Combined quantum state preparation and laser cooling of a continuous beam of cold atoms
Gianni Di Domenico, Laurent Devenoges, Claire Dumas, Pierre Thomann
CCombined quantum state preparation and laser cooling of a continuous beam of coldatoms
Gianni Di Domenico, ∗ Laurent Devenoges, Claire Dumas, and Pierre Thomann
Laboratoire Temps-Fr´equence, Universit´e de Neuchˆatel,Avenue de Bellevaux 51, CH-2009 Neuchˆatel, Switzerland (Dated: 7 November 2010)We use two-laser optical pumping on a continuous atomic fountain in order to prepare cold cesiumatoms in the same quantum ground state. A first laser excites the F = 4 ground state to pump theatoms toward F = 3 while a second π -polarized laser excites the F = 3 → F (cid:48) = 3 transition of theD2 line to produce Zeeman pumping toward m = 0. To avoid trap states, we implement the firstlaser in a 2D optical lattice geometry, thereby creating polarization gradients. This configurationhas the advantage of simultaneously producing Sisyphus cooling when the optical lattice laser istuned between the F = 4 → F (cid:48) = 4 and F = 4 → F (cid:48) = 5 transitions of the D2 line, which isimportant to remove the heat produced by optical pumping. Detuning the frequency of the second π -polarized laser reveals the action of a new mechanism improving both laser cooling and statepreparation efficiency. A physical interpretation of this mechanism is discussed. PACS numbers: 37.10.De,37.10.Jk,37.10.Vz,32.80.Xx,32.60.+i
I. INTRODUCTION
Quantum state preparation, i.e. preparation of an en-semble of quantum systems in a given quantum state, isuseful in many physics experiments ranging from quan-tum information science to quantum metrology, not tomention fundamental physics experiments. For examplein cesium atomic clocks, one needs to create a popula-tion inversion on the clock transition ( F = 3 , m = 0 → F = 4 , m = 0) before being able to probe the transi-tion probability with Ramsey microwave spectroscopy.Indeed, when the microwave is tuned to the clock tran-sition, the population of the F = 4 state increases by∆ P = P , − P , where P F,m denotes the relative pop-ulation of state | F, m (cid:105) . Therefore, any increase of P , will result in a corresponding increase of the clock reso-nance signal and as a consequence of the clock stability,the maximum being reached when P , = 1 which corre-sponds to a pure quantum state preparation. The samediscussion applies to cold atom interferometers since theirprinciple of operation is very similar to atomic clocks.The only difference being that the atom’s motion is en-tangled with their internal quantum state in such a waythat one can probe their motion by measuring the abovementioned clock transition probability. In this case,quantum state preparation will result in increased signal-to-noise ratio and thus increased sensor sensitivity. Otherexamples where quantum state preparation plays a cru-cial role include masers, lasers, and all quantum informa-tion science experiments.Many different approaches have been used to prepareatomic samples in a given quantum state. The first ther-mal beam cesium clocks used Stern-Gerlach magnets toselect atoms in one of the hyperfine ground states [1]. ∗ Electronic address: [email protected]
Later, this selection process was advantageously replacedby laser optical pumping to transfer all the atoms fromone hyperfine ground state to the other [2–6]. Then inthe 1980s, two-laser optical pumping was proposed toproduce both hyperfine and Zeeman pumping toward oneof the clock state Zeeman sub-levels [7–9]. However noimprovement of the signal-to-noise ratio was observed be-cause of the increased noise introduced by the pumpingprocess [10, 11]. With the advent of cold atoms andatomic fountain clocks, two-laser optical pumping wasabandoned because the number of pumping cycles is high( >
10 on average with cesium) and thus produces an un-wanted heating of the atomic cloud. Importance was puton purity of state preparation, and selection methods in-volving selective excitation followed by optical blowingof unwanted atoms were introduced [12]. Nowadays, thesame methods are being used in cold atom interferome-ters [13–15]. Many other original state preparation meth-ods have been presented in the literature, for exampleoptical pumping followed by rf transfer in a vapour cell[16, 17], optical pumping via incoherent Raman transi-tions in cavity QED [18], preparation of pure superposi-tion states with push-pull pumping [19] or via ellipticallypolarized bichromatic fields [20].In our experiment we want to prepare the quantumstate of a continuous beam of cold atoms with a trans-verse temperature between 3 and 4 µ K. Therefore we can-not afford the reheating due to spontaneous emission and,as a consequence, methods which combine state prepa-ration and laser cooling are of particular interest for us.Moreover, we need a method which can be adapted tothe continuous beam case. The first striking exampleof such combined internal and external state prepara-tion was the observation of velocity selective coherentpopulation trapping [21] in 1988 with metastable heliumatoms. Other schemes involving alkali-metal atoms in-clude Zeeman-shift-degenerate Raman sideband coolingwhich prepares the atoms in the stretched state ( m = F ) a r X i v : . [ phy s i c s . a t o m - ph ] N ov [22–24]. In Ref. [25], we showed that it is possible toadapt this scheme to a continuous beam of cold atoms.However, the resulting quantum state | F = 3 , m = 3 (cid:105) isnot useful for an atomic clock and thus we would needto replace in this sideband cooling scheme the Zeemanshift by a Stark shift to accumulate all the atoms in | F = 3 , m = 0 (cid:105) as proposed in Ref. [26]. Even thoughthis scheme seems very promising for cold atom clock ap-plications, at present we are not aware of any experimen-tal realization and we suppose that this may be due tothe technical challenge of tailoring the required AC Starkshifts. As a first step toward this Stark-shift-degenerateRaman sideband cooling scheme, in this work we startby realizing optical pumping toward m = 0 in parallelwith Sisyphus cooling, similarly to the work described inRef.[27] which was realized with Doppler cooling in a 3Dmagneto-optical trap.This article is organized as follows. In section II we willstart by presenting the principle of our optical pumpingscheme and explain how one can combine both quantumstate preparation and laser cooling in the same interac-tion zone. In section III, we will describe the experimen-tal setup that we use to produce the continuous beamof cold atoms, to prepare the atoms in one of the clockstates, and then to characterize both state preparationand laser cooling efficiency. In section IV we will presentour experimental results and discuss their physical inter-pretation in section V, and finally we will conclude insection VI. II. QUANTUM STATE PREPARATIONPRINCIPLE
Our objective is to prepare a continuous beamof cold cesium atoms in the ground state sub-level | F = 3 , m = 0 (cid:105) by making use of two-laser optical pump-ing. A first laser is used to excite the F = 4 groundstate in order to pump the atoms into F = 3 while asecond π -polarized laser excites the F = 3 → F (cid:48) = 3transition. As a consequence, | F = 3 , m = 0 (cid:105) is the onlyground state sub-level which is not excited by laser lightand therefore all atoms will accumulate into that state [7–11]. Therefore, the effect of the second laser is to produceZeeman pumping toward m = 0. Since we are workingwith a beam of cold atoms, we have to care about heatingproduced by the optical pumping cycles. Indeed, start-ing with a uniform population distribution among the F = 3 , | F = 3 , m = 0 (cid:105) . Every pumping cy-cle transfers a random recoil to the atom and therebyincreases the temperature of the atomic sample. In ourcase, this increase of temperature will result in an in-creased divergence of the atomic beam.The main idea, which distinguishes this proposal fromprevious work, is to adjust the frequency, polarization,and geometry of the laser beams in such way as to pro- (cid:72) a (cid:76) optical latticeatomic beampump laser (cid:72) b (cid:76) FIG. 1: (a) Configuration of laser beams used for simultane-ous quantum state preparation and laser cooling. The atomicbeam is vertical and all laser beams are in the same horizontalplane. The optical lattice beam is folded and retro-reflectedon a prism to create a phase-stable 2D optical lattice (seeRef. [28]). It has an incoming polarization which is linear at45 ◦ with respect to the vertical direction, and subsequent mul-tiple reflexions on metallic mirrors introduce some ellipticity.The Zeeman pump beam is linearly polarized in the verticaldirection and retro-reflected with a mirror. It makes a smallangle of approximately 5 ◦ with the lattice incoming direction.In principle this angle doesn’t play any role, its value is sim-ply restricted by the optical access to the vacuum system.All mirrors are metallic. A vertical magnetic field is used tostabilize the atomic polarization. (b) Frequencies of the laserbeams with respect to the cesium D2 line transitions. In orderto produce optical pumping toward | F = 3 , m = 0 (cid:105) the opti-cal lattice beam should excite either the F = 4 → F (cid:48) = 4 orthe F = 4 → F (cid:48) = 3 transition, and the Zeeman pump beamshould excite the F = 3 → F (cid:48) = 3 transition. As explainedin the text, both lasers are detuned in order to produce lasercooling simultaneously with state preparation. duce Sisyphus cooling in order to compensate heatingproduced by the optical pumping cycles. The config-uration of laser beams is shown in Fig. 1(a) and theirfrequencies relative to the cesium D2 lines are definedin Fig. 1(b). The atomic beam is vertical and all laserbeams are in the same horizontal plane.The first laser beam, which produces hyperfine pump-ing from F = 4 to F = 3, is folded and retro-reflected ona prism to create a phase-stable 2D optical lattice (seeRef. [28]). Its incoming polarization is chosen linear at45 ◦ with respect to the vertical direction in such way as toproduce a strong polarization gradient in the intersectionregion. With this polarization gradient and a carefullychosen laser frequency, one can observe Sisyphus coolingin the transverse directions. Another benefit of a po-larization gradient is to avoid Zeeman coherent trappingstates thanks to the motion of the atoms in the differentpolarization sites. In order to produce optical pumpingtoward F = 3, the optical lattice beam should excite ei-ther the F = 4 → F (cid:48) = 4 or the F = 4 → F (cid:48) = 3transitions, and by detuning its frequency on the blueside of those transitions, one can simultaneously produceSisyphus cooling. (cid:72) a (cid:76) (cid:72) b (cid:76) FIG. 2: (a) Scheme of the experimental setup. One cansee the two-dimensional magneto-optical trap (2D-MOT)where the atomic beam is produced, the 3D moving mo-lasses (MM) which cools and launches the atoms at a speed of4 m/s, the transverse laser cooling (TLC) which collimates theatomic beam, the quantum state preparation stage (QSP), thedepumper (DEP), the microwave cavity (MWC), and finallythe fluorescence detection (DET). (b) Scheme of Zeeman sub-level showing how the population distribution of the F = 3ground state is measured in two steps: (1) selective microwaveexcitation followed by (2) detection of atoms in F = 4. SeeSection III for details. The second laser beam, which produces Zeeman pump-ing toward m = 0, is linearly polarized in the verti-cal direction and retro-reflected with a mirror. Thanksto its polarization and according to selection rules, thislaser will excite all sub-levels of F = 3 except m = 0.As a consequence, the atoms will tend to accumulate in | F = 3 , m = 0 (cid:105) . During this process, each time an atomis pumped back into F = 4, it couples again to the opti-cal lattice and thus restarts to be laser cooled. Thereforethe global picture is as follows: either the atoms are in F = 3 and they are pumped toward m = 0, or they arein F = 4 and they are laser cooled.As a final remark, let’s note that simultaneous excita-tion with two lasers detuned by the hyperfine frequencymay, in principle, produce hyperfine coherent dark states.This would reduce the efficiency of state preparation andcause unwanted frequency shifts in the subsequent Ram-sey resonance. However, in our experiment the two lasersare not phase locked and their linewidths are approxi-mately 500 kHz. These linewidths in conjunction withthe long interaction time (a few ms) does not allow forthe creation of coherent dark states. III. EXPERIMENTAL SETUP
The scheme of the experiment is presented in Fig. 2(a).It shows the main elements necessary to produce thecontinuous beam of cold cesium atoms, followed by the quantum state preparation region, the fountain parabolawhich passes two times through the microwave cavity,and finally the probe laser beam used for fluorescencedetection.The source of the atomic beam is a two-dimensionalmagneto-optical trap (2D-MOT) followed by a 3D mov-ing molasses (MM) previously described in Refs [29, 30]respectively. In the MM beams, the atoms are continu-ously captured, cooled and launched upward by the mov-ing molasses technique. We thus obtain a continuousatomic beam with a total flux of 10 at/s, a temperaturebetween 50 and 100 µ K and an adjustable velocity, setat 4 m/s in our experiment [31].We further collimate the atomic beam in order to re-duce the loss of atoms due to thermal expansion duringthe time of flight. To this end, we implement trans-verse laser cooling (TLC) in a two-dimensional opticallattice perpendicular to the atomic beam as describedin Ref. [28]. By making use of Sisyphus cooling, we re-duce the transverse temperature down to 4 µ K with anefficiency close to 100% [32]. At this level, the atomicbeam has a diameter of approximately 10 mm, and sub-sequently the effective fountain beam diameter is limitedby the microwave cavity openings to 9 mm.Quantum state preparation (QSP) takes place after thecollimation stage, 2 . F = 4 to F = 3,is implemented as a folded optical lattice as shown inFig. 1(a). Its power (2 . F = 4 → F (cid:48) = 4 transition) are chosen to op-timize the cooling, and therefore limit the heating pro-duced by the optical pumping cycles. Its incoming polar-ization is chosen linear at 45 ◦ with respect to the verticaldirection in order to produce a strong polarization gra-dient, with the double benefit to enable Sisyphus coolingand avoid the formation of Zeeman coherent states at thesame time. The second laser, used for Zeeman pumpingto the | F = 3 , m = 0 (cid:105) ground state, is superposed uponthe optical lattice as described in Fig. 1(a). It is retro-reflected to minimize pushing of the atomic beam. Inthe basic configuration, its frequency is tuned on reso-nance with the F = 3 → F (cid:48) = 3 transition and its power(3 µ W) has been adjusted to optimize the flux of atomsin | F = 3 , m = 0 (cid:105) . Moreover, we observed that it is pos-sible to improve quantum state preparation by detuningthis laser a few MHz above the F = 3 → F (cid:48) = 3 tran-sition and adjusting its power accordingly. This will bediscussed in detail in Sections IV and V. Both the op-tical lattice and Zeeman pump laser beams are gaussianwith a waist of 5 . . π -pumping toward m = 0 re-quires a linear polarization aligned with the quantizationaxis which is determined by the magnetic field direction.Therefore, not only the direction of the polarization vec-tor has to be finely tuned to that of the magnetic field,but also any ellipticity should be avoided. Since the laserlight is transported to the vacuum chamber with polar-ization maintaining fibers, special care was taken to op-timize the extinction ratio, which was measured to be40 dB after the first passage through the vacuum chamberi.e. before the retro-reflection mirror. Any component ofthe magnetic field perpendicular to the laser polarizationdestroys the atomic alignment. Therefore a small com-ponent of the magnetic field parallel to the laser polar-ization is necessary to stabilize the created atomic align-ment against the depolarization effect of the unavoid-able residual magnetic field inhomogeneities in the trans-verse directions. In our experiment, we use three pairsof coils mounted in Helmoltz-like configuration to controlthe magnetic field in this region. The external fields arethus compensated and one can finely tune the value anddirection of the resulting magnetic field ( ≈ µ T) to alignit with the pump laser polarization.After state preparation, a small fraction of the atoms(of the order of 10%) remain in the F = 4 ground state,probably due to experimental imperfections like the dif-ficulty to perfectly superpose the optical lattice and Zee-man pump lasers, stray light scattered and reflected bythe windows of the vacuum system, and fluorescencelight from the laser cooling regions. Therefore we usea depumper laser (DEP) to completely depopulate the F = 4 ground state before performing Ramsey spec-troscopy. This laser is tuned on the F = 4 → F (cid:48) = 4transition with a power of 0 . . ◦ with re-spect to the horizontal, in such way that the atoms de-scribe an open parabola passing through the microwavecavity before reaching the detection region.In order to characterize the state preparation efficiency,we use Ramsey spectroscopy to measure the populationdistribution on the Zeeman sub-levels of the F = 3ground state. The measurement principle, illustrated inFig. 2(b), proceeds in two steps: firstly the populationof | F = 3 , m (cid:105) is transferred into | F = 4 , m (cid:105) by selectivemicrowave excitation, and then the F = 4 populationis measured by fluorescence detection. More precisely,the microwave excitation exchanges the populations of | F = 3 , m (cid:105) and | F = 4 , m (cid:105) , and thus this method gives ameasurement of P ,m − P ,m where P F,m denotes the pop-ulation of state | F, m (cid:105) . However, thanks to the depumperlaser, the total population of F = 4 is smaller than 3 %.Therefore, with the resonable hypothesis that at the endof the two-laser pumping process the population distribu-tion of both ground states are similar ( P ,m ≈ . P ,m ),this method gives an estimate of P ,m with a relative error smaller than 3%. In practice, the two Rabi interac-tions are spatially separated in our continuous fountain,and the atoms pass through the microwave cavity in theupward direction, having a first π/ ≈ .
5s and turn back into the cavity where the second π/ | F = 3 , m (cid:105) → | , m (cid:105) transitions asshown in Figure 2(b). The experimental value of 77 nTshifts the seven resonances by m ×
540 Hz around the | F = 3 , (cid:105) → | F = 4 , (cid:105) clock transition. By scanningthe microwave frequency around each transitions, one canthen probe their population by fluorescence detection ofthe F = 4 atomic flux. To this end, a retro-reflectedprobe laser beam (14 mm diameter and 1 mW of power)tuned 2 - 5 MHz below the F = 4 → F (cid:48) = 5 transition issent through the atomic beam, and the fluorescence lightis collected and measured with a low noise photo-detector(DET).Finally, by superposing a repumper laser (30 µ W tunedto the F = 3 → F (cid:48) = 4 transition) on the probe laser, wecan detect all the atoms in both ground states and thusobtain a measure of the total fountain flux. In our exper-iment, we use the total flux as an indirect measurementof the cooling efficiency of quantum state preparation.Indeed, the fountain flux is reduced by the losses due tothe thermal expansion of the atomic beam during thetime of flight. Since the atomic beam section expandsproportionally to the square of the transverse velocity,the total fountain flux is inversely proportional to thetransverse temperature. This will be discussed in moredetail in Section V. IV. EXPERIMENTAL RESULTSA. Efficiency of state preparation
The microwave spectra measured with and withoutstate preparation are presented in Fig. 3. They wereobtained by scanning the microwave frequency and mea-suring the number of atoms detected in F = 4. As ex-plained in section III, before the microwave interrogationa depumper is used to depopulate the F = 4 groundstate, and the microwave power is optimized for each Zee-man component in order to produce two π/ < F = 3 population distribution.The microwave spectrum obtained without statepreparation is shown in Fig. 3(a). It is composed ofseven Rabi resonances corresponding to the transitions | , m (cid:105) → | , m (cid:105) for m = − , ..., +3. Ramsey fringes areclearly visible on the central resonance ( m = 0) but noton the other Zeeman components. This is partly dueto magnetic field inhomogeneities in the microwave in-terrogation region but mostly to the frequency samplingresolution which was adjusted to observe the Rabi res- m (cid:61)(cid:45) m (cid:61)(cid:45) m (cid:61)(cid:45) m (cid:61) m (cid:61) m (cid:61) m (cid:61) (cid:45) (cid:45) (cid:45) (cid:64) Hz (cid:68) D e t ec t e d s i gn a l (cid:64) m V (cid:68) (cid:72) a (cid:76) m (cid:61)(cid:45) m (cid:61)(cid:45) m (cid:61)(cid:45) m (cid:61) m (cid:61) m (cid:61) m (cid:61) (cid:45) (cid:45) (cid:45) (cid:64) Hz (cid:68) D e t ec t e d s i gn a l (cid:64) m V (cid:68) (cid:72) b (cid:76) FIG. 3: Microwave spectra measured (a) without state prepa-ration and (b) with state preparation. These spectra repre-sent the number of atoms detected in F = 4 as a functionof the microwave frequency. When the microwave is reso-nant with one Zeeman component ( | , m (cid:105) → | , m (cid:105) ) atomsare transferred in F = 4 and detected. The microwave poweris adjusted for each Zeeman component in order to producetwo π/ onances. Indeed, for practical reasons the frequency issampled every 4 Hz except in a band of ±
25 Hz around9192631770 Hz where the resolution is increased to 0 . F = 3is not uniform, it is asymmetric, and only 8.7% of theatoms are in m = 0 (see also Table I).Both the non-uniformity and the asymmetry of thepopulation distribution may be explained by the polar-ization of the folded optical lattice used for transversecooling which contains some ellipticity induced the mul-tiple reflexions on metallic mirrors. For our application,namely a primary frequency standard, both the asymme-try and the small number of atoms in m = 0 are prob-lematic. Indeed, the asymmetry may produce Rabi andRamsey pulling [33], and the small atom number will re-duce the signal-to-noise ratio and thus degrade the clockstability. Without Withstate preparation state preparation P , − . . P , − . . P , − . . P , . . P , . . P , . . P , . . − . − . F = 3 ground state Zeemansub-levels measured with and without state preparation (dataobtained from Fig. 3). Orientation is given by O ∝ Σ mP ,m and it is normalized relative to its maximum value. It is ameasure of the population distribution dipole moment andthus of its asymmetry. Perfect state preparation should give P , = 100 % and orientation = 0 %. These two problems can be solved simultaneously byintroducing state preparation. Indeed, the microwavespectrum measured with state preparation is shown inFig. 3(b). On this graph, one can see that the popula-tion distribution is quite symmetric and that 56.6% ofthe atoms accumulated in m = 0. More precisely andfor comparison, the populations obtained from both mi-crowave spectra are summarized in Table I. In order toquantify the gain in symmetry, we calculated the orien-tation from both population distributions.Finally, we can conclude that state preparation de-creased the orientation (asymmetry) by a factor twelve,and it increased the population of m = 0 by a factorsix. In section V, we will discuss with more detail thefactors limiting the purity of state preparation in our ex-periment. B. Evidence of laser cooling during statepreparation
In order to observe laser cooling we scanned the op-tical lattice laser frequency and measured the total fluxat the end of the fountain i.e. in the detection zone. Asexplained in section III, a decrease of the atomic beamtransverse temperature will lead to a decrease of thelosses during the parabolic flight and thereby to a higherflux measured in the detection zone. More precisely,the total flux is inversely proportional to the transversetemperature and therefore it can be used as an indirecttemperature measurement. During this measurement werecorded both the total flux ( F = 3 ,
4) and the flux in | F = 3 , m = 0 (cid:105) and the results are shown in Fig. 4. Onecan see that the total flux does not change symmetricallyaround the F = 4 → F (cid:48) = 3 and F = 4 → F (cid:48) = 4 transi-tions. It increases on the blue side, which is a signatureof Sisyphus cooling, and decreases on the red side, which (cid:45) (cid:45) (cid:45)
100 0 100 200 300 4000.0.10.20.30.40.50.60.70.8 Optical lattice frequency detuning (cid:68)Ν lattice (cid:64)
MHz (cid:68) F oun t a i n f l ux (cid:144) Φ i t o t A B C cooling cooling (cid:174)
3' 4 (cid:174)
4' 4 (cid:174) FIG. 4: Measurement of the total fountain flux in F = 3 , | F = 3 , m = 0 (cid:105) (dashed line) as afunction of the optical lattice laser frequency. ∆ ν lattice is thelattice frequency detuning from the F = 4 → F (cid:48) = 4 transi-tion of cesium D2 line. The Zeeman pump laser is tuned onthe F = 3 → F (cid:48) = 3 transition. The vertical axis is normal-ized to the total flux obtained without state preparation φ tot i .The dotted horizontal lines A, B and C indicate the total fluxlevels when the optical lattice laser is on resonance with thethree transitions respectively. Cooling is observed on the blueside of those transitions, see text for details. is a signature of Sisyphus heating. This point will bediscussed in more detail in section V. Finally, we notethat approximately 55% of the atoms are in m = 0 overa large range of frequencies. This indicates that statepreparation is not limited by the optical lattice hyperfinepumping rate and that it takes place independently ofthe laser cooling processes. C. Role of the pump laser frequency
In principle, the role of the pump laser should be lim-ited to optical pumping and thus we expected a frequencydependence symmetric around the atomic transitions, incontrast with the optical lattice laser. However, the inter-play between two lasers may bring surprises and thereforewe repeated the measurement of previous section but thistime by scanning the pump laser frequency and measur-ing the fountain flux in | F = 3 , m = 0 (cid:105) . The results arepresented in Fig. 5 for two different values of the pumppower.Firstly, we observe that the m = 0 fountain flux is in-creased on the 3 → (cid:48) transition because π -optical pump-ing populates m = 0, and decreased on the 3 → (cid:48) transi-tion because π -optical pumping populates m = ±
3. Sec-ondly, one can see that the m = 0 flux can be notablyincreased by detuning the pump laser a few MHz to theblue side of the 3 → (cid:48) transition. It is important to em-phasize that the asymmetry observed around the 3 → (cid:48) transition cannot be explained by optical pumping mech-anisms only. The light shift produced by the pump lasershould be accounted for in order to explain such an asym- (cid:45) (cid:45)
100 0 100 2000.0.10.20.30.40.50.6 Pump laser frequency detuning (cid:68)Ν pump (cid:64)
MHz (cid:68) F oun t a i n f l ux (cid:144) Φ i t o t (cid:174)
2' 3 (cid:174)
3' 3 (cid:174) FIG. 5: Measurement of the | F = 3 , m = 0 (cid:105) fountain flux asa function of the Zeeman pump laser frequency. ∆ ν pump isthe pump laser frequency detuning from the F = 3 → F (cid:48) = 3transition of cesium D2 line. The pump laser power valuesare 3 µ W (solid line) and 30 µ W (dashed line). The verticalaxis is normalized to the total flux obtained without statepreparation φ tot i . The horizontal axis is calibrated using asaturated absorption signal obtained from a small fraction ofthe pump laser. metry. Our interpretation is that we observe the firstsigns of Stark-shift-degenerate Raman sideband coolingas proposed in Ref. [26]. This point will be discussedfurther in section V.We repeated this measurement for a higher power ofthe pump laser and the maximum of the m = 0 flux isshifted toward higher frequencies. For a pump power of250 µ W this maximum approaches 25 MHz above the3 → (cid:48) transition, which is interesting for a practicalimplementation of this scheme (laser locking) since it co-incides with the saturated absorption cross-over between3 → (cid:48) and 3 → (cid:48) transitions. By locking the pump laserto this point, we observed that laser cooling compensatescompletely the heating produced by state preparation.Finally, we observe a small increase of the m = 0 fluxon the 3 → (cid:48) transition. According to our simulations,this is due to a concentration of the atoms in the low m values provoked by this optical pumping configuration. D. Atomic beam noise measurement
As mentioned in the introduction, two-laser opticalpumping was implemented for state preparation in athermal cesium beam resonator but a degradation of thesignal-to-noise ratio was observed due to the presenceof excess noise on the fluorescence signal [11]. Laterthis additional noise was analyzed and attributed to thepresence of residual unpumped atoms combined with fre-quency fluctuations of the pumping laser [10]. Giventhat our experimental conditions are very different (loweratomic flux, longer interaction time, low frequency noiselasers) our hope was to observe an improvement of thesignal-to-noise ratio by introducing state preparation.In order to demonstrate that, we measured the signal-to-noise ratio of the m = 0 fountain flux with and with-out state preparation. More precisely, in our measure-ment the signal S (A) is obtained from the DC currentof the fluorescence detection photodiode, and the noise N (A / √ Hz ) from the linear spectral density of the pho-todiode current at 1 Hz (the fountain clock modulationfrequency). In order to observe any departure from theatomic shot noise level, we repeated the measurement fordifferent values of the total flux. Given that the fountainis shot-noise limited at low flux, this measurement hasthe advantage to allow for an absolute calibration of thedetection efficiency [32, 34]. The results are presented inFig. 6 where we reported φ eq = 2( S/N ) as a function ofthe atomic flux. Here φ eq is the shot-noise limited equiv-alent flux i.e. which would give the same signal-to-noiseratio.One can see in Fig. 6 that the signal-to-noise ratiois improved thanks to state preparation. Indeed, withstate preparation (point B) we measure a threefold im-provement of the maximum equivalent flux compared tothe situation without state preparation (point A). Thegraph also shows a departure from the shot-noise limitfor higher values of the flux. However this behavior isalso observed on the total flux and thus cannot be at-tributed to state preparation. Indeed, the maximum S/N of the m = 0 flux with state preparation (point B) is ap-proximately equal to the maximum S/N of the total flux(point C). Therefore, our interpretation is that this noiseis imprinted on the atomic flux during the laser coolingstages before state preparation.
V. DISCUSSION OF RESULTSA. What limits state preparation purity
As reported in Table I, we were able to accumulate56.6% of the atoms in | F = 3 , m = 0 (cid:105) with our statepreparation scheme. This is a sixfold improvement overthe situation without state preparation, however thisscheme should allow us to reach 100% in m = 0, at leastin principle. In order to understand what is the limitingfactor in our experiment, we developed a numerical modelof optical pumping based on the rate equations presentedin Ref. [8] with the notable difference that we took intoaccount off-resonance excitation of all transitions.We performed numerical simulations of state prepa-ration with the same parameters as in the experiment.The optical lattice laser frequency is 125 MHz abovethe F = 4 → F (cid:48) = 4 transition and has a power of2 . . on average). TheZeeman pump laser frequency is on resonance with the F = 3 → F (cid:48) = 3 transition and it has a power of 3 µ W(i.e. 3 µ W/cm on average). The length of the statepreparation zone, where all the laser beams are super-posed to the atomic beam, is estimated to be of 11 mm.With an atomic beam velocity of 3 . A Bshot (cid:45) noise limit10 Atomic flux in (cid:160) F (cid:61) m (cid:61) (cid:92) (cid:64) s (cid:45) (cid:68) Φ e q (cid:61) (cid:72) S (cid:144) N (cid:76) (cid:64) s (cid:45) (cid:68) (cid:72) a (cid:76) Cshot (cid:45) noise limit10 Total atomic flux (cid:72) F (cid:61) (cid:76) (cid:64) s (cid:45) (cid:68) Φ e q (cid:61) (cid:72) S (cid:144) N (cid:76) (cid:64) s (cid:45) (cid:68) (cid:72) b (cid:76) FIG. 6: Evolution of φ eq = 2( S/N ) where S is the detectedsignal and N its noise spectral density, as a function of theatomic flux. (a) The triangles are measurements of the atomsin | F = 3 , m = 0 (cid:105) obtained without state preparation. Thesquares are measurements of the atoms in | F = 3 , m = 0 (cid:105) ob-tained with state preparation. (b) The circles are measure-ments of the total flux in F = 3 and 4. As explained in thetext, the fountain is shot-noise limited at low flux and thusallows a calibration of the photodetector signal in atoms / sec(horizontal axis). On these graphs the shot-noise limit is a linewith unit slope. Note that the three points marked by lettersA, B, C were obtained with the maximum total fountain flux. transit time of 3 ms. The laser light is transported tothe vacuum system with polarization maintaining fibers.The extinction ratio of the pump laser beam was opti-mized with care and was measured to be 40 dB. There-fore the intensity of the circular polarization componentof the pump beam is smaller than 1 / µ T with rectangular coils inHelmoltz-like configuration. However, the presence of µ -metal material close to this region (the magnetic shieldsof the interaction zone) deforms significantly the mag-netic field lines and thus the field homogeneity is difficultto evaluate. After inserting these experimental parame-ters in our optical pumping model, we found that 6% oftransverse residual magnetic field can explain our exper-imental results. Indeed, the populations obtained fromour simulations are in good agreement with the popula-tions measured in Table I with state preparation. Moreprecisely, the populations are 59.5% in m = 0, 11.7%in m = ±
1, 5.8% in m = ±
2, and 2.7% in m = ± B. Laser cooling compensates the heat producedby state preparation
State preparation with optical pumping heats theatomic beam due to the random recoils generated byspontaneous emission during optical pumping cycles. Inorder to understand the role of laser cooling, we calcu-lated the number of spontaneously emitted photons gen-erated by state preparation when both the Zeeman pumpand the optical lattice lasers are on resonance. By nu-merical integration of rate equations, and with the sameconditions as in the experiment, we obtain the numberssummarized in Table II. From the number of recoils, wecalculated the elevation of transverse temperature of theatomic beam. The initial value of T i = 4 µ K was mea-sured in Ref. [32] and the final value T f is calculatedaccording to statistical kinetic theory:12 k B T f = 12 k B T i + 12 M v r N photons / k B is the Boltzmann constant, v r = 3 . · − m/sis the recoil velocity and M the cesium atomic mass.From the elevation of the atomic beam transverse tem-perature we calculated the decrease of the detected foun-tain flux according to φ tot f /φ tot i = T i /T f and the resultsare summarized in Table II. These calculated fluxes ratiosare in good agreement with the levels indicated by thehorizontal lines A, B and C in the measurement of Fig. 4.They give us reference levels with respect to which lasercooling can be observed in the measurement of Fig. 4.Now let’s discuss the optical lattice frequencies atwhich we can expect laser cooling to operate in our exper-iment. A condition for Sisyphus cooling to work is thatoptical pumping always populates the Zeeman sub-levelswith minimum energy. According to transitions oscilla-tor strengths, this condition is fulfilled on the blue sideof transitions F → F (cid:48) ≤ F and on the red side of transi-tions F → F (cid:48) > F . Indeed, Sisyphus cooling was demon-strated to work for a positive detuning when F (cid:48) ≤ F andfor a negative detuning when F (cid:48) > F [35, 36]. This isexactly what we observe in Fig. 4: cooling operates onthe blue side of transitions 4 → (cid:48) and 4 → (cid:48) , but onthe red side of transition 4 → (cid:48) . Let’s remark that inthe first case, i.e. F (cid:48) ≤ F and positive detuning, atomsare pumped toward states weakly coupled to laser light(gray optical lattice) and thus heating by photon scat-tering is reduced. While in the second case, i.e. F (cid:48) > F and negative detuning, atoms are pumped toward states Optical lattice frequency 4 → (cid:48) → (cid:48) → (cid:48) N photons
37 44 31000 T i ( µ K) 4 4 4 T f ( µ K) 6 . . φ tot f /φ tot i N photons emitted spontaneouslyduring state preparation by numerical integration of the rateequations. T i is the atomic beam transverse temperature be-fore state preparation measured in previous work [32]. T f is the temperature after state preparation calculated from T i and N photons using Eq. 1. φ tot f /φ tot i is the ratio of the totalfountain flux with and without state preparation lasers. Thiscalculated ratio is in good agreement with the measured levelsA,B and C shown in Fig. 4 maximally coupled to laser light (bright optical lattice)and thus photon scattering is increased. This explainsthat the minimum temperatures, corresponding to max-imum total fluxes, are observed close to the 4 → (cid:48) and4 → (cid:48) transitions.To conclude this discussion, let’s emphasize that theatoms entering the state preparation region are alreadycold ( T i = 4 µ K) but they are heated by the opticalpumping cycles necessary for state preparation. In thissituation, we observe that Sisyphus cooling removes mostof the heat produced by state preparation, and thereforehelp us to take the best advantage of state preparation.
C. Effect of the Zeeman pump laser frequency onstate preparation
As shown in Fig. 5, detuning the Zeeman pump laser onthe blue side of the 3 → (cid:48) transition can lead to a notableimprovement of state preparation, which manifests itselfas an increase of the | F = 3 , m = 0 (cid:105) flux. As mentionedin section IV, the asymmetry observed around the 3 → (cid:48) transition cannot be explained by optical pumping mech-anisms only. Here the light shift produced by the pumplaser plays an important role and our interpretation isthat we observe the first signs of Stark-shift-degenerateRaman sideband cooling, as proposed in Ref. [26].This cooling mechanism, which is briefly recalled inFig.7, is crucially dependent on the light-shift inducedby the pump laser. Indeed, it produces cooling cycles fora positive light shift, and conversely heating cycles fora negative light shift. As a consequence, it will inducean asymmetric variation of state preparation efficiencyaround the 3 → (cid:48) transition as observed in Fig.5.In order to check the relevance of this explanation, wecalculated the depth of the potential wells created by theoptical lattice laser in our experimental conditions. Weused the procedure developped in Ref. [25] to obtain the n=0n=1n=2m F = -1 m F = 0m F’ = -1 π F’=3 m F = +1m F’ = 0 m F’ = +1 π∆ E Stark
F=3
FIG. 7: Principle of Stark-shift-degenerate Raman sidebandcooling [26]. Cold atoms are trapped in the potential wellsof an optical lattice, their motion is quantized, and n isthe vibrational quantum number. Ground state Zeemansub-levels with m (cid:54) = 0 are light-shifted by the blue de-tuned π -pumping laser. When degeneracy is reached between | F = 3 , m = 0 , n (cid:105) and | F = 3 , m = ± , n − (cid:105) , degenerate Ra-man sideband transitions can take place (black horizontal ar-rows), followed by π -pumping which closes the cooling cycle.Every cooling cycle removes one vibrational quantum untilthe atoms reach | F = 3 , m = 0 , n = 0 (cid:105) . See Ref. [26] for moredetails. potential depth ∆ U ≈ E r where E r = ¯ h k / (2 M ) is therecoil energy ( E r ≈ h × ω v ≈ π × m = ± hω v we calculated that 15 µ W of pump power are neces-sary. This is close to the experimental values, which arebetween 3 µ W and 30 µ W, and therefore the mechanismof Stark-shift-degenerate Raman sideband cooling shouldbe effective in our experiment. Here, one should keep inmind that these numbers represent average values, butthe experimental situation is more complex since all thelaser beams are gaussian with an intensity which variesby a factor 10 from the center to the edge.The Stark-shift-degenerate Raman sideband coolingmechanism not only prepares the atoms in m = 0 butalso cools them. Therefore, it should also have an effecton the total flux which gives an indirect measurementof the atomic beam transverse temperature. Indeed, wemeasured the total flux as a function of the pump laserfrequency and the results are shown in Fig. 8. One can seethat the variation of the total flux is asymmetric aroundthe 3 → (cid:48) and 3 → (cid:48) transitions, the cooling efficiencybeing higher on the blue side of those transitions as ex-pected according to sideband cooling. This observationdeserves two comments. Firstly, Stark-shift-degenerateRaman sideband cooling is also possible on the 3 → (cid:48) transition where it accumulates the atoms in m = ± m = 0 as shown in Fig. 5. Secondly, theasymmetry in cooling efficiency is much more visible on (cid:45) (cid:45)
100 0 100 200 3000.0.20.40.60.81. Pump laser frequency detuning (cid:68)Ν pump (cid:64)
MHz (cid:68) F oun t a i n f l ux (cid:144) Φ i t o t (cid:174)
2' 3 (cid:174)
3' 3 (cid:174) FIG. 8: Measurement of the total fountain flux as a function ofthe Zeeman pump laser frequency. ∆ ν pump is the pump laserfrequency detuning from the F = 3 → F (cid:48) = 3 transition ofcesium D2 line. The pump laser power is 3 µ W. The verticalaxis is normalized to the total flux obtained without statepreparation φ tot i . The horizontal axis is calibrated using asaturated absorption signal obtained from a small fraction ofthe pump laser. the 3 → (cid:48) transition which can be attributed to the factthat this transition is closed, i.e. the atoms are rarelypumped into the other ground state F = 4 where Sisy-phus cooling takes place. This is yet another evidencethat a cooling mechanism is active on the F = 3 groundstate and that this mechanism involves light shifts.To conclude this discussion, we remark that Stark-shift-degenerate Raman sideband cooling mechanism ex-plains all the experimental features observed by varyingthe frequency of the Zeeman pump laser in our statepreparation experiment. Nevertheless, present limita-tions of our experimental setup do not allow us to con-clude with certainty. Further studies are necessary inorder to confirm this interpretation. VI. CONCLUSION
In this work, we demonstrate that quantum statepreparation can be combined with laser cooling toprepare a continuous atomic fountain of cold atoms.More precisely, we use two-laser optical pumping toprepare cold cesium atoms in the same ground state | F = 3 , m = 0 (cid:105) . A first laser, in a folded optical lat-tice configuration, couples to the F = 4 ground stateand transfers the atoms in F = 3, while a second π -polarized laser, the so-called Zeeman pump, excites the F = 3 → F (cid:48) = 3 transition of the D2 line to pump theatoms toward m = 0. When both lasers are on reso-nance, we observe a notable heating of the cold atomicbeam produced by the optical pumping cycles. On theother hand, we demonstrate that it is possible to com-bine state preparation with Sisyphus cooling by detuningthe optical lattice laser frequency, and thereby to removemost of the heat produced by optical pumping.0Using this technique, we were able to prepare 56.6%of the atoms in | F = 3 , m = 0 (cid:105) , limited by technical im-perfections, without degrading the total fountain flux.Moreover the atomic orientation (asymetry of the popu-lation distribution among Zeeman sub-levels) of the pre-pared beam has been reduced by a factor twelve which isan advantage for its use in a primary frequency standard.Furthermore, we showed that state preparation improvesthe signal-to-noise ratio of the | F = 3 , m = 0 (cid:105) fountainflux by a factor √ F = 3 → F (cid:48) = 3 transition. In contrast, the effi- ciency decreases on the red side of the same transition,which reveals the role of the light-shift produced by theZeeman pump laser. Similar observations where madearound the F = 3 → F (cid:48) = 2 transition, even though thefinal Zeeman states are m = ± m = 0. We at-tribute this improvement of both laser cooling and statepreparation efficiency to the first observation of Stark-shift-degenerate Raman sideband cooling. Acknowledgments
This work was supported by the Swiss NationalScience Foundation (grant 200020-121987/1 and Eu-roquasar project 120418), the Swiss Federal Office ofMetrology (METAS), and the University of Neuchˆatel.We acknowledge helpful comments from the referee. [1] W. Gerlach and O. Stern, Ann. Phys. , 673 (1924).[2] A. Kastler, J. Phys. Radium , 255 (1950).[3] J.-L. Picqu´e, Metrologia , 115 (1977).[4] M. Arditi and J.-L. Picqu´e, J. Physique Lett. , 379(1980).[5] S.-I. Ohshima, Y. Nakadan, T. Ikegami, Y. Koga,R. Drullinger, and L. Hollberg, IEEE Transactions onInstrumentation and Measurement , 533 (1989).[6] P. Cerez, G. Theobald, V. Giordano, N. Dimarcq, andM. de Labachelerie, IEEE Transactions on Instrumenta-tion and Measurement , 137 (1991).[7] E. de Clercq, M. de Labachellerie, G. Avila, P. Cerez,and M. Tetu, J.Physique , 239 (1984).[8] G. Avila, V. Giordano, V. Candelier, E. de Clercq,G. Theobald, and P. Cerez, Phys. Rev. A , 3719 (1987).[9] P. Tremblay and C. Jacques, Phys. Rev. A , 4989(1990).[10] G. Lucas-Leclin, P. C´erez, and N. Dimarcq, Journal ofPhysics B: Atomic, Molecular and Optical Physics ,327 (1999), URL http://stacks.iop.org/0953-4075/32/i=2/a=014 .[11] N. Dimarcq, V. Giordano, C. Jacques, G. Th´eobald, andP. C´erez, in Proc. 5th European Frequency and Time Fo-rum (1991), pp. 187–189.[12] G. Santarelli, P. Laurent, P. Lemonde, A. Clairon, A. G.Mann, S. Chang, A. N. Luiten, and C. Salomon, Phys.Rev. Lett. , 4619 (1999).[13] A. Peters, K. Y. Chung, and S. Chu, Metrologia , 25(2001).[14] T. Mller, M. Gilowski, M. Zaiser, P. Berg, C. Schubert,T. Wendrich, W. Ertmer, and E. M. Rasel, The Euro-pean Physical Journal D - Atomic, Molecular, Opticaland Plasma Physics , 273 (2009).[15] A. Gauguet, B. Canuel, T. L´ev`eque, W. Chaibi, andA. Landragin, Phys. Rev. A , 063604 (2009).[16] N. D. Bhaskar, Phys. Rev. A , R4559 (1993).[17] S. Micalizio, A. Godone, F. Levi, and C. Calosso, Phys.Rev. A , 023419 (2009).[18] A. D. Boozer, R. Miller, T. E. Northup, A. Boca, andH. J. Kimble, Phys. Rev. A , 063401 (2007). [19] Y.-Y. Jau, E. Miron, A. B. Post, N. N. Kuzma, andW. Happer, Phys. Rev. Lett. , 160802 (2004).[20] S. A. Zibrov, V. L. Velichansky, A. S. Zibrov, A. V.Taichenachev, and V. I. Yudin, Opt. Lett. , 2060(2006), URL http://ol.osa.org/abstract.cfm?URI=ol-31-13-2060 .[21] A. Aspect, E. Arimondo, R. Kaiser, N. Vansteenkiste,and C. Cohen-Tannoudji, Phys. Rev. Lett. , 826(1988).[22] S. E. Hamann, D. L. Haycock, G. Klose, P. H. Pax, I. H.Deutsch, and P. S. Jessen, Phys. Rev. Lett. , 4149(1998).[23] V. Vuleti´c, C. Chin, A. J. Kerman, and S. Chu, Phys.Rev. Lett. , 5768 (1998).[24] P. Treutlein, K. Y. Chung, and S. Chu, Phys. Rev. A ,051401 (2001).[25] G. Di Domenico, N. Castagna, G. Mileti, P. Thomann,A. V. Taichenachev, and V. I. Yudin, Phys. Rev. A ,063403 (2004).[26] A. V. Taichenachev, A. M. Tumaikin, V. I. Yudin, andL. Hollberg, Phys. Rev. A , 033402 (2001).[27] J. M. Choi, J. M. Kim, S. Y. Jeong, and D. Cho, Journalof the Korean Physical Society , 425 (2005).[28] G. Di Domenico, N. Castagna, M. Plimmer, P. Thomann,A. Taichenachev, and V. Yudin, Laser physics , 1674(2005).[29] N. Castagna, J. Gu´ena, M. D. Plimmer, and P. Thomann,Eur. Phys. J. Appl. Phys. , 21 (2006).[30] P. Berthoud, E. Fretel, and P. Thomann, Phys. Rev. A , R4241 (1999).[31] P. Thomann, M. Plimmer, G. D. Domenico, N. Castagna,J. Gu´ena, G. Dudle, and F. F¨uzesi, Applied Physics B:Lasers and Optics , 659 (2006).[32] G. D. Domenico, Ph.D. thesis, Universit´e de Neuchˆatel(2004), URL doc.rero.ch/record/3641/files/these_DiDomenicoG.pdf .[33] J. Vanier and C. Audoin, The Quantum Physics ofAtomic Frequency Standards (Adam Hilger, 1989).[34] J. Gu´ena, G. Dudle, and P. Thomann, Eur. Phys. J.Appl. Phys. , 183 (2007), URL http://dx.doi.org/ .[35] M. Drewsen, U. Drodofsky, C. Weber, G. Schreiber, andJ. Mlynek, J. Phys. B: At. Mol. Opt. Phys , L843(1996). [36] H. Ellmann, J. Jersblad, and A. Kastberg, Eur. Phys. J.D13