PHYSICAL REVIEW A 71, 023610 共2005兲

Bose-Hubbard model with attractive interactions

Michael W. Jack1,2 and Makoto Yamashita1
1
NTT Basic Research Laboratories, NTT Corporation, 3-1 Morinosato Wakamiya, Atsugi-shi, Kanagawa 243-0198, Japan
2
Department of Physics and Astronomy and Rice Quantum Institute, Rice University, Houston, Texas 77251, USA
共Received 26 May 2004; published 24 February 2005兲

We consider the Bose-Hubbard model of atoms in an optical lattice potential when the atom-atom interac-
tions are attractive. If the lowest-energy lattice sites are degenerate 共such as in the homogeneous case兲, then, at
a critical value of the interaction strength, a phase-coherent condensate becomes unstable to a quantum super-
position such that the number distribution of each of the degenerate sites becomes double peaked. In the limit
when the interaction dominates, the superposition becomes macroscopic and has the form 兩␺典
⬀ 兺 jei␾ jb̂†N
j 兩vac典, where N is the total number of atoms and the sum ranges over the energy-degenerate sites.

DOI: 10.1103/PhysRevA.71.023610 PACS number共s兲: 03.75.Lm

An optical lattice potential loaded with a Bose-Einstein Hubbard model can be deduced from the results for a simple
condensate of neutral atoms was predicted by Jaksch et al. two-mode model of a condensate in a double-well potential
关1兴 and recently confirmed by Greiner et al. 关2兴 to be a real- 关13,14兴. In contrast to the repulsive case, it is predicted that
ization of the Bose-Hubbard model of condensed matter the number fluctuations will increase 共␦ni ⬎ 冑ni兲 as the mag-
physics 关3兴. In addition to exemplifying a quantum phase nitude of the interaction energy is increased 关13兴. The behav-
transition, this system is ideal for creating and controlling the ior as the magnitude of the interaction energy is further in-
quantum states of the atoms at the sites of the lattice poten- creased depends crucially on the single-particle energies of
tial 关2,4兴. The Bose-Hubbard model is described by a Hamil- the two wells, ⑀1 and ⑀2. In the case where the wells are
tonian of the form 关1兴 asymmetric, ⑀1 ⫽ ⑀2, the energy is minimized by all atoms
accumulating in the lower-energy site. In the case of sym-
1
H=−J 兺 b̂†j b̂i + 兺i ⑀in̂i + 2 Un̂i共n̂i − 1兲
具j,i典
共1兲 metric wells, ⑀1 = ⑀2, Cirac et al. 关14兴 and Steel et al. 关13兴
have shown that the system is unable to choose which site to
accumulate in and will form a quantum superposition of the
where b̂i and n̂i = b̂†i b̂i are the annihilation and number opera- two possibilities. This superposition state is associated with
tors of the mode localized at the ith lattice site. J is the very large number fluctuations: ␦ni → ni. In the multiwell
hopping matrix to neighboring sites, U ⬀ as is the strength of case considered here, we expect this superposition state to
the on-site interactions due to s-wave scattering, and ⑀i is an form between all sites that are degenerate with the lowest-
energy offset due to an additional confining magnetic trap. In energy site.
the large hopping regime, J Ⰷ 兩U兩, the ground state of this These results show that if the site energies are degenerate,
system is well described by a phase-coherent condensate and then a phase-coherent con-densate will become unstable to a
the number fluctuations are approximately Poissonian: ␦ni superposition state. On the other hand, in the absence of a
⬅ 冑具n̂2i 典 − n2i ⬇ 冑ni, where ni = 具n̂i典. For U ⬎ 0, increasing the lattice potential, a condensate with attractive interactions is
interaction energy reduces the on-site number fluctuations known to become unstable as the interaction strength is in-
共␦ni ⬍ 冑ni兲 and, at a critical value of U / J, the system under- creased 关7–12兴. It is convenient to consider two distinct types
goes a Mott-insulator phase transition 共␦ni → 0兲 and all phase of instability: 共I兲 a global implosion of the condensate wave
function confined in a harmonic trap 关7,8兴 and 共II兲 local in-
coherence vanishes 关2兴. In the experimental realization, the
stabilities of an unconfined condensate accompanied by large
simple global phase-transition picture is complicated by the
density fluctuations 关9,12兴. In the present case we find no
presence of the confining trap which produces local Mott
evidence of an instability of type I, even with the addition of
domains 关5兴.
a harmonic confining trap, and conclude that the Bose-
The use of neutral atoms opens up the possibility of ex-
Hubbard model becomes invalid before this type of instabil-
ploring the Bose-Hubbard model with attractive interactions
ity can occur. However, the instability described here where
共U ⬍ 0兲 as certain species of atom interact via a negative
a condensate gives way to a superposition state, shows many
s-wave scattering length and many atoms can be made to
similarities to an instability of type II.
interact via a negative scattering length by using the tech-
We can analyze the stability of a condensate in a lattice
nique of Feshbach resonance to alter the interaction potential
potential via a Bogoliubov-type treatment 关15,16兴: in the
关6兴. Attractive interactions are particularly interesting be-
limit of large hopping, J Ⰷ 兩U兩, the ground state is well ap-
cause they lead to an instability of a phase-coherent conden-
proximated by a phase-coherent condensate described by a
sate 关7–12兴. The aim of this article is to describe the nature
mean field, and we can consider small fluctuations about this
of this instability for the Bose-Hubbard model and its rela-
tionship to the formation of unusual quantum states. mean field by making the replacement b̂i共t兲 = e−i␮t/ប关␤i
Certain aspects of the behavior of the attractive Bose- + ␦ˆ i共t兲兴 in the Heisenberg equations of motion for b̂i and

1050-2947/2005/71共2兲/023610共4兲/$23.00 023610-1 ©2005 The American Physical Society

M. W. JACK AND M. YAMASHITA PHYSICAL REVIEW A 71, 023610 共2005兲

neglecting all terms except those linear in the fluctuations ␦ˆ i.

This results in the equations

0=−J 兺 ␤ j + 共⑀i − ␮ + U兩␤i兩2兲␤i ,

具j,i典
共2兲

d␦ˆ i
dt
兺
= − J ␦ˆ j + 共2U兩␤i兩2 + ⑀i − ␮兲␦ˆ i + U␤2i ␦ˆ †i
具j,i典
共3兲

for the mean field and fluctuations, respectively. After solv-

ing Eq. 共2兲 for ␤i and ␮, Eq. 共3兲 can be solved by making the
Bogoliubov transformation ␦ˆ i = 兺kui,ke−i␻kt␦ˆ k + vi,k
* i␻kt ˆ †
e ␦k , with
the normalization condition 兺i兩ui,k兩 − 兩vi,k兩 = 1, and solving
2 2

the resulting equations for ␻k, ui,k, and vi,k. FIG. 1. Ground state of a homogeneous one-dimensional lattice.
Assuming a homogeneous one-dimensional lattice of M This figure shows ␦ni 共solid line兲, C1 共dashed line兲, and CN 共dash-
sites 共with periodic boundary conditions兲 containing N at- dotted line兲 as a function of U / J. The inset shows the energy dif-
oms, Eqs. 共2兲 and 共3兲 can be solved analytically 关15,16兴 such ference between the first excited state and the ground state. The
that ␮ = Un − 2J, where n = 兩␤i兩2 = N / M is the mean-field solu- dotted lines correspond to a Bogoliubov treatment.
tion and the quasiparticle energies have the form
M

ប␻k = 冑␧k共␧k + 2Un兲

1
冑MN! 兺
i␾ j †N
共4兲 兩␺典 = e b̂ j 兩vac典, 共6兲
j=1
where ␧k = 4J sin2共ak / 2兲. Here k = 共2␲ / aM兲m for m
= −M / 2 , ¯ , M / 2. This shows that as U becomes increas- and the number fluctuations become ␦ni = N冑M − 1 / M or
ingly negative, the first quasiparticle energy 共corresponding ⬇N / 冑M for many sites.
to k = 2␲ / aM兲 drops to zero at Un = −2J sin2共␲ / M兲 and then In the current experiments 关2,4兴, an inhomogeneity is in-
becomes imaginary, signaling a critical point beyond which troduced to the lattice system by an harmonic magnetic trap
the lattice system is unable to support a condensate. An ana- which is used to confine the atoms in space. In the one-
lytical expression for the on-site number fluctuations can dimensional case considered here, this gives rise to the
also be determined from this treatment 关16兴 共see also 关17兴兲 as single-particle energies ⑀i = ␭关i − 共M + ⌬兲 / 2兴2, where i ranges
from 1 to M, ␭ is a measure of th.e curvature, and 0 艋 ⌬
n ␧ ⬍ 1 is the offset of the lattice from the center of the confining
␦n2i =
M
兺k ប␻k k . 共5兲 potential. Figure 2 shows the results of exact calculations in

It is evident that at the critical point the number fluctuations

共5兲 diverge. Comparing with Ref. 关9兴 we see that a type-II
instability is formally very similar to the present case.
In order to treat the strong interaction regime beyond the
instability 共where the Bogoliubov treatment breaks down兲,
we have numerically calculated the exact ground state. In the
Fock state basis the state space of the system is large: 共N
+ M − 1兲 ! / 关N ! 共M − 1兲 ! 兴. But the Hamiltonian 共1兲 is a very
sparse matrix and so for small numbers of atoms and sites we
can calculate the lowest few eigenvalues and eigenstates by
the Lanczos method 关18兴. The results of these calculations
are presented in Fig. 1. After the critical point of the Bogo-
liubov treatment the exact calculations show that the number
distribution becomes double peaked 共corresponding to the
formation of a superposition state兲 which gives rise to the
sudden increase in the number fluctuations shown in Fig. 1.
The two peaks of the distribution move further apart and
narrow as the interaction is increased, which reduces the
single-particle correlation between neighboring sites, C1 FIG. 2. Ground state in the presence of a harmonic confining
trap with ␭ = 0.1J and ⌬ = 0.9. 共a兲–共c兲 show ni 共white bars兲 and ␦ni
= 具b̂†i b̂i+1典, but increases the N-particle correlation CN
共black bars兲 at each site for various values of U / J. 共d兲 shows n4
i b̂i+1典 / N!. Finally, in the strong attractive interaction
= M具b̂†N N
共dashed line兲 and ␦n4 共solid line兲 as a function of U / J. The inset in
limit, the results confirm that the ground state is a macro- 共d兲 shows the energy difference between the first excited state and
scopic superposition of the form the ground state.

023610-2
BOSE-HUBBARD MODEL WITH ATTRACTIVE INTERACTIONS PHYSICAL REVIEW A 71, 023610 共2005兲

FIG. 3. Minimum energy gap between the ground state and first FIG. 4. Plot of the critical value of U / J as a function of the
excited state as a function of ⌬ for M = 50 sites. The values of the curvature of the confining trap, ␭, in the case of two degenerate
curvature are ␭ = 10−3 共solid line兲, ␭ = 10−2 共dashed line兲, and ␭ sites at the center of the trap. The inset shows the width w of the
= 10−4 共dotted line兲. condensate at the critical point.

the case of nondegenerate sites. Note that the number fluc- tively “capture” the superposition before the atoms are able
tuations become large close to where the critical point would to tunnel into the lowest-energy site. To illustrate this point
have occurred if the sites were degenerate 共the system gets we have numerically solved the Schrödinger equation for a
close to forming a superposition at this point兲 and then de- two-mode model with the time-dependent interaction
crease as the atoms accumulate into a single site. strength U共t兲 = U f 共1 − e−␥t兲. The results of this simulation are
We can also calculate the critical behavior of the inhomo- shown in Fig. 5. The large number fluctuations for a suffi-
geneous system via a Bogoliubov treatment, which requires ciently large ␥ 共dotted line兲 confirm the formation of a su-
the numerical solution of the mean-field equations 共2兲 before perposition even though the single-particle energies in the
solving the linear equations 共3兲 for the quasiparticles. In the two wells differ.
nondegenerate case when 0 ⬍ ⌬ ⬍ 1, for M odd, there is al- Experimental realization of the interaction-dominated re-
ways one site, i = 共M + 1兲 / 2, with the lowest single-particle gime of the Bose-Hubbard model with a large number of
energy. In this case, no critical point is seen and as the inter- atoms is complicated by the fact that the localization of all N
action becomes more attractive, the condensate—which has atoms at a single lattice site may render the Bose-Hubbard
an approximately Gaussian spatial profile of width model invalid unless the magnitude of the scattering length is
w—simply decreases in width until all atoms accumulate in small. Approximating the potential at each lattice site by a
site i = 共M + 1兲 / 2. In Fig. 3 we have plotted the minimum harmonic potential of length aho, the interaction strength
energy gap to the first excited state, ⌬Emin 共found by varying must satisfy niaho Ⰷ 具n̂i共n̂i − 1兲典兩as兩, such that the interaction
U / J over a broad range of values兲 as a function of ⌬. The does not alter the shape of the localized mode functions at
finite value of this gap for ⌬ ⫽ 0 demonstrates the stability of
the condensate in the nondegenerate case. We find no evi-
dence of an instability of type I which would be expected to
occur even in the nondegenerate case.
Figure 4 shows the dependence of the critical point
共where the first excitation energy becomes imaginary兲 on the
curvature in the degenerate case: ⌬ = 0. The inset shows that
for a broad range of trap curvatures, the critical point occurs
when the condensate width is of the order of the site spacing.
In a realistic situation it may be difficult 共if not impos-
sible兲 to create a lattice with exactly degenerate sites. How-
ever, in current experiments, the high interaction regime is
reached by a dynamic process whereby the relevant param-
eter ␬共t兲 = 兩U共t兲兩 / J共t兲 is increased at a certain rate ␥ 共this is
normally in the adiabatic regime so the system remains in the FIG. 5. Nonadiabatic evolution of a two-mode model from the
ground state兲. In this case, close to ␬共t兲 ⬃ 1 / N 共where the ground state at U共t = 0兲 = 0 with the parameters U f = −0.05J, ⑀2 − ⑀1
critical point would be if the sites were degenerate兲, if one = 0.0005J. This figure shows the time evolution of the number fluc-
increases ␬共t兲 at a rate faster than the oscillation frequency tuations ␦n1 for the rates ␥ ⬍ 0.005J / ប 共solid line兲 and ␥
between two different sites, ␥ Ⰷ 共⑀1 − ⑀2兲 / ប 共but still slower = 0.01J / ប 共dashed line兲. The inset shows the two-peaked structure
than the tunneling rate兲, then, for short times, the system will 共coresponding to a superposition state兲 of the number distribution
be unaware of the inhomogeneity and it is possible to effec- P共n1兲 = 円具␺共t兲 兩 n1 , N − n1典円2 at the point corresponding to the arrow.

023610-3
M. W. JACK AND M. YAMASHITA PHYSICAL REVIEW A 71, 023610 共2005兲

each site. 共In fact, we can continue to use the Bose-Hubbard choose which site due to the energy degeneracy and so form
model beyond this inequality with renormalized parameters a superposition of all the possibilities. Interestingly, our re-
关19兴, the ultimate limit being the stability of the localized sults suggest that the Bose-Hubbard model becomes invalid
mode function against a type-I instability: aho ⬎ ni兩as兩.兲 A before an instability of type I 共for a confined condensate兲
possible method to overcome this is to load a very deep occurs and that an instability of type II 共for an unconfined
optical lattice 共so that J is small兲 with repulsively interacting condensate兲 corresponds to the formation of superposition
atoms and use Feshbach resonance to slowly tune the inter-
states due to the homogeneity of free space. In an experimen-
action through zero so it becomes just slightly negative, as
described in Ref. 关20兴 for 7Li. This method has the additional tal realization, atom loss 共or absorption imaging兲 will destroy
advantage that it will minimize three-body loss of atoms, the superposition by tending to localize the atoms at one site.
which scales as as4 关21兴. Superpositions such as 共6兲 can be destroyed by the loss of
In conclusion, the ground state of the attractive Bose- just one atom which will “collapse” the quantum state of the
Hubbard model displays behavior fundamentally different atoms to one of the degenerate lattice sites. Less macroscopic
from the repulsive case. In particular, if the lowest-energy superpositions, such as those formed just after the critical
sites are degenerate then, at a critical value of the interaction point, will be more robust against loss 关14兴. Methods of non-
strength, a phase-coherent condensate becomes unstable to a destructive detection of these superposition states will be the
quantum superposition such that the number distribution at topic of future work.
each degenerate site becomes double peaked. The atoms
have a tendancy to accumulate at a single site in order to The authors would like to thank A. Kawaguchi, Y. Tokura,
minimize the interaction energy but they are unable to and R. Hulet for useful discussions.

关1兴 D. Jaksch, C. Bruder, J. I. Cirac, C. W. Gardiner, and P. Zoller, A. Cornell, and C. E. Wienman, Nature 共London兲 412, 295
Phys. Rev. Lett. 81, 3108 共1998兲. 共2001兲.
关2兴 M. Greiner, O. Mandel, T. Esslinger, T. W. Hänsch, and I. 关12兴 J. K. Chin, J. M. Vogels, and W. Ketterle, Phys. Rev. Lett. 90,
Bloch, Nature 共London兲 415, 39 共2002兲. 160405 共2003兲.
关3兴 M. P. A. Fisher, P. B. Weichman, G. Grinstein, and D. S. 关13兴 M. J. Steel and M. J. Collett, Phys. Rev. A 57, 2920 共1998兲.
Fisher, Phys. Rev. B 40, 546 共1989兲. 关14兴 J. I. Cirac, M. Lewenstein, K. Molmer, and P. Zoller, Phys.
关4兴 C. Orzel, A. K. Tuchman, M. L. Fenselau, M. Yasuda, and M. Rev. A 57, 1208 共1998兲.
A. Kasevich, Science 291, 2386 共2001兲. 关15兴 D. van Oosten, P. van der Straten, and H. T. C. Stoof, Phys.
关5兴 G. G. Batrouni, V. Rousseau, R. T. Scalettar, M. Rigol, A. Rev. A 63, 053601 共2001兲.
Muramatsu, P. J. H. Denteneer, and M. Troyer, Phys. Rev. Lett. 关16兴 K. Burnett, M. Edwards, C. W. Clark, and M. Shotter, J. Phys.
89, 117203 共2002兲. B 35, 1671 共2002兲.
关6兴 S. Inouye, M. R. Mathews, J. Stenger, H. J. Miesner, D. M. 关17兴 J. Javanainen, Phys. Rev. A 60, 4902 共1999兲.
Stamper-Kurn, and W. Ketterle, Nature 共London兲 392, 151 关18兴 Templates for the Solution of Algebraic Eigenvalue Problems:
共1998兲. A Practical Guide, edited by Z. Bai, J. Demmel, J. Dongarra,
关7兴 P. A. Ruprecht, M. J. Holland, K. Burnett, and, M. Edwards A. Ruhe, and H. van der Vorst, 共SIAM, Philadelphia, 2000兲.
Phys. Rev. A 51, 4704 共1995兲. 关19兴 D. van Oosten, P. van der Straten, and H. T. C. Stoof, Phys.
关8兴 G. Baym and C. J. Pethick, Phys. Rev. Lett. 76, 6 共1996兲. Rev. A 67, 033606 共2003兲.
关9兴 V. A. Yurovsky, Phys. Rev. A 65, 033605 共2002兲. 关20兴 K. E. Strecker, G. B. Partridge, A. G. Truscott, and R. G.
关10兴 J. M. Gerton, D. Strekalov, I. Prodan, and R. G. Hulet, Nature Hulet, Nature 共London兲 417, 150 共2002兲.
共London兲 408, 692 共2000兲. 关21兴 B. D. Esry, C. H. Greene, and J. P. Burke, Phys. Rev. Lett. 83,
关11兴 E. A. Donley, N. R. Claussen, S. L. Cornish, J. L. Roberts, E. 1751 共1999兲.

023610-4