Frequency Comparison of Two High-Accuracy Al+ Optical Clocks

C. W. Chou, D. B. Hume, J. C. J. Koelemeij, D. J. Wineland, and T. Rosenband

Time and Frequency Division, National Institute of Standards and Technology, Boulder, Colorado 80305 (Dated: February 2, 2010) We have constructed an optical clock with a fractional frequency inaccuracy of 8.6 1018 , based on quantum logic spectroscopy of an Al+ ion. A simultaneously trapped Mg+ ion serves to sympathetically laser-cool the Al+ ion and detect its quantum state. The frequency of the 1 S0 3 P0 clock transition is compared to that of a previously constructed Al+ optical clock with a statistical measurement uncertainty of 7.0 1018 . The two clocks exhibit a relative stability of 2.8 1015 1/2 , and a fractional frequency dierence of 1.8 1017 , consistent with the accuracy limit of the older clock.

arXiv:0911.4527v2 [quant-ph] 2 Feb 2010

Optical clocks based on petahertz (1015 Hz) transitions in isolated atoms have demonstrated signicant improvements over the current cesium primary-frequencystandards at 9.2 GHz. They also shed light on fundamental physics, such as the possible variation of physical constants. While the merits of laser-cooled ion optical frequency standards were known [1, 2], further developments were required to permit their use. Sub-hertz linewidth lasers [3] enabled single ions to be probed with sucient resolution for high-stability clock operation, and control of external-eld perturbations allowed such clocks to operate with an inaccuracy below 1016 [4, 5]. For comparison, cesium standards that realize the SI second have reached an inaccuracy of 3 1016 [6], and an optical lattice clock based on Sr atoms has been reported [7] with an inaccuracy of 1.5 1016. Here we describe an Al+ ion clock with an inaccuracy of 8.61018. The 1 S0 3 P0 transition in Al+ at 1.121 PHz is of interest due to its low sensitivity to electromagnetic perturbations and its narrow natural linewidth of 8 mHz [8, 9]. Al+ has the smallest sensitivity to blackbody radiation [10, 11] among atomic species currently under consideration for clocks, thus relaxing the requirement on ambient temperature control. However, the absence of an accessible allowed optical transition prevents the internal state of Al+ ions from being detected by conventional methods, and the ion cannot currently be directly laser-cooled. Quantum logic spectroscopy (QLS) [12] overcomes these diculties by trapping a logic ion that can be directly laser-cooled together with the Al+ clock ion. The coupled motion of the two ions allows for sympathetic cooling, as well as the transfer of the clock ions quantum state to the logic ion, where the state can be measured. The clock described here shares features with our previously-constructed Al-Be clock [5, 9], but also includes many changes, making a comparison of the two clocks a valuable test of systematic errors. In the new clock, the 9 Be+ logic ion has been replaced by 25 Mg+ , whose mass closely matches that of 27 Al+ . Laser-cooling ineciencies due to mass-mismatch are thus suppressed. The Al-Mg ion trap is a linear Paul trap built from tool-machined all-metal electrodes (Fig. 1). This construction diers from the Al-Be trap that was built from

1070 nm fiber laser

AOM

x2

BS BS AOM

x2

AOM 2

Clock servo

Ultra-stable Cavity

x2 AOM 1 Be
+

Al

Al+

Mg

Frequency recording

FIG. 1: Setup for comparing the frequencies of the two Al+ clocks. The 4th harmonic of a ber laser is locked to the Al+ 1 S0 3 P0 clock transition in the Al-Mg apparatus with a xed oset frequency (applied to AOM1). Another laser beam derived from the same laser probes the clock transition in the Al-Be apparatus, where the laser frequency is locked to the clock transition in a separate digital feedback loop that controls AOM2. The record of frequencies applied to AOM2 represents the dierence in clock frequencies. BS: beam splitter; AOM: acousto-optic modulator; x2: frequency doubler.

laser-machined and gold-coated alumina electrodes [13], and it exhibits reduced RF-micromotion-inducing electric elds. QLS with a Mg+ logic ion proceeds in the same way as with Be+ , but the ground-state-cooling process [14] has been modied to require only two lasers rather than three [15]. This cools the out-of-phase axial motional mode to an average quantum number of n < 0.05. It also enables quantum-non-demolition transfer (QNDT) of the Al+ -clock-ion state to Mg+ with approximately 80 % delity in a single QNDT repetition and over 99 % delity after typically ve QNDT repetitions [16]. The trap utilized in the new Al+ clock has bladeshaped gold-coated beryllium-copper electrodes (Fig. 1) whose edges are approximately 400 m from the ions. The 25 Mg+ logic ion (I = 5/2) is manipulated with light of 279.5 nm wavelength from two independent lasers. A frequency-doubled dye laser resonantly drives |2 S1/2 , F {2, 3} 2 P3/2 cycling and repumping transitions, while a frequency-quadrupled ber laser is detuned by 40-60

2
TABLE I: Systematic eects that shift the clock from its ideal unperturbed frequency. Shifts and uncertainties given are in fractional frequency units (/). See text for discussion. Eect Shift Uncertainty (1018 ) (1018 ) Excess micromotion -9 6 Secular motion -16.3 5 Blackbody radiation shift -9 3 Cooling laser Stark shift -3.6 1.5 Quad. Zeeman shift -1079.9 0.7 Linear Doppler shift 0 0.3 Clock laser Stark shift 0 0.2 Background gas collisions 0 0.5 AOM freq. error 0 0.2 Total -1117.8 8.6 TABLE II: Motional modes of the Al-Mg ion pair. For each of the six normal modes, the oscillation frequency and zeropoint motional amplitude for each ion is listed. The trap axis corresponds to z , and x, y are two orthogonal radial di rections whose orientation is determined by the trap geometry. Also shown are the calculated ( C ) and measured ( M ) n n Doppler-cooled average motional quantum numbers, and the time-dilation (TD) per motional quantum as well as the total TD per mode. f Mg+ 27 Al+ nM nC TD/quantum Total TD
25

[MHz] 6.53 5.66 5.20 [nm] 4.9 2.9 4.6 y y z [nm] 2.6 -5.0 -4.1 y y z 2.9 4.5 3.4 3.3 3.8 3.4 [1018 ] 0.226 0.731 0.197 [1018 ] 0.77 3.66 0.77

4.64 5.5 x 3.5 x 6.3 5.9 0.290 1.97

3.41 -4.2 x 6.2 x 10.0 8.0 0.771 8.10

3.00 5.6 z 5.8 z 7.0 5.9 0.133 1.00

GHz to drive Raman transitions between the 25 Mg+ hyperne qubit states. Both ions are created via multiphoton-ionization of neutral atoms from ovens. A 396 nm diode laser produces 27 Al+ [17], while a frequencydoubled dye laser at 285 nm creates 25 Mg+ [18]. Typically the clock transition is probed with 150 ms duration -pulses with a duty cycle of approximately 65 %. The remaining 35 % is occupied with state preparation and state detection functions as well as interleaved experiments that allow real-time measurement and reduction of micromotion. Systematic shifts of the Al-Mg clock are listed in Table I. Two types of residual motion cause timedilation shifts: micromotion near the trap drive frequency of RF = 59 MHz, and harmonic-oscillator (secular) motion at the ions normal mode frequencies (Table II). In both cases the clock frequency shifts by v2 f 1 + ( 400MHz )2 , where we add to the rel = 2c2 ativistic time-dilation v 2 /(2c2 ) a frequency-dependent term that corresponds to the Stark shift from the motioninducing electric elds. Here v 2 is the mean-squared Al+ -ion velocity and f is the frequency of motion. For the highest motional frequencies the Stark shift correction is 2 %. Excess micromotion (EMM) refers to the rapid ion motion at RF [19]. It is caused by electric elds that force the ion away from the RF-minimum of the ion trap, or phase shifts between trap electrodes that cause the RF-elds to be non-zero at the pseudo-potential minimum. We measure the amplitude of this motion at RF by observing the motional-sideband strength of the Al+ 1 S0 3 P1 transition in three orthogonal directions. For small amplitude of motion, the time dilation shift is / = |RF /L |2 = 2.8 1015 ||2 , where L = 1.12 PHz is the probe laser frequency, and = (1 , 2 , 3 ) is the measured EMM Lamb-Dicke parameter (the ratio of sideband and carrier Rabi rates)

for the three orthogonal directions. Typical values for 1,2,3 are 0.01 to 0.04. We note that this EMM measurement method detects slow electric eld uctuations such as those caused by the migration of photo-electrons, as well as faster uctuations caused by periodic line noise (50 to 60 Hz). Fluctuations that are shorter than the laser probe period of 0.05 to 0.1 ms will not be detected. In a perfect linear Paul trap EMM along the trap axis does not occur, but imperfections in the ion-trap geometry can lead to axial EMM . For the Al-Mg trap there exists a sharp minimum of axial EMM at one spatial location. The Al+ ion is held at this point, but random background-gas collisions cause the Al-Mg ion pair to spontaneously re-order, which causes the Al+ ion to move by 3 m every 200 s, on average. The Al+ ion is maintained at the position of minimal micromotion by adjusting the electrode voltages every 10 s to force the ion-pair into the desired order. When the ions are in the wrong order (about 5 % of the time), the Al+ ion experiences excess axial micromotion and a corresponding clock shift of / = 2.7 1017. This additional shift is included in Table I. During each Al+ clock interrogation pulse, the Mg+ ion is simultaneously Doppler cooled by a laser that is tuned 21 MHz below the |2 S1/2 , F = 3, m = 3 |2 P3/2 , F = 4, m = 4 cycling transition. The amplitude of secular motion (corresponding to the motional temperature) is extracted from the ratio of amplitudes for the red- and blue-sidebands of a Raman transition for all six normal modes [14]. Measured values are shown in Table II, together with values calculated from laser-cooling theory for a single ion [20]. The single-Mg+ -ion Doppler cooling limit is also valid when applied to each of the six normal modes of the Al-Mg ion pair. The measured and calculated motional quantum numbers agree within the measurement uncertainty, and we consider the stated 30 % uncertainty for this shift to be a conservative limit. The Mg+ Doppler cooling laser beam maintains the

3 Al-Mg ion pair at a constant motional temperature during the clock interrogation pulse, but because it also overlaps the Al+ ion, it causes an AC Stark shift by coupling o-resonantly to allowed transitions that connect to the ground (3s2 )1 S0 and excited (3s3p)3 P0 clock states. Following the evaluation of the blackbody radiation shift [10], we estimate the dierential clock polarizability at 279.5 nm as / = (3.5 0.6) 1017 S, where (S = I/IS ) is the saturation parameter for 25 Mg+ (IS 2470 W/m2 ). The intensity I of the Mg+ Doppler cooling laser is estimated from the rate at which this laser repumps the |2 S1/2 , F = 2, m = 2 dark hyperne ground state. The ion uorescence photo-multiplier counts F (t) collected in a duration t due to repumping of the dark state may be written as F (t) = b(t+ (et/ 1)) where = (0.217/S) ms and b is the bright-state counting rate. We extract S by tting F (t) to the observed ion uorescence. Typically, we measure = 2.1 0.8 ms, and nd / = (3.6 1.5) 1018 . Another Stark shift is caused by thermal blackbody radiation. The temperature of the Al-Mg ion trap is measured with two platinum sensors at opposite ends of the trap structure which are expected to be at temperature extremes. Heat is removed primarily through thermal conduction at one end of the trap where we measure a temperature of 35 C. At the higher thermal resistance trap-end the temperature is 40 C, and the thermal radiation eld impinging upon the ion is bounded by this maximum temperature and the laboratory room temperature of 22 C: Tion = (31 9) C. During operation, the Al+ clock servo alternates between probing of the |1 S0 , mF = 5/2 |3 P0 , mF = 5/2 and |1 S0 , mF = 5/2 |3 P0 , mF = 5/2 transitions every 5 s, and the apparatus synthesizes an average of these two frequencies to eliminate rst-order Zeeman shifts [9, 21]. Each transitions resonance is probed several times at the high- and low-frequency half-maximum points to derive a frequency-correction signal. The frequency-dierence between the transitions is proportional to the mean magnetic eld B , which allows an accurate estimate of the quadratic Zeeman shift due to the quasi-static quantization eld of typically B = 0.1 mT. However, the quadratic Zeeman shift is proportional 2 2 to B 2 = B 2 + BAC , where BAC is the variance of the magnetic eld about its mean. The dominant sources of varying magnetic elds are currents at RF in conductors near the ion. We vary the trap RF drive power P and measure the frequency of the hyperne clock transition in 25 Mg+ |F = 3, mF = 0 |F = 2, mF = 0 near 1.789 GHz, which has a strong quadratic dependence on 2 the magnetic eld, to nd BAC = 1.451012(P/W) T2 . + 2 For Al clock operation P = 15 W, and BAC = 2.17 11 2 10 T , which alters the quadratic Zeeman shift by / = (1.4 0.3) 1018 . Other potential systematic shifts are listed in Table I. When stabilizing the clock laser to the ion we probe
1015
/ 2.8x10 Nsample std. dev. Overl. Allan dev.
15 1/2

1016

10

100 [seconds]

1 000

10 000

FIG. 2: Clock stability. Fractional frequency uncertainty vs. averaging period ( ) for a comparison between the two Al+ clocks (10700 s duration). Overlapping Allan deviation and N-sample standard deviation are shown [22]. For each comparison measurement the coecient of the 1/2 asymptote is estimated and used to derive the measurements statistical uncertainty. The 2.8 1015 1/2 asymptote is reached for averaging periods that are longer than the servo time constant of 10 s.

the clock transition alternately with counter-propagating laser beams to observe and cancel potential rst-order Doppler shifts [5]. For the clock comparison described below, we observe a dierential shift for the two probe directions of (1.2 0.7) 1017 but this eect is suppressed by taking the average. The suppression factor is limited because the atomic line-shape and hence the servo gains dier slightly for the two probe directions. During clock operation the mean fractional gain imbalance was 1.5 %, thereby reducing the possible rst order Doppler shift to 3 1019 . We have looked for Stark shifts due to the clock pulse itself by raising the intensity of the probe beam in one clock, and comparing the frequency to that produced by the other Al+ clock. With an increase of 40 dB in the clock probe intensity, we observe no statistically significant frequency dierence with a fractional uncertainty of 2 1015 , constraining any eect that scales linearly with this intensity to 2 1019 . We observe a rate of background gas collisions similar to the Al-Be clock and assign a 5 1019 uncertainty to this potential shift [5]. Thermally induced frequency errors from beam switching AOMs (Fig. 1) have been evaluated previously [5], and this uncertainty is reduced to 2 1019 by operating the AOMs at less than 1 mW. We directly compared the Al-Mg clock with the AlBe clock to perform independent tests that could reveal unaccounted-for clock shifts. The Al-Be clock was evaluated with an accuracy of 2.3 1017 [5], and this evaluation remained valid during the two-clock comparison. The two clocks were compared in 56 separate measurements, each of duration 1000 11000 s. As shown in Fig. 1, the frequency of the 1 S0 3 P0 probe laser was actively steered to the Al+ ion in the Al-Mg apparatus

4 Future frequency ratio measurements of the Al+ clock and the NIST Hg+ optical clock would enable improved constraints on present-era changes in the ne-structure constant [5]. We thank J. C. Bergquist for use of his stable FabryPerot cavities, T. M. Fortier for optical frequency measurements, W. M. Itano for atomic structure calculations, and S. Bickman and D. Leibrandt for helpful comments on the manuscript. This work was supported by ONR. J.C.J.K. acknowledges support from the Netherlands Organisation for Scientic Research (NWO). Contribution of NIST, not subject to U.S. copyright.

1016 0 1016
0 10 20 measurement # 40 30 50

FIG. 3: Measurements of the fractional frequency dierence between the two Al+ clocks (blue points). Error bars represent the statistical uncertainty (see Fig. 2). The horizontal line shows the weighted mean of 1.8 1017 with an overall statistical uncertainty of 7.0 1018 (shaded band).

with a servo time-constant of about 10 s. A portion of this laser light simultaneously probed the Al+ ion in the Al-Be clock, where it was servoed to the clock transition in a separate digital feedback loop. The frequency produced by this feedback loop represents the frequency dierence between the two clocks, and was recorded and analyzed for stability (Fig. 2). Average frequencies of the individual measurements corrected for known shifts are shown in Fig. 3, where the overall weighted mean is (AlMg AlBe )/ = (1.8 0.7) 1017 . This value is consistent with the 1- error of 2.5 1017 that is calculated by adding in quadrature the inaccuracies of the two clocks and the statistical uncertainty. The reduced-2 for this data set is 1.02, indicating that the error bars derived from the estimated stability correctly capture the scatter of the data. For this data set the total in-loop servo error for the Al-Mg clock was / = 6 1019 , and for a well-designed servo loop this error declines faster than the statistical error. It is therefore not included as a systematic shift in Table I. In summary, we have built an Al+ ion clock with a fractional frequency inaccuracy of 8.6 1018 . Its frequency is compared to that of a previously constructed Al+ clock, and the measured fractional frequency dierence of (1.8 0.7) 1017 is consistent with the inaccuracy of the previous clock. Signicantly, the statistical uncertainty in the frequency comparison of 7.0 1018 is smaller than the inaccuracy of either clock, and the average measurement stability was 2.8 1015 1/2 (total measurement duration: 164 967 s). This result may be compared to other direct same-species atomic clock comparisons, where two Cs microwave clocks have reached an agreement of (4 3) 1016 [23], and two Yb+ single ion clocks showed agreement of (3.8 6.1) 1016 [24].

[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15]

[16] [17] [18] [19] [20] [21] [22] [23] [24]

chinwen@[Link] Present address: Laser Centre Vrije Universiteit Amsterdam, The Netherlands H. G. Dehmelt, IEEE Trans. Inst. Meas. 31, 83 (1982). D. J. Wineland, W. M. Itano, J. C. Bergquist, and R. G. Hulet, Phys. Rev. A 36, 2220 (1987). B. C. Young, F. C. Cruz, W. M. Itano, and J. C. Bergquist, Phys. Rev. Lett. 82, 3799 (1999). W. H. Oskay et al., Phys. Rev. Lett. 97, 020801 (2006). T. Rosenband et al. Science 319, 1808 (2008). T. E. Parker, in Proc. 2008 EFTF Conf. (2008). G. K. Campbell et al. Metrologia 45, 539 (2008). N. Yu, H. Dehmelt, and W. Nagourney, in Proc. Nat. Acad. Sci (1992) p. 7289. T. Rosenband et al. Phys. Rev. Lett. 98, 220801 (2007). T. Rosenband et al. arXiv:physics/0611125 (2006). J. Mitroy, J. Y. Zhang, M. W. J. Bromley, and K. G. Rollin, European Physical Journal D 53, 15 (2009). P. O. Schmidt et al. Science 309, 749 (2005). M. A. Rowe et al. Quant. Inform. Comp. 2, 257 (2002). C. Monroe et al. Phys. Rev. Lett. 75, 4011 (1995). The entropy-reducing laser that repumps the states |2 S1/2 , F {2, 3}, m = 2 via |2 P1/2 , F = 3, m = 3 into |2 S1/2 , F = 3, m = 3 is replaced by the repeated application of a pulse sequence that rst drives the |2 S1/2 , F = 2, m = 2 |2 P3/2 , F = 3, m = 3 transition and subsequently the transition |2 S1/2 , F = 3, m = 2 |2 S1/2 , F = 2, m = 2 . D. B. Hume, T. Rosenband, and D. J. Wineland, Phys. Rev. Lett. 99, 120502 (2007). G. S. Hurst, M. G. Payne, S. D. Kramer, and J. P. Young, Rev. Mod. Phys. 51, 767 (1979). N. Kjaergaard et al. Appl. Phys. B 71, 207 (2000). D. J. Berkeland et al. J. Appl. Phys. 83, 5025 (1998). W. M. Itano and D. J. Wineland, Phys. Rev. A 25, 35 (1982). J. E. Bernard, L. Marmet, and A. A. Madej, Opt. Comm. 150, 170 (1998) . W. J. Riley, NIST Special Publication 1065 (2008). C. Vian et al. IEEE Trans. Inst. Meas. 54, 833 (2005) . T. Schneider, E. Peik, and C. Tamm, Phys. Rev. Lett. 94, 230801 (2005).