a

r
X
i
v
:
c
o
n
d
-
m
a
t
/
0
2
0
7
1
7
2
v
1


[
c
o
n
d
-
m
a
t
.
s
t
a
t
-
m
e
c
h
]


6

J
u
l

2
0
0
2
Dynamical Superuid-Insulator Transition
in a Chain of Weakly Coupled Bose-Einstein Condensates
A. Smerzi
1
, A. Trombettoni
1,2
, P.G. Kevrekidis
3
and A.R. Bishop
1
1
Theoretical Division and Center for Nonlinear Studies, Los Alamos National Laboratory, Los Alamos, NM 87545, USA
2
Istituto Nazionale di Fisica per la Materia and International School for Advanced Studies, via Beirut 2/4, I-34014, Trieste,
Italy
3
Department of Mathematics and Statistics, University of Massachusetts, Amherst, MA 01003-4515, USA
(February 6, 2008)
We predict a dynammical classical superuid-insulator transition (CSIT) in a Bose-Einstein con-
densate (BEC) trapped in an optical and a magnetic potential. In the tight-binding limit, this
system realizes an array of weakly-coupled condensates driven by an external harmonic eld. For
small displacements of the parabolic trap about the equilibrium position, the BEC center of mass
oscillates with the relative phases of neighbouring condensates locked at the same (oscillating) value.
For large displacements, the BEC remains localized on the side of the harmonic trap. This is caused
by a randomization of the relative phases, while the coherence of each individual condensate in the
array is preserved. The CSIT is attributed to a discrete modulational instability, occurring when
the BEC center of mass velocity is larger than a critical value, proportional to the tunneling rate
between adjacent sites.
PACS: 63.20.Pw, 05.45.-a
The recent experimental investigations of the dynam-
ical properties of a Bose-Einstein condensate trapped in
optical potentials [14], have led to a rapidly growing
interest in this topic [510]. The spatial and temporal
coherence of matter waves emitted at dierent heights
of the gravitational eld has been proven in [1], after
loading a condensate in a vertical optical trap. Num-
ber squeezed (non-classical) states have been realized in
[2]. In [3] Bloch oscillations and interband transitions in
an accelerating lattice have been observed. In [4], the
optical potential was superimposed on a harmonic mag-
netic trap, realizing a chain of weakly coupled conden-
sates (i.e., a Josephson junction array) driven by an ex-
ternal parabolic eld. For small initial displacements, the
condensate center of mass oscillated symmetrically with
the relative phases among adjacent sites locked togheter.
Here we demonstrate that for large displacements,
when the velocity of the center of mass reaches a crit-
ical value (proportional to the tunneling rate beteen ad-
jacent sites), the BEC abruptly stops on the side of the
harmonic trap (i.e., without reaching its center). We de-
ne an order parameter for the system, and show that
this dynamical transition from a superuid to an in-
sulator regime is associated with a randomization of
the relative phases among dierent wells. As we will
discuss below, this transition has a classical (mean-eld
or Gross-Pitaevskii) nature, and it is dierent (but with
some analogies) from the quantum superuid-insulator
(Mott) transition caused by the number squeezing of
the quantum states in each well [9]. It also diers from
the Landau dissipation mechanism, occurring in (quasi-
)homogeneous systems when the velocity of the conden-
sate is larger than the sound speed [6]. Rather, the CSIT
is driven by a modulational instability (MI) that causes
an exponential growth of small perturbations of a carrier
wave, as a result of the interplay between dispersion and
nonlinearity. The MI is a general feature of discrete as
well as continuum nonlinear wave equations. Its demon-
strations span a diverse set of disciplines ranging from
uid dynamics [11] (where it is usually referred to as the
Benjamin-Feir instability) and nonlinear optics [12] to
plasma physics [13]. One of the early contexts in which
its signicance was appreciated was the linear stability
analysis of deep water waves. It was only much later rec-
ognized that the conditions for MI would be signicantly
modied for discrete settings relevant to, for instance, the
local denaturation of DNA [14] or coupled arrays of opti-
cal waveguides [15]. In the latter case, the relevant model
is the discrete nonlinear Scr odinger equation (DNLS),
and its MI conditions were discussed in [16]. In this let-
ter we propose an experiment to observe a superuid-
insulator mean-eld dynamical transition (as a conse-
quence of the MI), with weakly coupled Bose-Einstein
condensates driven by an external harmonic eld.
The BEC dynamics is governed by the Gross-Pitaevskii
equation (GPE):
ih

t
=
h
2
2m

2
+ [V
ext
+ g
0
| |
2
] (1)
where g
0
=
4 h
2
a
m
, a is the s-wave scattering length, m
the atomic mass and the chemical potential. The con-
densate wave function (r, t) is normalized to the total
number of condensate atoms N
T
, and we consider a re-
pulsive interatomic interaction a > 0. The external po-
tential V
ext
is given by the sum of the harmonic conning
potential V
M
=
m
2
[
x
x
2
+
2
r
(y
2
+z
2
)] and the optical po-
tential V
L
= V
0
cos
2
(kx). The valleys of the potential are
1
separated by a lattice spacing of /2, with = 2/k.
We consider a chemical potential << V
0
and the trans-
verse degrees of freedom to be frozen by a tight magnetic
connement, so as to justify the study of the system in
an eective one-dimensional geometry.
In the tight-binding approximation (r, t) =

N
T

n

n
(t)
n
(r), with the wavefunction
j
of the
condensate in the j-th site of the array, weakly coupled
in the barrier region with the wavefunctions
j1
of the
condensates in the neighbour sites. It is then possible to
map the GPE onto the DNLS [7]:
ih

n
t
= K(
n1
+
n+1
) + (
n
+ U |
n
|
2
)
n
(2)
with K

dr

h
2
2m
(

n+1
) +
n
V
ext

n+1

pro-
portional to the microscopic tunneling rate between adja-
cent sites, U = g
0
N
T

dr
4
n
and
n
=

dr[
h
2
2m
(

n
)
2
+
V
ext

2
n
] = n
2
, with =
1
2
m
2
x

2
. The DNLS Hamil-
tonian is
H =

n
[K(
n

n+1
+

n+1
) +
n
|
n
|
2
+
U
2
|
n
|
4
]
(3)
with i

n
,
n
canonically conjugate variables. Both H
and the norm

n
|
n
|
2
= 1 are integrals of the motion.
Let us consider, rst, the case
n
= 0 (which corre-
sponds to neglecting the eect of the harmonic trap).
Stationary solutions of Eq. (2) are plane waves =

0
exp[i(knt)], of frequency = 2K cos(k)+U|
0
|
2
.
The stability analysis of such states can be carried out
by perturbing the carrier wave with small amplitude
phonons:
n
=
0
e
i(knt)
+ ue
i(qnt)
+ v

e
i(qn+t)
.
The DNLS excitation spectrum (for
n
= 0) is then given
by:

= 2K sin(k) sin(q)
2

4K
2
cos
2
(k) sin
4
(
q
2
) + 2KU|
0
|
2
cos(k) sin
2
(
q
2
). (4)
The carrier wave becomes modulationally unstable when
the eigenfrequency in Eq.(4) becomes imaginary:
U|
0
|
2
> 2K cos(k) sin
2
(q/2), (5)
namely, when cos(k) < 0. Therefore, if the interatomic
interaction is repulsive (U > 0), the system suers an
exponential growth of perturbations when /2 < k <
3/4. This result will remain valid in the case of non-
homogeneous travelling wave-packets driven by external
elds, when their width is much larger than the wave
length associated with the collective motion. This con-
clusion can be further understood in the light of the col-
lective coordinate equations of motion developed in [7].
Generally speaking, the mapping of the GPE into the
DNLS allows for the study of solitons and localized exci-
tations as well as dynamical instabilities in the framework
of the lattice theory [1719]. We remark, however, that
the MI is a general feature of the GPE with a periodic
external potential, and not necesserly in the tight bind-
ing limit. In the perturbative limit, >> V
0
, and in
absence of external driving elds (i.e., with V
M
= 0), the
MI has been studied in [6,8] (of course, in this limit the
MI condition doers from Eq.(5)).
The eect of the exponential growth of phonon modes
of arbitrary momenta in the DNLS leads to an eective
dephasing among dierent sites of the lattice. Indeed the
phases of each condensate enter into a running regime,
with an angular velocity dierent from site to site and
proportional to the local (on-site) eective chemical po-
tential. The complete delocalization in momentum space
leads to a strong localization in real space, hence to the
appearance of localized structures of large amplitude (see
Fig. 1). This localization has also been attributed (in the
absence of any external potential
n
= 0) to the presence
of the so-called Peierls-Nabarro barrier [20], which pins
such large amplitude solutions [21], not allowing them to
propagate. The excess kinetic energy is partially stored
to high-frequency internal ac oscillations among ad-
jacent wells (see also [22,23]), and partially converted
to wakes of small amplitude extended wave radiation
[20,25].
The CSIT can be observed experimentally by condens-
ing, rstly, the atomic gas in both the magnetic and the
optical trap, and, then, adiabatically displacing the mag-
netic eld from its initial position. For small displace-
ments, in line with the ndings of [4], the system coher-
ently oscillates about the center of the potential. If we
rewrite
j
=

n
j
e
ij
, this implies that the phase dier-
ence between sites is given by
j+1
(t)
j
(t) = (t).
The center of mass =

j
jn
j
and the phase dierence
will then satisfy
h
d
dt
(t) = 2K sin (t)
h
d
dt
(t) = 2 (t). (6)
Eqs. (6) have the usual form of the Josephson equa-
tions [24] and indicate that the overall array of bosonic
Josephson junctions behaves as a single Josephson junc-
tion, whose critical current is 2K/h. The collective co-
herence was experimentally demonstrated in [4] by the
interference pattern obtained upon releasing the conden-
sates from the optical and magnetic traps.
To monitor the dynamical transition of interest, we
dene < k >=

k
k|

k
|
2
= , with

k
the Fourier
transform of the condensate wave-function. In the co-
herent, small amplitude oscillations regime, the quasi-
momentum < k > exhibits regular oscillations (see Fig.
2). However, for < k > /2, the system becomes mod-
ulationally unstable and localization ensues. The critical
initial displacement
cr
can therefore be obtained from
Eqs. (6) with = /2. In lattice units:
2

cr
=

2K

. (7)
In Fig. 2 we plot < k > vs. time for three initial displace-
ments. When (t = 0) is smaller than
cr
, the average
momentum < k > oscillates in accordance with Eq. (6).
Note that for (t = 80) the system approaches very close
to the instability line. When the initial displacement is
larger than the critical value, < k > abruptly drops as
soon as it crosses the critical point. This is accompa-
nied by the sudden arresting of the BEC center of mass
(cf. Fig. 1) and by a collective dephasing (cf. Fig. 3).
The key experimental signature would be the disappear-
ance of the interference fringes after the expansion of the
BEC, while the center of mass of the resulting cloud will
be resting on the side of the traps center.
The values of the parameters in DNLS are V
0
= 5E
R
with = 795 nm and the recoil energy E
R
=
h
2
2m
2
,
N
T
= 50000, K = 5.5 10
2
E
R
and = 1.5 10
5
E
R
:
the critical displacement from Eq. (7) is
cr
84 sites,
in good agreement with our numerical ndings (cf. Figs
2 and 3). The loss of coherence is highlighted in Fig.
3, where we plot the temporal evolution of the modulus
squared of an eective complex order parameter measur-
ing the overall coherence of the system, dened as:
=

j+1
. (8)
When the collective oscillations are coherent, the value
of the order parameter is |(t)|
2
= 1 (see the cases
(t = 0) = 40, 80 in Fig. 3). On the other hand,
a complete dephasing is characterized by |(t)|
2
= 0,
and occurs for (t = 0)
cr
, or, equivalently, when
< k > /2 (cf. Fig. 2). It is worth noting that such
randomization takes place between the phases of conden-
sates localized in dierent wells, even though each one
remains internally coherent. This is shown in Fig. 1,
where we plot the density for dierent times below (a,b)
and above (c,d) the critical displacement. The numer-
ical solutions of the DNLSE and the GPE are in good
agreement. The motion of the center of mass in the su-
percritical case is reported in Fig. 4, where the numer-
ical solutions of the DNLS and the full one-dimensional
GPE (1) are compared. In both cases the system stops
at < x > 35 m (with a slight dierence between the
DNLS and the GPE predictions), while the center of the
harmonic trap is located at x = 0 m. From Eqs. (6)
we can calculate the critical current, i.e., the maximum
allowed velocity in the coherent transmission of matter
waves: by setting = /2 we readily see that the crit-
ical velocity of the center of mass,

max
is equal to the
critical current per particle I
c
= 2K/h. In dimensional
units:
v
c
=
K
h
(9)
which gives v
c
= 0.98 m/ms, in agreement with the
DNLS numerical result, and close to the numerical GPE
value v
c
= 1.18 m/ms.
From the above ndings, we can conclude that the ef-
fect of the MI is to dephase the system. In the eec-
tive 1D geometry we have considered, such dephasing is
strong enough to stop the falling condensate. In higher
dimensions, the dephasing can be partial, and will only
damp the BEC motion. Yet, its onset will still be given
by Eq. (9). The CSIT regards a classical eld (the solu-
tion of the GPE) and it is qualitatively dierent from the
quantum Mott insulator-superuid transition in meso-
scopic Josephson junction chains, which is driven by the
competition between zero-point quantum phase uctua-
tions and the Josephson coupling energy. Yet, it is pos-
sible to draw an analogy. In the former CSIT case, the
insulator regime is associated with a vanishing temporal
correlation among the phases of each site, each phase still
being meaningful in the GPE sense. The quantum transi-
tion is also driven by a loss of phase correlations induced
by the localization of atoms in each site, which, however,
arises from the non-commuting nature of the number-
phase observables. Clearly, such quantum uctuations
cannot be captured within the GPE framework. Also,
the latter transition is reversible (i.e., long-range phase
coherence is restored upon adiabatically increase of the
tunnel coupling), while the former is not. It is worth
noting that very recent experimental works [26] have il-
lustrated the existence (and reversibility) of the quantum
phase transition, rendering the experimental verication
of the classical dynamic transition suggested herein, a
natural next step for experimental studies. In conclu-
sion, we notice that the MI can be studied in term of one
(or several) bifurcation points in an eective stationary
Hamiltonian. Such bifurcation points separate regions
with dierent symmetries, and it is common in the liter-
ature to study such dynamical transitions in terms of an
order parameter [27], borrowing the language and con-
cepts of statistical phase transitions. This mapping, in
the specic case of MI, deserves further studies.
The modulational instability (and the consequent
superuid-insulator transition) studied here can also be
observed with dierent experimental setups (e.g., with
the condensate at the center of the harmonic trap and
with the laser moving across). In fact, similar MI and
pinning results have been obtained in the case in which
the harmonic trap is displaced at a constant speed ex-
ceeding a critical value [28]. These results illustrate the
generality and importance of the eects of the MI mech-
anism in the motion of Bose-Einstein condensates and
underscore its potential in inducing localization and de-
phasing of such coherent structures.
[1] B.P. Anderson and M.A. Kasevich, Science 282, 1686
3
(1998).
[2] C. Orzel, A. K. Tuchman, M. L. Fenselau, M. Yasuda,
M. A. Kasevich, Science 291, 2386 (2001).
[3] O. Morsch, J.H. M uller, M. Cristiani, D. Ciampini and
E. Arimondo, Phys. Rev. Lett. 87, 140402 (2001)
[4] F.S. Cataliotti, S. Burger, C. Fort, P. Maddaloni, F. Mi-
nardi, A. Trombettoni, A. Smerzi and M. Inguscio, Sci-
ence 293, 843 (2001).
[5] M.L. Chiofalo and M.P. Tosi, Phys. Lett. A 268, 406
(2000).
[6] B. Wu and Q. Niu, Phys. Rev. A 64, 061603(R) (2001).
[7] A. Trombettoni and A. Smerzi, Phys. Rev. Lett. 86,
2353, (2001).
[8] V.V. Konotop and M. Salerno, cond-mat/0106228.
[9] D. Jaksch et al., Phys. Rev. Lett. 81, 3108 (1998)
[10] P. Pedri et al., Phys. Rev. Lett. 87, 220401 (2001)
[11] T.B. Benjamin and J.E. Feir, J. Fluid. Mech. 27, 417
(1967).
[12] L.A. Ostrovskii, Sov. Phys. JETP 24, 797 (1969).
[13] T. Taniuti and H. Washimi, Phys. Rev. Lett. 21, 209
(1968); A. Hasegawa, Phys. Rev. Lett. 24, 1165 (1970).
[14] M. Peyrard, T. Dauxois, H. Hoyet and C.R. Willis, Phys-
ica 68D, 104 (1993).
[15] R. Morandotti, U. Peschel, J.S. Aitchison, H.S. Eisenberg
and Y. Silberberg, Phys. Rev. Lett. 83, 2726 (1999).
[16] Yu.S. Kivshar and M. Peyrard, Phys. Rev. A 46, 3198
(1992).
[17] A.C. Scott, Nonlinear Science: Emergence and Dynam-
ics of Coherent Structures, Oxford Univ. Press, Oxford,
1999
[18] D. Hennig and G.P. Tsironis, Phys. Rep. 307, 333 (1999).
[19] P.G. Kevrekidis, K.. Rasmussen and A.R. Bishop, Intn.
J. Mod. Phys. B 15, 2833 (2001).
[20] M. Peyrard and M. Kruskal, Physica 14D, 88 (1984).
[21] T. Dauxois and M. Peyrard, Phys. Rev. Lett. 70, 3935
(1993).
[22] Yu.S. Kivshar, D.E. Pelinovsky, T. Cretegny and M.
Peyrard, Phys. Rev. Lett. 80, 5032 (1998).
[23] P.G. Kevrekidis and C.K.R.T. Jones, Phys. Rev. E 61,
3114 (2000).
[24] A. Barone and G. Paterno, Physics and Applications of
the Josephson Effect (Wiley, New York, 1982).
[25] P.G. Kevrekidis and M.I. Weinstein, Physica 142D, 113
(2000).
[26] M. Greiner et al., Nature 415, 39 (2002)
[27] H. Haken, Synergetics, Springer-Verlag Berlin (1977);
G. Venkataraman and V. Balakrisnan, Sinergetica ed
Instabilita dinamiche, Course IC, Proceedings of the
International School of Physics Enrico Fermi, Varenna,
Caglioti, Haken and Lugiato Ed.s, (1988), 175
[28] A. Smerzi, A. Trombettoni, P.G. Kevrekidis and A.R.
Bishop (unpublished).
20 0 20 40 60
x (m)
0.0
0.2
|

|
2
0.00
0.04
0.08
|

n
|
2
20 0 20 40 60
x (m)
0.0
0.2
|

|
2
0.00
0.04
0.08
|

n
|
2 (a) (c)
(b) (d)
FIG. 1. The density calculated at dierent times 0, 20,
40 ms (from the right of each gure to the left) as a func-
tion of the position, with initial displacements
(
0) = 50, 120
sites, which are, respectively, below and above the critical
value xicr 84 (7). The GPE (b,d), and the DNLS (a,c)
wavefunctions normalized to 1 are compared. (a,b)
(
0) = 50
sites; (c,d)
(
0) = 120 sites. The external parabolic potential,
which drives the oscillations, is centered at x = 0.
0 100 200
time (ms)
2
1
0
1
2
<
k
>
/2
40
80
90
FIG. 2. The quasi-momentum < k > vs. time for three
dierent initial displacements: 40, 80 and 90 sites. When
< k > reaches /2 (i.e., for an initial displacement greater
than cr 84 calculated with Eq.(7) the system becomes
modulationally unstable.
4
0 100 200
time (ms)
0.0
0.5
1.0
|

|
2
90
40
80
FIG. 3. The modulus square of the order parameter
dened in Eq. (8) as a function of time, for three dierent
initial displacements (40, 80 and 90 sites) and with the same
parameters as in Fig. 2. When the quasi-momentum < k >
reaches /2 (i.e., for an initial displacement greater than cr),
the order parameter drops to 0; cf. Fig. 2.
0 100 200
time (ms)
20
30
40
50
60
<
x
>

(

m
)
FIG. 4. The center of mass vs. time for a supercritical
initial displacement of 150 sites. Solid line: Gross-Pitaevskii
equation; dashed line: discrete nonlinear Schrodinger equa-
tion. The (displaced) center of the trap is at x = 0 m.
5