Fermionic Atoms in a Three Dimensional Optical Lattice:

Observing Fermi Surfaces, Dynamics, and Interactions

Michael Kohl,
*
Henning Moritz, Thilo Stoferle, Kenneth Gunter, and Tilman Esslinger
Institute of Quantum Electronics, ETH Zurich, Honggerberg, CH-8093 Zurich, Switzerland
(Received 12 October 2004; published 4 March 2005)
We have studied interacting and noninteracting quantum degenerate Fermi gases in a three-dimensional
optical lattice. We directly image the Fermi surface of the atoms in the lattice by turning off the optical
lattice adiabatically. Because of the conning potential, gradual lling of the lattice transforms the system
from a normal state into a band insulator. The dynamics of the transition from a band insulator to a normal
state is studied, and the time scale is measured to be an order of magnitude larger than the tunneling time
in the lattice. Using a Feshbach resonance, we increase the interaction between atoms in two different spin
states and dynamically induce a coupling between the lowest energy bands. We observe a shift of this
coupling with respect to the Feshbach resonance in free space which is anticipated for strongly conned
atoms.
DOI: 10.1103/PhysRevLett.94.080403 PACS numbers: 03.75.Ss, 05.30.Fk, 34.50. s, 71.18.+y
The exploration of quantum degenerate gases of fermi-
onic atoms is driven by the ambition to get deeper insight
into long-standing problems of quantum many-body phys-
ics, such as high temperature superconductivity. Very re-
cently, the crossover regime between a strongly interacting
two-component Fermi gas and a molecular Bose-Einstein
condensate has been studied in harmonic traps [15].
These experiments mark a milestone towards the under-
standing of superuidity of fermionic atoms. However, the
analogy to an electron gas in a solid is limited since there
the electrons experience a periodic lattice potential. The
lattice structure is, in fact, a key ingredient for most models
describing quantum many-body phenomena in materials. It
has been suggested that strongly interacting fermionic
atoms in optical lattices could be employed for studies of
high-T
c
superconductivity [6], Mott-insulating phases [7],
Bose condensation of fermionic particle-hole pairs [8], or
interacting spin systems [9].
Here we report on an experiment bridging the gap
between current ultracold atom systems and fundamental
concepts in condensed matter physics. A quantum degen-
erate Fermi gas of atoms is prepared in the crystal structure
of a three-dimensional optical lattice potential created by
three crossed standing laser waves. The unique control
over all relevant parameters in this system allows us to
carry out experiments which are not feasible with solid-
state systems.
It was conceived by Jaksch et al. that ultracold atoms
exposed to the periodic potential of an optical lattice are an
almost ideal realization of a Hubbard model [10]. This
model is elementary to describe the quantum physics of
many electrons in a solid. It takes into account a single
band of a static lattice potential and assumes the interac-
tions to be purely local [11]. Ultracold atoms in an optical
lattice give a very direct access to the underlying physics.
The fundamental parameters include the tunnel coupling
between adjacent lattice sites, the atom-atom interactions,
and the dimensionality of the system. Previous experi-
ments with far-detuned three-dimensional optical lattices
[1214] were always carried out with bosonic atoms, and
experiments with fermions were restricted to a single
standing wave [15]. In the latter situation many atoms
can reside in each standing wave minimum, but formation
of a band insulator is prevented by the weak transverse
connement. The observed inhibition of transport [16] is
due to localized states and therefore differs qualitatively
from the band insulator which we create in the three-
dimensional optical lattice.
The experiments are performed in a modied apparatus
previously used to study bosonic rubidium atoms in optical
lattices [14,17]. A mixture of bosonic
87
Rb and fermionic
40
K atoms is captured in a magneto-optical trap. For mag-
netic trapping we optically pump the potassium atoms into
the jE 9,2, m
E
9,2i and the rubidium atoms into the
jE 2, m
E
2i hyperne ground state, with E being the
total angular momentum and m
E
the magnetic quantum
number. The mixture is evaporatively cooled using micro-
wave radiation to selectively remove the most energetic
rubidium atoms from the trap. The potassium cloud is
sympathetically cooled by thermal contact with the rubid-
ium atoms [18]. After reaching quantum degeneracy for
both species with typically 6 10
5
potassium atoms at a
temperature of T,T
F
0.32 (T
F
260 nK is the Fermi
temperature of the noninteracting gas), we remove all the
rubidium atoms from the trap. The potassium atoms are
then transferred from the magnetic trap into a crossed beam
optical dipole trap whose laser beams possess a wavelength
of 826 nm and are focused at the position of the Fermi
gas to 1,e
2
radii of 50 m (x axis) and 70 m (y axis).
The initial trapping frequencies are w
x
2r 93 Hz,
w
y
2r 154 Hz, and w
z
2r 157 Hz. When load-
ing the optical trap, we turn off the magnetic connement
in such a way that a variable homogeneous magnetic eld
remains present. In the optical trap we prepare a spin
mixture with 50 4% in each of the jE 9,2, m
E

9,2i and jE 9,2, m
E
7,2i spin states using a
PRL 94, 080403 (2005)
P HYS I CAL RE VI E W L E T T E RS
week ending
4 MARCH 2005
0031-9007,05,94(8),080403(4)$23.00 080403-1 2005 The American Physical Society
sequence of two radio-frequency pulses. By lowering the
depth of the optical trap on a time scale of 2 s we further
evaporatively cool the potassium gas. This is done at a bias
magnetic eld of B 227 G, which is well above the
magnetic Feshbach resonance centered at B
0
202.1 G
[1], and the s-wave scattering length between the two
fermionic spin states is o 118o
0
(o
0
is the Bohr radius).
At the end of the evaporation we reach temperatures be-
tween T,T
F
0.2 and 0.25 with 5 10
4
to 2 10
5
par-
ticles, respectively.
Prior to loading the atoms into the optical lattice we tune
the magnetic eld to B 210 0.1 G, such that the
s-wave scattering length between the two states vanishes.
Then the standing wave laser eld along the vertical z axis
is turned on. Subsequently, the optical dipole trap along the
y axis is turned off and a standing wave laser eld along the
same axis is turned on, followed by the same procedure
along the x axis. In order to keep the loading of the atoms
into the lattice as adiabatic as possible the intensities of the
lasers are slowly increased (decreased) using exponential
ramps with time constants of 10 ms (25 ms) and durations
of 20 ms (50 ms), respectively.
In its nal conguration the optical lattice is formed by
three orthogonal standing waves with mutually orthogonal
polarizations and 1,e
2
radii of 50 m (x axis) and 70 m
(y axis and z axis), which are derived from the same lasers
as for the optical dipole trap. The laser elds of the three
beams have a linewidth of the order of 10 kHz and their
frequencies are offset with respect to each other by be-
tween 15 and 150 MHz. The resulting optical potential
depth V
x,y,z
is proportional to the laser intensity and is
conveniently expressed in terms of the recoil energy E
r

h
2
k
2
,2m, with k 2r, and m being the atomic mass.
The lattice depth was calibrated by modulating the laser
intensity and studying the parametric heating. The calibra-
tion error is estimated to be <10%.
The potential created by the optical lattice results in a
simple cubic crystal structure and the Gaussian intensity
proles of the lattice beams give rise to an additional
conning potential which varies with the laser intensity.
As a result, the sharp edges characterizing the T 0
distribution function for the quasimomentum in the homo-
geneous case [19] are expected to be rounded off. A
quantitative picture can be obtained by considering a
tight-binding Hamiltonian to describe noninteracting fer-
mions in an optical lattice with an additional harmonic
connement [20]. At T 0 the inhomogeneous system
is characterized by the total atom number N and by the
characteristic length [ over which the potential shift due to
the harmonic connement equals the tunnel coupling ma-
trix element 1. One nds [
n


21,mw
2
n
p
, with the fre-
quencies of the external harmonic connement given by
w
n
(n x, y, z). The density distribution scaled by [
n
and
the momentum distribution of the atoms in the lattice
depend only on the characteristic density
c

NJ
3
[
x
[
y
[
z
,
where J is the lattice spacing [7]. For a three-dimensional
lattice with 20 20 20 sites we have numerically calcu-
lated the characteristic density for the onset of a band
insulator to be
c
60. For this value of
c
the occupation
number at the center of the trap is larger than 0.99. It has
been pointed out that a fermionic band insulator in an
optical lattice with conning potential constitutes a high
delity quantum register [21].
In the experiment we probe the population within the
rst Brillouin zones by ramping down the optical lattice
slowly enough for the atoms to stay adiabatically in the
lowest band while quasimomentum is approximately con-
served [22]. We lower the lattice potential to zero over a
time scale of 1 ms. After 1 ms we switch off the homoge-
neous magnetic eld and allow for a total of 9 ms of
ballistic expansion before we take an absorption image of
the expanded atom cloud. The momentum distribution
obtained from these time of ight images, shown in
Fig. 1, reproduces the quasimomentum distributions of
the atoms inside the lattice. With increasing characteristic
density the initially circular shape of the Fermi surface
develops extensions pointing towards the Bragg planes and
nally transforms into a square shape completely lling the
rst Brillouin zone deeply in the band insulator. We have
observed population of higher bands if more atoms are
lled into the lattice initially. In Fig. 2 the experimental
data for momentum distributions along the line with qua-
simomentum q
y
0 are compared to the results of nu-
merical simulations using the same characteristic densities.
When imaging the cloud along the x direction we nd a
homogeneous lling of the band in the vertical (z) direc-
tion, probably due to the change in the harmonic conne-
ment while loading the lattice combined with the presence
of gravity. This asymmetry between the horizontal x, y, and
the vertical z directions vanishes when the gas approaches
the band insulating regime. We have examined the level of
adiabaticity of our loading scheme into the optical lattice
by transferring the atoms from the band insulator back into
the crossed beam dipole trap. There we nd a temperature
of 0.35T
F
when the initial temperature prior to loading into
the lattice was 0.2T
F
.
We have studied the dynamic response of the noninter-
acting Fermi gas to a change in the characteristic density
a c b d
OD
0
0.04
e
x
y
2hk
FIG. 1. Observing the Fermi surface. Time of ight images
obtained after adiabatically ramping down the optical lattice.
The characteristic density increases from left to right. (a) Image
of 3500 atoms per spin state and a potential depth of the optical
lattice of 5E
r
. Images (b)(e) were obtained with 15 000 atoms
per spin state. The potential depths of the optical lattices were
(b) 5E
r
, (c) 6E
r
, (d) 8E
r
, and (e) 12E
r
. The images show the
optical density (OD) integrated along the vertically oriented z
axis after 9 ms of ballistic expansion.
PRL 94, 080403 (2005)
P HYS I CAL RE VI E W L E T T E RS
week ending
4 MARCH 2005
080403-2
from a value deep in the band insulating regime to a value
below. In the latter regime the fermions are delocalized
over several sites of the optical lattice and an interference
pattern is observed when the atoms are abruptly released
from the lattice. The width of the interference peaks is a
measure of the length scale over which the atoms are
delocalized in the lattice or, equivalently, their coherence
length. We change the characteristic density in situ by
varying the strength of the lattice laser beams. Starting
from an initial characteristic density of
c
16 in an
optical lattice with a potential depth of 5E
r
we create a
band insulator with a characteristic density of
c
2700
at a potential depth of 15E
r
. After holding the atoms for
5 ms we reduce the potential depth back to 5E
r
, using an
exponential ramp with duration and time constant i
r
. This
is followed by a rapid switch-off of the lattice [see
Fig. 3(a)]. We measure the width of the central momentum
peak in the time of ight images for different durations i
r
and obtain the time scales r
x
2.7 0.4 ms and r
y

3.8 0.3 ms in the x and the y directions, respectively.
This corresponds to approximately 10 times the time scale
for tunneling given by h,2z1 at a potential depth of 5E
r
,
where z is the coordination number of the lattice. This
nontrivial dynamics appears to be signicantly slower than
the time scale measured for the transition of a Mott-
insulating state to a superuid state using bosonic atoms
in an optical lattice [13]. The comparatively slowdynamics
of delocalization of the fermions when approaching the
normal state is most likely due to Pauli blocking which
prevents tunneling of atoms in regions where the lowest
band is full and the atoms are well localized.
We investigate the interacting regime in the lattice start-
ing from a noninteracting gas deep in a band insulator with
V
x
12E
r
and V
y
V
z
18E
r
, and corresponding trap-
ping frequencies of w
x
2r 50 kHz and w
y
w
z

2r 62 kHz in the individual minima. A short radio-
frequency pulse is applied to transfer all atoms from the
jE 9,2, m
E
7,2i state into the jE 9,2, m
E

5,2i state, with the atoms in the jE 9,2, m
E

9,2i state remaining unaffected. We ramp the magnetic
eld with an inverse sweep rate of 12 s,G to different
nal values around the Feshbach resonance [see Fig. 4(a)]
located at B 224 G [23]. The sweep across the Feshbach
resonance goes from the side of repulsive interactions
towards the side of attractive interactions. When using
this direction of the sweep there is no adiabatic conversion
to molecules. After turning off the optical lattice adiabati-
cally and switching off the magnetic eld, we measure the
momentum distribution. To see the effect of the interac-
tions we determine the fraction of atoms transferred into
higher bands. For nal magnetic eld values well above the
Feshbach resonance we observe a signicant increase in
the number of atoms in higher bands along the weak axis of
the lattice, demonstrating an interaction-induced coupling
between the lowest bands. Since the s-wave interaction is
changed on a time scale short compared to the tunneling
0 5 10 15 20
30
40
50
60
R
M
S

w
i
d
t
h

[

m
]
Rampdown time t
r
[ms]
0 5 10
5
10
15
tr
Time [ms]
L
a
t
t
i
c
e

d
e
p
t
h

[
E
r
]
a
c
0
x-direction
y-direction
b
FIG. 3 (color online). Restoring phase coherence. (a) Control
sequence for the depth of the optical lattice. (b) Pseudocolor
image of the momentum distribution after releasing the atoms
from the initial optical lattice of 5E
r
and 6 ms ballistic expan-
sion. It reveals the central momentum peak and the matter wave
interference peaks at 2 hk. The data are averaged over ve
repetitive measurements. (c) Width of the central momentum
peak obtained from Gaussian ts to the atomic density distribu-
tion. The initial width is determined by the momentum spread of
an atom localized in the vibrational ground state of a lattice well.
The 10% difference in this size comes from slightly different
magnications of the imaging system in the two orthogonal
directions. The difference in the asymptotic values of the width
can most likely be attributed to the loading sequence of the
lattice and to the asymmetry of the conning potentials due to
the different beam waists. The error bars show the statistical
error of four repetitive measurements.
O
p
t
ic
a
l
d
e
n
s
it
y

[
a
r
b
.

u
n
it
s
]
-2 -1 0 1 2
40
50
60
70
-2 -1 0 1 2
40
80
120
160
Quasimomentum [hk]
-
a b
-2 -1 0 1 2
40
50
60
c
FIG. 2 (color online). Analysis of the density distributions.
The dots are cuts through the measured density distribution for
quasimomentum q
y
0 after adiabatically ramping down the
optical lattice. (a) Normal state with
c
14.5, (b) band insu-
lator with
c
137, and (c) band insulator with
c
2500. We
have numerically calculated the momentum distribution function
of fermions in the lowest band of a three-dimensional lattice with
20 20 20 sites and characteristic lengths [
x
,J 3.2,
[
y
,J 2.6, [
z
,J 2.5 [(a),(b)] and [
x
,J 1, [
y
,J 0.8,
[
z
,J 0.8 (c), assuming zero temperature (solid lines).
Experimental data of (c) are averaged over ve images.
Imperfect adiabaticity during the switch-off of the optical lattice
may cause the rounding off of the experimental data at the edge
of the Brillouin zone in (b) and (c). The calculated momentum
distribution function is scaled to match the experimental data
using identical scale factors for all graphs.
PRL 94, 080403 (2005)
P HYS I CAL RE VI E W L E T T E RS
week ending
4 MARCH 2005
080403-3
time between adjacent potential minima we may regard the
band insulator as an array of harmonic potential wells. It
has been shown that increasing the s-wave scattering
length for two particles in a harmonic oscillator shifts the
energy of the two-particle state upwards until the next
oscillator level is reached [24]. In our case this leads to a
population of higher energy bands. The fraction of atoms
transferred could be limited by the number of doubly
occupied lattice sites and tunneling in the higher bands.
The number of doubly occupied sites could be measured
by studying the formation of molecules in the lattice. In
addition, we observe a shift of the position of the Feshbach
resonance from its value in free space to larger values of
the magnetic eld [see Fig. 4(a)], which has been predicted
for tightly conned atoms in an optical lattice [25]. This
mechanism for a connement induced resonance is related
to the phenomenon predicted for one-dimensional quan-
tum gases [26] which has as yet escaped experimental
observation. For a quantitative description of this strongly
interacting Fermi gas on a lattice a multiband Hubbard
model could be considered, but these are even in the static
case notoriously difcult or even impossible to solve with
the present methods [27].
In conclusion, we have created a fermionic many-
particle quantum system on a lattice. We have demon-
strated the dynamic control over the parameters of the
system such as lling and interactions which is not feasible
in solid-state systems. For the noninteracting static regime
we nd good agreement between our measurements and a
theoretical model. Both the dynamic measurements and the
strongly interacting case pose challenges for the present
theoretical understanding of many-particle fermionic sys-
tems on optical lattices.
We thank G. Blatter, C. Bruder, H. P. Buchler, S. Jonsell,
A. Muramatsu, M. Rigol, C. Schori, P. Torma, and M.
Troyer for insightful discussions, and SNF, SEP
Information Sciences, and QSIT for funding.
*Electronic address:[email protected]
[1] C. A. Regal, M. Greiner, and D. S. Jin, Phys. Rev. Lett. 92,
040403 (2004).
[2] M. Bartenstein et al., Phys. Rev. Lett. 92, 120401
(2004).
[3] M. W. Zwierlein et al., Phys. Rev. Lett. 92, 120403 (2004).
[4] J. Kinast et al., Phys. Rev. Lett. 92, 150402 (2004).
[5] T. Bourdel et al., Phys. Rev. Lett. 93, 050401 (2004).
[6] W. Hofstetter, J. I. Cirac, P. Zoller, E. Demler, and M. D.
Lukin, Phys. Rev. Lett. 89, 220407 (2002).
[7] M. Rigol, A. Muramatsu, G. G. Batrouni, and R. T.
Scalettar, Phys. Rev. Lett. 91, 130403 (2003).
[8] C. Lee, Phys. Rev. Lett. 93, 120406 (2004).
[9] L. Santos et al., Phys. Rev. Lett. 93, 030601 (2004).
[10] D. Jaksch, C. Bruder, J. I. Cirac, C. W. Gardiner, and P.
Zoller, Phys. Rev. Lett. 81, 3108 (1998).
[11] J. Hubbard, Proc. R. Soc. London A 276, 238 (1963).
[12] M. T. DePue, C. McCormick, S. L. Winoto, S. Oliver, and
D. S. Weiss, Phys. Rev. Lett. 82, 2262 (1999).
[13] M. Greiner, O. Mandel, T. Esslinger, T. W. Hansch, and I.
Bloch, Nature (London) 415, 39 (2002).
[14] T. Stoferle, H. Moritz, C. Schori, M. Kohl, and T.
Esslinger, Phys. Rev. Lett. 92, 130403 (2004).
[15] G. Modugno, F. Ferlaino, R. Heidemann, G. Roati, and M.
Inguscio, Phys. Rev. A 68, 011601(R) (2003).
[16] L. Pezze et al., Phys. Rev. Lett. 93, 120401 (2004).
[17] H. Moritz, T. Stoferle, M. Kohl, and T. Esslinger, Phys.
Rev. Lett. 91, 250402 (2003).
[18] G. Modugno et al., Science 297, 2240 (2002).
[19] N. W. Ashcroft and N. D. Mermin, Solid States Physics
(Saunders College Publishing, Philadelphia, PA, 1976).
[20] M. Rigol and A. Muramatsu, Phys. Rev. A 69, 053612
(2004).
[21] L. Viverit, C. Menotti, T. Calarco, and A. Smerzi, Phys.
Rev. Lett. 93, 110401 (2004).
[22] M. Greiner, I. Bloch, O. Mandel, T. W. Hansch, and T.
Esslinger, Phys. Rev. Lett. 87, 160405 (2001).
[23] C. A. Regal and D. S. Jin, Phys. Rev. Lett. 90, 230404
(2003).
[24] T. Busch, B.-G. Englert, K. Rzazewski, and M. Wilkens,
Found. Phys. 28, 549 (1998).
[25] P. O. Fedichev, M. J. Bijlsma, and P. Zoller, Phys. Rev.
Lett. 92, 080401 (2004).
[26] M. Olshanii, Phys. Rev. Lett. 81, 938 (1998).
[27] M. Troyer and U. Wiese, cond-mat/0408370.
F
r
a
c
t
i
o
n

o
f

a
t
o
m
s

i
n

h
i
g
h
e
r

b
a
n
d
s
Final magnetic field [G]
0 40 80
0.0
0.1
0.2
0.3
Inverse sweep rate [s/G]
b
B
final
=233.8 G
220 230 240
0.0
0.1
0.2
0.3
a free space
Feshbach
resonance
c
a
3D
=0
FIG. 4 (color online). Interaction-induced transition between
Bloch bands. (a) Transferring fermions into higher bands using a
sweep across the Feshbach resonance (lled symbols). The
inverse magnetic eld sweep rate is 12 s,G. The line shows
a sigmoidal t to the data. The open symbols show a repetition of
the experiment with the atoms prepared in the spin states jE
9,2, m
E
9,2i and jE 9,2, m
E
7,2i where the scat-
tering length is not sensitive to the magnetic eld. The magnetic
eld is calibrated by rf spectroscopy between Zeeman levels.
Because of the rapid ramp, the eld lags behind its asymptotic
value and the horizontal error bars represent this deviation.
(b) Fraction of atoms in higher bands for a nal magnetic eld
of 233 G for different magnetic eld sweep rates. The vertical
error bars show the statistical error of four repetitive measure-
ments. (c) Momentum distribution for a nal magnetic eld of
B 233 G and a 12 s,G sweep rate. Arrows indicate the
atoms in the higher bands.
PRL 94, 080403 (2005)
P HYS I CAL RE VI E W L E T T E RS
week ending
4 MARCH 2005
080403-4