PHYSICAL REVIEW A VOLUME 59, NUMBER 5 MAY 1999

Spatial fragmentation of a Bose-Einstein condensate in a double-well potential

R. W. Spekkens and J. E. Sipe
Department of Physics, University of Toronto, 60 St. George Street, Toronto, Ontario, Canada M5S 1A7
~Received 30 October 1998!
We present a theoretical study of the ground state of a Bose-Einstein condensate with repulsive interparticle
interactions in a double-well potential, using a restricted variational principle. Within such an approach, there
is a transition from a single condensate to a fragmented condensate as the strength of the central barrier of the
potential is increased. We determine the nature of this transition through an approximate analytic solution as
well as a numerical solution of our model, in the regime where the interparticle interactions can be treated
perturbatively. The degree of fragmentation of the condensate is characterized by the degrees of first- and
second-order spatial coherence across the barrier. @S1050-2947~99!06605-6#

PACS number~s!: 03.75.Fi, 05.30.Jp, 32.80.Pj

I. INTRODUCTION Fock state, and that can be said to describe ‘‘partial fragmen-
tation’’ of the condensate. The equations that such a state
The recent experimental demonstration of interference must satisfy are derived within a variational approach. The
phenomena in Bose-condensed atomic gases @1# motivates a second goal of the paper is to solve these equations in a
study of the spatial coherence of a condensate in a double- regime where the interparticle interactions can be treated per-
well potential. In particular, we are interested in the loss of turbatively. Numerical solutions of the equations and ana-
spatial coherence that can occur at zero temperature due to lytic approximations to these solutions are obtained within
fragmentation of the condensate. A fragmented condensate is this regime. It should be noted that this limit is inappropriate
one for which there is a macroscopic occupation of two or for the description of the MIT condensate interference ex-
more orthogonal single-particle wave functions. If the occu- periment @1#, and, consequently, our results do not specify
pied single-particle wave functions are spatially well sepa- the nature of the transition for this experimental setup. None-
rated, coherence over the spatial extent of the entire system theless, we expect the generic features of the transition to
will be lost, persisting only over the spatial extent of each persist in the experimentally relevant regime.
fragment. We pause to consider previous treatments of this topic
As Nozières @2# has pointed out, for repulsive interparticle and their relation to this work. Röhrl et al. @4# provided a
interactions it is the exchange energy that typically prevents model of the MIT condensate interference experiment; how-
fragmentation into a number of degenerate ~or nearly degen- ever, it is a mean-field analysis and therefore cannot describe
erate! single-particle wave functions. However, this argu- fragmentation. Fragmentation in the case of attractive inter-
ment is inapplicable for bosons in an external potential with particle interactions has been considered by Wilkin, Gunn,
several local minima, since single-particle wave functions and Smith @5#, but this effect is qualitatively different from
that are localized about these minima may have very little that of the repulsive case. Finally, Milburn et al. @6# consid-
overlap with one another, thereby leading to a very small ered the energy eigenstates of a Bose-Einstein condensate in
exchange energy. Moreover, since the self-interaction energy a double-well potential, and predicted fragmentation when
of such a fragmented condensate is smaller than that of a the interparticle interactions are sufficiently strong. How-
single condensate, it is possible for the total interaction en- ever, these authors considered only traps with weakly
ergy to be smaller as well. Although every particle in the coupled wells, and therefore could not determine the degree
fragmented condensate will pay a price in kinetic energy to of fragmentation of the ground state in the regime of low
occupy localized wave functions, the overall energy may still barrier strengths where the coupling between the wells is
be less than that of a single condensate. Indeed, it can be strong. Moreover, this paper did not address the issue of the
shown that in the limit of a symmetric double-well potential spatial coherence of the ground state. Since the presence of
with an infinitely strong central barrier, one can always find long-range order is a defining characteristic of Bose-Einstein
a fragmented state that has a total energy lower than any condensation, it is critical to understand the manners in
single condensate @3#. which this spatial coherence can be lost.
Thus we have the following situation in a double-well There has also been theoretical work on the problem of
potential: in the absence of any central barrier the ground Bose condensates containing atoms in two different internal
state is well approximated by a single condensate, while in states, which is analogous to the two well problem. Esry
the presence of an infinitely strong barrier it is well approxi- et al. @7# determined the probability distributions and life-
mated by a fragmented condensate. It is clear, therefore, that times of two interacting condensates in different internal
there must be a transition between these two extremes as one states confined to the same trap, while Cirac et al. @8# as well
increases the strength of the barrier. The first goal of this as Steel and Collett @9# considered the ground states of such
paper is to propose a theoretical model for describing this a system when the internal atomic states can be controlled by
transition. Specifically, we argue for an approximation of the a Josephson-like laser coupling. In this case, the distinguish-
fully interacting ground state that is more general than a ability of atoms in the two condensates is ensured by their

1050-2947/99/59~5!/3868~10!/$15.00 PRA 59 3868 ©1999 The American Physical Society

PRA 59 SPATIAL FRAGMENTATION OF A BOSE-EINSTEIN . . . 3869

internal state rather than their single-particle wave functions. State ~2! is certainly not the most general state one can
However, the requirement of orthogonality in the spatial de- consider. Indeed, such a state would be a poor choice if one
grees of freedom is indispensable in a multiple-well problem, were interested in studying the depletion of a single conden-
since it is precisely the shape of the single-particle wave sate due to interactions, since one there expects a certain
functions that determines the degree of fragmentation in the fraction of the particles to be distributed among a macro-
system. scopic number of single-particle states. However, in this pa-
The remainder of the paper is organized as follows. In per we are interested in the possibility of the particles being
Sec. II we present our model and define some useful mea- redistributed into a few single-particle states that are each
sures of spatial coherence of the condensate. In Sec. III we macroscopically occupied. We restrict ourselves to two
present approximate analytic solutions of our model in the single-particle states because the double-well geometry we
regime of nearly noninteracting particles, and compare these are considering encourages fragmentation into two pieces
to a numerical solution for a particular choice of the external @3#. Although it may be energetically favorable to fragment
potential. The experimental signature of fragmentation and into more than two pieces at very high particle densities, we
finite temperature effects are discussed in Sec. V, followed defer consideration of this possibility to a later work.
by our concluding remarks in Sec. VI. Among the many-body states defined by Eq. ~2!, we con-
sider only those which have the same symmetry as the
II. MODEL Hamiltonian under reflections about x50. This implies that
the single-particle wave functions are mirror images of one
A. Basic approach another across x50 within a phase factor, f 1 (2x,y,z)
Our system consists of an even number N of spinless 5e i u f 2 (x,y,z), and that the coefficients satisfy C N 1
bosons at zero temperature. We model the interactions by a 5C N2N 1 . With this assumption, and choosing f 1 and f 2 to
two-particle pseudopotential in the shape-independent ap- be real, the Hamiltonian takes the form
proximation, V(r,r8 )5g d (r2r8 ) with an interaction
strength g54 p a sc\ 2 /m, where a sc is the s-wave scattering Ĥ 2 5 e 11N̂1„e 121gT 1 ~ N̂21 ! …~ a †1 a 2 1a †2 a 1 !
length, and m is the mass of the bosons. The Hamiltonian is
given by @10# gT 0 2
1 ~ N̂ 1 1N̂ 22 2N̂ !

E F
2
\2 †
Ĥ5 d 3r 2 Ĉ ~ r! “ 2 Ĉ ~ r! 1U ~ r! Ĉ † ~ r! Ĉ ~ r!
2m gT 2 † †
1 ~ a 1 a 1 a 2 a 2 1a †2 a †2 a 1 a 1 14N̂ 1 N̂ 2 ! , ~3!

G
2
g
1 Ĉ † ~ r! Ĉ † ~ r! Ĉ ~ r! Ĉ ~ r! , ~1!
2
where N̂ 1 5a †1 a 1 , N̂ 2 5a †2 a 2 , N̂5N̂ 1 1N̂ 2 , and where

where Ĉ(r) is the quantum field operator, and U(r) is the

external potential. The external potential is taken to exhibit a
single minimum along the y and z axes, and a double mini-
e 115 E S
d 3 r f 1 ~ r! 2
\2 2
2m
“ 1U ~ r! f 1 ~ r! ,D
E S D
mum along the x axis. It is also taken to be symmetric about
\2 2
x50. e 125 d 3 r f 1 ~ r! 2 “ 1U ~ r! f 2 ~ r! ,
In order to capture the phenomenon of fragmentation in 2m
our model of the ground state, we must go beyond a mean-
field analysis. Specifically, we consider arbitrary superposi-
tions of Fock states where up to two single-particle states are T 05 E d 3 r f 41 ~ r! ,
occupied. This corresponds to postulating a state vector of
the form
N
T 15 E d 3 r f 31 ~ r! f 2 ~ r! ,

uc&5 ( C N 1 u N 1 ,N 2 & ~ f 1 , f 2 ! , ~2!

N 1 50
T 25 E d 3 r f 21 ~ r! f 22 ~ r! .
where
These quantities have the following physical interpretation:
~ a †1 ! N 1 ~ a †2 ! N 2 e 11 is the single-particle energy for the state f 1 ; e 12 is pro-
u N 1 ,N 2 & ~ f 1 , f 2 ! [ u vac&
AN 1 AN 2 portional to the inversion frequency of a single particle in the
external potential; finally, T 0 quantifies the self-interaction
energy, while T 1 and T 2 both quantify the cross-interaction
is the Fock state in which N 1 particles occupy the single-
energy.
particle state f 1 (r)5 ^ ru a †1 u vac& and N 2 particles occupy In order to facilitate comparison of our work with earlier
f 2 (r)5 ^ ru a †2 u vac& . The total number of particles is fixed, studies @6,9#, we re-express the Hamiltonian in terms of op-
N 2 [N2N 1 , the vector consisting of the set of coefficients erators satisfying angular momentum commutation relations,
C N 1 is normalized, and the single-particle wave functions f 1 rather than in terms of the creation and annihilation operators
and f 2 are both normalized and orthogonal to one another. we have employed thus far. We introduce the operators
3870 R. W. SPEKKENS AND J. E. SIPE PRA 59

1 B. A restricted variational principle

Ĵ z 5 ~ a †2 a 1 1a †1 a 2 ! ,
2 We now turn to the problem of identifying the ground
state of our model. This can be achieved by minimizing the
i expectation value of Ĥ 2 with respect to variations in both the
Ĵ y 5 ~ a †2 a 1 2a †1 a 2 ! , coefficients C N 1 and the single-particle wave function f 1 ,
2
subject to the constraints that the set of coefficients is nor-
1 malized, and f 1 is both normalized and orthogonal to f 2 , its
Ĵ x 5 ~ N̂ 2 2N̂ 1 ! , mirror image about x50. However, it is in fact more conve-
2
nient to minimize the expectation value of Ĥ 2 with respect to
which form an angular momentum algebra with total angular variations in f s and f a rather than f 1 . The reason for this is
momentum j5N/2 @6#. In terms of these operators, the that no constraint corresponding to orthogonality is required
Hamiltonian can be rewritten as when working with f s and f a , since they are orthogonal by
construction; as a result the analysis is simplified.
Ĥ 2 5E 0 12„e 121gT 1 ~ N21 ! …Ĵ z 12gT 2 Ĵ 2z 1g ~ T 0 2T 2 ! Ĵ 2x , We begin with the variation of f s and f a , implementing
~4! the normalization constraints through Lagrange multipliers
where E s and E a , respectively. This results in two coupled nonlin-
ear Schrödinger equations for f s and f a ,

F G
E 0 [ e 11N1 41 N ~ N22 !~ gT 0 1gT 2 ! ,
\ 2“ 2
2 1U ~ r! 1gG a° f a2 ~ r! 1gG ax f b2 ~ r! f a ~ r!
and where we have used Ĵ 2x 1Ĵ 2y 1Ĵ 2z 5 j( j11) to eliminate 2m
Ĵ 2y from the expression. The observable corresponding to Ĵ x 5E a f a ~ r! , ~5!
is the particle number difference between the localized states
f 1 and f 2 . Defining the wave functions f s 52 21/2( f 1 where
1 f 2 ) and f a 52 21/2( f 1 2 f 2 ), which are, respectively,
G a° 5 ^ ~ a a† a a ! 2 2a a† a a & / ^ a a† a a & ,
symmetric and antisymmetric about x50, one sees that Ĵ z
~6!
can be rewritten as 21 (a †s a s 2a †a a a ), where a †s and a †a are the G ax 5 ^ a a† a a† a b a b 1a b† a b† a a a a 14a a† a a a b† a b & / ^ a a† a a & ,
creation operators associated with f s and f a , respectively.
Thus, Ĵ z corresponds to the particle number difference be- and where the indices ( a , b ) take the values (s,a) and (a,s).
tween the symmetrized states f s and f a . Finally, Ĵ y corre- We now minimize the expectation value of Ĥ 2 with re-
sponds to the condensate momentum. Since we are consid- spect to variations in the coefficients C N 1 , and implement the
ering the ground state, it follows that ^ J y & 50, and since the normalization constraint on the C N 1 through a Lagrange mul-
ground state is symmetric under reflections about x50 it tiplier E. This results in a five-term recurrence relation for
follows that ^ J x & 50. the coefficients,

F N e 111
gT 0 2
2 G
~ N 1 1N 22 2N ! 12gT 2 N 1 N 2 2E C N 1 1 @ e 121gT 1 ~ N21 !#@ AN 1 ~ N 2 11 ! C N 1 21 1 AN 2 ~ N 1 11 ! C N 1 11 #

gT 2
1 @ A~ N 1 21 ! N 1 ~ N 2 11 !~ N 2 12 ! C N 1 22 1 A~ N 2 21 ! N 2 ~ N 1 11 !~ N 1 12 ! C N 1 12 # 50, ~7!
2

for each value of N 1 ; E is immediately identified as the ex- also minimizes E; thus it may be necessary to compare the
pectation value of Ĥ 2 . The latter set of equations forms a energies of many solutions in order to find the ground state.
matrix eigenvalue equation for the N-element vector of co- C. Regime of nearly noninteracting particles
efficients C N 1 . Given values for e 11 , e 12 , T 0 , T 1 , and T 2 ,
The full problem outlined above is rather complex. In this
we can solve this equation by diagonalizing an N3N matrix
paper, we consider only perturbative solutions of Eq. ~5! in
with nonzero entries along five diagonals, a problem which the nearly noninteracting regime, which we here define as
is numerically tractable if the number of nonzero coefficients the regime where the interaction energy is small compared to
is not too large. the difference between e (1)
s , the energy of the first symmetric
Since Eqs. ~5! and ~7! form a coupled set of equations for single-particle excited state of the external potential, and e s ,
f s , f a , and C N 1 , we must in general solve these self- the energy of the ground state. This is ensured by the crite-
consistently. Of the many solutions thus obtained, the ground rion
state is the one which minimizes the value of E. However, it
is not obvious that the solution that minimizes E s and E a gNT 0 ! e ~s1 ! 2 e s . ~8!
PRA 59 SPATIAL FRAGMENTATION OF A BOSE-EINSTEIN . . . 3871

Since T 0 is on the order of the inverse of the volume of the The ground state is u N/2,N/2& ( f 1 , f 2 ) , which describes two
trap, this criterion places an upper limit on the density of the independent condensates, or, in other words, a condensate
condensate. which is fragmented into two pieces. Since we are consider-
In this regime, we can treat the nonlinear terms in Eq. ~5! ing a potential well that is symmetric about x50, the two
perturbatively. To obtain the expectation value of Ĥ 2 to first fragments are equally populated. It seems appropriate to re-
order in the perturbation, we need only solve for the eigen- fer to any state of the form u N 1 ,N 2 & ( f 1 , f 2 ) where N 1 and N 2
functions of Eq. ~5! to zeroth order. Thus, we need only are macroscopic and f 1 and f 2 are orthogonal, as a ‘‘dual
solve the two linear Schrödinger equations condensate’’ @11#. In this paper we will be concerned only
with dual condensates wherein N 1 5N 2 5N/2, and f 1 and

F 2
\ 2“ 2
2m G
1U ~ r! 2 e a f a ~ r! 50 ~9!
f 2 are mirror images of one another across x50.
The analysis above confirms, for the limit of nearly non-
interacting particles, the results of an earlier study @3#: the
ground state is well approximated by a single condensate at
for a 5s, a. In this case, the wave functions f s and f a are zero barrier strength, and a dual condensate at infinite barrier
simply the two lowest single-particle energy eigenfunctions strength. At intermediate barrier strengths, we keep all the
of the external potential, and the assumption of a state of the
form of Eq. ~2! corresponds to a two mode approximation. terms in Ĥ 2 for our calculations. Although the cross-
The solutions of Eq. ~9! determine the magnitudes of interaction terms are typically found to be small for generic
e 11 , e 12 , T 0 , T 1 , and T 2 , and these subsequently define shapes of the double-well potential, it is not obvious that
the form of the recurrence relation ~7! that must be solved to these terms are negligible for an arbitrary potential, and thus
obtain the coefficients C N 1 . Although the full results are pre- we include them in our analytic results wherever possible.
sented in Sec. III, it is illustrative to consider the ground state D. Measures of the degree of fragmentation
of Ĥ 2 in two particularly simple limits: that of no barrier and
that of an infinitely strong barrier. Finally, in order to facilitate the interpretation of our re-
sults, we highlight some observables that reveal the degree
In the absence of any barrier, u e 12u . e (1) s 2 e s , and since of spatial fragmentation of the condensate. The most useful
u T 1 u and T 2 are on the order of T 0 or less, it follows from
observables for this purpose are those that probe the spatial
criterion ~8! that u e 12u @NgT 0 ,Ng u T 1 u ,NgT 2 . If we provi-
coherence of the condensate across the barrier. In analogy to
sionally assume that the ground state fulfills the conditions
measures of optical coherence @12#, we normalize the first-
that N u ^ Ĵ z & u * ^ Ĵ 2z & , ^ Ĵ 2x & , then we are led to approximate the
order correlation function r 1 (r,r8 )5 ^ Ĉ † (r)Ĉ(r8 ) & to ob-
Hamiltonian by tain the degree of first-order spatial coherence between
points r and r8 ,
Ĥ 2 .E 0 12 e 12Ĵ z . ~10!
r 1 ~ r,r8 !
g ~ 1 ! ~ r,r8 ! 5 . ~12!
The ground state of Eq. ~10! is simply the Fock state u N & f s @ r 1 ~ r,r! r 1 ~ r8 ,r8 !# 1/2
which describes N particles occupying the single-particle
Considering points r5(x,y,z) and r8 5(2x,y,z) where x is
ground state f s . Since this solution satisfies our provisional
positive and chosen to be sufficiently large so that u f 1 (r) u
assumption, the approximation is consistent. Such a Fock
! u f 2 (r) u and u f 1 (r8 ) u @ u f 2 (r8 ) u , for any state of the form
state is of course what one would expect for the ground state
of a single well in the limit of nearly noninteracting particles. of Eq. ~2! that is symmetric under reflection about x50, the
When this state is written in the form of Eq. ~2!, that is, in quantity g (1) (r,r8 ) is in fact independent of r and r8 , and has
the basis of u N 1 ,N 2 & ( f 1 , f 2 ) states, rather than the basis of the value
u N s ,N a & ( f s , f a ) states, the coefficients C N 1 form a binomial ^ a †1 a 2 1a †2 a 1 &
distribution over N 1 , centered at N/2. It seems appropriate to C ~ 1 !5 . ~13!
N
refer to any state of the form u N & f 0 for macroscopic N and
arbitrary f 0 , as a ‘‘single condensate.’’ In this paper, we are We refer to C (1) simply as the degree of first-order spatial
concerned only with single condensates wherein the single- coherence across the barrier. It is straightforward to verify
particle wave function f 0 is symmetric about x50. We do that it attains its maximum value of 1 for a single condensate
not introduce any additional terminology to distinguish such and a value of 0 for a dual condensate.
a state from one with arbitrary f 0 , since no confusion is The second-order correlation function r 2 (r,r8 )
likely to arise. 5 ^ Ĉ † (r)Ĉ † (r8 )Ĉ(r8 )Ĉ(r) & , which is simply the normally
In the limit of infinite barrier strength, the amplitudes of ordered density-density correlation, can be normalized to ob-
f s and f a at x50 are necessarily zero, while f s and f a at tain the degree of second-order spatial coherence between
xÞ0 satisfy the same equation. Consequently, f s and f a points r and r8 ,
differ only in their symmetry under reflection about x50 and
thus e 125T 1 5T 2 50. The Hamiltonian of Eq. ~4! then re- r 2 ~ r,r8 !
g ~ 2 ! ~ r,r8 ! 5 . ~14!
duces to @ r 2 ~ r,r! r 2 ~ r8 ,r8 !# 1/2
Defining C (2) in a manner completely analogous to C (1) , we
Ĥ 2 5E 0 1gT 0 Ĵ 2x . ~11! find
3872 R. W. SPEKKENS AND J. E. SIPE PRA 59

124
N S D
DN 1 2 a Taylor expansion of C(u1 p/N) to second order in p/N. In
this way, the recurrence relation ~7! for the C N 1 can be recast

S D
C ~ 2 !5 2, ~15! as a second-order differential equation for the function C(u).
N22 DN 1
14 Moreover, any sum over N 1 can be approximated by an in-
N N tegral over u. In particular, the constraint of normalization
for the coefficients is replaced by the constraint that the in-
where DN 1 [( ^ N̂ 21 & 2 ^ N̂ 1 & 2 ) 1/2 is the variance in the number tegral of C 2 (u) over all u is 1/N. The assumption of smooth-
of particles occupying the localized state f 1 . We refer to ness is readily verified to be appropriate for a single conden-
C (2) as the degree of second-order spatial coherence across sate, and we therefore expect it to continue to hold for
the barrier. The variance in N 1 for a single condensate u N & f s solutions over a range of small barrier heights.
is that of a binomial distribution over N 1 , namely, AN/2. For We also make use of the fact that the coefficients C N 1 are
a dual condensate the number of particles in a well is fixed, significant only in the region where N 1 & AN, or, equiva-
so that DN 1 50. As a consequence, C (2) 2150 for a single lently, that the function C(u) is significant only where u
condensate u N & f s , and C(2) 2152/(N22) for a dual conden- & A1/N. This follows from the fact that any set of coeffi-
sate. Since DN 1 is sufficient to specify C (2) , while being cients that has significant amplitude outside the range N 1
simpler to interpret, we use DN 1 together with C (1) to char- & AN also has an energy that is larger than a single conden-
acterize our results. sate; the interaction energy is greater since it scales with
DN 1 , and the single-particle energy is greater since it is a
III. ANALYTIC APPROXIMATIONS AND NUMERICAL minimum for a single condensate. It is therefore appropriate
SOLUTIONS to expand each of the N 1 -dependent terms as a power series
in u,
For a given shape of the double-well potential, it is
straightforward to obtain the single-particle ground state and `
2
first excited state by solving the linear Schrödinger equation
N (
AN 1 ~ N 2 11 ! 5 I n u n ,
~9!. The localized single-particle states f 1 and f 2 are simply n50

the sum and difference of the single-particle ground and first

`
excited states. Using these wave functions, the coefficients 4
C N 1 can be obtained by solving the recurrence relation ~7!. 2
A~ N 1 21 ! N 1 ~ N 2 11 !~ N 2 12 ! 5 ( J nu n,
N n50
Here we present approximate analytic solutions for the
C N 1 given an arbitrary shape of the double-well potential. which implicitly defines the I n and J n .
Subsequently, we consider a particular form of the external The second-order differential equation for C(u) we obtain
potential for which the single-particle ground, and first ex- is found to have a first-order term which can be eliminated
cited states are obtained numerically. This allows us to ob- by the substitution C̄(u)5C(u)exp(2(na2nu2n) with an ap-
tain a numerical solution for C N 1 and to compare this solu- propriate choice of the constants a 2n . It then follows that
tion to the analytic approximations. C̄(u) satisfies a second-order differential equation identical
to that of a particle with position u in a one-dimensional
A. Continuum approximation potential well of even powers of u. Since the solution is only
Suppose the coefficients for the ground state satisfy the significant in the range u& A1/N, we make the approxima-
condition tion that terms in the potential that are quartic or of higher
order in u are small perturbations upon the quadratic term,
u C N 1 11 2C N 1 u !C N 1 . and can be neglected. In this case, the function C̄(u)
5C(u)exp(2tu2/2) satisfies the equation for the modes of a
It is then useful to construct a function C(u), defined over simple harmonic oscillator:
the real numbers, such that C(u)5C N 1 at the discrete points
u5N 21 (N 1 2N/2), and such that u C 8 (u) u !NC(u). Given 2C̄ 9 ~ u ! 1 ~ n 2 1 t 2 ! u 2 C̄ ~ u ! 5 ~ h 1 t ! C̄ ~ u ! , ~16!
this assumption of smoothness, a coefficient of the form
C N 1 1p , where p is a small integer, is well approximated by where, after expanding the I n and J n to leading order in 1/N,

h 52N 2
F
E/N2 e 111 e 121Ng S1
4
3
T 0 1T 1 1 T 2
4 DG ,
2 e 122Ng ~ T 1 1T 2 !

n 25
2N 2
2 e 122Ng ~ T 1 1T 2 ! FS D
22
N
h
2
X S D CG
~ 2 e 122NgT 1 ! 1Ng T 0 1 2
h
N2
23 T 2 , ~17!

t 52.
PRA 59 SPATIAL FRAGMENTATION OF A BOSE-EINSTEIN . . . 3873

Thus the solution for C(u) that minimizes h , thereby mini- B. Two-coefficient approximation
mizing the energy E, is a Gaussian, At large barrier strengths, we make use of the following
conditions:
1 1 2 /4s 2
C~ u !5 e 2u , ~18! u g u !1,
AN ~2p! As
1/4

with width where ~22!

s5
1 1
, ~19!
AS N N
2 2 D
11 „2 e 122g ~ N21 ! T 1 …
2 AN A2A1 AA 2 1B g[
gT 0

where and

A5
3
4
2 e 122Ng T 1 1 T 2
3
,
S D where
u z u !1,

~23!
2N 2 e 122Ng ~ T 1 1T 2 !

S D
N 2 gT 2
3 1 z[
AS D
.
2 e 122Ng T 1 1 T 2 2 T 0 N N
2 2
B5 , 11 „2 e 122g ~ N21 ! T 1 …
2 e 122Ng ~ T 1 1T 2 ! 2 2
The first of these conditions is always satisfied for suffi-
and with energy
ciently strong barriers, since in the limit of an infinitely

E5N F 2 e 122Ng ~ T 1 1T 2 !
4N 2 s 2
1 ~ e 111 e 12!
strong barrier, e 125T 1 50, while T 0 is finite. Moreover, we
have numerically verified that the second condition holds at
sufficiently large barrier strengths for a variety of external

S DG
potentials. Within the domain of applicability of these con-
1 3 ditions, we seek coefficients C N 1 that satisfy the recurrence
1Ng T 0 1T 1 1 T 2 . ~20!
4 4 relation ~7! to first order in g and z . Dividing the recurrence
relation by gT 0 , one finds that all terms involving N 2 gT 2
The variance in N 1 for this solution is simply N s , while the have a magnitude on the order of g z and can therefore be
degree of first-order spatial coherence is given by neglected. In this limit, the following set of coefficients are a

S D
solution:
2N2 1
C ~ 1 ! 5e 21/8s 11 22 s 2 . ~21! N
N C N 1 5 A122 g 2 for N 1 5
2
In the absence of any barrier, u e 12u
@NgT 0 ,Ng u T 1 u ,NgT 2 , and it can be verified that Eq. ~19! N N
5g for N 1 5 11, 21 ~24!
predicts the appropriate value for this limit, namely, DN 1 2 2
. AN/2, the value for a single condensate. As the barrier
strength is increased, the magnitudes of e 12 , T 1 , and T 2 50 otherwise.
decrease, while the magnitude of T 0 does not vary signifi-
cantly; it therefore follows from Eq. ~19! that DN 1 will de- The energy in this case is
crease with barrier strength. When DN 1 falls below 1, the
assumption of smoothness breaks down. Thus the range of
validity of the continuum approximation is DN 1 *1 or,
E5N e 111 S N ~ N22 !
4
22 g 2 gT 0 . D ~25!

equivalently, C (1) *0.88.

We refer to this approximation as the two-coefficient ap-
If the potential is such that Ng u T 1 u ,NgT 2 ! u e 12u continues
proximation, and we dub any state of the form of Eq. ~24! a
to hold as the barrier strength is raised from zero, then to
perturbed dual condensate. For such a state, C (1) and DN 1
leading order in 1/N the expression for s simplifies to
are given by

s5
1 1
C ~ 1 ! 52 g A122 g 2 A 2

A
, ~26!
2 AN N gT 0
11 ,
N
11
2 ~ 2 e 12!
DN 1 5 A2 g . ~27!
depending only on the ratio of the interaction energy to the
splitting between the symmetric and antisymmetric levels of Keeping terms to first order in g , and to leading order in
the trap. powers of 1/N, we have C (1) 52 g . The range of validity of
3874 R. W. SPEKKENS AND J. E. SIPE PRA 59

the two-coefficient approximation is the range of barrier

strengths for which C (1) !1 and DN 1 !1.
An alternative manner of deriving this solution that is
perhaps more physically intuitive, is to begin by assuming a
state of the form of Eq. ~24! and showing that the value of g
that minimizes the energy is indeed the value given in Eq.
~22!. We begin by recalling the form of Ĥ 2 , exhibited in Eq.
~3!. Assuming that u z u !1, the T 2 term in (Ĥ 2 ) can safely be
neglected. If we introduce the operators

N
n̂ 1 5N̂ 1 2 ,
2

N
n̂ 2 5N̂ 2 2 ,
2

then we find that

^ Ĥ 2 & .E dual1„e 121gT 1 ~ N21 ! …^ a †1 a 2 1a †2 a 1 &

1 12 gT 0 ~ ^ n̂ 21 & 1 ^ n̂ 22 & ! ,

where E dual is the energy of the dual condensate. Since the

magnitude of the cross-interaction term involving T 1 only
depends on the many-body state through expectation values
of bilinear operators, this term, together with the e 12 term,
can be considered as an effective single-particle energy. To FIG. 1. Numerical solutions for the single-particle wave func-
first order in g and to leading order in 1/N, the perturbed tions f 1 (r) and f 2 (r) and the coefficients C N 1 for N5100 particles
dual condensate has ^ a †1 a 2 1a †2 a 1 & 52 g N, and consequently and double-well potentials with barrier strengths of a
it has an effective single-particle energy benefit over the dual 50, 15, 30, 45, and 60 in units of A\ 3 v x /m. The remaining pa-
condensate of 2 g N(2 e 122gNT 1 ) ~this quantity is positive rameter values are specified in the text. The dotted curve is the
at the barrier strengths of interest, since e 12,0 and typically external potential along the x axis U(x,y50,z50) in units of \ v x .
T 1 ,0). On the other hand, ^ n̂ 21 & 5 ^ n̂ 22 & 52 g 2 for such a The solid and dashed curves are f 1 (x,y50,z50) and f 2 (x,y
state, corresponding to a self-interaction energy cost of 50,z50), respectively, in arbitrary units.
2 g 2 gT 0 . Thus the effective single-particle energy of the per-
turbed dual condensate decreases linearly with g , while the a 2 /2d 2
self-interaction energy increases quadratically with this pa- U ~ r! 5m ~ v 2x x 2 1 v 2y y 2 1 v 2z z 2 ! 1 e 2x . ~28!
rameter. The minimum occurs precisely when g has the A2 p d
value N(2 e 122gNT 1 )/2gT 0 , which approximates the value
in Eq. ~22! for N@1.
The parameter values appropriate for Ref. @1# are v x 52 p
For typical double-well potentials, where Ng u T 1 u ! u e 12u ,
319 Hz, v y 5 v z 52 p 3250 Hz, and d 56 m m. Within
g is well approximated by N(2 e 12)/2gT 0 to leading order in
such a trap, criterion ~8! for the applicability of our pertur-
1/N. In this case, the transition from a completely frag-
bative approach becomes N!100, which is much smaller
mented condensate to one that shows some coherence across
than the number of condensate atoms in their experiment.
the barrier occurs when the number of particles times the
We consider instead a larger trap, specifically, one which is
ratio of the inversion frequency to the self-interaction energy
isotropic with the trapping frequency of the axis of weakest
becomes non-negligible. Thus, for a given shape of the trap
and a fixed barrier strength, the degree of fragmentation de- confinement in the MIT trap, v x 5 v y 5 v z 52 p 319 Hz. In
creases as the number of particles is increased but increases this case our perturbative approach is good for up to approxi-
as the strength of the interaction is increased. mately N5100 particles, and this is the example we con-
sider. The scattering length of 23Na is taken to be a sc53 nm
@13#.
C. Numerical solutions In Fig. 1 we plot the profile along the x axis of the exter-
The form of the double-well potential in the MIT conden- nal potential U(r) and the wave functions f 1 (r) and f 2 (r),
sate interference experiment @1# is well modeled by a term together with the coefficients C N 1 for several values of the
that is harmonic along all three Cartesian axes, with frequen- barrier strength. Figure 2 displays the degree of first-order
cies v x , v y , and v z , respectively, to which is added a spatial coherence, C (1) , and the variance, DN 1 , in the number
Gaussian barrier of width d and strength a centered at x of particles occupying the localized state f 1 as a function of
50: the barrier strength. The generic features of these results per-
PRA 59 SPATIAL FRAGMENTATION OF A BOSE-EINSTEIN . . . 3875

gravity through a pair of slits, located symmetrically about

x50. These slits can be formed by changing the shape of the
trapping potential, as long as this change is not so rapid that
excitations are induced, and not so slow that the system has
time to relax to a new many-body ground state. For simplic-
ity of the analysis, we also assume that the particles on the
left and right are each given momentum translations of mag-
nitude \k toward one another @14#. In the absence of such
translations, the interference pattern is simply more compli-
cated, and has been studied by Rörhl et al. @4#. We make the
approximation that the interparticle interactions are insignifi-
cant during this expansion period. In this case, only the
single-particle wave functions evolve, while the coefficients
C N 1 in the many-body state ~2! remain unchanged.
Suppose the slits are centered at points r5(x,y,z) and
r8 5(2x,y,z), where x is positive and chosen to be suffi-
ciently large so that u f 1 (r) u ! u f 2 (r) u and u f 1 (r8 ) u
@ u f 2 (r8 ) u . In this case, the single-particle wave functions
f 1 (r) and f 2 (r) evolve to wave functions localized entirely
at just one of the slits. After the momentum translation and a
FIG. 2. ~a! The degree of first-order spatial coherence across the period of free expansion the single-particle wave functions
barrier, C ( 1 ) , and ~b! the variance in the occupation number of one originating from the left and right of the barrier acquire com-
of the wells, DN 1 , as a function of the barrier strength, a , in units plex phase factors of e ikx and e 2ikx and magnitudes we de-
of A\ 3 v x /m for N5100 particles and the parameter values speci- note by f̃ 1 (r) and f̃ 2 (r), respectively. For many-body states
fied in the text. The solid curve is the numerical solution, the dashed that are symmetric under a reflection about x50, these mag-
curve is the continuum approximation, and the dotted curve is the nitudes are roughly uniform and equal in the far field of the
two-coefficient approximation. double slit, so that the many-body state in the far field can be
approximated by the many-body state prior to removal of the
sist for different choices of parameters in Eq. ~28! as well as trap, with f 1 (r) and f 2 (r) replaced by e ikx and e 2ikx , re-
for different choices of the form of U(r). spectively.
Also displayed in Fig. 2 are the values for C (1) and DN 1 We now imagine detectors in the far field that are as-
given by the continuum approximation, as specified by Eqs. sumed to remove atoms from the condensate @15#. The prob-
~19! and ~21!, and given by the two-coefficient approxima- ability distribution over the position r1 of the first detection
tion, as specified by Eqs. ~26! and ~27!, for the same choice is given by the expectation value of normally ordered field
of external potential. For their respective ranges of validity, operators P 1 (r1 )5(1/N) ^ Ĉ † (r1 )Ĉ(r1 ) & . The probability
the analytic approximations are found to fit the numerical distribution over the positions r1 ,r2 , . . . ,rm of the first m
results extremely well. detections is given by P m (r1 ,r2 , . . . ,rm )5 @ (N2m)!/
From these calculations arise the following picture of the
transition between a single condensate and a fragmented N!] ^ Ĉ † (r1 )Ĉ † (r2 )•••Ĉ † (rm )Ĉ(rm )•••Ĉ(r2 )Ĉ(r1 ) & . The
condensate. Moving up from zero barrier strength, there is a density of detection events that emerges in a single run of the
range of barrier strengths over which C (1) is close to unity, double-slit experiment has the form r m (r)5(N/m) ( m i51 d (r

while DN 1 falls from its single condensate value of AN/2 to 2ri ) where the set of positions $ ri % is obtained from the
a value of 1. Moving down from infinite barrier strength, probability distribution P m (r1 ,r2 , . . . ,rm ). Of course, the
there is a range of barrier strengths over which C (1) and DN 1 finite resolution of any realistic detector can be accounted for
are both much less than 1. Between these two domains, there by replacing the d function in this expression with a suitably
is a narrow range of barrier strengths wherein the greatest broadened distribution; as long as the resolution is finer than
the distance between the fringes of the interference pattern,
part of the transition in C (1) is made. The barrier strengths
the difference will not be significant.
delimiting these domains can be estimated analytically using
In a single run of the double-slit experiment, both the
the approximations presented in this section.
single and dual condensates typically yield a distribution
r m (r) with essentially the maximum possible fringe visibil-
IV. DISCUSSION ity. This is obviously true for a single condensate, and has
been shown to be true for a dual condensate in the seminal
A. Experimental signature of fragmentation
paper of Javanainen and Yoo @15#. Thus the mere presence
Herein we consider a measurement of the first-order de- of such interference is not indicative of a nonzero degree of
gree of spatial coherence. This is accomplished by a type of first-order coherence. However, suppose the experiment is
interference experiment that has been widely discussed in the repeated many times with the same initial many-body state.
literature @1,14–17#. Essentially, it constitutes a double-slit In this case, the spatial phase of the interference pattern will
experiment for Bose condensates. The thought experiment vary randomly from one run to the next if the initial state is
runs as follows. After preparation of the condensate, the trap a dual condensate, while it will remain fixed if the initial
potential is removed and the atoms fall under the force of state is a single condensate @15#. We therefore expect the
3876 R. W. SPEKKENS AND J. E. SIPE PRA 59

degree of first-order spatial coherence to be revealed by the ages the pattern of detections from an infinite number of runs
variance in the spatial phase of the interference pattern over of the double-slit experiment, all prepared initially in the
many runs, or, equivalently, the fringe visibility of the aver- same many-body state, and each involving m detection
age detection pattern over many runs. Indeed, if one aver- events, one obtains

r̄ m ~ r! 5 E d 3 r 1 •••d 3 r m S N
m
m

( d ~ r2ri !
i51
D P m ~ r1 ,r2 , . . . ,rm !

E
m
N
5
m (
i51
d 3 r 1 •••d 3 r i21 d 3 r i11 •••d 3 r m P m ~ r1 , . . . ,ri21 ,r,ri11 , . . . ,rm ! 5 ^ Ĉ † ~ r! Ĉ ~ r! & . ~29!

The final equality follows from the fact that each element of V. CONCLUSIONS
the sum is simply equal to P 1 (ri ). If the fringe visibility of
A theoretical treatment of fragmentation in Bose conden-
the average detection pattern, r̄ m (r), is evaluated for the sates must go beyond a mean-field analysis. For the case of
many-body state in the far field, it is found to be precisely repulsive interparticle interactions and a double-well trap-
equal to our definition of C (1) , the degree of first-order spa- ping potential, we have proposed an approach wherein an
tial coherence across the barrier, and thus enables a measure- approximation to the many-body ground state is obtained by
ment of the latter. a restricted variational principle. We have implemented this
Another possibility for an experimental study of fragmen- proposal for the case of nearly noninteracting particles.
tation is a measurement of the degree of second-order spatial The coherence properties that we have considered are the
coherence; this may be accessible through nonresonant im- degrees of first- and second-order coherence across the cen-
aging @18#. tral barrier of the potential. The first of these quantifies the
variance over many runs in the spatial phase of the fringe
B. Finite temperatures
pattern arising from the interference of atoms on either side
of the barrier. The second is essentially the density-density
We close this section with a few comments on the effect correlation across the barrier, and for the states we consider
of finite temperatures on the coherence properties of the con- it is a simple function of the variance in the number of par-
densate. The excited states of the system are not in general ticles in one of the wells. We find that as the barrier strength
well approximated by a state of the form of Eq. ~2!. None- is increased, this variance is continuously squeezed down
theless it can be shown that at infinite barrier strength the from its value for a single condensate. The degree of first-
first-order degree of spatial coherence across the barrier is order spatial coherence is close to unity when this variance is
zero for thermal equilibrium at any finite temperature. The greater than 1, but thereafter drops off rapidly. Above a cer-
proof is as follows. For an infinitely strong barrier the Hamil- tain critical barrier strength we find that both quantities be-
tonian is separable into two terms, each involving only op- come much less than 1, indicating that the condensate is
erators pertaining to particles on one side of the barrier. Con- essentially completely fragmented.
sequently, any nondegenerate energy eigenstates are tensor We have discussed how the degree of first-order coher-
products of states describing particles on one side of the ence might be measured through interference experiments,
barrier only. For any set of energy eigenstates that are de- and argued that a significant effect should be present even at
generate, the subspace of Hilbert space spanned by this set finite temperatures. A concern, however, is that the ground
has a basis of such product states. Thus one can always iden- state might be difficult to prepare if the relaxation time of the
tify a basis of energy eigenstates that are product states of system is long compared to the lifetime of the condensate.
this sort. Since the density operator that represents thermal This could arise if the only way for the particles to be redis-
equilibrium, r̂ , is a mixture of these energy eigenstates, we tributed across the barrier is by tunneling through it. How-
will necessarily have Tr„r̂ Ĉ † (r)Ĉ(r8 )…50 if r and r8 are ever for numbers of particles that are not too large, this tun-
on opposite sides of the barrier, and consequently the degree neling time need not be restrictive. For instance, in the
of first-order spatial coherence across the barrier for such a example presented in Sec. III C, the single-particle tunneling
mixed state is zero as well. In the absence of any barrier, as time at the barrier strength where C (1) 50.88 is approxi-
long as the temperature is small enough that most of the mately 1 min, while at the barrier strength where C (1) 50.1 it
particles are in the lowest single-particle energy level, the is roughly 1 h.
first-order degree of spatial coherence for the thermal state Our variational approach can be extended in a straightfor-
should be close to unity. For such temperatures, if the barrier ward manner to the determination of the many-body ground
strength is varied from zero to infinity, the first-order degree state in systems where the external potential has an arbitrary
of spatial coherence of the thermal state will vary from number n of minima. In such cases, one would simply intro-
nearly unity to zero. Thus a significant transition must still duce states that are arbitrary superpositions of Fock states
occur at such temperatures. where up to n single-particle wave functions are occupied.
PRA 59 SPATIAL FRAGMENTATION OF A BOSE-EINSTEIN . . . 3877

Such an analysis is of relevance to the determination of the ACKNOWLEDGMENTS

coherence properties of Bose condensates in optical lattices
@19#. Moreover, at extremely high densities, where the inter- We wish to thank Allan Griffin, Aephraim Steinberg,
action energy is dominant, it may become energetically fa- Mike Steel, and Wolfgang Ketterle for helpful discussions.
vorable for a condensate to begin to fragment even in the This work was supported by the National Sciences and En-
presence of a perfectly uniform potential. We hope to ad- gineering Research Council of Canada, and Photonics Re-
dress this possibility in future work. search Ontario.

@1# M. R. Andrews, C. G. Townsend, H.-J. Miesner, D. S. Durfee, @11# The term ‘‘double condensate’’ has been used to denote two
D. M. Kurn, and W. Ketterle, Science 275, 637 ~1997!. condensates of atoms of different internal spin states @8#. We
@2# P. Nozières, in Bose-Einstein Condensation, edited by A. Grif- use the term ‘‘dual condensate’’ for two condensates that are
fin, D. W. Snoke, and S. Stringari ~Cambridge University only distinguished by the spatial wave functions of the atoms.
Press, Canbridge, 1995!, Chap. 2. @12# R. Loudon, The Quantum Theory of Light (Oxford University
@3# R. W. Spekkens and J. E. Sipe, Prog. Phys. 46, 873 ~1998!. Press, New York, 1983!, Chap. 3.
@4# A. Röhrl, M. Naraschewski, A. Schenzle, and H. Wallis, Phys. @13# See Ref. @4# for a justification of this choice.
Rev. Lett. 78, 4143 ~1997!. @14# J. Javanainen and M. Wilkens, Phys. Rev. Lett. 78, 4675
@5# N. K. Wilkin, J. M. F. Gunn, and R. A. Smith, Phys. Rev. Lett. ~1997!.
80, 2265 ~1998!. @15# J. Javanainen and S. M. Yoo, Phys. Rev. Lett. 76, 161 ~1996!.
@6# G. J. Milburn, J. Corney, E. M. Wright, and D. F. Walls, Phys.
@16# J. I. Cirac, C. W. Gardiner, M. Naraschewski, and P. Zoller,
Rev. A 55, 4318 ~1997!.
Phys. Rev. A 54, R3714 ~1996!.
@7# B. D. Esry, Chris H. Greene, James P. Burke, Jr., and John L.
@17# M. Naraschewski, H. Wallis, A. Schenzle, J. I. Cirac, and P.
Bohn, Phys. Rev. Lett. 78, 3594 ~1997!.
Zoller, Phys. Rev. A 54, 2185 ~1996!.
@8# J. I. Cirac, M. Lewenstein, K. Mo” lmer, and P. Zoller, Phys.
@18# E. V. Goldstein and Pierre Meystre, Phys. Rev. Lett. 80, 5036
Rev. A 57, 1208 ~1998!.
@9# M. J. Steel and M. J. Collett, Phys. Rev. A 57, 2920 ~1998!. ~1998!.
@10# See, e.g., A. L. Fetter and J. D. Walecka, Quantum Theory of @19# D. Jaksch, C. Bruder, J. I. Cirac, C. W. Gardiner, and P. Zoller,
Many-Particle Systems ~McGraw-Hill, New York, 1971!. Phys. Rev. Lett. 81, 3108 ~1998!.