week ending

PRL 101, 130401 (2008) PHYSICAL REVIEW LETTERS 26 SEPTEMBER 2008

Experimental Observation of Oscillating and Interacting Matter Wave Dark Solitons

A. Weller, J. P. Ronzheimer, C. Gross, J. Esteve, and M. K. Oberthaler
Kirchhoff Institute for Physics, University of Heidelberg, INF 227, 69120 Heidelberg, Germany

D. J. Frantzeskakis
Department of Physics, University of Athens, Panepistimiopolis, Zografos, Athens 157 84, Greece

G. Theocharis and P. G. Kevrekidis

Department of Mathematics and Statistics, University of Massachusetts, Amherst Massachusetts 01003-4515, USA
(Received 7 April 2008; published 22 September 2008)
We report on the generation, subsequent oscillation and interaction of a pair of matter-wave dark
solitons. These are created by releasing a Bose-Einstein condensate from a double well potential into a
harmonic trap in the crossover regime between one dimension and three dimensions. Multiple oscillations
and collisions of the solitons are observed, in quantitative agreement with simulations of the Gross-
Pitaevskii equation. An effective particle picture is developed and confirms
pffiffiffi that the deviation of the
observed oscillation frequencies from the asymptotic prediction
z = 2, where
z is the longitudinal
trapping frequency, results from the dimensionality of the system and the soliton interactions.

DOI: 10.1103/PhysRevLett.101.130401 PACS numbers: 03.75.Lm, 05.45.Yv

Solitons are one of the most prominent features of non- Different methods have been explored to create dark
linear dynamics emerging in diverse fields extending from solitons in Bose-Einstein condensates [9–15]. In our ex-
hydrodynamics to solid state physics and from nonlinear periment, the solitons are generated by merging two co-
optics to biophysics. Dark solitons are the fundamental herent condensates initially prepared in a double well
excitations of the defocusing nonlinear Schrödinger equa- potential. This formation process can be regarded as a
tion [1], and have the form of a localized ‘‘dip’’ on a back- consequence of matter-wave interference of the two con-
ground wave, accompanied by a phase jump [2]. These densates [21–24]. The further evolution of the created
localized waveforms have been demonstrated experimen- solitons in the trap is shown in Fig. 1(a). Our procedure
tally in different contexts, including liquids [3], discrete is very similar to the recently reported generation of vor-
mechanical systems [4], thin magnetic films [5], optical tices out of a triple well potential [25].
media [6–8], and, more recently, Bose-Einstein conden- Since the two dominant solitons are created with a
sates (BECs) [9–15]. The possibility of creating pairs of distance of a few healing lengths  ( is on the order of
dark solitons [7] has stimulated considerable interest in the 250 to 400 nm), which defines the range of the repulsive
repulsive [16] short-range interactions between them soliton interaction, the collisions between them lead to a
[17,18]. The resulting collisions, during which the solitons significant modification of the oscillation frequency. The
approach within a distance comparable to their width, have measured frequencies deviate up to 16% from the single
a universal character and thus, e.g., optical solitons interact soliton
pffiffiffi asymptotic Thomas-Fermi 1D (TF1D) prediction of
essentially the same way as matter-wave solitons.
z = 2 [26], where
z is the longitudinal trapping fre-
In this Letter we report on the systematic generation of a quency. Our experimental results are in quantitative agree-
pair of matter-wave dark solitons, which is subsequently ment with numerical simulations of the Gross-Pitaevskii
oscillating and colliding in a harmonic trap. Our experi- equation (GPE). They reveal that dark solitons can behave
ment is performed in the crossover regime between one very similar to particles. This is confirmed by explaining
and three dimensions [19], where dark solitons exist and the essential features of the dynamics within a simple
are robust [20]. This allows us to monitor multiple oscil- physical picture regarding the dark solitons as particles in
lations and collisions of dark solitons, permitting the pre- an effective potential due to the external trap and their
cise measurement of their oscillation frequency and their mutually repulsive interactions. Being in the crossover
mutual repulsive interactions. Previous experiments have regime, the role of the transverse degrees of freedom has
been performed in a genuine 3D regime where dark sol- to be included in the effective potential [27].
itons are unstable due to the so-called snaking instability Before elaborating on the theoretical models and sys-
and eventually decay into vortex rings [11,20]. In these tematic studies we will briefly describe the details of the
experiments, solely their translation in the trap has been experimental setup. We prepare a BEC of 87 Rb in the
shown [9–11]. Only very recently dark solitons have been jF ¼ 2; mF ¼ 2i state, containing about N ¼ 1500 atoms
reported to undergo a single oscillation period in a har- in a double well potential. This potential is realized by
monic trap [15]. superimposing a far detuned crossed optical dipole trap

0031-9007=08=101(13)=130401(4) 130401-1 Ó 2008 The American Physical Society

 week ending
PRL 101, 130401 (2008) PHYSICAL REVIEW LETTERS 26 SEPTEMBER 2008

be approximately 1 kHz and the lattice spacing is 5:7 m.

This results in a double well potential with a well distance
of 5:4 m.
In order to start with a well-defined phase between the
two condensates, the barrier height is chosen to be low
enough such that thermal phase fluctuations are negligible
for the measured temperature of T  10 nK [28] (the
critical temperature for condensation is Tc  110 nK)
and high enough so that high contrast solitons are formed.
The solitons are created by switching off the optical lattice
and merging the two condensates in the remaining har-
monic potential. After the switching off, the trap frequen-
cies are ramped to the parameters of interest (
z ,
? ). The
distance between the formed solitons is adjusted by choos-
ing different sets of final frequencies and different atom
numbers. For each parameter set, the ramping time is
empirically optimized to minimize the excitation of the
quadrupole mode [e.g., from ð
z ;
? Þ ¼ ð63 Hz; 408 HzÞ
to (53 Hz, 890 Hz) within 10 ms for N ¼ 1700 atoms, or to
(58 Hz, 408 Hz) within 3 ms for N ¼ 950]. The atomic
density after a certain evolution time in the harmonic trap
is obtained using standard absorption imaging with an
optical resolution of approximately 1 m. We use a short
time of flight between 0.6 and 0.9 ms to enhance the
contrast.
In our experiment, the initial distance D ¼ 5:4 m
between the two colliding condensates is well within the
regime where the interaction energy exceeds the kinetic
energy and thus the formation of dark solitons is expected
due to nonlinear interference. This regime is reached if D is
FIG. 1. Observation of the time evolution of dark solitons in a smaller than the critical distance Dc ¼ ð6 N@a s 1=3

z m Þ ¼
harmonic trap. The dominant soliton pair is indicated by arrows.
25:8 m with as being the s-wave scattering length,
z
(a) Experimental observation of the dynamics of the longitudinal
atomic density. Each longitudinal density profile (vertical lines),
the longitudinal trap frequency and m the atomic mass
corresponding to a given evolution time, is deduced from typi- [22]. The formation of dark solitons for our experimental
cally 10 experimental realizations. The obtained absorption parameters is confirmed by 3D GPE simulations as shown
images of the condensate at each time step are averaged and in Fig. 1. Including the optical and time resolution, the
integrated over their transverse direction. The number of atoms experimentally observed density profile evolution is well
in the shown case is N ¼ 1700 and the trapping frequencies are reproduced. A dominant pair of solitons oscillates close to
ð
z ;
? Þ ¼ ð53 Hz; 890 HzÞ. (b) Result of the numerical integra- the center of the cloud and we can also distinguish addi-
tion of the 3D GPE taking into account the full preparation tional pairs of solitons with much lower contrast. In the
process of the solitons. (c) Same as (b), taking into account the following, we focus on the dynamics of the dominant
finite spatial (1 m) as well as temporal resolution (1 ms) of the
central pair and show that its oscillation frequency is
experiment. The loss of contrast due to the convolution process
explains the experimentally observed fading out of the solitons well described within a two soliton approximation.
with time. We experimentally investigate the oscillation frequency
of the dominant soliton pair for different trap parameters
and different intersoliton distances. A typical data set con-
( ¼ 1064 nm) and a one dimensional optical lattice ( ¼ sists of 50 time steps and 10 pictures per time step. The
843 nm). The first beam of the dipole trap has a Gaussian numerical simulations predict that the solitons do not cross
waist of 5 m and results in a strong transverse and weak each other at the collision points [see inset of Fig. 3(c)], but
longitudinal confinement. The second beam orthogonally our finite resolution does not allow us to distinguish
crosses the first one and has an elliptic shape (60 m
whether this is actually the case in the experiment. In order
230 m waist) leading to an extra adjustable confinement to extract the oscillation frequency of the solitons, we fit
only in the longitudinal direction of the trap. We start our the time evolution of the intersoliton distance as shown in
experiments with a transverse frequency of the total har- the inset of Fig. 2. The obtained frequency is divided by
monic trap of
? ¼ 408 Hz and a longitudinal one of
z ¼ two in order to compare it to the oscillation frequency
63 Hz. The barrier height of the optical lattice is chosen to expected for a single trapped soliton. The shot to shot
130401-2
 week ending
PRL 101, 130401 (2008) PHYSICAL REVIEW LETTERS 26 SEPTEMBER 2008

reproducibility of the soliton dynamics up to 100 ms allows perimental parameter range pffiffiffiffiffi is
 0:06–0:14, N
as =
the observation of up to 7 oscillation periods. The typical a?  1:2–1:8 and ððN=
Þas =a? Þ1=3  2:8–4:4, pffiffiffi which
statistical experimental error in the frequency measure- sets us out of the validity domain of the
z = 2 prediction.
ment is 1:5%. Figure 2 shows the results of our frequency We now give a theoretical description of the different
measurements and their comparison with numerical simu- effects leading to the observed upshift. We consider the
lations for the motion of two trapped solitons using the two solitons as particles moving in an effective potential
nonpolynomial Schrödinger equation (NPSE) [29], which which arises from the combination of a harmonic potential
is an excellent approximation to the 3D GPE in the dimen- due to the trap [26] [see Fig. 3(a)] and a repulsive potential
sionality crossover regime [27]. due to the interaction between the solitons [30]. Because of
In order to capture the essentials of the dynamics of the the spatially symmetric preparation, the effective potential
experimentally realized soliton pairs in the simulations, we is a symmetric double well potential which is depicted in
initialize the condensate with two solitons such that the rms Fig. 3(b). This potential can be expressed as a function of
amplitude of their oscillating motion matches the one the distance z of each of the solitons from the trap center
observed experimentally. The good agreement between and its time derivative z:_
numerics and experiments shows that the dynamics pro-
z2 B2
duced by our experimental method is well described within _ ¼ ð2
1s Þ2
Vðz; zÞ þ ; (1)
a two soliton approximation even though extra solitons are 2 2mcosh2 ð2Bz=Þ
produced. From our experiment and the NPSE simulations,
pffiffiffi
we observe an upshift up to 16% from the
z = 2 predic-
tion which was the first value theoretically derived for the
oscillation frequency of a single trapped soliton [26]. It is
expected to be valid in a 1D trap in thep
asymptotic
ffiffiffiffiffi Thomas-
Fermi limit (N
as =a?  1 and ððN=
Þas =a? Þ1=3  1)
[19], where
¼
z =
?  1 is the aspect ratio of the trap
and a? the transverse harmonic oscillator length. Our ex-

FIG. 3. The oscillation dynamics of dark solitons in a trapped

BEC is well captured in an effective particle picture. (a) For one
soliton, the particle moves in a harmonic trap. (b) For two
solitons, an additional barrier due to the repulsive interaction
appears. (c) For the one and two soliton case, the dependence of
the oscillation frequencies on the oscillation amplitude from the
FIG. 2. Comparison between experimentally obtained soliton trap center is shown for one experimental parameter set p with
ffiffiffi
oscillation frequencies and NPSE simulation for one and two
¼ 0:06. The dashed line shows the TF1D prediction (
z = 2).
solitons. Each frequency point is deduced from the temporal The thin solid line indicates the upshift of the single soliton
evolution of the soliton distance as shown in the inset. Different frequency mainly due to dimensionality. For the case of two
symbols correspond to different aspect ratios
of the trap. For solitons, the thick solid line also includes the upshift due to the
each aspect ratio the oscillation amplitude is varied as explained intersoliton interaction deduced from the NPSE. The dotted line
in the text. NPSE simulations are represented by solid lines for represents the result of the simple effective particle model from
the two soliton case, and by dashed lines for the respective single Eq. (1). Density profile evolutions obtained from the NPSE are
soliton oscillations. The error bars on the measured frequencies shown in the insets. A collision between the two solitons is also
account for statistical errors on the measured soliton and trap shown in detail, demonstrating that they do not cross each other.
frequencies and systematic errors on the atom number used to The white lines correspond to the trajectories of the density
calculate the healing length. minima.
130401-3
 week ending
PRL 101, 130401 (2008) PHYSICAL REVIEW LETTERS 26 SEPTEMBER 2008

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi simple effective particle picture confirms that the oscilla-

where B ¼ 1  ðz=Þ _ 2 ð@=Þ2 denotes the darkness of the
tion frequency of two solitons in a harmonic trap is affected
solitons,  is a typical interaction energy on the order of
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi by two effects, namely, the single soliton frequency upshift
the chemical potential,  ¼ @=ðmÞ the associated heal- and the intersoliton interaction. The presented robust
ing length and
1s the oscillation frequency of a single method for preparing solitonic excitations will be a starting
trapped soliton. The frequency of the motion is obtained by point for further studies towards multisoliton interactions
solving the Euler-Lagrange equation associated with the and perhaps even dark soliton gases.
Lagrangian Lðz; zÞ _ ¼ z_ 2 =2  Vðz; zÞ. _ To obtain quantita- We thank P. Schmelcher for useful discussions as well as
tive agreement, the model has to take into account cor- B. Hemmerling, R. Gati, and T. Ottenstein. We acknowl-
rectly both the free propagation of the solitons in the trap edge support from NSF, DFG, and AvH. J. E. acknowl-
when they are far away from each other (z  ) and the edges support from the EC MC-EIF program.
repulsive interaction when they approach.
Good estimates for the single soliton frequency
1s are
obtained by numerical integration of the NPSE describing
a single soliton. Because our experimental parameters are
both in the crossover regime and slightly out of the [1] V. E. Zakharov and A. B. Shabat, Sov. Phys. JETP 37, 823
Thomas-Fermi limit, corrections to the asymptotic value (1973).
pffiffiffi [2] Y. S. Kivshar and B. Luther-Davies, Phys. Rep. 298, 81

z = 2 are expected. Therefore the oscillation frequency
(1998).
of a single dark soliton is upshifted by a few percent from
[3] B. Denardo, S. Wright, and W. Putterman, Phys. Rev. Lett.
the asymptotic value as discussed in detail using the 64, 1518 (1990).
Bogoliubov–de Gennes analysis of the NPSE in [27] [see [4] B. Denardo et al., Phys. Rev. Lett. 68, 1730 (1992).
Fig. 3(a)]. The simulation results for the three different [5] M. Chen et al., Phys. Rev. Lett. 70, 1707 (1993).
parameter sets used in the experiment are shown in Fig. 2. [6] P. Emplit et al., Opt. Commun. 62, 374 (1987).
This upshift for the single soliton case can be decomposed [7] D. Krökel et al., Phys. Rev. Lett. 60, 29 (1988).
into two contributions. For example, considering one spe- [8] A. M. Weiner et al., Phys. Rev. Lett. 61, 2445 (1988).
cific parameter set with
 0:06, the upshift is 5% [see [9] S. Burger et al., Phys. Rev. Lett. 83, 5198 (1999).
Fig. 3(c)]. Predictions using the 1D GPE already give a [10] J. Denschlag et al., Science 287, 97 (2000).
value approximately 2% higher than the asymptotic limit [11] B. P. Anderson et al., Phys. Rev. Lett. 86, 2926 (2001).
because the Thomas-Fermi limit is not reached [31]. The [12] Z. Dutton et al., Science 293, 663 (2001).
[13] P. Engels and C. Atherton, Phys. Rev. Lett. 99, 160405
effect of dimensionality, i.e., the role of the transverse
(2007).
degrees of freedom which is captured only by the NPSE [14] G.-B. Jo et al., Phys. Rev. Lett. 98, 180401 (2007).
or the 3D GPE, accounts for the remaining pffiffiffi3%. Figure 3(c) [15] C. Becker et al., Nature Phys. 4, 496 (2008).
shows the comparison between the
z = 2 prediction and [16] K. J. Blow and N. J. Doran, Phys. Lett. A 107, 55 (1985).
the single soliton NPSE simulation for the considered [17] D. Foursa and P. Emplitt, Phys. Rev. Lett. 77, 4011 (1996).
parameter set. [18] A. Dreischuh et al., Phys. Rev. Lett. 96, 043901 (2006).
As also shown in Fig. 3(c), the repulsive interaction [19] C. Menotti and S. Stringari, Phys. Rev. A 66, 043610
between the solitons results in an additional upshift of (2002).
the oscillation frequency compared to the single soliton [20] A. E. Muryshev, H. B. van Lindenvan den Heuvell, and
case that strongly depends on the oscillation amplitude. G. V. Shlyapnikov, Phys. Rev. A 60, R2665 (1999).
[21] W. P. Reinhardt and C. W. Clark, J. Phys. B 30, L785
Our effective particle model accurately reproduces the
(1997).
upshift if the interaction parameter  is set to be the [22] T. F. Scott, R. J. Ballagh, and K. Burnett, J. Phys. B 31,
chemical potential of the condensate obtained from the L329 (1998).
3D GPE equation. In our experimentally accessible pa- [23] C. Lee, E. A. Ostrovskaya, and Y. S. Kivshar, J. Phys. B
rameter range, the agreement of the model with NPSE 40, 4235 (2007).
simulations is better than 5%. This allows us to clearly [24] Since the submission of the manuscript we learned about a
identify the significant role of the repulsive interactions related experiment: I. Shomroni et al., arXiv:0805.3263v1.
and shows that the effective repulsive potential in Eqn. (1) [25] D. R. Scherer et al., Phys. Rev. Lett. 98, 110402 (2007).
obtained in the 1D homogeneous case is a good approxi- [26] T. Busch and J. R. Anglin, Phys. Rev. Lett. 84, 2298
mation to our complex situation. (2000).
In conclusion, we controllably create pairs of dark sol- [27] G. Theocharis et al., Phys. Rev. A 76, 045601 (2007).
[28] R. Gati et al., Phys. Rev. Lett. 96, 130404 (2006).
itons by colliding two atomic clouds released from a
[29] L. Salasnich, A. Parola, and L. Reatto, Phys. Rev. A 65,
double well potential in a harmonic trap. The full dynamics 043614 (2002).
of multiple dark soliton oscillations and collisions is ob- [30] Y. S. Kivshar and W. Królikowski, Opt. Commun. 114,
served, allowing for precise frequency measurements and 353 (1995). This potential is valid provided that the
showing that dark solitons are still stable after several solitons do not overlap.
collisions. The experimentally observed total upshifts [31] V. A. Brazhnyi and V. V. Konotop, Phys. Rev. A 68,
from the TF1D frequency prediction are up to 16%. A 043613 (2003).

130401-4