Interference pattern and visibility of a Mott insulator

Fabrice Gerbier, Artur Widera, Simon Fölling, Olaf Mandel, Tatjana Gericke and Immanuel Bloch
Institut für Physik, Johannes Gutenberg-Universität, 55099 Mainz, Germany.
(Dated: February 2, 2008)
We analyze theoretically the experiment reported in [F. Gerbier et al, cond-mat/0503452], where the in-
terference pattern produced by an expanding atomic cloud in the Mott insulator regime was observed. This
interference pattern, indicative of short-range coherence in the system, could be traced back to the presence
arXiv:cond-mat/0507087v2 [cond-mat.other] 5 Jul 2005

of a small amount of particle/hole pairs in the insulating phase for finite lattice depths. In this paper, we an-
alyze the influence of these pairs on the interference pattern using a random phase approximation, and derive
the corresponding visibility. We also account for the inhomogeneity inherent to atom traps in a local density
approximation. The calculations reproduce the experimental observations, except for very large lattice depths.
The deviation from the measurement in this range is attributed to the increasing importance of non-adiabatic
effects.

PACS numbers: 03.75.Lm,03.75.Hh,03.75.Gg

The superfluid to Mott insulator (MI) transition undergone The paper is organized as follows. In section I, we recall
by an ultracold Bose gas in an optical lattice has attracted the description of ultracold atoms in an optical lattice by the
much attention in the recent years as a prototype for strongly Bose-Hubbard model, and discuss the inhomogeneous shell
correlated quantum phases [1, 2, 3, 4]. A key observable in structure that develops in an external confining potential. Sec-
these systems is the interference pattern observed after releas- tion II presents the calculation of the interference pattern ob-
ing the gas from the lattice and letting it expand for a certain served after free expansion of the atom cloud and its link with
time of flight. Monitoring the evolution of this interference the quasi-momentum distribution. The main results are pre-
pattern not only reveals the superfluid-to-MI transition [2, 4], sented in sections III and IV, where we respectively present
but also allows for example the detection of number-squeezed the calculation of the interference pattern in the uniform case
states in the lattice [5, 6], or the observation of collapse and using the RPA, and extend it to the inhomogeneous case to
revivals of coherence due to atomic interactions [7]. Because compare to the experimental data of [8]. Details of the calcu-
of its experimental importance, a quantitative understanding lation are described in the appendix.
of this interference signal is crucial to characterize quantum
phases of bosons in optical lattices.
I. BOSE-HUBBARD HAMILTONIAN
Although no interference pattern is expected for a uniform
array of Fock states (what we call a “perfect” Mott Insula-
tor) [29], a finite visibility is nevertheless observed in exper- In this section, we briefly recall the theoretical description
iments above the insulator transition [2, 4, 8], in agreement of an ultracold atomic gas trapped in an optical lattice. The
with numerical calculations [9, 10]. We have studied this phe- optical lattice potential, which results from the superposition
nomenon experimentally, and shown that despite its insulating of three orthogonal and independent pairs of counterpropagat-
nature that forbids long-range coherence, a MI still exhibits ing laser beams, can be written as
short-range coherence at the scale of a few lattice sites [8].  
This can be attributed to the structure of the ground state for VOL (r) = V0 sin2 (kL x) + sin2 (kL y) + sin2 (kL z) , (1)
finite lattice depths, which consists of a small admixture of
Here V0 is the lattice depth, kL = 2π/λL is the laser wavevec-
particle/hole pairs on top of a perfect MI. A qualitative model
tor, λL is the laser wavelength and m is the atomic mass.
based on a lowest-order calculation of the ground state wave-
As usual, we measure V0 in units of the single-photon recoil
function was also presented in our previous work, which re-
energy ER = h2 /2mλ2L . The lattice potential has a simple
produced the main trend and order of magnitude of the ob-
cubic periodicity in three dimensions, with a lattice spacing
served visibility.
d = λL /2 ≈ 425 nm in our case. As shown in [1], the behavior
In the present paper, we would like to present a more pre- of the atomic system in such a potential can be described by
cise calculation that includes higher order corrections (see the Bose-Hubbard model, defined by the hamiltonian
also [11]). We describe a MI state at zero temperature using
the Random Phase Approximation (RPA), already introduced X XU
in Refs. [12, 13, 14, 15, 16]. Instead of the path integral ap- H = −t â†i â j + n̂i (n̂i − 1) . (2)
proach used by these authors, we obtain here the RPA Green’s hi, ji i
2
function using a different method inspired by Hubbard’s orig-
inal treatment of the fermionic model [17, 18]. Taking the Here the operator â†i creates an atom at site i, n̂i = â†i âi is the
experimental geometry and the inhomogeneous particle distri- on-site number operator, and the notation hi, ji restricts the
bution into account, we find good agreement with our exper- sum to nearest neighbors only. The relative strength between
imental data, which provides further support for the physical the tunneling matrix element t and the on-site interaction en-
picture presented above. ergy U is controlled by the depth V0 of the periodic potential
 2

which confines the atoms [30]. The phase diagram of this where z = 6 is the number of nearest neighbors in 3 dimen-
hamiltonian is well known: The system is in a MI state within sions. In the specific example shown in Fig. 1, we have chosen
characteristic lobes in a t/U versus chemical potential µ phase V0 = 18 ER and N = 2.2 × 105 atoms, so that both n0 = 1 and
diagram, and in a superfluid state outside of these lobes [19]. n0 = 2 MI are present. Similarly, we calculate for our experi-
In the experiments, an additional potential Vext (r) is super- mental parameters that an n0 = 3 shell is also present for atom
imposed to the lattice potential, leading to a spatially varying numbers larger than 2.7 × 105 , and lattice depths larger than
chemical potential across the cloud. This favors the formation 16 ER .
of a “wedding cake” structure of alternating MI and superfluid
shells, which reflects the phase diagram of the Bose-Hubbard
model [1, 9, 19, 20]. The external potential is due to a combi-
nation of a magnetic potential in which the condensate is ini-

Density nd 3
tially formed and of an optical potential due to the Gaussian 2
shape of the lattice beams. To a good approximation, it can be
considered as a harmonic potential with trapping frequency
r 1
8V0
Ω= ω2m + , (3)
mw2

where ωm is the frequency of the magnetic trap, assumed

isotropic, and where w is the waist (1/e2 radius of the intensity
0 10 20 30
profile) of the lattice beams, assumed identical for all axes. radius r/d
For large lattice depths, the confinement is mainly due to the
optical part.
In the current experiments, this external potential varies FIG. 1: Calculated density profile for a lattice depth V0 = 18 ER and
slowly across the lattice. In this limit, the shell structure N = 2.2 × 105 atoms. Here d is the lattice spacing. The upper solid
can be calculated in a local density approximation, which as- line indicates the numerical calculation of the total density, and the
sumes a known relation nh [µ] between the density n and the dotted line is the U → ∞ extrapolation (see section IV).
chemical potential µ for the homogeneous system. Then, the
coarse-grained density [31] for the inhomogeneous system is
calculated as n(r) = nh [µ − VRext (r)]. The chemical potential
is fixed by the relation N = d(3) r n(r). For a fixed lattice II. INTERFERENCE PATTERN
depth and atom number, we calculate numerically the relation
nh [µ] using mean-field theory at zero temperature, i.e. in the We now turn to the description of the interference pattern
mean-field ground state [12, 21]. We then repeat the steps observed after release of the atom cloud from the optical lat-
outlined above, varying the chemical potential until the target tice and a period of free expansion. From an absorption image
atom number is obtained within 0.1 %. For all calculations, of such a pattern, the phase coherence of the atomic sample
the values ωm = 2π × 15 Hz and w = 130 µm are used, which can be directly probed. The density distribution of the expand-
match our experimental parameters. We show in Fig. 1 an ex- ing cloud after a time of flight tf can be calculated as [3, 9, 24]
ample of such a calculation for a lattice depth of V0 = 18 ER
and N = 2.2 × 105 atoms. !3 !2 !
m mr mr
The presence of the external potential significantly affects n(r) = w̃ S k= , (5)
the atom distribution in the lattice, which is determined by ~tf ~tf ~tf
the competition between interaction and potential energy. On which mirrors the momentum distribution of the original
the one hand, expanding the cloud minimizes the density and cloud. Momentum space and real space in the image plane
the interaction energy, and on the other, contracting it mini- are related by the scaling factor ~tf /m - independent of the
mizes the potential energy, as in conventional harmonic traps lattice parameters. The envelope |w̃|2 is the Fourier transform
[22, 23]. The latter is favored at low atom numbers, where of the Wannier function in the lowest Bloch band. When each
only a n0 = 1 shell forms. When more atoms are added, the potential well is approximated as an harmonic potential, the
radius of the unity filled MI region increases until a critical Wannier function is the corresponding gaussian ground state
atom number, which we estimate to be 7 × 104 atoms for our wavefunction. The envelope function in Eq. (5) then reads
parameters, is reached. Above this critical atom number, a !2  2
higher density core appears near the trap center. A MI with 2 mr 1  r 
w̃ ≈ 3/2 exp − 2 , (6)
atoms per site is then obtained near the trap center if the lattice ~tf π wtf wtf
depth is above the critical value V0 ≈ 14.7 ER . This value has
where wtf = ~tf /mw0 , and where w0 is the size of the on-site
been calculated using the boundary derived in [12],
Wannier function. Finally, we have defined the quantity
X
eik·(ri −r j ) hâ†i â j i.
U  p
= 2n0 + 1 + 2 n0 (n0 + 1), (4) S(k) = (7)
zt n0 i, j
 3

When k is restricted to the first Brillouin zone, S(k) is nothing even “stretch” over a few lattice sites. This mobility acquired
else than the quasi-momentum distribution. Information about by particle/hole pairs is reflected in the modified dispersion re-
the many-body system is contained in this quantity, which is lation (9), which explicitly includes the band structure. Note
periodic with the periodicity of the reciprocal lattice 2π/d. finally that higher order excitations, corresponding to occupa-
Thus, to predict the interference pattern and compare to the tion numbers n0 ± 2, n0 ± 3, ... are neglected. At zero tempera-
experiments, our goal is to calculate S(k) for a given lattice ture, such excitations become important only very close to the
depth and density. superfluid transition where the MI is destroyed.
The quasi-momentum distribution can be di-
rectly deduced
R using the general relation S(k) =
III. QUASI-MOMENTUM DISTRIBUTION IN THE −i limδt→0+ dω 2π G(k, ω)e −iωδt
. Using (8), one has [15]
HOMOGENEOUS MOTT INSULATOR
 
For simplicity, we consider first the case of uniform filling  tk 1 
 2 + U(n 0 + 2 ) 1 
in the lattice, i.e. an integer number n0 of atoms per site, and S(k) = n0  q −  . (10)
2 
tk + 4tk U(n0 + 12 ) + U 2
 2
we assume the system to be at zero temperature and in the
insulating phase. In the limit of zero tunneling, the ground
state wavefunction is a perfect MI, i.e. a product of number To first order in t/U, this reduces to
states at each site, and its Green function G0 can be calcu-
lated exactly (see appendix). The lowest-lying excited states S(k) ≈ n0 − 2n0 (n0 + 1)tk /U, (11)
of the system are “particle” and “hole” states, where a supple-
mentary particle is added (respectively removed) at one lat- also obtained in [8] by calculating the many-body wave func-
tice site. Creating these excitations costs a finite interaction tion perturbatively. We find that the two predictions rapidly
energy, respectively E (+) = Un0 and E (−) = U(n0 − 1) [19]. converge. For example, they differ by less than 10 % for
To calculate the quasi-momentum distribution for a finite U/zt > 6.6, 11.6 and 16.9 for n0 = 1, 2 and 3, respectively.
tunneling t, many-body techniques can be applied to obtain These values have to be compared to the respective critical
the single-particle Green function, G(k, ω). Using a path in- values for MI formation, U/zt = 5.83, 9.89, 13.93. This in-
tegral approach, several authors [12, 13, 14, 15, 16] have dicates that the coherence beyond nearest neighbors is rather
been able to calculate the Green function of the Mott insulator rapidly lost as one goes further into the MI phase. However,
within the RPA, the visibility itself remains finite in a substantial range of U/zt,
implying a persistent short-range coherence.
1 Zk 1 − Zk
G(k, ω) = (+)
+ . (8)
~ ~ω + µ − Ek ~ω + µ − Ek(−)
IV. COMPARISON WITH THE EXPERIMENTS
The poles Ek(±) of the Green function are the quasi-particle
energies [12] To compare with the experiments reported in [8], several
tk 1 1 features have to be taken into account. First, only the col-
Ek(±) =+ U(n0 − ) ± Dk [n0 ], (9) umn density n is accessible experimentally, i.e. the density
2 2 2
P integrated along the probe line-of-sight (which we take here
In Eq. (9), tk = −2t ν=x,y,z cos(kν d) is the dispersion re- parallel to the z axis). Second, the visibility is experimentally
lation for q
a free particle in the tight-binding limit, and deduced from two points according to
Dk [n0 ] = tk2 + 4tk U(n0 + 12 ) + U 2 . The particle weight is
nmax − nmin
Vexp = . (12)
 
Zk = Ek(+) + U /Dk [n0 ]. In the Appendix, we present an nmax + nmin
alternative derivation of (8) based on the equation of motion
method, which follows closely Hubbard’s method [17, 18]. To eliminate the Wannier envelope, nmax and nmin are mea-
Here we will simply comment on the physical picture behind sured from two points at the same distance from the cloud
this approach. The RPA considers that the particle/hole na- center (see Fig. 2), so that the envelope automatically can-
ture of the low-lying excitations is not significantly changed cels out. For√ example, √ nmax is found at point (2π/d, 0) and
by introducing a finite tunneling (in technical terms, the self nmin at (2π/ 2d, 2π/ 2d). This reduces the visibility com-
energy remains approximately the same as in the t → 0 limit). pared to the usual definition. In the theoretical calculation it
The first effect of tunneling is to introduce a finite amount of is straightforward to account for these two effects.
particle/hole components in the the many-body ground state The third effect, the shell structure of the MI, is handled
wavefunction. In the form given in [8], corresponding to a here in an approximate way. In the numerical calculations,
first order calculation, a particle/hole pair necessarily occu- the shell distribution always includes small regions with non-
pies two neighboring lattice sites due to the particular form integer filling, which the theory above cannot handle. How-
of the tunneling hamiltonian. Through higher-order tunneling ever, these domains are small, and have a strongly depleted
processes captured by the Green function (8), the particle and superfluid component, so that we do not expect them to have
the hole forming the pair can tunnel independently. As a re- a large effect on the visibility. Therefore, we approximate the
sult, the pair acquires a mobility through the lattice, and may density distribution by a “ziggurat”-like profile, where only
 4

Visibility V
10-1


π
a
FIG. 2: Measurement of visibility. The interference pattern shown in
10 15 20 25
the left graph corresponds to a lattice depth of 8 ER , in the superfluid
regime. The right graph indicates the geometry of the reciprocal lat-
tice. Gray areas are the first and second Brillouin zones (projected
Lattice depth (ER)
in the image plane), and the white dot indicate the position of the
maxima of the interference pattern. Along the circle, the Wannier
1
function envelope takes the same value, and we measure the interfer-
ence “minimum” at the intersection of this circle and of the diagonal
of the lattice square lattice, indicated by the black dot.

Visibility V
MI shells are present. The actual extension of each shell is
calculated as if t were zero, taking the external potential into 10-1
account [25]. In Fig. 1, we compare the profile in this ap-
proximation (dotted line) with the numerically calculated one
(solid line). For large lattice depths, both agree reasonably. b
Note that the density profile still depends weakly on the lat-
10 15 20 25
tice depth through the external confinement [see Eq. (3)].
The momentum distribution deduced from Eqs. (5,10) is Lattice depth (ER)
averaged over the distribution of atoms to compare with the
experimental data (see [8] for details on the experiment). The
results are plotted versus lattice depth in Fig. 3, for two differ-
FIG. 3: Comparison between the measured and the calculated vis-
ent atom numbers in the lattice. For the lowest atom number
ibility. The upper and lower graphs correspond to atom numbers
N = 2.2 × 105 , we calculate that only n0 = 1 and n0 = 2 shells N = 2.2× 105 and N = 5.6× 105 , respectively. The dotted and dashed
are present. For the largest N = 5.6 × 105 , a core with n0 = 3 lines indicate the calculated visibility for homogeneous MI with fill-
atoms per site is also present. Note that in the latter case, the ing factor n0 = 1, 2. The solid lines are calculations including the
actual density distribution might deviate more from the cal- inhomogeneous shell distribution. Typical standard deviations for
culated one, due to three-body losses in the n0 = 3 region. the experimental data are 1 % or below. Our calculation of the equi-
We find that the calculation agrees with the measured visibil- librium distribution at zero temperature indicate that in case a, only
ity within 20 % for V0 ≤ 22 ER . The theory curves terminate MI regions with n0 = 1 and n0 = 2 atoms per site form, whereas in
when the MI shell with highest filling disappears, as it is re- case b, a core with n0 = 3 is also present.
placed by a large superfluid core not described by our theory.
Note that the calculation does not include any free parameter.
However, we consistently find that the calculated value lies visibility in the MI may be a sensitive probe of the many-body
below the measured visibility for large lattice depths V0 ≤ dynamics of the superfluid-to-insulator transition.
22 ER . Moreover, the deviation increases with increasing lat-
tice depths, which shows that the superfluid shells play little
role in determining the visibility for such large lattice depths, V. CONCLUSION
as assumed in our calculation. In Fig. 4, the fractional de-
viation of the calculated visibility from the measured one is In conclusion, we have derived in this paper a theoretical
plotted versus lattice depth for four data sets. Remarkably, al- expression for the interference pattern of a Mott insulator af-
though the atom numbers are rather different from one data set ter release from the optical lattice and a time of flight. Our
to another we find a common trend in the data. On the other calculations take deviations from perfect filling due to a finite
hand, this observation also suggests that a breakdown of adia- tunneling into account, and use a simplified but realistic model
baticity occurs for the particular ramp used in the experiments of the shell structure of the MI. Good agreement with our ex-
to increase the laser intensity to its final value, a point already perimental data reported in [8] is found, at least for moderate
identified in [8]. We conclude that, perhaps surprisingly, the lattice depths. For very large lattice depths, an increasing de-
 5

APPENDIX A: GREEN FUNCTION OF THE

HOMOGENEOUS MOTT INSULATOR IN THE RANDOM
Fractional deviation 0.6 PHASE APPROXIMATION
from measurement
0.4 In this Appendix, we present a derivation of Eq. (8) using
the equation of motion approach. The single-particle Green
0.2 function is defined at zero temperature as

Gi j (t) = −ihT âi (0)â†j (t)i (A1)

0
= −iθ(t)hâi (0)â†j (t)i
-0.2 −iθ(−t)hâ†j (t)âi (0)i,
15 20 25 30
Lattice depth (ER) where T is the time-ordering operator and θ is the Heaviside
step function. Since we consider a time-independent and ho-
mogeneous system, we take a Fourier transformation of this
FIG. 4: Fractional deviation of the calculated visibility from the mea-
equation with respect to space and time (denoted by the sym-
surements. Symbols denote different data sets with different atom
numbers and densities (: N = 2.2 × 105 , : N = 3.6 × 105 , :
bol F ), and define
N = 4.3 × 105 , : N = 5.9 × 105 ). h i
G(k, ω) = F Gi j (t) (A2)

In the frequency-momentum representation,

h i the Heisenberg
∂
equation of motion i~ ∂t Gi j (t) = H, Gi j (t) takes the form

(~ω + µ)G(k, ω) = 1 + tk G(k, ω)

viation points to non-adiabatic effects in the conversion from h i
−iF hT ni âi (0)â†j (t)i . (A3)
a condensate to an insulating state, which could in principle
be studied by the method presented here. Nevertheless, in
The last term on the right hand side of (A3) is usually rewritten
view that no free parameter is included in the theory, we con-
as Σ(k, ω)G(k, ω), where Σ is the self-energy. This gives the
clude that the momentum distribution (10) describes the sys-
expression
tem well. This supports the physical picture of the system as
a (dilute) gas of partice/hole pairs, mobile through the lattice, 1
on top of a regularly filled Mott insulator. Furthermore, the G(k, ω) = . (A4)
~ω + µ − tk − Σ(k, ω)
validity of the RPA to describe their behavior is qualitatively
verified. Let us first assume that no tunneling is present (t = 0). In
this case, the Green function can be obtained exactly from
Our calculation neglects entirely the superfluid component, Q definition (A1) and the ground state wave function |Ψi =
its
i |n0 ii , where each site is in the Fock state |n0 i. The result
which is correct only for large lattice depths where the sys- G0 (ω) is independent of momentum, and reads
tem is almost completely insulating. Recently, several authors
[15, 26, 27] have proposed to modify the standard mean-field n0 + 1 n0
description [19, 21] to better account for long- and short-range G0 (ω) = − . (A5)
~ω + µ − Un0 ~ω + µ − U(n0 − 1)
coherence. It would be interesting to compare the predictions
of those approaches with our data for lower lattice depths, In this self-interacting limit, we can rewrite Eq. (A5) as
where the system is expected to be a strongly depleted su- G0 (ω)−1 = ~ω + µ − Σ0 (ω), with the self energy [16]
perfluid, and therefore amenable neither to a Bogoliubov-like
description nor to a strongly interacting one as provided in U 2 n0 (n0 + 1)
Σ0 (ω) = 2Un0 − . (A6)
this paper. Also, an investigation of finite temperature effects ~ω + µ + U
[13] would be useful. In particular, an interesting question is
to know whether the visibility measurements presented here This expression is exact in the t → 0 limit, and coincides with
could be used for thermometry in the lattice. the one found in [16]. The first term is simply the Hartree-
Fock energy per particle for uncondensed atoms (hence the
factor of 2), whereas the second term - which has the same
We would like to thank Dries van Oosten, Paolo Pedri order of magnitude at low energy - accounts for the correla-
and Luis Santos for useful discussions. Our work is sup- tions between particle that drive the system into the perfectly
ported by the Deutsche Forschungsgemeinschaft (SPP1116), ordered ground state.
the European Union under a Marie-Curie Excellence Grant If we now restore a finite tunneling, but still consider a sys-
and AFOSR. FG acknowledges support from a Marie-Curie tem in the insulating phase, a reasonable approximation is to
Fellowship of the European Union. assume that the self energy is not changed with respect to the
 6

strongly interacting limit. We comment on this approximation which has a typical RPA form. Using the explicit result for
in the test. Making this approximation yields G0 , we obtain after some algebra Eq. (8) in the text, which
explicitly displays particle and hole components.
1
G(k, ω) ≈
~ω + µ − tk − Σ0 (ω)
G0 (ω)
= , (A7)
1 − t(k)G0 (ω)

[1] D. Jaksch, C. Bruder, J. I. Cirac, C. W. Gardiner, and P. Zoller, 117203 (2002).

Phys. Rev. Lett. 81, 3108 (1998). [21] K. Sheshadri, H. R. Krishnamurthy, R. Pandit, and T. V. Ra-
[2] M. Greiner, O. Mandel, T. Esslinger, T. W. Hänsch, and makrishnan, Euro. Phys. Lett. 22, 257 (1993).
I. Bloch, Nature 415, 39 (2002). [22] F. Dalfovo, S. Giorgini, L. P. Pitaevskii, and S. Stringari, Rev.
[3] W. Zwerger, J. Opt. B: Quantum Semiclass. Opt. 5, S9 (2003). Mod. Phys. 71, 463 (1999).
[4] T. Stöferle, H. Moritz, C. Schori, M. Köhl, and T. Esslinger, [23] F. Gerbier, J. H. Thywissen, S. Richard, M. Hugbart, P. Bouyer,
Phys. Rev. Lett. 92, 130403 (2004). and A. Aspect, Phys. Rev. A 70, 013607 (2004).
[5] C. Orzel, A. K. Tuchman, M. L. Fenselau, M. Yasuda, and [24] P. Pedri, L. Pitaevskii, S. Stringari, C. Fort, S. Burger, F. S.
M. K. Kasevich, Science 291, 2386 (2001). Cataliotti, P. Maddaloni, F. Minardi, and M. Inguscio, Phys.
[6] Z. Hadzibabic, S. Stock, B. Battelier, V. Bretin, and J. Dalibard, Rev. Lett. 87, 220401 (2001).
Phys. Rev. Lett. 93, 180403 (2004). [25] B. DeMarco, C. Lannert, S. Vishveshwara, and T.-C. Wei, Phys.
[7] M. Greiner, O. Mandel, T. W. Hänsch, and I. Bloch, Nature 419, Rev. A 71, 063601 (2005).
51 (2002). [26] J. J. Garcia-Ripoll, J. I. Cirac, P. Zoller, C. Kollath,
[8] F. Gerbier, A. Widera, S. Fölling, O. Mandel, T. Gericke, and U. Schollwöck, and J. von Delft, Optics Express 12, 42 (2004).
I. Bloch, cond-mat/0503452 (2005). [27] C. Schroll, F. Marquardt, and C. Bruder, Phys. Rev. A 70,
[9] V. A. Kashurnikov, N. V. Prokof’ev, and B. V. Svistunov, Phys. 053609 (2004).
Rev. A 66, 031601(R) (2002). [28] M. Krämer, L. Pitaevskii, and S. Stringari, Phys. Rev. Lett. 88,
[10] R. Roth and K. Burnett, Phys. Rev. A 67, 031602(R) (2003). 180404 (2002).
[11] Y. Yu, cond-mat/0505181 (2005). [29] A recent experiment has shown that this is not necessarily true
[12] D. van Oosten, P. van der Straten, and H. T. C. Stoof, Phys. Rev. in single-shot experiments for a large number of atoms per site
A 63, 053601 (2001). [6]. Our experimental parameters are such that no residual in-
[13] D. B. M. Dickerscheid, D. van Oosten, P. J. H. Denteneer, and terference is expected, in agreement with our observations.
H. T. C. Stoof, Phys. Rev. A 68, 043623 (2003). [30] In subsequent calculations we use the approxi-
[14] D. M. Gangardt, P. Pedri, L. Santos, and G. V. Shlyapnikov, mate expressions U ≈ √5.97(a/λL )(V0 /ER )0.88 , and
cond-mat/0408437 (2004). t = 1.43(V0 /ER )0.98 exp(−2.07 V0 /ER ). We have obtained
[15] K. Sengupta and N. Dupuis, Phys. Rev. A 71, 033629 (2005). these formula, accurate within 1 % in the range V0 = 8 − 30 ER ,
[16] D. van Oosten, D. B. M. Dickerscheid, B. Farid, P. van der by numerically solving for the band structure and performing a
Straten, and H. T. C. Stoof, Phys. Rev. A 71, 021601(R) (2005). fit to the calculated curves.
[17] J. Hubbard, Proc. Roy. Soc. A276, 238 (1963). [31] The coarse-grained density is understood here as an average of
[18] J. Hubbard, Proc. Roy. Soc. A281, 401 (1964). the discrete atomic density over a volume of linear size large
[19] M. P. A. Fisher, P. B. Weichman, G. Grinstein, and D. S. Fisher, compared to the lattice spacing, but small compared to the over-
Phys. Rev. B 40, 546 (1989). all extent of the cloud [28].
[20] G. G. Batrouni, V. Rousseau, R. T. Scalettar, M. Rigol, A. Mu-
ramatsu, P. J. H. Denteneer, and M. Troyer, Phys. Rev. Lett. 89,

View publication stats