P HYSICAL R EVIEW

LETTERS

VOLUME 75 27 NOVEMBER 1995 NUMBER 22

Bose-Einstein Condensation in a Gas of Sodium Atoms

K. B. Davis, M.-O. Mewes, M. R. Andrews, N. J. van Druten, D. S. Durfee, D. M. Kurn, and W. Ketterle
Department of Physics and Research Laboratory of Electronics, Massachusetts Institute of Technology,
Cambridge, Massachusetts 02139
(Received 17 October 1995)
We have observed Bose-Einstein condensation of sodium atoms. The atoms were trapped in a novel
trap that employed both magnetic and optical forces. Evaporative cooling increased the phase-space
density by 6 orders of magnitude within seven seconds. Condensates contained up to 5 3 105 atoms at
densities exceeding 1014 cm23 . The striking signature of Bose condensation was the sudden appearance
of a bimodal velocity distribution below the critical temperature of , 2 mK. The distribution consisted
of an isotropic thermal distribution and an elliptical core attributed to the expansion of a dense
condensate.

PACS numbers: 03.75.Fi, 05.30.Jp, 32.80.Pj, 64.60.–i

Bose-Einstein condensation (BEC) is a ubiquitous phe- rubidium [11], lithium [12], and, as reported in this paper,
nomenon which plays significant roles in condensed mat- in sodium. Our results are distinguished by a production
ter, atomic, nuclear, and elementary particle physics, as rate of Bose-condensed atoms which is 3 orders of
well as in astrophysics [1]. Its most striking feature is a magnitude larger than in the two previous experiments.
macroscopic population of the ground state of the system Furthermore, we report a novel atom trap that offers a
at finite temperature [2]. The study of BEC in weakly superior combination of tight confinement and capture
interacting systems holds the promise of revealing new volume and the attainment of unprecedented densities of
macroscopic quantum phenomena that can be understood cold atomic gases.
from first principles, and may also advance our under- Evaporative cooling requires an atom trap which is
standing of superconductivity and superfluidity in more tightly confining and stable. So far, magnetic traps and
complex systems. optical dipole traps have been used. Optical dipole traps
During the past decade, work towards BEC in weakly provide tight confinement, but have only a very small
interacting systems has been carried forward with excitons trapping volume (1028 cm3 ). The tightest confinement in
in semiconductors and cold trapped atoms. BEC has a magnetic trap is achieved with a spherical quadrupole
been observed in excitonic systems, but a complete potential (linear confinement); however, atoms are lost
theoretical treatment is lacking [1,3]. The pioneering from this trap due to nonadiabatic spin flips as the atoms
work towards BEC in atomic gases was performed with pass near the center, where the field rapidly changes
spin-polarized atomic hydrogen [4,5]. Following the direction. This region constitutes a “hole” in the trap of
development of evaporative cooling [6], the transition was micrometer dimension. The recently demonstrated “TOP”
approached within a factor of 3 in temperature [7]. Laser trap suppresses this trap loss, but at the cost of lower
cooling provides an alternative approach towards very low confinement [8].
temperatures, but has so far been limited to phase-space We suppressed the trap loss by adding a repulsive
densities typically 105 times lower than required for BEC. potential around the zero of the magnetic field, literally
The combination of laser cooling with evaporative cooling “plugging” the hole. This was accomplished by tightly
[8–10] was the prerequisite for obtaining BEC in alkali focusing an intense blue-detuned laser that generated a
atoms. This year, within a few months, three independent repulsive optical dipole force. The optical plug was
and different approaches succeeded in creating BEC in created by an Ar1 -laser beam (514 nm) of 3.5 W focused

0031-9007y95y75(22)y3969(5)$06.00 © 1995 The American Physical Society 3969

VOLUME 75, NUMBER 22 PHYSICAL REVIEW LETTERS 27 NOVEMBER 1995

to a beam waist of 30 mm. This caused 7 MHz (350 mK) were first pumped to the F ­ 2 state by switching on a
of light shift potential at the origin. Heating due to photon 10 mWycm2 laser beam in resonance with the F ­ 1 !
scattering was suppressed by using far-off-resonant light, F ­ 2 transition. 10 ms later the atoms were concur-
and by the fact that the atoms are repelled from the region rently exposed to a 100 ms, 0.25 mWycm2 probe laser
where the laser intensity is highest. pulse in resonance with the F ­ 2 ! F ­ 3 transition,
The experimental setup was similar to that described propagating along the trap’s y direction. This probe laser
in our previous work [9]. Typically, within 2 s 109 beam was imaged onto a charge-coupled device sensor
atoms in the F ­ 1, mF ­ 21 state were loaded into a with a lens system having a resolution of 8 mm. Up to
magnetic trap with a field gradient of 130 Gycm; the peak 100 photons per atom were absorbed without blurring the
density was , 1011 cm23 , the temperature , 200 mK, image due to heating.
and the phase-space density 106 times lower than required At temperatures above 15 mK the observed trapped
for BEC. The lifetime of the trapped atoms was , clouds were elliptical with an aspect ratio of 2:1 due to the
30 s, probably limited by background gas scattering at a symmetry of the quadrupole field. At the position of the
pressure of , 1 3 10211 mbar. optical plug they had a hole, which was used for fine align-
The magnetically trapped atoms were further cooled ment. A misalignment of the optical plug by , 20 mm
by rf-induced evaporation [8,9,13]. rf-induced spin flips resulted in increased trap loss and prevented us from cool-
were used to selectively remove the higher-energy atoms ing below 50 mK. This is evidence that the Majorana spin
from the trap resulting in a decrease in temperature for flips are localized in a very small region around the center
the remaining atoms. The total (dressed-atom) potential of the trap. At temperatures below 15 mK, the cloud sepa-
is a combination of the magnetic quadrupole trapping rated into two pockets at the two minima in the potential of
potential, the repulsive potential of the plug, and the Fig. 1. The bottom of the potential can be approximated
effective energy shifts due to the rf. At the point where as a three-dimensional anisotropic harmonic oscillator po-
atoms are in resonance with the rf, the trapped state tential with frequencies vy2 ­ mB0 ys2mx0 d, vz2 ­ 3vy2 ,
undergoes an avoided crossing with the untrapped states vx2 ­ vy2 fs4x02 yw02 d 2 1g, where m is the atom’s magnetic
(corresponding to a spin flip), and the trapping potential moment, m the mass, B0 the axial field gradient, w0 the
bends over. As a result, the height of the potential barrier Gaussian beam waist parameter (1ye2 radius) of the opti-
varies linearly with the rf frequency. The total potential is cal plug, and x0 the distance of the potential minimum from
depicted in Fig. 1. Over 7 s, the rf frequency was swept the trap center. x0 was directly measured to be 50 mm by
from 30 MHz to the final value around 1 MHz, while the imaging the trapped cloud, w0 (30 mm) was determined
field gradient was first increased to 550 Gycm (to enhance from x0 , the laser power (3.5 W), and B0 (180 Gycm).
the initial elastic-collision rate) and then lowered to 180 With these values the oscillation frequencies are 235, 410,
Gycm (to avoid the losses due to inelastic processes at the and 745 Hz in the y, z, and x directions, respectively.
final high densities). When the final rf frequency nrf was lowered below
Temperature and total number of atoms were deter- 0.7 MHz, a distinctive change in the symmetry of the
mined using absorption imaging. The atom cloud was velocity distribution was observed [Figs. 2(a) and 2(b)].
imaged either while it was trapped or following a sud- Above this frequency the distribution was perfectly spher-
den switch-off of the trap and a delay time of 6 ms. ical as expected for a thermal uncondensed cloud [14]. Be-
Such time-of-flight images displayed the velocity dis- low the critical frequency, the velocity distribution con-
tribution of the trapped cloud. For probing, the atoms tained an elliptical core which increased in intensity when

FIG. 2 (color). Two-dimensional probe absorption images,

after 6 ms time of flight, showing evidence for BEC. (a) is
the velocity distribution of a cloud cooled to just above the
transition point, (b) just after the condensate appeared, and
(c) after further evaporative cooling has left an almost pure
FIG. 1. Adiabatic potential due to the magnetic quadrupole condensate. (b) shows the difference between the isotropic
field, the optical plug, and the rf. This cut of the three- thermal distribution and an elliptical core attributed to the
dimensional potential is orthogonal to the propagation direction expansion of a dense condensate. The width of the images
(y) of the blue-detuned laser. The symmetry axis of the is 870 mm. Gravitational acceleration during the probe delay
quadrupole field is the z axis. displaces the cloud by only 0.2 mm along the z axis.
3970
VOLUME 75, NUMBER 22 PHYSICAL REVIEW LETTERS 27 NOVEMBER 1995

the rf was further swept down, whereas the spherical cloud cubic thermal de Broglie wavelength exceeds 2.612 at the
became less intense. We interpret the elliptical cloud as bottom of the potential [2]. For a harmonic oscillator po-
due to the Bose condensate, and the spherical cloud as due tential this is equivalent to Nc ­ 1.202skB T d3 yh̄ 3 vx vy vz
to the normal fraction. [16]. For our trap and 2.0 mK, Nc ­ 2 3 s1.2 3 106 d,
In the region just below the transition frequency one where the factor of 2 accounts for the two separated clouds.
expects a bimodal velocity distribution: a broad distribu- The predicted value for Nc depends on the sixth power of
tion due to the normal gas and a narrow distribution due the width of a time-of-flight image and is only accurate
to the condensate. The cross sections of the time-of-flight to within a factor of 3. We determined the number of
images (Fig. 3) indeed show such bimodal distributions atoms by integrating over the absorption image. At the
in this region. Figure 4 shows how suddenly the time- transition point, the measured number of 7 3 105 agrees
of-flight image changes below nrf ­ 0.7 MHz. The ef- with the prediction for Nc . The critical peak density nc
fective area of the observed cloud becomes very small at 2.0 mK is 1.5 3 1014 cm23 . Such a high density ap-
[Fig. 4(a)], while the velocity distribution is no longer pears to be out of reach for laser cooling, and demonstrates
Gaussian [Fig. 4(b)] and requires different widths for the that evaporative cooling is a powerful technique to obtain
condensate and the normal fraction [Fig. 4(c)]. not only ultralow temperatures, but also extremely high
At the critical frequency, a temperature of (2.0 6 0.5) densities.
mK was derived from the time-of-flight image. An inde- An ideal Bose condensate shows a macroscopic popu-
pendent, though less accurate estimate of the temperature T lation of the ground state of the trapping potential. This
is obtained from the dynamics of evaporative cooling. Ef- picture is modified for a weakly interacting Bose gas. The
ficient evaporation leads to a temperature which is about 10 mean-field interaction energy is given by nŨ, where n is
times smaller than the depth of the trapping potential [15]. the density and Ũ is proportional to the scattering length
Since the speed of evaporation depends exponentially on a: Ũ ­ 4p h̄2 aym [2]. Using our recent experimental
the ratio of potential depth to temperature, we expect this result a ­ 4.9 nm [9], ŨykB ­ 1.3 3 10221 K cm3 .
estimate of T ­ 2 mK to be accurate to within a factor At the transition point, nc ŨykB Tc ­ 0.10. Conse-
of 2. quently, above the transition point, the kinetic energy
The critical number of atoms Nc to achieve BEC is de- dominates over the interaction energy, and the velocity
termined by the condition that the number of atoms per

FIG. 4. Further evidence for a phase transition is provided

by the sudden change of observed quantities as the final rf
frequency nrf is varied. (a) Area of the cloud in the time-of-
flight image versus nrf . The area was obtained as the ratio
of the integrated optical density and the peak optical density.
FIG. 3. Optical density as a function of position along the z The area changes suddenly at nrf ­ 0.7 MHz. Below the
axis for progressively lower values of the final rf frequency. same frequency, the velocity distributions (Fig. 3) cannot be
These are vertical cuts through time-of-flight images like those represented by a single Gaussian, as demonstrated by the x 2
in Fig. 2. For nrf , 0.7 MHz, they show the bimodal velocity for a single Gaussian fit (b), and required different widths for
distributions characteristic of the coexistence of a condensed the condensate (full circles) and noncondensate fraction (c). In
and uncondensed fraction. The top four plots have been offset (a) and (c) the lines reflect the behavior of a classical gas with
vertically for clarity. a temperature proportional to the trap depth.
3971
VOLUME 75, NUMBER 22 PHYSICAL REVIEW LETTERS 27 NOVEMBER 1995

distribution after sudden switch-off of the trap is isotropic. titatively for the observed distributions because we have
For the condensate, however, the situation is reversed. two separated condensates which overlap in the time-of-
As we will confirm below, the kinetic energy of the flight images, and also because of some residual horizon-
condensate is negligible compared to its interaction tal acceleration due to the switch-off of the trap, which is
energy [17]. Furthermore, well below the transition negligible for the thermal cloud, but not for the conden-
point, the interaction with the noncondensate fraction sate [20].
can be neglected. In such a situation, the solution The lifetime of the condensate was about 1 s. This
of the nonlinear Schrödinger equation reveals that the lifetime is probably determined either by three-body
condensate density n0 srd is a mirror image of the trapping recombination [21] or by the heating rate of 300 nKys,
potential V srd: n0 srd ­ n0 s0d 2 V srdyŨ, as long as this which was observed for a thermal cloud just above Tc .
expression is positive, otherwise n0 srd vanishes (see, This heating rate is much higher than the estimated
e.g., Refs. [5,18]). For a harmonic potential, one obtains 8 nKys for the off-resonant scattering of green light and
the peak density n0 s0d for N0 atoms in the condensate may be due to residual beam jitter of the optical plug.
n0 s0d ­ 0.118sN0 m3 vx vy vz yh̄ 3 a3y2 d2y5 . In conclusion, we were able to Bose-condense 5 3 105
Typically, we could cool one-fourth of the atoms at the sodium atoms within a total loading and cooling cycle
transition point into a pure condensate. For an observed of 9 s. During evaporative cooling, the elastic collision
N0 ­ 1.5 3 105 , we expect the condensate to be 2 times rate increased from 30 Hz to 2 kHz resulting in a mean
more dense than the thermal cloud at the transition point, free path comparable to the dimensions of the sample.
and about 6 times larger than the ground-state wave Such collisionally dense samples are the prerequisite for
function. The kinetic energy within the condensate is studying various transport processes in dense ultracold
, h̄2 ys2mR 2 d, where R is the size of the condensate matter. Furthermore, we have reached densities in excess
[18], while the internal energy is 2n0 Ũy7. Thus the of 1014 cm23 , which opens up new possibilities for
kinetic energy of the condensed atoms is around 1 nK, studying decay processes like dipolar relaxation and three-
much less than the zero-point energy of our trap (35 body recombination, and for studying a weakly interacting
nK) and the calculated internal energy of 120 nK. This Bose gas over a broad range of densities and therefore
estimate is consistent with our initial assumption that the strengths of interaction.
kinetic energy can be neglected compared to the interaction We are grateful to E. Huang and C. Sestok for im-
energy. portant experimental contributions, to M. Raizen for the
The internal energy is , 25 times smaller than the loan of a beam pointing stabilizer, and to D. Kleppner for
thermal energy s3y2dkB Tc at Tc . Consequently, the width helpful discussions. We are particularly grateful to D. E.
of the time-of-flight image of the condensate is expected Pritchard, who not only contributed many seminal ideas
to be about 5 times smaller than at the transition point. to the field of cold atoms, but provided major inspira-
This is close to the observed reduction in the width tion and equipment to W. K. This work was supported by
shown in Fig. 4(c). This agreement might be fortuitous ONR, NSF, JSEP, and the Sloan Foundation. M.-O. M.,
because we have so far neglected the anisotropy of the K. B. D., and D. M. K. would like to acknowledge support
expansion, but it indicates that we have observed the from Studienstiftung des Deutschen Volkes, MIT Physics
correct magnitude of changes which are predicted to occur Department Lester Wolfe fellowship, and NSF Gradu-
at the BEC transition of a weakly interacting gas. In ate Research Fellowship, respectively, and N. J. v. D. from
several cooling cycles, as many as 5 3 105 condensed “Nederlandse Organisatie voor Wetenschappelijk Onder-
atoms were observed; we estimate the number density in zoek (NWO)” and NACEE (Fulbright fellowship).
these condensates to be 4 3 1014 cm23 .
A striking feature of the condensate is the nonisotropic
velocity distribution [11,19]. This is caused by the “ex-
plosion” of the condensate due to repulsive forces which [1] A. Griffin, D. W. Snoke, and S. Stringari, Bose-Einstein
are proportional to the density gradient. The initial accel- Condensation (Cambridge University Press, Cambridge,
eration is therefore inversely proportional to the width of 1995).
the condensate resulting in an aspect ratio of the velocity [2] K. Huang, Statistical Mechanics (Wiley, New York,
distribution, which is inverted compared to the spatial dis- 1987), 2nd ed.
tribution. When we misaligned the optical plug vertically, [3] J. L. Lin and J. P. Wolfe, Phys. Rev. Lett. 71, 1222 (1993).
[4] I. F. Silvera and M. Reynolds, J. Low Temp. Phys. 87,
the shape of the cloud changed from two vertical crescents
343 (1992); J. T. M. Walraven and T. W. Hijmans, Physica
to a single elongated horizontal crescent. The aspect ratio (Amsterdam) 197B, 417 (1994).
of the time-of-flight image of the condensate correspond- [5] T. Greytak, in Ref. [1], p. 131.
ingly changed from horizontal to vertical elongation. In [6] N. Masuhara et al., Phys. Rev. Lett. 61, 935 (1988).
contrast, just above the transition point, the velocity dis- [7] J. Doyle et al., Phys. Rev. Lett. 67, 603 (1991).
tribution was found to be spherical and insensitive to the [8] W. Petrich, M. H. Anderson, J. R. Ensher, and E. A.
alignment of the plug. However, we cannot account quan- Cornell, Phys. Rev. Lett. 74, 3352 (1995).

3972
VOLUME 75, NUMBER 22 PHYSICAL REVIEW LETTERS 27 NOVEMBER 1995

[9] K. B. Davis et al., Phys. Rev. Lett. 74, 5202 (1995). [15] K. B. Davis, M.-O. Mewes, and W. Ketterle, Appl. Phys.
[10] C. S. Adams et al., Phys. Rev. Lett. 74, 3577 (1995). B 60, 155 (1995).
[11] M. H. Anderson et al., Science 269, 198 (1995). [16] V. Bagnato, D. E. Pritchard, and D. Kleppner, Phys. Rev.
[12] C. C. Bradley, C. A. Sackett, J. J. Tollett, and R. G. Hulet, A 35, 4354 (1987). This formula is derived assuming
Phys. Rev. Lett. 75, 1687 (1995). kB Tc ¿ h̄vx,y,z , which is the case in our experiment.
[13] D. E. Pritchard, K. Helmerson, and A. G. Martin, in [17] Note that already for about 200 atoms in the ground state,
Atomic Physics 11, edited by S. Haroche, J. C. Gay, and the interaction energy in the center of the condensate
G. Grynberg (World Scientific, Singapore, 1989), p. 179. equals the zero-point energy.
[14] The measured 1ye decay time for the magnet current is [18] G. Baym and C. Pethick (to be published).
100 ms, shorter than the v 21 of the fastest oscillation [19] M. Holland and J. Cooper (to be published).
in the trap (210 ms). We therefore regard the switch- [20] These effects do not affect the vertical velocity distribu-
off as sudden. Any adiabatic cooling of the cloud during tions shown in Fig. 3.
the switch-offtibs would result in a nonspherical velocity [21] A. J. Moerdijk, H. M. J. M. Boesten, and B. J. Verhaar,
distribution due to the anisotropy of the potential. Phys. Rev. A (to be published).

3973