Arxiv:1510.05744V2 [Physics.Atom-Ph] 29 Jun 2016 Mrvsteat-Eaainpromneo H Coatings

Home , Mean free path

arXiv:1510.05744v2 [physics.atom-ph] 29 Jun 2016 mrvsteat-eaainpromneo h coatings. the of ripening performance that anti-relaxation confirmed the also improves and “ripen- rate, as VCC walls known the process cell increases heating ing” the a between that found freely we travel addition, not In do se atoms the in alkali negligible path that not free is mean gas background short the This that atoms. indicates rubidium of path free mean okws1 was work aei para in rate Tel: Agricul of e-mail: University Tokyo Physics, Applied of Department Hatakeyama Atsushi Sekiguchi, Naota para with coated cells iin VC)btenakl tm n akrudgsin gas background and atoms bu alkali between (VCCs) lisions Abstract 2016 13, January Fax: o-elgbecliin fakl tm ihbackground with atoms alkali of collisions Non-negligible hc assteakl tm obuc aytmswith- times surface, Para many glass polarization. bounce spin bare to of the relaxation atoms of out alkali that the than causes weaker which al much with is coating anti-relaxation atoms to the c of is anti-relaxation interaction so-called first The a ing. The with bu surfaces cell. glass inert glass inner an the a with in cell atoms the fill alkali of spin-polar state a preserve ground to used techniques major two are There Introduction 1 m ftecl,adas ti osbet oaieadde- and polarize to possible is e it more vol- also atoms entire alkali and the tect cell, sample the atoms of inhomogeneity alkali ume field because to reduced due be lines can resonance magnetic bu of relax- over ing advantages wall-induced of before number times a 000 Bu 10 2]. [1, to ation up atoms bounce can kali materials, coating anti-relaxation used widely ewe laiaosadtebcgon a.Hwvr the However, gas. background collisions the ignore and to enough atoms low alkali be between backgroun to the thought is experiments, of pressure these studies gas In and [9]. [8], propagation memory ultra-sensi light quantum 4], [3, [5–7], clocks magnetometry atomic including m in applications, used been therefore have Bu cells line. anti-relaxation-coated resonance optical the of broadening sure wl eisre yteeditor) the by inserted be (will aucitNo. manuscript ff rgsfe nirlxto-otdcls h vrg VC average The cells. anti-relaxation-coated er-gas-free + + 1()4 8 7554 388 42 (0) 81 1()4 8 7554 388 42 (0) 81 [email protected] ffi emaue h aeo eoiycagn col- velocity-changing of rate the measured We × -otdrbdu ao el rprdi this in prepared cells vapor rubidium n-coated 10 ff rgsfe nirlxto-otdclshave cells anti-relaxation-coated er-gas-free 6 s − 1 hc orsod to corresponds which , ff cieybcueteei opres- no is there because ectively ff rgs h eodi ocover to is second The gas. er ffi ff rgscls h broaden- the cells: er-gas n ffi ,oeo h most the of one n, ∼ mi the in mm 1 ff er-gas-free ueadTcnlg,Kgni oy 8-58 Japan 184-8588, Tokyo Koganei, Technology, and ture oat- kali tive any nse al- C ized d . reybtentewalls? the between cell freely coated the Are “bu reactions. really chemical generation by the gases to lead additional would atoms alkali of reactivity high h rudsaehprn rniinfeunyb 4Hz 14 by frequency transition hyperfine state shi [10–12], ground gas molecules, the background the hydrogen of species and typical example, are For atoms which account. cesium into between taken be collisions influenc should the collisions walls, the the of between fo freely high travel too to is re- atoms pressure alkali gas been background have the If outgassing [10–13]. ported resulting and silic with coatings atoms alkali compound of reactions chemical and [2] cerned reclshv o enrpre xetfravr ihpres- high hundred very a a for of except ( order reported sure been bu the not in have on pressures cells gas is free background cell However, 4]. coated [3, a Hz i of by induced surfaces shift ner frequency the considering applications, ao el ihapara (Rb) a rubidium with in cells demonstrated vapor was VCC measurement measured the This from rate. estimated pres- be can cause free atoms to mean alkali of and not path pressure enough gas background low The is broadening. pressure sure gas the the breaking be- without when gas (VCCs) cell background collisions and atoms velocity-changing alkali of tween rate the suring [10]. pressure surfa gas cell background the the with increasing atoms by of rate collision ma the ripening reduce has that simply concerns It are perf there density. although anti-relaxation vapor [10,18], improves mance saturation ripening its that den reported than vapor been lower alkali much initial vapor is the alkali sity because increase 16–18], to 10, [2, experiments density before cells coated perfor generally on is Ripening reactions. ac th chemical may over celerate which pressure “ripening,” the called of sometimes history process, the heating track to possible it makes [15]. coating octadecyltrichlorosilane an 1]a 30 at [14] nfc,otasn npara in outgassing fact, In ee erpr ipeade and simple a report we Here, cell the breaking without measurement pressure situ In ≤ 0 a akrudgsi irfbiae elwith cell microfabricated a in gas background Pa) 600 ◦ ff .Ti hf a epolmtci tmcclock atomic in problematic be may shift This C. rgsfe”i h es htakl tm travel atoms alkali that sense the in er-gas-free” ffi a nbu in gas otn.W on htteVCC the that found We coating. n ffi otdclshsbe con- been has cells coated n ff ciemto o mea- for method ective ff er-gas-free ff er-gas- med / or- on Pa ce of n- es ft y e s - r - 2 Naota Sekiguchi, Atsushi Hatakeyama

5 1 rates were higher than a few 10 s− . The mean free path of resulting in fewer atoms within a probed velocity range (Fig. 1b). Rb atoms and the background gas pressure were estimated The signal S (t) is thus described by to be a few mm and several Pa, respectively. In addition, by measuring the VCC rates before and after ripening, we S (t) S esc(t) S c(t), (1) ∝ × found that ripening increased the background gas pressure. The spin-relaxation times were also measured before and af- where the contributions of the ballistic escape and the VCCs ter ripening, and the number of bouncesbefore relaxation was are denoted as S esc(t) and S c(t), respectively. Detailed discus- evaluated. We found that ripening increased the number of sions about S esc(t) and S c(t) will be provided in Sec. 2.3. bounces.

2.2 Experimental setup 2 VCC-rate measurement Measurement of the VCC rate was carried out for paraffin- 2.1 Method coated Rb vapor cells at room temperature, the fabrication procedure of which will be described in Sec. 3. A longitudi- Our method to measure the VCC rate γ is based on velocity- nal magnetic field of 40 µT was applied to the vapor cell with selective optical pumping and VCC-induced redistribution of Helmholtz coils in a permalloy magnetic shield. The frequen- the velocity of spin-polarized alkali atoms. A resonant, circu- cies of the pump and the probe beams were stabilized to the larly polarized pump laser pulse selectively polarizes alkali F = 3 F = 4 transition frequency of the 85Rb D2 line. → ′ atoms whose velocity components vz along the laser propa- The profiles of the beams were measured with a CMOS cam- gation direction (the z direction) lie within a narrow velocity era and fitted by Gaussian functions. The diameters of the distribution centered at vz = 0. The pump pulse irradiates probe and the pump beams at which their intensities were re- alkali atoms for only a short duration. Spin polarization is de- duced by 1/e2 compared to the beam center intensity were tected with a counterpropagating collinear probe laser beam 0.38 cm and 0.053 cm, respectively. The probe beam power (see Fig. 1a). The diameter of the probe beam is much larger was 10 µW, and the pump beam power was 0.3 mW. than that of the pump beam. A similar optical configuration The pump pulse has a finite duration τ, and τ must be has been adopted to measure diffusion coefficients [19]. much longer than the lifetime of the Rb D2 line to polarize The signal S (t) measured with the probe beam indicates Rb atoms. At the same time, τ shouldnot be muchlonger than the polarization of atoms with vz 0 in the probe beam re- the typical time constants of the atoms’ ballistic escape and ∼ gion and decreases with time t due to two different mecha- the decrease in S (t) due to the VCCs. In this experiment, the nisms. First, the ballistic escape of the polarized atoms from pump beam irradiated the cell for 1 µs, and was then shut off the probe beam region decreases S (t). Second, the VCCs be- with an acousto-optic modulator at t = 0. Measurements were tween alkali atoms and the background gas causes a decrease performed for the left- and right-circularly polarized probe in S (t), as the VCCs redistribute the atomic velocity of the beams, and the signal S (t) was derived from the difference polarized atoms toward the Maxwell-Boltzmann distribution, between the transmitted powers for the two polarizations.

(a) Probe beam Cell Pump beam 2.3 Analysis

Experimental data for an uncoated cell and a paraffin-coated cell are shown in Fig. 2 by blue and red points, respectively. It is clear that the decrease in signal S (t) for the paraffin- (b) coated cell was more rapid than that for the uncoatedcell. The paraffin-coated cell, therefore, contained more background

Fsp(vz,0) Fprobe(vz) gas than the uncoated cell. To derive the VCC rate, we simu- lated the signal S (t) with the following model.

Fsp(vz,∞) First, we considered the contribution S esc(t) of the ballistic escape to the signal S (t). The polarized atoms’ velocity vz 0 component vr perpendicularto the beams obeys the Maxwell- Boltzmann distribution. For simplicity, we assumed that the Fig. 1 (Color online) Schematic illustration of VCC rate measure- pump beam diameter was infinitely small and atoms were po- ment. (a) Spin polarization produced by the pump pulse is detected = with the probe beam, which is counterpropagating and collinear to larized at the center of the probe beam at t 0. The probabil- the pump beam. (b) The probe beam detects the spin polarization of ity G(vr)dvr that a polarized atom has a velocity between vr + atoms within the velocity distribution Fprobe(vz). The initial veloc- and vr dvr was yielded as ity distribution Fsp(vz, 0) of spin polarization, which can be broader 2 that F (v ) because of power broadening, is redistributed toward 2v v probe z G(v )dv = r exp r dv , (2) F (v , ). r r 2 r sp z v − vD !  ∞ D     Non-negligible collisions of alkali atoms with background gas in buffer-gas-free cells coated with paraffin 3

!" laser. In the steady state, F (v , t = ) should be propor- sp z ∞ Uncoated cell tional to the Maxwell-Boltzmann distribution. The probe beam !# γ = 0 Paraffin-coated cell detects the polarization within the velocity distribution Fprobe(vz), #!$ Fit curve which is proportional to the absorption cross-section for the #!% probe beam. The signal detected with the probe beam is pro- portionalto the integralof the productof Fsp(vz, t) and Fprobe(vz) #!&

(arb. units) over the velocity vz, S #!" ∞ #!# S c(t) Fsp(vz, t) Fprobe(vz)dvz. (7) - ' ∝ Z × γ)*)+!"),) # ). −∞ '#!" # # "# (# &# We assumed the time variation of Fsp(vz, t) as Time (µs) ∂Fsp(vz, t) = γ F (v , t) F (v , ) , (8) ∂t sp z sp z Fig. 2 (Color online) Signals S of the VCC rate measurement for − h − ∞ i an uncoated cell shown by blue points and a paraffin-coated cell where the steady-state velocity distribution Fsp(vz, ) was (#2, post-ripening) shown by red points. Both data are normalized given by ∞ at 0 µs. The dashed line shows a simulation of the S without VCCs. The solid line shows a fit curve to the data for the coated cell. 1 v 2 F (v , ) = ∞ F (v , 0)dv exp z . (9) sp z sp z′ z′   ∞ Z × vD √π − vD ! −∞   where the most probable speed vD is given by vD = √2kBT/m   We then calculated S c(t) from Eqs. (7), (8), and (9).  with the Boltzmann constant kB, the absolute temperature T, and the atomic mass m. The polarized atoms move at a con- As the pump pulse has a finite duration τ = 1 µs, the 0 stant velocity during the ballistic escape. Hence, the proba- experimental data were compared with τ S esc(t t′) S c(t − − × − bility ρ(r, t)dr that a polarized atom is at a distance from the t′)dt′. The curve calculated with the VCCR rate γ = 0 (the center between r and r + dr at time t was derived from Eq. (2) dashed line in Fig. 2) in the absence of VCCs is in reasonable as agreement with the experimental data for the uncoated cell. 2r r 2 The experimental data for the coated cell were well fitted by ρ(r, t)dr = exp dr. (3) 5 1 2 the curve with γ = 7.2 10 s (the solid line in Fig. 2). (vDt) − vDt !  −   In this work, we consider× that dominant uncertainties in the   As the optical depths of the vaporcells were muchsmaller VCC rate γ come from our adopted model or uncertainties in than unity at room temperature, the signal is proportional to the experimental conditions, while the fitting errors are on the the product of ρ(r, t)dr and the intensity of the probe beam at order of 1%. the position of a polarized atom:

∞ S esc(t) Iprobe(r) ρ(r, t)dr. (4) 2.4 Estimation of background gas pressure and mean free ∝ Z0 × path of alkali atoms Here, the beam profile I (r) of the probe beam was de- probe The VCC rate γ is given by γ = nσ v¯. Here, n is the number scribed by a Gaussian function c density of background gas, σc is the cross-section of VCCs, 2 √2r andv ¯ is the mean relative velocity. The mean free path λ of Iprobe(r) exp , (5) Rb atoms is then λ = 1/(nσ ) = v¯/γ. Although the back- ∝ − φ /2  c  probe  ground gas species were unknown in this work, we assumed         v¯ = 1 105 cm/sas most componentsof the backgroundgasin where the probe beam diameter φprobe = 0.38 cm, as men- some× silicon compound-coated cells are light [10–12]. From tioned above. By substituting Eqs. (3), (5) into Eq. (4), S esc(t) was obtained as the ideal gas law, the pressure p of the background gas was estimated by p = kBT/(λσc), where σc was assumed to be 1 14 2 S (t) . (6) σc = 1 10− cm by reference to Ref. 20 regarding Na-Ne esc 2 2 2 × ∝ φprobe + 8vDt collisions. The mean free path and background gas pressure for the paraffin-coated cell shown in Fig. 2 were estimated to Next, we considered the contribution S c(t) of the VCCs be 1.4 mm and 3.0 Pa, respectively. to the signal S (t). The polarization of the alkali atoms has a velocity distribution Fsp(vz). The velocity distribution Fsp(vz) are redistributed toward the Maxwell-Boltzmann distribution 3 Cell preparation and spin-relaxation measurement due to the VCCs, while the polarization is preserved. The velocity distribution Fsp(vz) of the spin polarization, thus, de- The above evaluations were performed for seven paraffin- pends on t and is described as Fsp(vz, t). The pump pulse coated cells madeof Pyrex glass. The dimensionsof the coated beam produces the initial Lorentzian distribution Fsp(vz, t = cells are shown in Table 1. Each cell had a stem with a con- 0), which corresponds to the absorption profile for the pump stricted capillary with a diameter of 1 mm to be connected ∼ 4 Naota Sekiguchi, Atsushi Hatakeyama

Table 1 Dimensions and slow spin-relaxation time τs of cylindrical of the polarization plane of the probe light for the unpolar- paraffin-coated cells. The coated cells are numbered from #1 to #7. ized atoms was adjusted to 0.0 mrad. The data were fitted by a double exponential function and indicated two relaxation Diameter Length slow spin-relaxation time τs (s) Cell components, a fast relaxation time τ f and a slow relaxation (mm) (mm) Pre-ripening Post-ripening time τs, with opposite signs as reported previously [18]. We #1 17.0 24.4 0.15 0.18 focused on the slow relaxation time τs, which is shown in #2 17.0 27.2 0.046 0.082 Table 1. #3 17.0 27.4 0.037 0.15 #4 17.0 25.0 0.040 0.19 #5 17.0 24.3 0.17 0.22 4 Results and Discussion #6 17.0 23.7 0.22 0.24 #7 17.0 24.3 0.081 0.13 VCC rates in all prepared coated cells are shown in Fig. 4. Although there was variation in VCC rate among the coated cells, VCC rates in the post-ripening coated cells were in- 5 1 to a vacuum system. The coating procedure was similar to creased to higherthan a few 10 s− . The correspondingmean those reported previously [16,17]. Coating material was frac- free paths were shorter than a few mm, which was much tional distilled between 220◦C and 240◦C from commercial smaller than the dimensions of a typical cell. It is clear that R paraffin wax Sasolwax H1 produced by Sasol Wax. The cell the ripeningprocess made the backgroundgas pressure higher 4 was evacuated below 10− Pa with a turbomolecular pump in all coated cells. and baked at 420◦C for 4 hours. After deposition of paraffin To test the generality of our results, we also measured 4 at 380◦C or 400◦C, the cell was re-evacuated below 10− Pa the VCC rate in a pwMB-coated cell fabricated in a differ- and then sealed with Rb metal (in natural isotopic abundance) ent laboratory. The coating material pwMB was mentioned in at the stem. In this work, initial Rb vapor density was suffi- Refs. 17,21,22. The pwMB-coated cell was spherical (47 mm ciently high to carry out the VCC rate and spin-relaxation in diameter) and filled with enriched 85Rb. The VCC rate in 5 1 time measurements. The measurements were also performed the pwMB-coated cell was measured as 3.8 10 s− . The × after ripening, during which the cell was heated at 80◦C for mean free path and the background gas pressure were esti- 10 hours while the stem was kept at 75◦C. mated to be 2.6 mm and 1.6 Pa, respectively. It is therefore The longitudinal spin-relaxation times of the fabricated very likely that paraffin-coated cells generally have relatively cells were measured in a manner similar to that described pre- high background gas pressures. viously [18]. The cell was placed in the magnetic shield and With the VCC rates in mind, we evaluated the number of immersed in a longitudinal magnetic field of 40 µT. A circu- bounces N before depolarization under the assumption that larly polarized laser light (pump light) produced spin polar- Rb atoms moved diffusively as described previously [23]. In ization of Rb atoms and was shut off at time t = 0. A linearly the evaluation, the diffusion coefficient D had two variations. polarized laser light (probe light) and a balanced polarime- One was estimated from the classical diffusion coefficient ter detected the optical rotation induced by spin polarization. D = λvm/3, where the mean speed vm of Rb atoms is given The power of the probe light was 5 µW, and the probe beam by vm = 2vD/ √π. The other was estimated from the diffusion profile was the same as that used in VCC-rate measurement. coefficient for Rb atoms in hydrogen molecules [24] and the Figure 3 shows the time dependence of the optical rotation estimated background gas pressure. In addition, the number angle in a paraffin-coated cell (# 1, post-ripening). The angle

3.0 6000 VCC rate (pre) VCC rate (post) 2.5 Number of bounces (pre) 2.5 Number of bounces (post) 5000

Data Number of bounces ) 2.0 Fit curve -1

s 2.0 4000 6 t f = 0.027 s 1.5 t 1.5 3000 s = 0.18 s

1.0 1.0 2000 VCC rate (10

0.5 0.5 1000

0.0 0.0 0 Optical rotation angle (mrad) #1 #2 #3 #4 #5 #6 #7 Coated cell -0.5 0.0 0.2 0.4 0.6 0.8 Fig. 4 VCC rates (left axis) and number of bounces (right axis) for Time (s) pre- and post-ripening coated cells. Hatched bars show VCC rates Fig. 3 (Color online) Spin relaxation time measurement in a for pre-ripening cells, and gray bars show those for post-ripening paraffin-coated cell (# 1, post-ripening). The experimental data were cells. The numbers of bounces for pre- and post-ripening cells are fitted by a double exponential function. shown by open circles and gray circles, respectively. Non-negligible collisions of alkali atoms with background gas in buffer-gas-free cells coated with paraffin 5 of bounces in the absence of VCCs was also calculated by 10. J.C. Camparo, R.P. Frueholz, and B. Jaduszliwer, J. Appl. Phys. 62, 676 (1987) vmA 11. Y.W. Yi, H.G. Robinson, S. Knappe, J.E. Maclennan, C.D. N = τs. (10) 4V × Jones, C. Zhu, N.A. Clark, and J. Kitching, J. Appl. Phys. 104, Here, A is the inner surface area and V is the volume of the 023534 (2008) cell. The numbers evaluated in these three ways were almost 12. S.N. Atutov, R. Calabrese, A.I. Plekhanov, and L. Tomassetti, Eur. Phys. J. D 68, 6 (2014) the same, as expected for the background gas pressure of a 13. M. Stephens, R. Rhodes, and C. Wieman, J. Appl. Phys. 76, few Pa, and are shown in Fig. 4. It is clear that the number of 3479 (1994) bounces, i.e., the anti-relaxation performance of the coating, 14. M. Arditi and T.R. Carver, Phys. Rev. 112, 449 (1958) was improved by ripening. 15. R. Straessle, M. Pellaton, C. Affolderbach, Y. Pétremand, D. Briand, G. Mileti, and N.F. de Rooij, Appl. Phys. Lett. 105, 043502 (2014) 5 Conclusions 16. E.B. Alexandrov, M.V. Balabas, D. Budker, D. English, D.F. Kimball, C.-H. Li, and V.V. Yashchuk, Phys. Rev. A 66, 042903 We measured the rates of VCCs between Rb atoms and back- (2002) groundgas in seven buffer-gas-free, paraffin-coated cells. The 17. S.J. Seltzer, M.-A. Bouchiat, and M.V.Balabas, in Optical Mag- 6 1 netometry, edited by D. Budker and D.F.J. Kimball, (Cambridge results indicated an average VCC rate of 1 10 s− . This VCC rate corresponds to the mean free path of× 1 mm and the University Press, Cambridge, 2013) pp. 205–224 background gas pressure of 4 Pa. This short mean free path 18. M.T. Graf, D.F. Kimball, S.M. Rochester, K. Kerner, C. Wong, D. Budker, E.B. Alexandrov, M.V. Balabas, and V.V. Yashchuk, compared to the cell dimensions means that Rb atoms in the Phys. Rev. A 72, 023401 (2005) coated cells do not travel freely between the cell walls. We 19. M. Parniak and W. Wasilewski, Appl. Phys. B 116, 415 (2014) also found that the so-called “ripening” process increased the 20. C.G. Aminoff, J. Javanainen, and M. Kaivola, Phys. Rev. A 28, background gas pressure. We confirmed improvement of the 722 (1983) anti-relaxation performance of the paraffin coatings by ripen- 21. S.J. Seltzer, D.J. Michalak, M.H. Donaldson, M.V. Balabas, ing. The relatively high VCC rates indicated that precise ex- S.K. Barber, S.L. Bernasek, M.-A. Bouchiat, A. Hexemer, A.M. periments using coated cells such as clock applications re- Hibberd, D.F.J. Kimball, C. Jaye, T. Karaulanov, F.A. Narducci, quire attention to the effects induced by collisions with back- S.A. Rangwala, H.G. Robinson, A.K. Shmakov, D.L. Voronov, ground gas. Models to describe atomic dynamics in coated V.V. Yashchuk, A. Pines, and D. Budker, J. Chem. Phys. 133, cells, such as discussed in Ref. 25 to reproduce electromag- 144703 (2010) netically induced transparency spectra, may require modifi- 22. A.M. Hibberd, S.J. Seltzer, M.V. Balabas, M. Morse, D. Bud- ker, and S.L. Bernasek, J. Appl. Phys. 114, 094513 (2013) cation to the assumption of the ballistic transport of atoms in 23. S.J. Seltzer, D.M. Rampulla, S. Rivillon-Amy, Y.J. Chabal, S.L. the cells. In particular, the background gas pressure shouldbe Bernasek, and M.V. Romalis, J. Appl. Phys. 104, 103116 (2008) checked when coated cells are heated. 24. R.J. McNeal, J. Chem. Phys. 37, 2726 (1962) 25. M. Klein, M. Hohensee, D.F. Phillips, and R.L. Walsworth, Acknowledgements We would like to thank Mikhail V. Balabas for Phys. Rev. A 83, 013826 (2011) fabrication of the pwMB-coated cell and for his helpful comments, and Antoine Weis for his important advice. This work was supported by a Grant-in-Aid for Scientific Research (No. 23244082) from the Japan Society for the Promotion of Science (JSPS).

References

1. H.G. Robinson, E.S. Ensberg, and H.G. Dehmelt, Bull. Am. Phys. Soc. 3, 9 (1958) 2. M.A. Bouchiat and J. Brossel, Phys. Rev. 147, 41 (1966) 3. H.G. Robinson and C.E. Johnson, Appl. Phys. Lett. 40, 771 (1982) 4. T. Bandi, C. Aﬀolderbach, and G. Mileti, J. Appl. Phys. 111, 124906 (2012) 5. J. Dupont-Roc, S. Haroche, and C. Cohen-Tannoudji, Phys. Lett. A 28, 638 (1969) 6. D. Budker, D.F. Kimball, S.M. Rochester, V.V. Yashchuk, and M. Zolotorev, Phys. Rev. A 62, 043403 (2000) 7. D. Budker and M. Romalis, Nat. Phys. 3, 227 (2007) 8. B. Julsgaard, J. Sherson, J.I. Cirac, J. Fiurasek, and E.S. Polzik, Nature 432, 482 (2004) 9. D. Budker, D. F. Kimball, S.M. Rochester, and V.V. Yashchuk, Phys. Rev. Lett. 83, 1767 (1999)