceleration is not cooling). Note, that in the decelerating atom frame, Zeeman
slowing looks like typical laser cooling [42]. Although strictly speaking, temperature
is not a good description because the atomic beam is not exchanging heat with its
surroundings, and thus is not coming to thermal equilibrium.
18
Our slower was originally configured to be a ?reverse? or ?
?
slower meaning
that the atomic beam enters the slower at a point where the magnitude of the field
starts small and increases down the length. The advantage of a ?
?
slower over
a ?traditional? ?
+
slower is that the laser frequency is not resonant with the low
velocity atoms exiting the slower [5]. Many slowers have a ?
?
and ?
+
portion, which
combines features from both types. Because table space was limited and this was
an add-on to the original design, we opted for the ?
?
slower. The optimum slower
laser detuning is set according to where the magnitude of the field is at a maximum.
In practice our reverse slowing technique did not work very well, so we adjusted
the currents to form a field profile optimum for a ??
+
? ?
?
? slower, which as a
zero-crossing in the field profile. The laser detuning, fixed by the value of the field
at the end of the slower, is aligned to pass through the MOT and focus on the oven
aperture. Because resonant light would push the MOT around as it was loaded, the
detuning of the slower frequency is chosen to be many linewidths off-resonant (?110
MHz). The final coil in the solenoid provides compensation for the slower field so
that there is not a significant gradient at the MOT.
The solenoid consists of 7 individual solenoids to allow us to easily modify the
field profile by adjusting each of the coil currents. A linear supply is used to drive
the current in the coils. There is a circuit which divides the current and farms it out
to each individual coil (Fig. 2.3 ). The coils are wound with 14 AWG kapton-coated
flat, copper wire. Even though our slower is quite short, we still get a reasonable
flux of atoms. Due to our ?mirror-MOT? plus Z-trap configuration (see MOT and
Z-trap sections) our capture volume is smaller than more traditional traps, thus we
19
do not require large numbers of atoms. In practice we load 0.5-1 ?10
9
atoms into
our MOT in 5 seconds of slowing. This is compared with 3D MOTs which typically
capture > 10
10
atoms.
2.4 Trapping atoms: Lasers and Magnetic Fields
Once there is a reliable source of atoms in a high vacuum environment, one
requires a method to capture them in free space. There are two main techniques
to do this: Magnetic fields and optical fields. The combination of near resonant
laser light and magnetic fields gives rise to the very popular magneto-optical-trap
or MOT. In this section I will discuss the MOT and the purely magnetic trap as
these are the methods used in the formation of the condensate. In a later section
about optical lattices, I will describe the features of optical traps.
2.4.1 Magneto-optical trap (MOT) and optical molasses
The MOT [50], one of the most powerful tools in experimental cold-atom
physics is a combination magnetic and optical fields to create a trap for neutral
atoms [42]. Our Zeeman slower used laser cooling in one direction. In principle,
one could cool in all three directions, potentially creating a ball of atoms instead
of a slow moving beam in primarily one direction. Suppose one has two counter-
propogating beams with oppositely circular polarized light relative to one another.
An atom moving from the center of the overlap region will preferentially undergo
multiple scattering events with the beam moving opposite to its direction of motion
20
C
E
G
C
E
G
Coil 1
Coil 2
Vcc Vcc
repeat to Coil 7,
 using Voltage 
dividers catered 
to Current needs
Slower 
power supply
*Transistor used: Darlington
R1=0.05 ?, 50 W, monitor  
R1R1
Zeeman Slower Current divider circuit
Figure 2.3: Circuit schematic for driving current in the Zeeman slower coils.
21
because the light will be Doppler shifted towards resonance. As with the case of the
Zeeman slower laser beam, the atom will be pushed. At some point it will see more
resonant photons from the other beam. This situation in three directions will create
a viscous damping of atoms at the overlap region, and is known as optical molasses.
This is not a trap, because an atom moving at non-zero velocity, but displaced from
the center of the overlap region, will not feel a force pushing it towards the middle.
A quadrupole configuration produces a field which increases in all directions
and has a zero at the center. The interaction of the atoms with the field, through the
Zeeman shift, provides a spatially dependent scattering rate resulting in the restoring
force necessary to create the trapping conditions. The combination of these two
techniques creates a harmonic trap (for small displacements) with damping, where
a large sample of atoms can be collected and continuously cooled. A schematic
representation of our MOT geometry and laser beams is shown in Fig. 2.5.
In the science side of the vacuum there are two wire traps: the U-trap and the
Z-trap. This type of design has become quite popular in the form of micro-fabricated
atom chips. We modeled our design after [67]. The U-trap is used in conjuction
with external coils to form a quadrupole field profile for the MOT. A schematic and
picture of this is shown in Fig. 2.4. If one examines a thin current-carrying wire,
the field decreases as 1/r. Adding a bias field, B
Bias
, perpendicular to the wire will
cancel the field due to the wire, thus creating a field minimum at a specific distance
in one direction. Along the line parallel to the wire, the field is a minimum, but
bending the leads to form a ?U? adds a gradient in this direction as well, creating
a three-dimensional trap. In the case of the U-trap, the field lines are at an angle
22
with respect to the idealized quadrupole and so to avoid this, the middle wire is
flattened into a plate [67].
We employ a mirror-MOT which is a modification of the traditional six-beam
MOT [52]. There is an in-vacuum gold-coated silicon wafer situated on top of the
wire-traps. Two beams enter the bottom window of the chamber and reflect from
this mirror at an angle. There are two additional beams counter-propagating, which
enter the chamber from side ports. The combination of these laser fields provides the
3D laser cooling in the MOT (Fig. 2.5) since the beams reflecting off of the mirror
overlap with the capture area prior to and after their reflections. The MOT beams
are detuned from resonance by approximately 20 MHz. This value was determined
empirically by optimizing atom number.
After up to 10
9
atoms are loaded into the MOT in 5-7 s, a second cooling stage
called optical molasses follows. The name molasses refers to the viscosity of the light
field (as described above and in Ref. [9]). The technique is simple: abruptly turn off
all magnetic fields. The atoms are very cold in the MOT and in zero-field, they will
take some time to move out of the region of interest. Additionally, we reduce the
radiation pressure pushing outwards (force due to re-scattering photons) on the atom
cloud by lowering the intensity of the light fields and detuning several linewidths
away from resonance. This stage cools the atoms to 30-40 ?K. The process is
necessary for efficient transfer to the purely magnetic trap (Z-trap), whose depth is
?mK. This step has been optimized to take a few ms. At times longer than a few
ms, atoms move away from the overlap region with the magnetic trap. Additionally,
the temperature at 10 ms is not measurably cooler than at 5 ms. We extract the
23
Wire-traps to form Magnetic f_ields 
A
B
C
D
A = 4 mm
B = 1 mm
C = 2.5 mm
D = 9.75 mm
E = 2 mm
Figure 2.4: Diagram depicting how magnetic traps can be formed from
bending wires. B
b
is a uniform bias field created by external coils (see
Ref. [67]). The dimensions of the wires used in our experiment are shown
on pictures of the traps.
24
temperature (in each direction, x and y) of the molasses by measuring the widths
of the atomic cloud (via absorption imaging, see Section 2.8), ?
x,y
, after different
expansion times, t
1
and t
2
, using
T
i
=
(?
2
2,i
??
2
1,i
)m
k
B
(t
2
2
?t
2
1
)
(2.2)
Here, M is the mass of
87
Rb and i is a direction, x or y. This equation assumes the
equipartition theorem in one-dimension,
1
2
Mv
2
i
=
1
2
k
b
T
i
[65].
2.4.2 Optical pumping
The atoms in the MOT are in a nearly equal distribution of |m
f
? states of the
|F =2? ground state. There are 2F + 1 sub-levels of this state. Only the low-field
seeking states are capable of being confined by the magnetic trap. At low fields
the possible states are |F =2,m
F
=2?, |F =2,m
F
=1?, and |F =1,m
F
= ?1?
of the 5
2
S
1/2
level. We choose the |F =2,m
F
=2? state and the atoms must be
transferred to this state. Presumably, doing this with 100% efficiency would yield
a factor of 5 increase in number of trapped atoms over not having this step. In
practice, we observe a factor of 3-4 in atom number and is probably limited due to
imperfect polarization. We optically pump using ?
+
polarized light (defined along
the quantization axis of the magnetic trap, which is in the B
y
direction) resonant
with the 5
2
S
1/2
|F =2? to 5
2
P
3/2
|F
prime
=2? transition. According to the selection rules
for this transition, once the atoms are in the |F =2,m
f
=2? state the light cannot
transfer them to another sub-level. The power and width of the optical pumping
pulse must be optimized for best results. We find that the optimum time and
25
power to maximize the factor of increase in number of atoms captured from optical
pumping, but minimize heating, is less than 500 ?s and 0.5 mW, respectively (for a
?1 cm size beam).
2.4.3 Magnetic confinement in a Ioffe-Pritchard type trap: Z-trap
At the minimum of a quadrupole field profile, there is a zero in the field and
atoms can be potentially lost from the trap by spin flipping (Majorana losses).
However, if the Larmor precession rate is much faster than relative changes in the
field during an atoms? orbital motion, then this is not a problem. When a large
fraction of the atoms occupy low-lying levels (e.g. due to strong cooling), then the
condition for adiabatic following is not satisfied. Thus, this loss mechanism is a
concern in the evaporative cooling regime near the condensation transition [42].
To avoid Majorana losses, the atoms must be confined to a non-zero B-field.
There are three major methods to accomplish this: an optical plug at the zero,
a time-orbiting potential which moves the zero, or a Ioffe-Pritchard trap which is
designed to remove the zero with an offset bias field. In our experiment, a Z-shaped
wire provides the necessary fields to produce a the Ioffe trap (Fig. 2.4). Starting with
the central bar, the current flowing through it yields a field that falls off as 1/r. The
external B
x
-field serves to cancel the field, creating a magnetic field minimum, which
is zero along the bar . However, the leads on each side (now bent into a Z-shape)
shift the minimum from zero, to prevent Majorana spin flips. This combination by
itself is enough to create a magnetic trap, however, we add a component, B
y
in order
26
to increase depth and field gradient. Our trap has a depth of around 1 mK and a
gradient (when compressed) of approximately 400 G/cm (for details about coils, see
Section 2.4.4) . The trap is anisotropic with a weak direction along the middle bar
of the wire. A diagnostic program calculating the parameters of the magnetic trap
was written by M. Beeler.
Physically, the Z-trap is situated on top of the U-trap, separated by a thin
layer of Kapton R? film (McMaster-Carr), and embedded into an aluminum nitride,
AlN, substrate (Stratamet). This substrate was chosen due to its high thermal
conductivity of 175 W/mK compared with ceramic (29.3 W/mK). It is a soapy
material so care must be taken to isolate it from water.
Following the optical pumping pulse, we simultaneously ramp on the three
required fields for the Ioffe-trap in 5-10 ms. The initial currents were chosen to match
the trap with the position of the molasses cloud. The initial trapping frequencies
are approximately ?
ho
=2pi ?20 Hz. The trap is immediately compressed during
250 ms to frequencies ?
x,y,z
=2pi?(410,120,410) Hz. The frequencies are measured
by abruptly changing the magnetic field, which induces oscillations of the atomic
cloud in the trap. The frequencies are extracted by varying the hold time following
the perturbation of the system. This method works well for the compressed cloud
after some evaporation. However, it cannot be used for the initial trap because of
the large extent of the cloud compared with the image area. Therefore, we use a
magnetic field calculation to determine these values. They are not critical because
all experiments are performed in traps that have been compressed and characterized.
This compression step serves to dramatically increase the collision rate, n?v.
27
Here, n is the density, ? is the s-wave scattering cross-section, and v is the thermal
velocity. This collision rate determines the rethermalization time after each step in
evaporative cooling, thus it is necessary to start with a high density of atoms. In our
system, compression also moves the trap closer to the wires/mirror because higher
frequencies are attained by approximate cancellation of the field at the wire.
There is noise which couples to the magnetic trap and heats the atoms. This
heating is trap bias dependent which suggests that it is rf noise. We have not
discovered the source of this noise and avoid it by changing the trapping parameters
prior to or after achieving BEC, depending on the application. It appears that traps
with a bias above 6-7 MHz heat at a rate of 10-100 nK/s compared to rates of 1-2
?K/s in the lower bias/higher [trapping] frequency traps.
Additionally, for experiments where one decompresses the magnetic confine-
ment, the trapping frequencies pass through 60 Hz and harmonics. This will drive
oscillations which will add position noise on the sample. To minimize this effect, we
abruptly pass through these resonances in a time step chosen to be faster than the
response of the atoms.
2.4.4 Coils: Bias and trimming fields
All coils in our system are external to the vacuum and are for creating ef-
fectively uniform fields at the location of the atoms. There are three main coils,
B
x
,B
y
,B
z
(see Fig. 2.5 for orientation). Both B
x
and B
y
are made of rectangular,
hollow copper tubing that are water cooled. The tube dimensions, including coating
28
are 4.4 ? 4.5 mm
2
. The coils for B
x
are insulated with a fiber-glass type coating
(tan-colored) and are comprised of two coils, each 42 turns (6 rows of 7 turns),
situated around a 2 3/4 con-flat flange on each side of the vacuum chamber. The
B
y
coils are kapton R? film coated (orange-colored), also water cooled with each coil
having 35 turns (5 rows of 7 turns) and oriented perpendicular in the plane with
respect to B
x
. In the original configuration we used air-cooled coils for creating the
B
y
field. After many duty cycles they would come to thermal equilibrium, however
this process took many hours. We attempted to add heat sinks to the coils, but this
did little to improve the time constant. For this reason, they were replaced with
water-cooled copper tubing. The B
z
coil is situated around the bottom window of
the chamber (6 inch con-flat flange), and has 25 turns (1 row, 25 turns) of 20 AWG
kapton R? coated wire (magnet wire, air-cooled).
We have wound ?trim? coils onto the chamber due to residual fields and for
finely tuning the fields at the location of the atoms during different stages, such as
optical molasses. These consist of a few turns (4-6) of 20 AWG magnet wire wound
around the conflat flanges in x, y, and slower?beam directions. The trim coils are
fixed to the chamber with epoxy.
2.4.5 Circuitry and power supplies
We use Sorensen DLM-600 power supplies to run up to 75 A of current through
our wire traps and water-cooled wires. There is an interlock which prevents the user
from running the supplies when there is no flow in the coils and/or chamber. The
29
Figure 2.5: View from the bottom of the UHV chamber showing coil
orientation and input beams for MOT. There are four beams total, two
from the side and two from the bottom. The reflection offthe gold mirror
gives the other two required beams for 3-D cooling in the MOT.
30
water pressure directly from the laboratory source is too low to be useful so we
installed a booster pump (McMaster-Carr) to bring it up to 100 psi. The water is
conveyed to the necessary places in the experiment by hoses. A flowmeter on the
output of the booster pump outputs a voltage proportional to the water flow. This
voltage is compared to a set point and when there is no flow, the circuit sends a
voltage to the rear connector of the supply which disables the output. The B
z
coil
is powered by a Sorenson XLT-60 supply.
Typical turn on and off times for the supplies driving the inductive load of
our coils is 5 ms and 10 ms , respectively. For our purposes the turn on time
is less important as we slowly ramp on the fields over the course of 25 ms. The
supply limited turn-off times are insufficient because we require abrupt shut-off of
our magnetic fields for sudden release without imparting velocity to the atomic
sample. For this reason, we use insulated gate bipolar transistors (IGBT) to switch
the current on and off. These transistors were chosen due to their high current
carrying capacity. With our relatively low currents compared with typical magnetic
trapping schemes which use a few hundred amperes, one could also implement a
high powered MOSFET. The IGBTs require a driver circuit to supply high current
to the gate to open the switch. The circuit diagram for these transistors is shown
in Fig. 2.6.
Because of the inductive load the abrupt shut-off can cause large voltage tran-
sients which potentially affect other electronics in the experiment and specifically
will damage the IGBT and IGBT driver circuitry. As a precaution we have varistors
(see Fig. 2.6) across the load and protection diodes across the collector-emitter of
31
IR 2127
dr
iv
er
Vcc R3
R2
R1
R4
R5
C1
NC
Pulseblaster IN
To Emitter
To Collector
To Gate
Vcc
Vcc=15 V
R1=390 ?
R2-270 ?, 0.5 W or 1 W
R3=1.5 k?
R4=15 ?, 5 W
R5=1 k?
C1=0.1 uni03BCF
C2=2 uni03BCF
IGBT driver circuit
Connection to IGBT :
driver
Varistor
C2
Coil
H11L1
optoisolator
C
E
G
DLM 600
+-
Figure 2.6: Driver circuit schematic.
32
each transistor. Both of these elements are crucial. We also have an optoisolator
preceding the driver to prevent large voltages making their way to the computer
control boards. Optoisolators create an electrical break. We monitor the current in
each of the magnetic trapping coils (B
x
,B
y
, Z-trap) using transducers (Danfysik).
The devices have a 100 kHz bandwidth and so are not suitable for examining short
transients. Measured with these monitors, the coils turn off times are 100 ?s or less,
depending on individual inductances. We have put capacitors in parallel with the
varistors in attempt to reduce ringing when the current is shut-off.
We use Kepco power supplies to run a maximum of 10 amps through each set
of trim coils. The value of the current was determined experimentally by measuring
the atom temperature of the optical molasses stage.
2.4.6 Lasers
We use semiconductor diode lasers for all applications related to forming a
BEC. Other lasers used in this experiment are a Ti-Sapphire laser (Tekhnoscan
Laser System) which outputs 1 W of tunable 800 nm light when pumped by a 10
W 532 nm laser (Coherent Verdi). This laser is used to form the optical lattices.
We also have a single frequency 15 W 1563.8 nm fiber laser (IPG photonics) for the
dipole trap.
Our MOT, optical pumping, Zeeman slower, and probe beams are derived
from a 750 mW diode laser (DLX-110 Toptica Photonics). For these applications
we require a narrow laser lock so we use phase-modulation on a saturated absorption
33
spectroscopy set-up which is similar to Pound-Drever-Hall lock [7]. The signal does
not come from a cavity reflection, but from an absorption cell. In short, a pick-
off beam is shifted to the |F =2? to |F =3? resonance using a double-pass 80
MHz acousto-optical modulator (AOM) (Isomet, Intraaction) set-up. A double-pass
scheme is generally preferable to decrease beam steering when the driving frequency
of the AOM is scanned. The output of the double-pass is split by a beam splitter and
one arm is modulated at 14 MHz by an electro-optic modulator (EOM) (New Focus).
This EOM puts sidebands on the laser that are 180
?
out-of-phase with each other.
The modulated beam is sent through a Rb cell and when scanned through resonance,
excites atoms. The unmodulated beam counter-propogates and burns a hole in the
Doppler-broadened absorption signal of the probe beam detected with a photodiode.
Without the EOM in place, this is saturated absorption spectroscopy. With the
additional phase modulation, the individual sidebands individually interfere with
the non-modulated pump beam to give a non-zero intensity modulation at the beat
frequency which increases or decreases away from the zero depending on the relative
phase. This signal has a zero-crossing which can be used for feedback to the laser.
We also use single and double-pass schemes to generate the desired frequencies
for the MOT, optical pumping, Zeeman slower, and probe beams. The magnitude
of the shift is dependent on the required atomic transition (Fig. 2.7). A diagram of
the optical table, separated into three parts, is shown in Fig. 2.8 (repumper laser),
Fig. 2.9 (trap laser, MOT light, optical pumping), and Fig. 2.10 (slower, probe,
and slower repumper).
The repumper beam, which is tuned 6.8 GHz away from the trapping light
34
optical pumping 
c
ooling ( MO
T
) ligh
t
r
epumper ligh
t 
pr
obe ligh
t
MOT light ~20 MHz 
below resonance 
5
2
S
1/2
5
2
P
3/2
780.241 209 686(13) nm
384.230 484 468 5(62) THz
12 816.549 389 93(21) cm
-1
1.589 049 462(38) eV
2.563 005 979 089 109(34) GHz
4.271 676 631 815 181(56) GHz
6.834 682 610 904 290(90) GHz
F = 2
F = 1
g
 F
= 1/2
(0.70 MHz/G)
g
 F
=-1/2
(-0.70 MHz/G)
193.7407(46) MHz
72.9112(32) MHz
229.8518(56) MHz
302.0738(88) MHz
266.6500(90) MHz
156.9470(70) MHz
72.2180(40) MHz
F = 3
F = 2
F = 1
F = 0
g
 F
= 2/3
(0.93 MHz/G)
g
 F
= 2/3
(0.93 MHz/G)
g
 F
= 2/3
(0.93 MHz/G)
Figure 2.7: D2 line for
87
Rb with transitions relevant to our experiment
labeled. Underlying diagram and spectroscopy from Ref. [62].
35
to the 5
2
S
1/2
|F =1? to 5
2
P
3/2
|F =2? atomic transition, does not have strict
requirements and so we use a dichroic atomic vapor laser locking (DAVLL) solu-
tion [10]. If a resonant laser beam is traveling through a sample of identical atoms,
one can in principle observe the linewidth of the transition because of the absorption
of photons. Since the atoms are moving with a distribution of velocities given by
their temperature, the distributed Doppler shift will broaden this absorption profile.
Other mechanisms can cause broadening; for a complete description of spectroscopy
techniques see Ref. [15]. This broadened feature can be positively or negatively
offset in frequency by magnetic fields if the incoming light is ?
+
or ?
?
polarized,
respectively. There is magnetic stripping cut into squares on each side of the cell to
provide the field. The output of the cell is split into two according to polarization
and subtracted (Nirvana detector, New Focus). The subtracted signal provides an
error signal for feedback to the laser. This scheme has the advantage of having a
large capture range, and not requiring modulation. We temperature stabilize the
cell and magnets because long term drift due to ambient fluctuations leads to drift
in the lock. A diagram of the set-up is shown in Fig. 2.8. The repumper beam is
coupled into the custom 4?4 fiber splitter (Canadian Inst.) and exits along only
two of the four outputs. It enters the chamber along the same path as the two MOT
beams entering through the bottom window (See Fig. 2.5).
There is a pick-offbeam in the repumper configuration for saturated absorption
spectroscopy. This serves as a reference for the user, but is not used for locking the
laser.
36
100 mW, 780 nm Sacher laser used for 
trap repumper, slower repumper
Rb cell-
DAVLL 
lock
Rb cell-
saturated 
absorption
Nirvana 
detector
magnet
subtract
 signal for
 Doppler-free 
spectrum
photodiodes
?/2
?/2
?/2
?/4
?/4
optical
isolator
To Fiber coupler
telescope for 
launching f_iber
To Table slower : 
beam for slower
repumper
Figure 2.8: Diagram of layout on optical table for the repumper lock
and necessary beams.
37
shutt
er
MO
T double pass upshif
t ~80 MH
z
750 mW, 780 nm Toptica laser used for 
trapping, probe, optical pumping, and slower light
?/2
opticalisola
t
or
shutt
er
?/2
?/2
optical pumping single pass 
downshift ~80 MHz
R
b c
ell 
L
ock double pass upshif
t ~80 MH
z
To Table 3, for
 slower and probe
EOM f
or lock ~14 MH
z
?
/2
?
/2
?/4
?/4
4 x 4 f_iber splitter
input -> 4 outs
?
/2
?
/2
f_iber for o.p.
?
/4
A
OM
AOM
AOM
DLX
-110
photodiode
glass slide
pick of_f for lock 
Figure 2.9: Diagram of layout on optical table for the trapping laser,
Pound-Drever lock, and necessary beams.
38
S
L
O
W
E
R
 
D
O
U
B
L
E
 
P
A
S
S
downshift ~59 MHz
f=10cm
lam/4
do
w
n
s
h
i
f
t
 
~
87
 
M
H
z
f
=
10
cm
l
a
m
/
4
F
R
O
M
 
T
A
B
L
E
 
1
Shutter
Shutter
notes:
1.  telescopes are either /2
or x2 depending on lens
orientation
2.  green = repump laser
    red = trap laser
d
o
w
n
s
h
i
f
t
 
~
8
0
 
M
H
z
 
(
n
o
t
 
w
e
l
l
 
d
e
f
i
n
e
d
)
f
=
1
0
c
m
l
a
m
/
4
Shutter
l
a
m
/
2
f
o
r
 
c
o
m
b
i
n
i
n
g
 
w
it
h
s
l
o
w
e
r
 
b
e
a
m
 
n
e
a
r
 
c
h
a
m
b
er
to chamber area
to chamber area
Repumper SLOWER DOUBLE PASS
PROBE DOUBLE PASS
F
R
O
M
R
e
p
u
m
p
 
Buy SmartDraw!- purchased copies print this 
document without a watermark .
Visit www.smartdraw.com or call 1-800-768-3729.
 
Figure 2.10: Diagram of layout on optical table for the Zeeman slower,
Probe beam, and Slower repumper. Shutters are either Uniblitz or dis-
assembled hard-drive.
39
2.5 Radio-frequency (rf) evaporation
The final stage of BEC formation is rf evaporation. As indicated in Fig. 2.12,
a PulseBlaster card is encoded with a set of rf frequency steps. This sequence is pro-
grammed onto a frequency synthesizer (PTS-Ebay), then amplified (Mini-Circuits)
and sent through an antenna (few loops of magnetic wire) situated externally on the
bottom window of the main UHV-chamber. The antenna is elliptically shaped so
as not to block the incoming MOT laser beams entering though the same window.
The antenna is terminated into a 50 ? attenuator.
The goal with rf evaporation is to remove the most energetic atoms that reside
in the higher magnetic field regions of the trap. After the remaining atoms rether-
malize through collisions, the cloud is colder. The rf field creates level crossings as a
route for hot atoms to escape via spin flipping. As the rf is swept to lower frequen-
cies, it accesses atoms further in from the edges of the trap. Because the rf field
creates a dominant loss mechanism, when its frequency is below the depth of the
trap determined by the B-field it defines the actual trap depth. During optimiza-
tion of evaporation, one maximizes the increase in phase-space density (psd) over
a given frequency ramp, which is given by n?
3
dB
, with the deBroglie wavelength,
?
dB
=
radicalBig
2pi?h
2
/Mk
b
T. n is the density, N/V. In a harmonic trap that does not
change volume during the evaporation process, the change in phase space density
(psd) is proportional to N/T
3
. The efficiency of the process can be monitored using
the parameter [36]
?
evap
=
ln(psd
final
/psd
init
)
ln(N
final
/N
init
)
. (2.3)
40
The preferred optimization of the rf frequency ramp is as follows. Starting with
a magnetically trapped sample that has been compressed to increase the density of
atoms and thus the collision rate, one uses the rf knife to determine the ?top? of
the trap in frequency. Starting at that frequency, in our case 40 MHz, we pick a
ramp step size and time, during which a measurable fraction of the atoms are lost,
e.g. 40-30 MHz in 5 seconds. After this step, there is a hold time of 1 second at the
final frequency, and then the atoms are released. The efficiency coefficient can be
calculated or one can monitor optical depth (see Section 2.8). In this experiment,
we find the atom cloud is not extremely sensitive to initial slopes of rf cuts. As
the atoms cool, they become more sensitive to the ramp steps. As we add steps
to the total ramp, previous steps are re-optimized. As a rule of thumb, the optical
depth of the cloud should go up. Our magnetic trap lifetime was so short when
the getters were used that the evaporation took 12 s and the BEC was small and
short-lived. Currently, the lifetime is >50 s and we find that optimum evaporation
times are 20-25 seconds depending on trapping frequencies. Weaker traps require
longer evaporation times because of the slower rethermalization rate. A typical rf
evaporation ramp (for a very tight trap and a decompressed trap) is shown in Fig.
2.11 with images of the cloud starting at 1 ?K. The signature of BEC formation is
the clear bimodal structure shown in the images (taken after expansion).
41
RF evaporation sequence
F
r
equenc
y (MH
z)
time (s)
 1 uni03BCK
 600 nK
 300 nK
 240 nK
 120 nK
a.
b. c.
Figure 2.11: Fig. depicts (a.) typical evaporation sequence and (b.)
absorption images of BEC formation. The red curve in (a.) is a ramp
for a decompressed trap and the blue curve is the ramp in a compressed
trap. The (b.) images and (c.) cross-sections were taken after the cloud
was released from the magnetic trap and allowed to expand for 17 ms.
The cross-sections in (c.) depict the clear bimodal distribution emerging
as the condensate is formed. This corresponds to a high-density core
shown in the absorption images in (b.).
42
2.6 Moving atoms: optical tweezer
After some evaporation, but prior to condensation, atoms can be moved to
the glass science cell using an optical tweezer. The advantage of this design is the
optical access to the atomic sample. Additionally, the proximity to the atoms allows
for high resolution imaging (large numerical aperture, (N.A.)). Although I initially
was involved in the design and use of this part of the apparatus, the experiments
in this thesis do not use the tweezers. Thus, I will only introduce the technique
here. The current lens design for our tweezer was determined and implemented by
M. Beeler.
Alternative to a magnetic trap, a dipole trap provides all optical confinement
for a neutral atom sample. An excellent review of this technique is Ref. [28]. Neutral
atoms do not have permanent dipoles, but acquire an induced dipole moment from
interaction with an oscillating electric field. This gives rise to an intensity-dependent
force attracting atoms to the minimum or maximum of the potential depending on
the sign of the detuning from resonance, ?, the power, P, and the minimum waist,
w
0
. For alkali atoms, this potential, due to a focused Gaussian laser beam, is
approximated as
U(r,z) ?
pic
2
2
parenleftbigg
2?
2
?
2,F
?
0,2
+
?
1
?
1,F
?
0,1
parenrightbigg
2P
piw
2
(z)
e
?2
r
2
w
2
(z)
(2.4)
Here, w(z)=w
0
radicalbig
1+(z/z
R
)
2
. z
R
is the Raleigh range, piw
2
0
/?.?
n,F
is the detuning
of the laser from a ground state level, F, (in our case, 2) to either the 5
2
P
3/2
(n=2) or 5
2
P
1/2
(n=1) excited state. ?
0,n
are the resonant frequencies of these
same transitions. In this equation the counter-rotating term is neglected and I
43
assume linear polarization. For far-off resonant traps, the counter-rotating term
contributes and is included in the calculation of trapping parameters. The scattering
rate depends on detuning as 1/?
2
so the laser frequency must be tuned many
linewidths away from resonance to attain long trapping times and low heating rates.
Using ?
i
=
radicalbig
k
s.c
/m and (?
2
U(0)/?i
2
)=k
s.c
(with i denoting direction x, y, or z
and s.c referring to the spring constant), the trapping frequencies are calculated.
In our experiment in excess 10
6
atoms can be loaded from the magnetic trap
into a beam of ?1567 nm light focused to a minimum waist of 40 ?m. The laser is a
15 W Erbium-doped single frequency, single polarization, fiber laser (IPG Photon-
ics). The magnetic trapping stage is necessary because 1567 nm is resonant with a
transition from the 5
2
P
3/2
state which prevents loading atoms from the MOT into
the dipole trap. The small detuning from this transition gives rise to a large excited
state light shift (hundreds of MHz using our powers). The level shifts closer to the
ground state which makes the MOT cooling light look blue-detuned and therefore
the atoms heat instead of cool.
The initial alignment of the dipole trap with the magnetic trap was difficult
because of the small overlap region. A tracer beam could not be used for alignment
because over the long pathlength 1550 nm mirrors transmit for 780 nm light (due to
the anti-reflection coatings). Instead, we created a MOT cloud that was low density
and spread over a large region. Due to the excited state light shift, the dipole beam
slightly deformed this cloud. The Pixelfly camera (see Section 2.8), running in ?live
feed? mode can detect this deformation. From there we could load first cold atoms
from a molasses and then walk the beam to the magnetic trap location near the
44
mirror surface.
The atoms, once transferred, are moved 25 cm to the glass cell by translating
the waist slowly enough that the atoms follow (1.5 s, in our experiment). The focus
lens is translated on an ultra-smooth air-bearing stage (Aerotech). This design was
modeled after the MIT Ketterle group who first pioneered moving BECs. With our
low density BECs, evaporation in a 1D dipole trap is too inefficient compared with
the vacuum limited lifetime (15 seconds in the dipole trap) to be successful. Thus,
when the atoms arrive in the cell, a secondary dipole trap intersects the cloud at a
shallow angle to increase longitudinal confinement. We did not appreciate the low
density problem in the initial design stages, and the experiment underwent many
improvements before BEC was achieved in the science cell. For a more in-depth
description of dipole trapping and its use in our current set-up, see the thesis of M.
Beeler, in addition to the Refs. given here.
2.7 Optical lattices
The dipole trap section lays the theoretical foundation for the understanding
of optical lattices. Lattices are a type of dipole trap using a standing wave, instead
of a traveling wave [28]. The physics is the same; only the equations for the potential
are slightly modified. The potential is
U(r,z) ? U
4P
piw
2
(z)
e
?2
r
2
w
2
(z)
(1 + cos2kz) (2.5)
Here, U is the pre-factor which is dependent on detuning and atomic parameters
[28]. The trapping frequency in the direction of propagation is approximately given
45
by ?
z
= (2U(0,0)k
2
/m)
1/2
and is typically in the tens of kHz range. The atoms are
attracted to the peaks or valleys of the standing wave depending on the sign of the
detuning. The distance between each well of the standing wave is ?/(2sin?), where
? is the angle at which the two beams forming the standing wave intersect. This
means that the condensate or thermal cloud is divided into pieces by the lattice. In
a 1D lattice, this gives an array of 2D clouds of atoms shaped like pancakes (in deep
lattices). All of the experiments described here use a 1D lattice. Experimentally,
the lattice beams are not approximately collimated, but instead, are focused, which
gives rise to additional dipole confinement. In principle one can design any arbitrary
lattice topology to suit the user?s needs. In the science cell, the optical access allows
for such versatility. In order to add together optical lattices for this purpose, one
just detunes the relative lattice beam frequencies by multiple MHz using an AOM.
The high frequency interference term between two lattices is time-averaged to zero
from the viewpoint of the atom sample.
All lattices in this experiment are derived from a tunable Ti-Sapphire laser
peaked at 800 nm (Tekhnoscan). After frequency shifting with double-pass AOM
set-ups, the individual beams pass through polarization maintaining fiber to clean
up the spatial mode. Following the output couplers, the beams are directed into
the chamber through the bottom window, where they are reflected off of the gold
mirror. The lattices used are described in more detail in Chapter 3.
46
2.8 Imaging atomic clouds and time-of-flight
No matter the experimental sequence, we detect atomic samples using a tech-
nique called absorption imaging. If resonant light illuminates a sample of atoms,
the amount of light remaining after absorption of photons in the cloud will depend
on the density of the sample via
I(x,y)=I
0
(x,y)exp
bracketleftbigg
??
abs
integraldisplay
n(x,y,z)dz
bracketrightbigg
(2.6)
Here, ?
abs
is the absorption cross-section. The important quantity is the optical
depth, O.D. (Eq. 2.7), which is extracted from this equation by taking two images:
one with atoms and one without.
O.D. = ?ln
parenleftbigg
I(x,y)
I
0
(x,y)
parenrightbigg
= ?
abs
integraldisplay
n(x,y,z)dz =?n(x,y)?l (2.7)
We integrate over the z-direction (direction of imaging) to get a 2D density profile,
?n(x,y). The imaging system currently in use was designed by M. Beeler. It has a
magnification of ?1. We acquire images of the sample in both the main chamber
and the science cell. There is a 12-bit PixelFly QE (Cooke Corp.) camera for the
main chamber (where the MOT/magnetic trap is formed) and a 10-bit Flea2 (Point
Grey) for the glass cell (where atoms are moved to using the dipole trap), although
in the future, the camera with higher dynamic range will be used in the chamber
where most experiments take place. In the main chamber, the numerical aperture
is limited by the radius of the vacuum chamber (9 cm). The imaging in the cell is
limited by lens size, since we can get as close as 1 cm from the sample.
Absorption imaging does not work well when the sample is very optically
47
thick, as is the situation for a very cold cloud of atoms. The cloud will act like a
lens and one will observe diffraction rings, distorting the image. For this reason,
the atoms are first released from the trap and allowed to expand while falling (time-
of-flight, TOF) prior to imaging. The resulting spatial distribution of the atoms
is related to the momentum distribution. For a thermal cloud in a harmonic trap,
regardless of the trapping frequencies, the cloud will be spherical after long TOF. In
the long expansion time limit the initial cloud size is neglected and the temperature
(measured along each direction of the cloud) is given by [38]
T
i
=
M
k
B
bracketleftbigg
?
i
t
TOF
bracketrightbigg
2
. (2.8)
Here, i denotes a direction and t
TOF
is the expansion time. Typical expansion times
range from 2 ms (for very hot or anisotropic clouds) to 25 ms (for BECs and atoms
loaded into lattices).
For diagnostic purposes, there are additional cameras, which give a continuous
live video feed to television monitors of the fluorescing atoms at the oven aperture
and in the MOT region.
2.9 Computer Control and Data Acquisition
Shown in Fig. 2.12 is the flow diagram for computer control of the experi-
ment. We use a PCI board called a PulseBlaster (SpinCore) to send digital signals
to all equipment requiring a simple on/off during an experimental sequence. The
PulseBlaster also triggers analog control boards to send specific voltages to types of
equipment whose outputs are multi-valued during a sequence. Both the analog and
48
digital boards are programmed using code written in LabView and inherited from
NIST (written by J.V. Porto and I.B. Spielman and modified for our experiment
by M. Beeler). A secondary PulseBlaster card is used for the evaporation sequence
described above in Section 2.5.
The camera takes an image with and without a cloud and the two are divided
in a program in LabView and analyzed in IGOR (Wavemetrics). Most of the data
acquisition software was also adapted from NIST by M. Beeler. For analyzing the
multipeak results for the experiment described in Chapter 2, there are additional
add-on procedure files, separate from the main analysis code.
49
Computer 2:
Data Acquisition
Software:
Labview/Igor
RF programming
board
Computer 1:
Control Software
PulseBlaster
24 channels
Analog Out
2 x 8 channels
trigger
digital outs
Data in
RF synthesizer
Camera(s)
Camera Card
analog breakouts
buf_fer circuit
Computer Control Flow Diagram
Voltage controls to Power Supplies, 
AOMs, EOMs, RF amplitude 
trigger
trigger
to RF antenna
Figure 2.12: Flow diagram for the computer control and data acquisition
of the experiment.
50
Chapter 3
Adiabaticity and Localization in Incommensurate lattices
3.1 Introduction and Motivation
Disorder plays an important role in many condensed matter systems,[22, 1]
with deep connections to quantum chaos [27], but can be difficult to systemati-
cally study due to the challenge of creating reproducible and quantifiable disorder.
Adding disorder to a lattice of bosons may give rise to a glassy phase (Bose glass)
predicted between the insulating and superfluid phases. [22]. It is also responsible
for exponentially localizing the wavefunction in a non-interacting system, so-called
Anderson localization [1].
The control available in ultra-cold atom systems [8] has historically made them
an attractive platform to study disorder [6, 53, 66, 61]. To date much of the work
adding disorder to ultra-cold atom systems has explored time-independent prop-
erties, but the long timescales associated with cold atoms allows investigation of
dynamical properties, as well (see [31, 58, 40, 6] and ref. therein). In these systems,
atoms, which are ? 10
5
? more massive than electrons, have ?s compared to ps
dynamical time-scales. In this work we examine the ability of a quasi-disordered
system to adiabatically follow changes in the Hamiltonian. The presence of disorder
produces a complicated eigenvalue spectrum, which greatly affects the adiabaticity
criteria. The physics of localization phenomena also has a significant impact on
51
time-dependent processes, such as adiabaticity. Small perturbations to the Hamil-
tonian can cause large changes to the ground state wavefunction over large length
scales, making it difficult for the system to adiabatically follow changes. One recent
theoretical study shows that adiabaticity in gapless systems is non-trivial, particu-
larly in lower dimensions[49]. Here we show that even in a system where there is a
well-defined separation between the ground and excited bands, such as is the case
in our experiment, adiabaticity is complicated by the presence of disorder [21]. The
following sections are an expanded description of the results which appeared in Ref.
[21].
3.2 Disorder
From the condensed matter point-of-view there are different types of disorder,
a few being dislocations, impurities, and random modulation of the energies, each
producing different physical effects. In this work, we focus on random modulation
of the periodic potential. Experimentally, the technique used to generate a random
potential is to focus an optical speckle pattern onto the sample. In practice, produc-
ing a grain size small compared to the underlying lattice constant is difficult [41, 43],
but can be accomplished by designing a chamber where a lens can be placed only
1-2 cm away from the sample (high numerical aperture) [66].
An alternative to approximating randomness is using incommensurate or two-
color lattices [54, 13, 57, 55]. In an infinite, non-interacting system, the Aubry-Andr?e
52
Hamiltonian [3] describes the system:
H = J
summationdisplay
m
(|w
m
??w
m+1
| + |w
m+1
??w
m
|)+?
summationdisplay
m
cos(2pi?m + ?)|w
m
??w
m
|, (3.1)
where J is the nearest neighbor tunneling and ? is the amplitude of the disorder
for an incommensurate lattice with the ratio of k-vectors, ? = k
2
/k
1
. The localized
Wannier states at site m are |w
m
?. This model predicts a crossover from a delo-
calized to an exponentially localized state for the ratio ?/J ? 2 if the lattices are
maximally incommensurate ([53, 27, 3] and Refs. therein). Presumably, deviations
from the ideal incommensurate ratio, increases in interactions, and finite-size effects
will change the predictions of this model [54, 57].
In this work, we study adiabaticity in a quasi-disordered system by adding one
or two weak incommensurate lattices to a one-dimensional optical lattice loaded
with a BEC. Localization occurs in both disordered and strictly incommensurate
potentials [54, 57, 13, 61, 55], and it is assumed that differences tend to disappear
in finite-sized systems such as ours. We observe a complex momentum distribution
of the atoms due to the presence of weak perturbing lattices following a ramped
loading process that would be nearly adiabatic for a single lattice. We gain insight
into the distributions from single-particle band structure and observe that the effects
of the perturbations disappear as interactions increase, as they suppress the long
wavelength density modulation of the wavefunction.
53
3.3 Band Structure for Single Lattices
An excellent place to begin a description of atoms loaded into a lattice is
single particle band structure [2]. The wavefunction of a particle in a given well of
a periodic potential is related to the wavefunction in a neighboring well by Bloch?s
theorem,
?(r) = exp(?iqr)?(r + R), (3.2)
Here, R is a translation in the lattice and q is the quasimomentum. Quasimomentum
is the phase of the wavefunction across adjoining sites in the lattice (it is equivalent
to the term crystal momentum found in most solid state physics texts [2]). Because
of the periodicity, the wavefunction and the potential can be rewritten as
?(r) = exp[?iqr]
summationdisplay
m
c
n
m
exp[?imGr] (3.3)
V(r)=
summationdisplay
m
V
m
exp[imGr], (3.4)
where G is the reciprocal lattice vector defined as 2pi/R. The index m refers to the
components of the momentum, n is the band index and V
m
are the Fourier compo-
nents of the potential. Using these substitutions, Schr?odinger?s equation becomes,
bracketleftbigg
?h
2
2M
(q?mG)
2
+ V
0
bracketrightbigg
c
q?mG
+ V
G
c
q?(m+1)G
+ V
?G
c
q?(m?1)G
= Ec
q?mG
. (3.5)
From 3.5 the energy bands are calculated. The energy as a function of quasimo-
mentum, q, is shown in Fig. 3.1 (presented in the reduced zone scheme) for the free
particle case (top) and a lattice of depth 5 E
rec
(bottom), where E
rec
=?h
2
k
2
/2M.
As shown in the free particle case (Fig. 3.1(a)), the ground and first excited
state are separated by 4E
rec
. Stationary, non-interacting atoms have no means by
54
quasimomentum (q), units of 2 k
1
band index=n
n=3
n=2
n=1
-1                                           0                                                      1  
Ener
gy (Er
ec)
8
6
4
2
0
band index=n
n=3
n=2
n=1
Ener
gy (Er
ec)
10
  8
  6
  4
  2
  
  0
-1                                          0                                                      1  
Turning up
 lattice potential
4 Erec
a.
b.
quasimomentum (q), units of 2 k
1
Figure 3.1: Energy bands as a function of quasimomentum for a free
particle (top) and a single lattice potential (bottom). The bands are
depicted in the reduced zone scheme. As the lattice is applied to the
particle, the energy bands separate and the degeneracy at the edge of
the Brillouin zone is lifted.
55
which to change q during the process of turning on or changing the amplitude of
the periodic potential (when there are no other forces present). In our system,
atom-atom interactions can change q, but the time scales are on the order of tens
of milliseconds (see Section 3.4). This means that the relevant time scale for adia-
batically loading atoms into the lowest energy band of the lattice is related to the
separation between the ground and excited band.
The generalized time-dependent equation for calculating an excited state con-
tribution is
?a
n,q
(t)=
a
1,q
(E
n,q
?E
1,q
)
angbracketleftbig
n,q|
?H
?t
|1,q
angbracketrightbig
e
i
R
t
0
E
n,q
?E
1,q
?h
dt
(3.6)
In order to remain adiabatic, the excited state population, |a
n,q
(t)|
2
, must stay lessmuch 1,
where n is the band index other than n = 1 and q is the quasimomentum. One can
calculate a corresponding time scale using Eq. 3.6 which, for loading a
87
Rb BEC
into a single lattice at k
1
? 2pi/?, is satisfied for times much greater than 5 ?s [59,
34, 47]. As the lattice depth increases, the bands continue to separate. Adiabaticity
is most difficult for q near the band edge, where the initial energy gap vanishes.
For our experiments, the initial momentum range is ?0.15 k
1
. Additionally, the
single particle band picture begins to break down as the BEC becomes strongly
interacting. The oft-cited adiabaticity expression [16, 59],
vextendsingle
vextendsingle
vextendsingle
vextendsingle
?n,q|
?H
?t
|1,q?
vextendsingle
vextendsingle
vextendsingle
vextendsingle
<<
?E
2
(q,t)
?h
(3.7)
makes the approximation that the Hamiltonian and eigenstates are slowly varying
and that the changes in energies are small. As we will see, these assumptions are
not valid in the incommensurate lattice case.
56
3.4 Experimental details and results
As described in Chapter 2, we collect 10
9
atoms in our MOT. An optical mo-
lasses stage follows by abruptly turning off all magnetic fields and carefully zeroing
the field at the atoms using the trim coils. During this phase, we reduce the in-
tensity and detune the MOT beams a few linewidths further from resonance (to
reduce the radiation pressure) using acousto-optic modulators (AOMs). This stage
serves to cool the atoms to ? 40?K. Because the atoms are so cold and do not
move very much during this ?5 ms stage, no additional forces are required to keep
them in place. At this point the atoms are in an equal distribution of m
F
levels,
and we need them in a low-field seeking state, so we optically pump the atoms into
the |F =2,m
F
=2? state. We do this with a 100 ?s pulse of circularly polarized
light tuned to the |F =2? to |F
prime
=2? transition. The polarization is given with
respect to a quantization axis defined by a small field turned on in the B
y
direction
during the pulse. The magnetic trap is slowly ramped on and then immediately
compressed to ?
x
? ?
z
? 2pi?410 Hz and ?
y
? 2pi?120 Hz to increase the density,
and therefore the collision rate. We evaporatively cool the atoms until we obtain a
condensate with less than 50 % thermal component (100 nK). To reduce the effects
of interactions, which are density and scattering length dependent (Section 3.6), the
trap is subsequently weakened giving final frequencies ?
x
? 2pi?40 Hz, ?
y
? 2pi? 20
Hz, and ?
z
? 2pi?30 Hz.
57
3.4.1 Optical Lattices
We load the BEC into a 1D optical potential (recall Section 2.7, Optical Lat-
tices), created by the addition of a primary and perturbing lattice(s). The lattices
are created by reflection off an in-vacuum gold mirror (used for the mirror-MOT)
situated approximately 2 mm from the magnetic trap center. As shown in Fig. 3.2,
each beam is reflected at a different angle (?
1
= 180
?
, ?
2
= 143
?
, and ?
3
= 155
?
,
defined as the angle between incoming and reflected beam) to yield a 1D pseudo-
disordered potential [29]. The effective wavevector of each lattice is k
eff
=2pi/(2l)
where the periodicity of the lattice, l = ?/(2sin(?)). Because the gold mirror sur-
face that defines the standing waves nodes is the same for all three lattices, they
are phase-locked together, which gives a very stable potential. To prevent interfer-
ence between the lattices, we detune the beams ?160 MHz from each other with
AOMs, so that the coupling terms between the beams oscillate rapidly compared to
the atomic motion, and are time-averaged to zero (from the reference frame of the
atoms). The 1/e
2
waists for the primary perturbing beams are approximately 400
?m and 3500 ?m, respectively. Fig. 3.2 depicts the lattice configuration.
The total potential is
V
tot
=
M
2
(?
2
x
x
2
+ ?
2
y
y
2
+ ?
2
z
z
2
)+V
1
sin
2
(k
1
z)+V
2
sin
2
(k
2
z)+V
3
sin
2
(k
3
z), (3.8)
where M is the atomic mass, k
1
=2pi/?, and ? = 796.6 nm. The ratios k
2
/k
1
=
0.806 ? 0.002 and k
3
/k
1
=0.919 ? 0.004 are extracted from absorption images of
atomic diffraction. The distance between momentum orders in each image, ?x, is
extracted from an image. Using ?x = vt and p = Mv =2n?hk, where v is velocity,
58
t is expansion time, we calculate the k of the standing wave. The error bars reflect
statistical uncertainty and include the error in position measurement. For the bulk
of the experiments described here, the lattice depths are V
1
=4.6 ? 0.3 E
rec
and
V
2,3
/V
1
=0.059 ? 0.003, where E
rec
=?h
2
k
2
1
/2M.
3.4.2 Lattice calibration
Although the potential is defined by clear parameters such as power, waist,
and detuning, there is significant error in the power and waist measurements using
standard equipment. The power can be measured with both a power meter having
a silicon detector, which is sensitive at 800 nm. We measure the waist with a meter
having a rotating silicon detector (Thorlabs) and also using a razor method. In the
razor method, a blade is placed at the waist and its position is scanned horizontally
through the beam in position steps much smaller than the beam diameter. The
transmitted power is plotted as a function of position and fit to an error function
curve. The waist is extracted from this curve fit. The errors arise from power
measurement, position measurement with a micrometer, and placement of the razor
blade.
The detuning is well known because we measure the wavelength on a waveme-
ter that is accurate to ? 30 MHz. For this reason, the individual lattice depths are
calibrated using two alternative methods. In the first method, the laser is tuned
close to resonance and a short pulse of light is imparted onto the sample. The
effect of the pulse is to write a periodic phase on the wavefuction. The length of
59
Figure 3.2: Beams propagating in the +z-direction are reflected off of an
in-vacuum gold mirror forming a standing wave (in the gravity direction).
They are reflected at different angles to give a 1D incommensurate lattice
potential to mimic disorder. The depiction of the cloud in the lattice is
not to scale. The approximate number of lattice sites occupied is ?70.
60
the pulse, ?, is chosen to fall in the phase grating limit, where the atomic spatial
wavefunction does not evolve in the wells of the standing wave. The condition for
this is that ? << 2pi/?
latt
. For a lattice with an estimated frequency of 20 kHz, ?
must be much less than 50 ?s. The resulting matter-wave diffraction pattern has
momentum peaks spaced at p
n
=2n?hk , with n being an integer. The sample is cold
enough (sub-1 ?K) or condensed so that the expansion of the cloud in position is
less than the expected spacing between momentum orders. The populations in the
peaks are P
n
= J
2
n
(epsilon1/2), where J
n
(z) are Bessel functions of the first kind [46]. The
argument of the Bessel function is related to the amplitude of the standing wave
by epsilon1 = ?U(0)/?h. U(0) is the depth of the standing wave potential [28]. For good
signal to noise, we choose a pulse which results in momentum components having
equal amplitudes. The error from intensity fluctuations and power measurements
does not contribute in this calibration measurement.
In the second calibration method, the atoms are loaded into a far-detuned
lattice, and abruptly released. A gaussian fit is applied to each of the peaks in the
resulting interference pattern. The relative populations in the ?1 order momentum
peaks (with respect to the 0
th
order peak), P
?1
, are extracted and the depth of the
lattice is determined from [11],
s =
16
[ln(P
?1
)]
2
P
1/4
?1
(3.9)
We load the atoms into the combined lattices by ramping up the intensities, keeping
a fixed ratio between V
2,3
and V
1
. A ramp time of 1 ms is chosen for most experiments
and is sufficiently long (compared to 5?s) to ensure loading a magnetically trapped
61
BEC into the lowest band of a single lattice (when no other lattices are present).
To assess our ability to adiabatically load a lattice, we perform two types of ex-
periments. In all experiments, we take an absorption image of the cloud after 22 ms
of expansion during time-of-flight. Following the ramped loading and a variable
hold time (typically 1 ms), we turn off both the lattice and the magnetic trapping
potentials. The image yields the momentum distribution of the atoms trapped in
the lattice (interactions during time-of-flight are negligible).
3.4.3 Sudden release of atoms from a lattice and incommensurate
potential
With a sudden release, the Bloch states are projected onto the free particle
plane wave states with momentum, p =?hq ? 2n?hk. For comparison, we present
absorption images after the sudden turn-off, with and without the perturbative
lattice(s) in Fig. 3.3. The existence of the weak perturbing lattice (Fig. 3.3(c-e))
markedly modifies the momentum distribution with the appearance of new peaks,
even though the perturbing lattice is so weak that by itself it does not produce
any diffraction with amplitude above our noise floor (Fig. 3.3(b)). In a system
with perfect resolution, the momentum peaks would be observable. There is an
enhancement of the peaks at ?2?hk
2
making them discernible above the image noise,
as well as new features at ?2?h|k
1
?k
2
|.
62
V1?4.6 Erec
V2?0.3 Erec
V1+V2
V1+V2+V3;V2?V3?0.15 Erec
momentum (?k) 
Amplitude (Arb. Units) 
(f)
(g)
(I) (II)
(i)
(h)
V1+V2; V1?7.5 Erec, V2?0.5 Erec
(a)
(b)
(c)
(d)
(e)
Figure 3.3: (I) Absorption images of a BEC loaded into different lattices
for ?
BEC
? 100 Hz after 20 ms TOF: (a)V
1
? 4.6 E
rec
and (b) V
2
? 0.3
E
rec
;(c) V
1
+V
2
; (d) V
1
+V
2
+V
3
, with V
2
? V
3
? 0.15 E
rec
; (e)V
1
=7.5
E
rec
and V
2
=0.5 E
rec
lattice. We turn on lattices using a 1 ms linear
ramp followed by a 1 ms hold and abrupt turn-off. (II) Using the data
from the absorption images ((I)-(a,c-e)) we extract the relative popu-
lation amplitudes (green bars) and overlay them with a band structure
calculation (no interactions) of the ground state momentum distribution
(black lines).
63
3.4.4 Adiabatic turn-off
In the second method, after the loading and holding process, we slowly ramp
down the lattice in 1-10 ms and then abruptly release from the magnetic trap (?100
?s). If the process is fully adiabatic, the cloud should return to the momentum
distribution of the original BEC. In Fig. 3.4, we present absorption images showing
the results of testing for adiabaticity with a ramp-up/hold/ramp-down sequence.
From Fig. 3.4a, one can see that the small perturbation prevents adiabaticity on
millisecond time scales. After ramping down, there are momentum peaks separated
by the difference in wavevectors, k
1
? k
2
. One has to use perturbations less than
?0.1 E
rec
to prevent excitations during the experimental sequence (Fig. 3.4(c-d)).
It should be noted that as the perturbation has an increasing effect on the
energy bands, it also becomes difficult to adiabatically turn-off the lattice.
3.5 Modeling results
We use single particle band structure and the 1D Gross-Pitaevskii equation to
model our system.
3.5.1 Band structure to approximate incommensurate lattices
In order to understand why such a small perturbation has dramatic effects on
the momentum structure and excited fraction we use single-particle band structure.
Even if the perturbing lattice is strictly incommensurate with the primary lattice
with k
2
/k
1
= ? equaling an irrational number, we can approximate ? as a ratio of
64
?
uni03BCBEC?100 Hz; V1+V2;
V1?5 Erec; V2?0.6 Erec
uni03BCBEC?100 Hz; V1+V2;
V1?5 Erec; V2?0.3 Erec
uni03BCBEC?100 Hz; V1+V2;
V1?5 Erec; V2?0.1 Erec
uni03BCBEC?100 Hz; V1+V2;
V1?5 Erec; V2?0 Erec
(a)
(b)
(c)
(d)
Figure 3.4: Results of adiabaticity check. (a)-(c) Absorption images
following ramping off primary plus one perturbing lattice of differing
strength. (d) Image of no lattices. These images illustrate the difficulty
of adiabatic loading of a condensate into a pseudo-disordered lattice.
65
two large integers, f/g. In a finite system such as ours where the BEC occupies
? 70 sites of the primary lattice and does not extend over many periods of the beat
frequency lattice (created by the combination of the lattice potentials) we expect
this approach to yield reasonable predictions [17]. This modification gives rise to
additional off-diagonal elements in the single period matrix (Eq. 3.5), which is then
re-diagonallized to calculate the eigenvectors and eigenenergies.
Assuming no excitations to higher bands, we predict the momentum distri-
bution for each lattice configuration using this modified band structure calculation
with f/g =9/11 and f
prime
/g = 10/11, approximations to the experimental ratios. The
integers in these ratios are not large (larger integers yield a more irrational num-
ber), but are chosen to give practical computation times. We increased the values
to confirm that our calculations captured the features of the system. We present the
results of this calculation overlaid with population amplitudes (green bars) extracted
from fits of our data in Fig. 3.3. In the case of three lattices, experimental resolu-
tion limits our ability to resolve the structure, and this is represented by wider bars.
The combination of the lattices creates a complex momentum structure, and the
band structure calculation is in good qualitative agreement with our measurements.
However, since the band-structure calculation assumes no excitations and a 1 ms
ramp causes depletion of the ground band, the predicted populations do not quan-
titatively agree with the data for multiple lattices. This mismatch increases as the
depths of both the primary and perturbing lattice(s) are increased. For deep lattices
(Fig. 3.3(i)), depending on the choice of f/g , band structure predicts peaks spaced
at k
1
/g, closer than the beat frequency, yet the envelope does not change. We find
66
these disappear much more rapidly than the beat frequency peaks as interactions
increase (see Section 3.6.1 below).
We also examine the changing wavefunction and the corresponding band-
structure in Fig. 3.5. Using the k?vector of the primary lattice to determine the
size of the first Brillouin zone, we plot the first two bands for the case of one and two
lattices. The dominant effect of the perturbing lattice is to open small gaps in the
ground band, as well as to slightly flatten the band between the new gaps. Starting
at q = 0 and traveling along the band, the first gap corresponds to quasimomentum
of (k
1
?k
2
). If instead one chooses the first Brillouin zone to span ?(k
1
?k
2
) which
is now much smaller, there exists a new band that is much lower in energy than the
first excited band in the single lattice case, which determines a new energy scale
relevant to adiabaticity. This new energy scale is also responsible for localizing the
wavefunction at the spatial period corresponding to the beat frequency (Fig. 3.5(c)
compared to Fig. 3.5(b)). The increased flatness of a band corresponds to a decrease
in the tunneling rate because the energy states in the neighboring wells are no longer
degenerate.
According to the new energy diagram, in order to satisfy the adiabaticity
condition (Eq. 3.6, 3.7) for two lattices with our parameters (V
2
? 0.059 V
1
,k
1
/k
2
?
0.8), the loading ramp must be much longer than 4 ms (a thousand times longer
than for a single lattice). This is shown in Fig. 3.6(b), where we calculate the
depletion of population in the ground band for different ramp rates as a function
of primary lattice depth for the case of three lattices. From Eq. 3.5, modified for
perturbing lattices when present, the coefficients, c
n,q
, can be calculated for any
67
depth potential. If the change in the potential as a function of time is known, then
the overlap of the eigenvectors can be used to calculate population as a function of
ramp time. Starting with the free particle solution, the instantaneous eigenfunctions
for a step in depth (time) are calculated and the overlap with the previous step is
calculated. The result of this calculation is shown in Fig. 3.6(b). The inset of
Fig. 3.6(b) depicts an expanded view of ramp times yielding less than 5 percent
excitation.
Fig. 3.6(c) shows the depletion of the ground state as a function of lattice depth
for a ramp time of approximately 2.5 ms for the case of one, two and three lattices.
Since it is often the case experimentally that we start with a BEC with non-zero
q (due to residual motion in the magnetic trap), we also show curves for loading a
BEC with q = 0.1 k
1
into a potential comprised of two and three lattices. This value
of q was chosen to lie within the new, smaller first Brillouin zone and to fall within
our experimental shot-to-shot fluctuations in q. These curves show that the effect of
additional perturbing lattices is the dominant effect on adiabaticity criteria. For q
= 0.1 k
1
, the adiabaticity time scale approaches 10 ms. We experimentally observe
that by ramping up the potential and then ramping down with additional lattices,
longer ramp times (5 ms compared to 250 ?s) are qualitatively better (less excita-
tions), but due to interactions, we are never able to be fully adiabatic. Although
excitation to higher bands may also produce new momentum peaks, for the data
presented in Fig. 3.3(c), we predict that between 1 and 5 percent of the population
is excited depending on initial quasimomentum. Thus most of the observed changes
in the momentum distribution can be ascribed to modification of the ground band
68
wavefunction, in this case localization.
The effects seen here are a result of the incommensurability of the lattices.
For experiments using commensurate lattices [47] where two lattices had a carefully
chosen ratio of 3:1, the criteria for adiabaticity with respect to the band structure is
not significantly altered. In this case the higher order momentum peaks due to the
perturbing lattice overlap the beat frequency peaks, which results in the wavefunc-
tion having modulation only at that spatial frequency. This difference can be seen
by calculating the ground state wavefunction for the two cases. For approximately
incommensurate lattices, the perturbation has a strong effect, with small spatial fre-
quencies appearing (Fig. 3.5(c)). In order to further quantify this dramatic change,
we plot the population of the p = 0 peak, p
0
, as a function of lattice depth for a
single lattice, and for a primary lattice with a perturbative lattice (with V
2
=0.059
V
1
, Fig. 3.6(a)). The black points are experimental measurements of the peak am-
plitude, which show a dramatic reduction of p
0
when the perturbing lattice is added.
We show two band structure calculations for f/g=4/5 (dashed) and 9/11 (solid),
both falling within the error bars of our experimental ratios. The significant differ-
ence between these cases is due to population growth in sub-beat frequency spaced
momentum peaks predicted for the ratio 9/11. The error bars represent statistical
uncertainties in populations and lattice depths (due to fluctuations in lattice beam
intensity). We expect that because the turn-on of two lattices is not fully adiabatic
there will be additional depletion of the central peak due to excitations into higher
bands, as observed.
69
Ener
gy (recoils) 
quasimomentum ( 2 k
1 
)
(a)
(b)
Amplitude  (Arb. Units) 
(c)
1 0.5 0 0.5 1
2
3
4
5
6
7
Figure 3.5: (a) Calculated first two bands for a single lattice and for
two incommensurate lattices. Note the gaps which form in the ground
band. (b-c) Plot of an example wavefunction multiplied by a Gaussian
envelope (b) without and (c) with a perturbing lattice
70
Ground band population (Arb. Units) 
(i) 
(ii) 
(iii) 
- 
 
 
  p
0
 (Arb. Units) 
(a)
(c)
Primary lattice final depth (E
rec
) 
(b)
(i) 
(ii) 
(iii) (iv) 
(v) 
(iv) 
(iii) 
Figure 3.6: (a) the probability amplitude, p
0
, as a function of lattice
depth for p = 0 peak for the case of a single lattice (V
1
) and a combined
potential (V
2
? 0.06V
1
). The changes in slope correspond to the depths
at which the atoms become localized at different spatial frequencies. (b)
Calculation of ground band population for differing ramp times for case
of three lattices ((i) 25 ?s, (ii) 250 ?s, (iii) 2.5 ms). The total pertur-
bation is V
2
+ V
3
? 0.06V
1
; (iv) Result of loading a single lattice with a
ramp time of 2.5 ms. Inset in (b) is a zoomed-in view for comparison of
(iii) and (iv). (c) Calculated ground state population for loading a (i)
single lattice, (ii) two lattices, (iii) two lattices starting with q =0.1 and
three lattices starting with (iv) q = 0 and (v) q =0.1 for a fixed ramp
time of 2.5 ms.
71
3.6 Localization
There are many types of phenomena classified as localization. In the context
of the experiments described here, the term localization refers to decreasing the
spatial extent of the wavefunction. In the limit of a non-interacting, infinite system
in the presence of a disordered potential, Anderson predicted [1] that the wavefunc-
tion would become exponentially localized through destructive interference of wave
reflections. In 1D systems, the wavefunction is always localized for a finite amount
of disorder, whereas in higher dimensions a transition from extended to localized
states is predicted. In practice cold atom experiments are finite systems and without
a Feshbach resonance to tune the scattering length, a, to zero (see for example [53]
and Refs. therein), these systems also have non-negligible interactions.
In this experiment the disorder is formed by an incommensurate lattice and
localization associated with this Hamiltonian is predicted using the Aubry-Andre
model [56, 53, 55]. Compared with Anderson-type localization, this model predicts a
transition from delocalized to localization even in 1D systems. In both the Anderson
and Aubry-Andr?e model interactions still compete and are predicted to destroy the
localization effects (recall Section 3.2).
In our weakly interacting system, we see the production of low spatial fre-
quency, long-wavelength components in the wavefunction with the application of
weak incommensurate lattices. This yields insight into the connection between dis-
order, localization, and incommensurate systems. In the canonical view of local-
ization, destructive interference due to reflections over long distances leads to the
72
localization of the wavefunction[1, 56]. Here we see that a single incommensurate
perturber yields a complex long wavelength structure. If we add a second perturber,
getting closer to true disorder, we see even more momentum components and calcu-
lations show the amplitude of the wavefunction is suppressed. Experimentally, due
to finite resolution, this appears as a broadening of the distribution (Fig. 3.3(d)).
Comparing three lattices to two, we observe a larger spread in the central feature,
as well as a corresponding decrease in optical depth. This indicates that there are
unresolved peaks beneath the overall envelope. In Ref.[54] it was suggested that an
indication of localization is the appearance of additional momentum peaks in the
matter-wave interference pattern. If something competes with these long range in-
terferences, such as atom-atom interactions, we can expect localization phenomena
to be drastically modified [40]. Indeed, until recently [53, 6] Anderson localization
had not been seen in cold atom systems due to the effect of atom-atom interactions.
3.6.1 Adding interactions: 1D Gross-Pitaevski equation
To study the effects of interactions in our system, we perform experiments in
three different magnetic traps of differing frequencies with chemical potentials of
?
BEC
/h?(2500,400,100) Hz. ?
BEC
is calculated from
?
BEC
=
1
2?h?
ho
parenleftbigg
15(1?T/T
c
)
3
Na
a
ho
parenrightbigg
2/5
(3.10)
Here, ?
ho
is the geometric mean of the harmonic oscillator frequencies, 2pi(f
x
f
y
f
z
)
1/3
,
a
ho
is the harmonic oscillator length equal to
radicalbig
?h/(M?
ho
). We observe (Fig. 3.7)
73
that as the overall confinement is increased, both the structure in the interference
pattern such as the k
1
?k
2
momentum peaks disappears and the overall size of the
cloud, after TOF, increases.
We compare the results of the non-interacting band-structure model to simu-
lations of our system using the Gross-Pitaevskii equation (GPE),
?
?h
2
2M
?
2
?(r)+V(r)?(r)+g|?(r)|
2
?(r)=??(r). (3.11)
This is Schr?odinger?s equation with a non-linear term with the coefficient, g =
4pi?h
2
a/M. Here, a is the s-wave scattering length. This equation assumes that the
mean field approximation is valid where the wavefunction is a product of single par-
ticle indistinguishable wavefunctions. It is most accurate when the system modeled
has only small deviations from BEC. In our system, this assumption may not be
true because of the localizing effects at weak interactions; however, the results of
the calculation match well with our observations. For more details on the GPE and
approximations for condensates, see Ref. [48, 12].
We reduce the 3D equation to 1D and solve using the split-operator method in
combination with imaginary time evolution [4]. Although our system is not 1D, but
an array of 2D pancakes, this reduction method captures the main features of our
results. The simulations using the GPE predict that in the moderately tight trap
(?
BEC
/h = 400 Hz), the k
1
? k
2
(beat frequency) peaks slightly persist, however
they disappear in the tightest trap. Fig. 3.7 shows absorption images with results
of GP simulation in three different traps. In our system, we cannot decouple the
size of the sample from interactions (since we increase interactions by compressing
74
the trap which decreases the size), which both contribute to the amount of spatial
localization. This can be done if one utilizes a Feshbach resonance, as in [53], to
vary the strength of the scattering length, a.
We calculate the effect of arbitrarily increasing the interactions in our weakest
trap (where the structure is most apparent) and see that the sub-beat frequency
peaks vanish rapidly, followed by the k
1
?k
2
peaks. This can be understood in that
interactions drive transitions between states of different quasimomentum, washing
out the discrete low momentum features, and destroying the long range spatial
periodicity of the wavefunction. The GP simulation also includes the harmonic
potential that is neglected in the band structure calculations. Simulations with our
experimental parameters show that in the limit that interactions do not dominate
(lattice depths<7E
rec
), the harmonic trap increases the effects the secondary lattice
has on the adiabaticity condition. For additional results of GPE calculations, see
thesis of M. Beeler.
3.7 Conclusions
In conclusion, we have presented data indicating that small perturbations to
a one-dimensional lattice system in the form of quasi-disorder, while leading to
localization of the wavefunction, also drastically change the dynamics of the system.
We present a theoretical treatment which suggests that although we are only slightly
modifying the energy of the system, the large alterations of the wavefunction demand
time scales for adiabaticity that are orders of magnitude longer than for loading a
75
3800
3600
3400
3200
3600350034003300320031003000
(I)
-3 -2 -1 0 1 2 3
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
(e)
(b)
-2                         0                         2   
0
0.1
0.2
0.3
0.4
uni03BCBEC?400 Hz; V1+V2;
V1?4.6 Erec; V2?0.6 Erec
-3 -2 -1 0 1 2 3
0
1
2
3
4
5
6
7
(d)
-2                         0                         2   
0
0.1
0.3
0.7
0.5
-3 -2 -1 0 1 2 3
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
2200
2000
1800
1600
Y Position 
(
m)
23002200210020001900180017001600
X Position (m)
?
momentum 
Amplitude (Arb. Units) 
(f)
(II)
(a)
(c)
-2                         0                         2   
0
0.1
0.2
0.3
0.4
uni03BCBEC?2500 Hz; V1+V2;
V1?4.6 Erec; V2?0.6 Erecc
0.5
uni03BCBEC?100 Hz; V1+V2;
V1?4.6 Erec; V2?0.3 Erec
(I)
Figure 3.7: (I) (a-c) Absorption images in traps with ?
BEC
?
(100,400,2500) Hz with primary lattice plus one perturbing lattice. (II)
(d-f) Results of GP simulation for (a-c) qualitatively showing the effects
of interactions.
76
single lattice. Because of the sensitivity of the adiabaticity criteria in the presence
of perturbations, one should be careful when studying disordered lattice systems to
identify and characterize any forms of disorder present in the potential, intentional or
otherwise. This work shows that disorder can have a strong influence on dynamics,
and that the long timescales of cold-atom optical lattice systems makes them ideal
for further explorations.
3.8 Comments and Future directions
3.8.1 Incommensurate lattices as a model for disorder and finite-size
effects
In our type of system, two important questions are (1) how well does the
system approximates true disorder and (2) how much does the finite-size of the
system matter. These are subtle questions, which remain open for further study.
Incommensurate lattices are defined as having a ratio of k-vectors that is an
irrational number. Two lattices that have a ratio that approaches the golden mean
are said to be maximally incommensurate (see [53, 55] and Refs. therein). In a
finite-size system, the effects of deviating away from the incommensurate lattice?s
ratio are not known. Intuitively, it seems that if the atomic cloud only extends over
a small length of the lattice, the potential it feels is pseudorandom and only in the
limit of infinitely long condensates do differences appear. This would be interesting
to be able to study, if one could arbitrarily control the k-vector ratio over a variety
of irrational numbers and characterize the irrationality of the ratio with respect to
77
the golden-mean.
An experiment such as this would also be able to investigate how closely a
finite-sized incommensurate system mimics random disorder, i.e. a potential formed
from speckle. We expect that adding an infinite number of incommensurate per-
turbers would perfectly mimic true disorder. Experimentally, this is not feasible,
but in a finite size system, we expect that many fewer perturbers would be necessary
to accomplish this. Although in [66], they are in the strongly interacting regime, it
is clear when comparing momentum distributions from their experiment and oth-
ers using speckle with our results [21] and those found in Ref. [53] that there is
a fundamental difference between the two types of randomness. It would be very
interesting to investigate the length scales over which this is experimentally valid.
3.8.2 Excited state populations
Using the adiabaticity condition, one can predict the growth of population
into higher bands of the lattice. This calculation is shown in Fig. 3.8. From the
calculation, it is predicted that the population of excited bands grows and different
bands dominate for different depths.
We previously attempted to characterize the excitations (with stronger per-
turbations) by letting the excited fraction relax into the thermal component. After
loading the BEC into the lattice(s) and holding for 1 ms, we ramp off the intensity
of the lattice(s) in 1-20 ms. The sample is held in the magnetic trap for a varying
amount of time. During this hold, we assume that the excited fraction (due to non-
78
popula
tion e
x
cit
ed band (
A
r
b
. Units)
Lattice depth (Erec)
population in 1st allowed 
excited state for a single lattice
population in 1
st 
allowed 
excited state for a combined lattice
population in 1
st
 allowed 
excited state for a single lattice
Figure 3.8: Comparing excited band populations for one and two total
lattices. The strength of the perturbation is 6 percent of the depth of the
primary lattice. After the strength of the primary lattice reaches a crit-
ical value, population growth accelerates in the band which corresponds
to the 1st excited state of the approximate k
1
?k
2
lattice.
79
adiabatic loading) will relax into the thermal component in the magnetic trap. We
do observe qualitatively that for very short loading times (compared to the band
separation) a large thermal fraction emerges. However, this was difficult to quantify
because the lattice turn-off induced breathing modes which did not damp on exper-
imental time-scales. Additionally, if the turn-off of the lattices induces additional
excitations, the two effects cannot be decoupled.
3.8.3 Band-mapping
Currently, we are working on a third, more comprehensive, experimental test
for adiabaticity. We will use a well-known technique called band-mapping in an
attempt to quantify excited state populations. After loading the atoms using the
same sequence as above, the lattice is turned off fast compared to many body in-
teractions, but slow compared to single particle excitations. The populations from
each band are mapped onto the Brillouin zones (ground band populations in first
Brillouin zone, first excited band populations map to second Brillouin zone, etc).
This is a tricky experiment, because as we have shown, the perturbing lattice dras-
tically modifies the energetics of the system. We anticipate that this technique will
break down as the band gaps close. We will start with a primary lattice and slowly
increase the perturbation to characterize this behavior.
80
Chapter 4
Final remarks
The study of BECs in disordered potentials is a rich field of study and there
are several experiments for which this apparatus is well-suited. I have described the
first results from this apparatus and this section will briefly describe a few future
experiments that I have had a role in planning.
4.1 Non-equilibrium dynamics and the effect of lattices on the for-
mation of BECs
Classical and quantum systems typically achieve equilibrium through exchange
of energy with the surroundings. For example, in a system of a dilute, thermal gas
of neutral atoms confined to a harmonic trap undergoing cooling to Bose-Einstein
condensation (BEC), these exchanges are in the form of collisions with atoms that
can subsequently leave the gas. In a crystal, there are exchanges of energy through
phonons and the atoms are ?pinned? to sites through chemical interactions. The
BEC-optical lattice is the atom-optic version of the solid-state crystal, but there
are no phonons and typically no evaporation so the system is isolated from the
environment. In an interacting system, particle-particle interactions play a role in
distributing energy among particles, even if the system is isolated from a reservoir.
Previous theoretical models of non-equilibrium dynamics have focused on de-
81
veloping a quantum picture of the Kibble-Zurek mechanism (KZM) [37, 68]. The
standard picture of this mechanism is that far away from a phase transition, the
system dynamics are adiabatic because the relaxation time, ?, is short, but as one
approaches a critical point, ? diverges. If one quenches the system, defined as cross-
ing the transition in a time, t
Q
, faster than ?, the system is driven out of equilibrium.
Thus, after the transition is passed, the order parameter will develop domains of
differing phase giving rise to topological defects such as vortices. Because this criti-
cal slowing down of the relaxation is inherent in the phase transition, the formation
of defects is inevitable; however, their persistence and density is dependent on the
speed of the quenching. This mechanism was initially developed in a classical system
with applications to defect formation in the early universe [37]. It has been shown
that features of the KZM can be mapped onto systems exhibiting quantum phase
transitions (QPT) such as the MI-SF transition [30, 20].
Quenching across a phase boundary has been used recently to observe vor-
tices during the BEC transition in an ultra-cold atom system [64]. We have begun
studying the dynamics of the condensate transition in the presence of a perturbing
potential. We evaporate to just above T
c
and subsequently turn on a 1D lattice po-
tential created in the manner described in Chapter 3. We continue the evaporation
process and observe the establishment of phase coherence between lattice sites. The
lattice potential is expected to slow down the re-thermalization process by affecting
the collision rate. Indeed, we have observed that evaporation in the presence of a
weak lattice (? 5 E
rec
) dramatically slows the equilibration process. The goal is
to measure this re-thermalization process and the dynamics that lead to relaxation
82
when the system is quenched into a regime that would be condensed. This class of
experiments investigating the KZM are interesting in terms of the defects and give
insight into resulting effects of a quench. The initial and final states of the system
are near-equilibrium, but the system is driven out of equilibrium near the phase
boundary.
Our system design, with the mirror defining the nodes of the lattice potential
being ?2 mm from the atomic sample, is ideal for the long time-scales for this study.
Recent observations suggest that the time for relaxation is on the order of seconds,
which translates into needing low loss and small heating rates. The versatility of our
system provides a platform for extending this study of non-equilibrium phenomena
in a single period lattice to the effects of disorder. Because we can load the atoms into
an incommensurate lattice and plan on implementing a speckle potential we can span
the range from quasi-periodic disorder to true disorder. As previous experiments
have shown, the signatures of disordered phases such as glasses can be difficult
to interpret. A potential method for further identifying a glassy phase may be
the observation of diverging equilibration time-scales through these non-equilibrium
experiments.
4.2 Bloch Oscillations
As described in Section 3.2, when a BEC is trapped in an optical lattice,
energy bands arise due to the periodicity in the system. In a perfect lattice without
defects, the application of a constant force, such as that due to gravity, causes the
83
atoms to oscillate in position and momentum, undergoing Bloch oscillations. The
period of oscillation is T
B
=2pi?h/Fd. Here, F is force and d is the lattice constant.
In our system, the lattice(s) are oriented in the direction of gravity, so we utilize
this as our force. Adding a perturbation to the lattice in the form of disorder, or
pseudodisorder (incommensurate lattice) will damp these oscillations. The disorder
breaks the symmetry and the energy levels become unequally spaced. This induces
dephasing of the matter-wave, thus damping the oscillations [60]. In solids, disorder
completely destroys Bloch oscillations which results in a finite resistance.
If instead, one makes the force time-dependent with an oscillatory frequency of
1/T
B
, one can observe coherent delocalization of the condensate [32]. In the presence
of both the disorder and the time-dependent force, there is a competition between
these two localizing effects [31]. We are studying this in our BEC-1D lattice system.
Thus far, we have observed that minimal disorder will lead to strong damping of
the Bloch oscillations.
4.3 Delta-kicked harmonic oscillator
Previous to experiments with disorder, we unsuccessfully tried to observe res-
onant features in a quantum ?-kicked harmonic oscillator. We hope to revisit this
project because we have made improvements to increase the stability of the system,
which may allow us to observe resonances.
The goal of this experiment was to study the delta-kicked harmonic oscilla-
tor in the context of a quantum mechanical system. Classically there are systems
84
exhibiting chaotic phenomena, such as the ?-kicked rotor. In order to better under-
stand the meaning of quantum chaos, one can choose the quantum analog of these
classical systems and observe their behavior. The delta-kicked rotor has been studied
in the context of cold atoms previously [44]. If one calculates the effect of adding in
a harmonic term, as exists for a magnetic trap, then under certain conditions there
are resonances (where the energy increases dramatically). The Hamiltonian for this
system, ignoring interactions due to a possible condensate fraction, is
H =
p
2
2m
+
M?
2
z
z
2
2
+ V
0
?
p
cos
2
(k
latt
z)
?
summationdisplay
n=0
?(t?nT
K
) (4.1)
Here, V
0
is the strength of the pulsed standing wave having k-vector, k, and pulse
width, ?
p
. n is the number of times the sample is kicked in a harmonic potential
with a trapping frequency, ?
z
.
The ?-kicked harmonic oscillator is also interesting because applications to
electronic transport and it exhibits a ?crystalline? phase space [35]. The effects of
non-linearity on this system due to the presence of a condensate has been investi-
gated theoretically [25] and to our knowledge, only one experiment has attempted
to study this system [19]. We performed calculations on our system to determine
the parameters for the resonances. We designed an experiment loading a cold cloud
(? 0.5?1?K) into an approximately cylindrically symmetric harmonic trap formed
by magnetic fields (see Chapter 2). We chose a non-condensed cloud because the
non-linear term does not contribute. We applied short kicks in the form of standing
waves to the atoms and monitored the energy as a function of kick number and kick
strength. These kicks were in the regime of a phase grating (Section 3.3.2).
85
While the ?-kicked rotor is a relatively simple experiment, the ?-kicked har-
monic oscillator, although seemingly a small extension, is much more sensitive to
experimental details. It requires exquisite stability in the harmonic trapping fre-
quencies. If these frequencies are changing during the kicking sequence, the res-
onances are predicted to broaden, or even disappear. Instabilities due to current
variations in the power supplies can cause large fluctuations in the magnetic fields.
Analysis showed that variations in the relative positions of the trap center and
standing wave washed out the resonances. We were not able to remove all the noise
from the system and thus did not observe resonances.
86
Appendix A
Inventory
The following is a list of the major components of our apparatus.
A.1 Vacuum
Part Number Part Description
Vendor
MCF600-SO200800-A Spherical Octagon vacuum chamber
Kimball Physics
9191145 Varian starcell diode ion pump
Varian Inc.
9290191 Mini-vac ion pump controller and cable
Varian Inc
274042 Dual filament nude bayard type ion gauge
Granville-Phillips
350501-A-T1 Ion gauge controller-350 and cable
Granville-Phillips
Custom Custom feed-throughs
Kurt Lesker Inc.
SG0150MCCF Manual gate valve
Kurt Lesker Inc.
B1381S Glass to metal seal 2.75 inch con-flat flange
Bomco
Rb/NF/6.4/17/FT (website) Rb-Alkali metal dispensers
SAES getters
37-3736 Argon packed Rb glass ampoule, 1g
87
Strem Chemical
MCF450-SS20400-A Spherical Square vacuum chamber
Kimball Physics Inc.
VA-30-DD-M, V-410-MS Varian-type 30 L/s ion pump and cable
Duniway stockroom
IPC-0062 Ion pump controller, diode or triode
Duniway stockroom
for custom flange rotary feedthrough piece (source)
Kurt Lesker
custom AlN mount for wire traps
Stratamet, Inc.
custom 780 nm and 1550 nm coatings on windows
Spectrum ThinFilms
A.2 Lasers, Fibers, AOMs, EOMs, Optoisolators
Part Number Part Description
Vendor
DLX-110 780 nm, 750 mW tapered amplifier diode laser
Toptica Photonics Inc.
?- laser controller
Toptica Photonics Inc.
ELR-15K-LP-SF 1550 nm, 15 W fiber laser
IPG Photonics
TEC A00-Littrow TEC series 780 nm, 100 mW diode laser
Sacherlaser
?- laser controller
Sacherlaser
TIS-SF-07 ring laser 800 nm tunable Ti-Sapphire Laser
Tekhnoscan
Verdi-10W 532 nm , 10 W Verdi, controller, cooler
88
Coherent
?- custom 4? 4 fiber splitter
Canadian Instrumentation
2007 Nirvana detector
New Focus
1206C-2-840B 80 MHz AOM, coating-780
Isomet
DE-802MN Deflection driver, 80 MHz
IntraAction
360-40, 302RM EOM, EOM driver, for dipole trap
ConOptics
4002-Visible EOM, laser lock
New Focus
A.3 Power Supplies
Part Number Part Description
Vendor
DLM-600 Sorensen 8V-75A power supply
Elgar Electronics
XLT-60 Sorensen 4V-15A power supply
Elgar Electronics
ATE 15-3 Kepco 1/4 rack supply
Test Equity
ATE 3-10 Kepco 1/2 rack supply
Test Equity
HP 6269B HP linear supply DC 0-40 V, 0-50 A
Ebay
A.4 Miscellaneous parts and notable electronic bits
Part Number Part Description
89
Vendor
LS6T2 mechanical shutter
Uniblitz
TIA 1000-1-R8 35 dB gain RF power amplifier
MiniCircuits
PTS 160 Frequency synthesizer
Ebay
ABL-200-40-10-LT40AS-NC air-bearing stage
Aerotech
LB1005 Lock box
Precision Photonics
09F2013, 09F2384, 09F2213 Varistor, littlefuse
Newark
96B6121 IGBT (GA200SA60U)
originally international rectifier,
but has been sold; check Newark
MJ4502G Darlington high current MOSFET, OnSemi
Newark
90
Bibliography
[1] M.H. Anderson, J.R. Ensher, M.R. Matthews, C.E. Wieman, and E.A. Cornell.
Observation of Bose-Einstein condensation in a dilute atomic vapor. Science,
269:198, 1995.
[2] Neil W. Ashcroft and N. David Mermin. Solid State Physics. Brooks Cole, 1st
edition, January 1976.
[3] S. Aubry and G. Andr?e. Analyticity breaking and Anderson localization in
incommensurate lattices. Ann. Israel. Phys. Soc., 3:133, 1980.
[4] Weizhu Bao, Dieter Jaksch, and Peter A. Markowich. Numerical solution of
the Gross-Pitaevskii equation for Bose-Einstein condensate. Journal of Com-
putational Physics, 187:318, 2003.
[5] Thomas E. Barrett, Samuel W. Dapore-Schwartz, Mark D. Ray, and Gregory P.
Lafyatis. Slowing atoms with ?
?
polarized light. Phys. Rev. Lett., 67(25):3483,
1991.
[6] Juliette Billy, Vincent Josse, Zhanchun Zuo, Alain Bernard, Ben Hambrecht,
Pierre Lugan, David Cl?ement, Laurent Sanchez-Palencia, Phillipe Bouyer, and
Alain Aspect. Direct oberservation of Anderson localization of matter waves in
a controlled disorder. Nature, 453:891, 2008.
[7] Eric D. Black. An introduction to Pound-Drever-Hall laser frequency stabiliza-
tion. Am. J. Phys, 69(1):79, 2001.
[8] Immanuel Bloch, Jean Dalibard, and Wilhelm Zwerger. Many-body physics
and ultracold gases. Rev. Mod. Phys., 80:885, 2008.
[9] Steven Chu, L. Hollberg, J.E. Bjorkholm, Alex Cable, and A. Ashkin. Three-
dimensional viscous confinement and cooling of atoms by resonance radiation
pressure. Phys. Rev. Lett., 55:48, 1985.
[10] Kristan L. Corwin, Zheng-Tian Lu, Carter F. Hand, Ryan J. Epstein, and
Carl E. Wieman. Frequency-stabilized diode laser with the Zeeman shift in an
atomic vapor. Applied Optics, 37(15):3295, 1998.
[11] M. Cristiani, O. Morsch, J.H. M?uller, D. Ciampini, and E. Arimondo. Experi-
mental properties of Bose-Einstein condensates in one-dimensional optical lat-
tices: Bloch oscillations, Landau-Zener tunneling and mean-field effects. Phys.
Rev. A, 65:063612, 2002.
[12] Franco Dalfovo, Stefano Giorgini, Lev. P. Pitaevskii, and Sandro Stringari.
Theory of Bose-Einstein condesation of dilute gases. Rev. Mod. Phys., 71:463?
512, 1999.
91
[13] B. Damski, J. Zakrzewski, L. Santos, P. Zoller, and M. Lewenstein. Atomic
Bose and Anderson glasses in optical lattices. Phys. Rev. Lett., 91(8):080403,
2003.
[14] K.B. Davis, M.-O. Mewes, M.R. Andrews, N.J. Van Druten, D.S. Durfee, D.M.
Kurn, and W. Ketterle. Bose-Einstein condensation in a gas of sodium atoms.
Phys. Rev. Lett., 75:3969, 1995.
[15] Wolfgang Demtroeder. Laser Spectroscopy: Basic concepts and instrumenta-
tion. Springer, 3rd edition, November 2002.
[16] J. Hecker Denschlag, J. E. Simsarian, H. Haffner, C. McKenzie, A. Browaeys,
D. Cho, K. Helmerson, S.L. Rolston, and W.D. Phillips. Bose-Einstein conden-
sate in an optical lattice. J. Phys. B., 35:3095, 2002.
[17] Roberto B. Diener, Georgios A. Georgakis, Jianxin Zhong, Mark Raizen,
and Qian Niu. Transition between extended and localized states in a one-
dimensional incommensurate optical lattice. Phys. Rev. A, 64:033416, 2001.
[18] Elizabeth A. Donley, Neil R. Claussen, Simon L. Cornish, Jacob L. Roberts,
Eric A. Cornell, and Carl E. Wieman. Dynamics of collapsing and exploding
Bose-Einstein condensates. Nature, 412(19):295, 2001.
[19] G.J. Duffy, A.S. Mellish, K.J. Challis, and A.C. Wilson. Nonlinear atom-optical
?-kicked harmonic oscillator using a Bose-Einstein condensate. Phys. Rev. A,
70:041602(R), 2004.
[20] J. Dziarmaga, A. Smerzi, W.H. Zurek, and A.R. Bishop. Dynamics of quantum
phase transition in an array of Josephson junctions. Phys. Rev. Lett., 88:167001,
2002.
[21] E. Edwards, M. Beeler, Tao Hong, and S.L. Rolston. Adiabaticity and local-
ization in incommensurate lattices. Phys. Rev. Lett., 101:260402, 2008.
[22] M.P.A. Fisher, P.B. Weichman, G. Grinstein, and D.S. Fisher. Boson localiza-
tion and the superfluid-insulator transition. Phys. Rev. B, 40:546, 1989.
[23] A. Flesch, M. Cramer, I.P. McCulloch, U. Schollw?ock, and J. Eisert. Probing
local relaxation of cold atoms in optical superlattices. Phys. Rev. A, 78:033608,
2008.
[24] J. Fortagh, A. Grossmann, T.W. H?ansch, and C. Zimmermann. Fast load-
ing of a magneto-optical trap from a pulsed thermal source. J. App. Phys.,
84(12):6499, 1998.
[25] S.A. Gardiner, D. Jaksch, R. Dum, J.I. Cirac, and P. Zoller. Nonlinear matter
wave dynamics with a chaotic potential. Phys. Rev. A, 62:023612, 2000.
92
[26] Markus Greiner et al. Quantum phase transition from a superfluid to a Mott
insulator in a gas of ultracold atoms. Nature, 415:39, 2002.
[27] D.R. Grempel, Shmuel Fishman, and R.E. Prange. Localization in an incom-
mensurate potential: An exactly solvable model. Phys. Rev. Lett., 49(11):833,
1982.
[28] Rudolf Grimm, Matthias Weidemuller, and Yurii B. Ovchinnikov. Optical
dipole traps for neutral atoms. Adv. At. Mol. Opt. Phys., 42:95, 2000.
[29] L. Guidoni, C. Triche, P. Verkerk, and G. Grynberg. Quasiperiodic optical
lattices. Phys. Rev. Lett., 79(18):3363, 1997.
[30] J. Dziarmaga H. Arodz and W.H. Zurek, editors. Patterns of symmetry break-
ing, Cracow, Poland, September 2003. Proceedings of the NATO Advanced
Study Institute on Patterns of Symmetry Breaking, Kluwer Academic Publish-
ers.
[31] M. Holthaus, G.H. Ristow, and D.W. Hone. Random lattices in combined a.c.
and d.c. electric fields: Anderson vs. Wannier-Stark localization. Europhys.
Lett., 32:241, 1995.
[32] V.V. Ivanov, A. Alberti, M. Schioppo, G. Ferrari, M. Artoni, M.L. Chiofalo,
and G.M. Tino. Coherent delocalization of atomic wave packets in driven lattice
potentials. Phys. Rev. Lett, 100:043602, 2008.
[33] B.D. Josephson. Coupled superconductors. Rev. Mod. Phys., 36:216, 1964.
[34] P.S. Julienne, C.J. Williams, Y.B. Band, and Marek Trippenbach. Loading
Bose-Einstein condensed atoms into the ground state of an optical lattice. Phys.
Rev. A, 72:053615, 2005.
[35] G.A. Kells, J. Twamley, and D.M. Hefferman. Dynamical properties of the
delta-kicked harmonic oscillator. Phys. Rev. E, 70:015203, 2004.
[36] Wolfgang Ketterle and N.J. Van Druten. Evaporative cooling of trapped atoms.
Adv. At. Mol. Opt. Phys., 37:181, 1996.
[37] T.W.B. Kibble. Topology of cosmic domains and strings. J. Phys. A, 9:1387,
1976.
[38] D. Kleppner, T.J. Greytak, T.C. Killian, D.G. Fried, L. Willmann, D. Landhuis,
and S.C. Moss. Bose-Einstein Condensation in Atomic Gases, Proceedings of
the international school of Physics ?Enrico Fermi,? Course CXL. IOS Press,
Amsterdam, 1999.
[39] P.J. Lee, M. Anderlini, B.L. Brown, J. Sebby-Strabley, W.D. Phillips, and J.V.
Porto. Sublattice addressing and spin-dependent motion of atoms in a double
well lattice. Phys. Rev. Lett., 99:020402, 2007.
93
[40] J.E. Lye, L. Fallani, C. Fort, V. Guarrera, Michele Modugno, D.S. Wiersma,
and Massimo Inguscio. Effect of interactions on the localization of a Bose-
Einstein condensate in a quasiperiodic lattice. Phys. Rev. A, 75:061603, 2007.
[41] J.E. Lye, L. Fallani, Michele Modugno, D.S. Wiersma, C. Fort, and Massimo
Inguscio. Bose-Einstein condensate in a random potential. Phys. Rev. Lett.,
95:070401, 2005.
[42] Harold J. Metcalf and Peter van der Straten. Laser Cooling and Trapping.
Graduate Texts in Contemporary Physics. Springer, New York, 1999.
[43] Michele Modugno. Collective dynamics and expansion of a Bose-Einstein con-
densate in a random potential. Phys. Rev. A, 73:013606, 2006.
[44] F.L. Moore, J.C. Robinson, C. Bharucha, P.E. Williams, and M.G. Raizen.
Observation of dynamical localization in atomic momentum transfer: A new
testing ground for quantum chaos. Phys. Rev. Lett., 73:2974, 1994.
[45] Oliver Morsch and Markus Oberthaler. Dynamics of Bose-Einstein condensates
in optical lattices. Rev. Mod. Phys., 78:179, 2006.
[46] Yu. B. Ovchinnikov, J.H. M?uller, M.R. Doery, E.J.D. Vrendenbregt, K. Helmer-
son, S.L. Rolston, and W.D. Phillips. Diffraction of a released Bose-Einstein
condensate by a pulsed standing light wave. Phys. Rev. Lett., 83(2):284, 1999.
[47] S. Peil, J.V. Porto, B. Laburthe Tolra, J.M. Obrecht, M. Subbotin, S. L. Rol-
ston, and W.D. Phillips. Patterned loading of a Bose-Einstein condensate into
an optical lattice. Phys. Rev. A, 67:051603, 2003.
[48] C.J. Pethick and H. Smith. Bose-Einstein Condensation in Dilute Gases. Cam-
bridge University Press, New York, 2002.
[49] Anatoli Polkovnikov and Vladimir Gritsev. Breakdown of the adiabatic limit
in low-dimensional gapless systems. Nature Physics, 4:478, May 2008.
[50] E. Raab, M. Prentiss, A. Cable, S. Chu, and D. Pritchard. Trapping of neutral-
sodium atoms with radiation pressure. Phys. Rev. Lett., 59:2631, 1987.
[51] Umakant D. Rapol, Ajay Wasan, and Vasant Natarajan. Loading of a Rb
magneto-optic trap from a getter source. Phys. Rev. A, 64:023402, 2001.
[52] J. Reichel, W. H?ansel, and T.W. H?ansch. Atomic micromanipulation with
magnetic surface traps. Phys. Rev. Lett., 83:3398, 1999.
[53] Giacomo Roati, Chiara D?Errico, Leonardo Fallani, Marco Fattori, Chiara Fort,
Matteo Zaccanti, Giovanni Modugno, Michele Modugno, and Massimo Ingus-
cio. Anderson localization of a non-interacting Bose-Einstein condensate. Na-
ture, 453:895, 2008.
94
[54] R. Roth and K. Burnett. Phase diagram of bosonic atoms in two-color super-
lattices. Phys. Rev. A., 68:023604, 2003.
[55] G. Roux, T. Barthel, I.P. McCulloch, C. Kollath, U. Schollw?ock, and T. Gia-
marchi. Quasiperiodic Bose-Hubbard model and localization in one-dimensional
cold atomic gases. Phys. Rev. A, 78:023628, 2008.
[56] L. Sanchez-Palencia, D. C
?
lement, P. Lugan, P. Bouyer, G.V. Shlyapnikov, and
A. Aspect. Anderson localization of expanding Bose-Einstein condensates in
random potentials. Phys. Rev. Lett., 98:210407, 2007.
[57] L. Sanchez-Palencia and L. Santos. Bose-Einstein condensate in optical qua-
sicrystal lattices. Phys. Rev. A, 72:053607, 2005.
[58] L. De Sarlo, L. Fallani, J.E. Lye, M. Modugno, C. Fort, and M. Inguscio.
Unstable regimes for a Bose-Einstein condensate in an optical lattice. Phys.
Rev. A, 72:013603, 2005.
[59] L.I. Schiff. Quantum Mechanics. McGraw-Hill, New York, 1955.
[60] T. Schulte, S. Drenkelforth, G. Kleine B?uning, W. Ertmer, J. Arlt, M. Lewen-
stein, and L. Santos. Dynamics of Bloch oscillations in disordered lattice po-
tentials. Phys. Rev. A, 77:023610, 2008.
[61] T. Schulte, S. Drenkelforth, J. Kruse, W. Ertmer, J . Arlt, K. Sacha, J. Za-
krzewski, and M. Lewenstein. Route towards Anderson-like localization of Bose-
Einstein condensates in disorder optical lattices. Phys. Rev. Lett., 95:170411,
2005.
[62] Daniel A. Steck. Rubidium 87 line data. available online at
http://steck.us/alkalidata, (revision 2.0.1, 2 May 2008).
[63] M.R. Walkiewicz, P.J. Fox, and R.E. Scholten. Candlestick rubidium beam
source. Rev. Sci. Instrum., 71:3342, 2000.
[64] C.N. Weiler, T.W. Neely, D.R. Scherer, A.S. Bradley, M.J. Davis, and B.P.
Anderson. Spontaneous vortices in the formation of Bose-Einstein condensates.
Nature, 455:948, 2008.
[65] David S. Weiss, Erling Riis, Yaakov Shevy, P. Jeffrey Ungar, and Steven Chu.
Optical molasses and multilevel atoms: experiment. J. Opt. Soc. Am. B,
6(11):2072, November 1989.
[66] M. White, M. Pasienski, D. McKay, S. Zhou, D. Ceperley, and B. Demarco.
Strongly interacting bosons in a disordered optical lattice. e-print cond-
mat/0807.0446, 2008.
[67] S. Wildermuth, P. Kr?uger, C. Becker, M. Brajdic, S. Haupt, A. Kasper, R. Fol-
man, and J. Schmiedmayer. Optimized magneto-optical trap for experiments
with ultracold atoms near surfaces. Phys. Rev. A, 69:030901, 2004.
95
[68] W.H. Zurek. Cosmological experiments in superfluid helium? Nature (London),
317:505, 1985.
96