Bothwell, T., Kennedy, C. J., Aeppli, A., Kedar, D., Robinson, J. M., Oelker, E., Staron, A. and Ye, J. (2022) Resolving the gravitational redshift across a millimetre-scale atomic sample. Nature, 620, pp. 420-424. (doi: 10.1038/s41586-021-04349-7) This is the Author Accepted Manuscript. There may be differences between this version and the published version. You are advised to consult the publisher’s version if you wish to cite from it. https://eprints.gla.ac.uk/264685/ Deposited on: 4 February 2022 Enlighten – Research publications by members of the University of Glasgow http://eprints.gla.ac.uk 1 Resolving the gravitational redshift within a millimeter atomic 2 sample 3 Tobias Bothwell1,*, Colin J. Kennedy1,2, Alexander Aeppli1, Dhruv Kedar1, John M. Robinson1, 4 Eric Oelker1,3, Alexander Staron1, and Jun Ye1,* 5 1JILA, National Institute of Standards and Technology and University of Colorado 6 Department of Physics, University of Colorado 7 Boulder, Colorado 80309-0440, USA 8 2Present Address: Honeywell Quantum Solutions, Broomfield, Colorado 80021 9 3Present Address: Physics Department, University of Glasgow 10 11 Einstein’s theory of general relativity states that clocks at different gravitational 12 potentials tick at different rates – an effect known as the gravitational redshift1. As 13 fundamental probes of space and time, atomic clocks have long served to test this 14 prediction at distance scales from 30 centimeters to thousands of kilometers2–4. 15 Ultimately, clocks will study the union of general relativity and quantum mechanics 16 once they become sensitive to the finite wavefunction of quantum objects oscillating in 17 curved spacetime. Towards this regime, we measure a linear frequency gradient 18 consistent with the gravitational redshift within a single millimeter scale sample of 19 ultracold strontium. Our result is enabled by improving the fractional frequency 20 measurement uncertainty by more than a factor of 10, now reaching 7.6×10-21. This 21 heralds a new regime of clock operation necessitating intra-sample corrections for 22 gravitational perturbations. 23 Modern atomic clocks embody Arthur Schawlow's motto to "never measure anything but 24 frequency." This deceptively simple principle, fueled by the innovative development of laser 25 science and quantum technologies based on ultracold matter, has led to dramatic progress in 26 clock performance. Recently, clock measurement precision reached the mid-19th digit in one 1 27 hour5,6, and three atomic species achieved systematic uncertainties corresponding to an error 28 equivalent to less than 1 s over the lifetime of the Universe7–10. Central to this success in 29 neutral atom clocks is the ability to maintain extended quantum coherence times while using 30 large ensembles of atoms5,6,11. The pace of progress has yet to slow. Continued improvement 31 in measurement precision and accuracy arising from the confluence of metrology and quantum 32 information science12–15 promises discoveries in fundamental physics16–20. 33 Clocks fundamentally connect space and time, providing exquisite tests of the theory of 34 general relativity. Hafele and Keating took cesium-beam atomic clocks aboard commercial 35 airliners in 1971, observing differences between flight-based and ground-based clocks 36 consistent with special and general relativity21. More recently, RIKEN researchers compared 37 two strontium optical lattice clocks (OLCs) separated by 450 m in the Tokyo Skytree, resulting 38 in the most precise terrestrial redshift measurement to date22. Proposed satellite-based 39 measurements23,24 will provide orders of magnitude improvement to current bounds on 40 gravitational redshifts3,4. Concurrently, clocks are anticipated to begin playing important roles 41 for relativistic geodesy25. In 2010 Chou et al.2 demonstrated the precision of their Al+ clocks by 42 measuring the gravitational redshift resulting from lifting one clock vertically by 30 cm in 40 43 hours. With a decade of advancements, today’s leading clocks are poised to enable local 44 geodetic surveys of elevation at the sub-centimeter level on Earth, complimenting spatial 45 averaging techniques26. 46 Atomic clocks strive to simultaneously optimize measurement precision and systematic 47 uncertainty. For traditional OLCs operated with one-dimensional (1D) optical lattices, achieving 48 low instability has involved the use of high atom numbers at trap depths sufficiently large to 49 suppress tunnelling between neighboring lattice sites. While impressive performance has been 2 50 achieved, effects arising from atomic interactions and AC Stark shifts associated with the 51 trapping light challenge advancements in OLCs. Here we report a new operational regime for 52 1D OLCs, both resolving the gravitational redshift across our atomic sample and 53 synchronously measuring a fractional frequency uncertainty of 7.6×10-21 between two 54 uncorrelated regions. Our system employs ~100,000 87Sr atoms at ~100 nK loaded into a 55 shallow, large waist optical lattice, reducing both AC Stark and density shifts. Motivated by our 56 earlier work on spin-orbit coupled lattice clocks27,28, we engineer atomic interactions by 57 operating at a ‘magic’ trap depth, effectively removing collisional frequency shifts. These 58 advances enable record optical atomic coherence (37 s) and expected single clock stability 59 (3.1×10-18 at 1 s) using macroscopic samples, paving the way toward lifetime limited OLC 60 operation. 61 Central to our experiment is an in-vacuum optical cavity (Fig. 1a and Methods) for 62 power enhancement of the optical lattice. The cavity (finesse 1300) ensures wavefront 63 homogeneity of our 1D lattice while the large beam waist (260 μm) reduces the atomic density 64 by an order of magnitude compared to our previous system10. We begin each experiment by 65 trapping fermionic 87Sr atoms into the 1D lattice at a trap depth of 300 lattice photon recoil 66 energies (Erec), loading a millimeter scale atomic sample (Fig. 1a). Atoms are simultaneously 67 cooled and polarized into a single nuclear spin before the lattice is adiabatically ramped to an 68 operational depth of 12 Erec. Clock interrogation proceeds by probing the ultranarrow 1S0 ( )→ 69 3P0 ( ) transition with the resulting excitation fraction measured by fluorescence spectroscopy. 70 Scattered photons are collected on a camera, enabling in-situ measurement with 6 μm 71 resolution, corresponding to ~15 lattice sites. 3 72 Quantum state control has been vital to recent advances in atom-atom and atom-light 73 coherence times in 3D OLCs and tweezer clocks5,11,15. Improved quantum state control is 74 demonstrated through precision spectroscopy of the Wannier-Stark states of the OLC29,30. The 75 1D lattice oriented along gravity has the degeneracy of neighboring lattice sites lifted by the 76 gravitational potential energy. In the limit of shallow lattice depths, this creates a set of 77 delocalized states. By ramping the lattice depth to 6 Erec, much lower than in traditional 1D 78 lattice operations7,8,10, clock spectroscopy probes this delocalization (Fig. 1d). The ability to 79 engineer the extent of atomic wavefunctions through the adjustment of trap depth creates an 80 opportunity to control the balance of on-site p-wave versus neighboring-site s-wave atomic 81 interactions. We utilize this tunability by operating at a ‘magic’ trap depth31, where the 82 frequency shifts arising from on-site and off-site atomic interactions cancel, enabling a 83 reduction of the collisional frequency shifts by more than three orders of magnitude compared 84 our previous work10. 85 Extended atomic coherence times are critical for both accuracy and precision. An 86 aspirational milestone for clock measurement precision is the ability to coherently interrogate 87 atomic samples up to the excited state’s natural lifetime. To evaluate the limits of our clock’s 88 atomic coherence, we perform Ramsey spectroscopy to measure the decay of fringe contrast 89 as a function of the free-evolution time. By comparing two uncorrelated regions within our 90 atomic sample, we determine the contrast and relative phase difference between the two sub- 91 ensembles (Fig. 2). The contrast decays exponentially with a time constant of 37 s (quality 92 factor of 3.6×1016), corresponding to an additional decoherence time of 53 s relative to the 3P0 93 natural lifetime (118 s)32. This represents the longest optical atomic coherence time measured 94 in any spectroscopy system to date. 4 95 We utilize Rabi spectroscopy in conjunction with in-situ imaging to microscopically 96 probe clock transition frequencies along the entire vertically oriented atomic ensemble. With a 97 standard interleaved probing sequence using the to transitions for 98 minimal magnetic sensitivity, we reject the first order Zeeman shifts and vector AC Stark shifts. 99 The in-situ imaging of atoms in the lattice allows measurement of unprocessed frequencies 100 across the entire atomic sample (Fig. 1a and Methods). The dominant differential perturbations 101 arise from atom-atom interactions (residual density shift contributions after we operate at the 102 ‘magic’ trap depth) and magnetic field gradients giving rise to pixel-specific 2nd order Zeeman 103 shifts. Using the total camera counts and mF-dependent frequency splitting, we correct the 104 density and 2nd order Zeeman shift at each pixel. These corrections result in the processed 105 frequencies per pixel shown in Fig. 3a, with error bars representing the quadrature sum of 106 statistical uncertainties from the center frequency, the density shift correction, and the 2nd order 107 Zeeman shift correction. Additional systematics are described in the Methods. This approach 108 demonstrates an efficient method for rapid and accurate evaluation of various systematic 109 effects throughout a single atomic ensemble. Unlike traditional 1D OLCs where systematic 110 uncertainties are quoted as global parameters, we now microscopically characterize these 111 effects. 112 This new microscopic in-situ imaging allows determination of the gravitational redshift 113 within a single atomic sample, probing an uncharacterized fundamental clock systematic. Two 114 identical clocks on the surface of a planet separated by a vertical distance will differ in 115 frequency ( as given by 116 (Eq. 1), 5 117 with the clock frequency, the speed of light, and the gravitational acceleration. The 118 gravitational redshift at Earth’s surface corresponds to a fractional frequency gradient of -1.09× 119 10-19/mm in the coordinate system of Fig. 1a. Measurement of a vertical gradient across the 120 atomic sample consistent with the gravitational redshift provides an exquisite verification of an 121 individual atomic clock’s frequency control. 122 Our intra-cloud frequency map (Fig. 3a) allows us to evaluate gradients across the 123 atomic sample. Over 10 days we performed 14 measurements (ranging in duration from 1-17 124 hours) to search for the gravitational redshift across our sample. For each we fit a linear slope 125 and offset after taking into account density shift and 2nd order Zeeman corrections, reporting 126 the slope in Fig. 3b. From this measurement campaign we find the weighted mean (standard 127 error of the weighted mean) of the frequency gradient in our system to be -1.00(12)×10-19/mm. 128 We evaluate additional differential systematics (see Methods) and find a final frequency 129 gradient of -9.8(2.3)×10-20/mm, consistent with the predicted redshift. 130 The ability to resolve the gravitational redshift within our system suggests a level of 131 frequency control beyond previous clock demonstrations, vital for the continued advancement 132 of clock accuracy and precision. Previous fractional frequency comparisons15 have reached 133 uncertainties as low as 4.2×10-19. Similarly, we perform a synchronous comparison between 134 two uncorrelated regions of our atomic cloud (Fig. 4a). By binning ~100 pixels per region, we 135 substantially reduce instability caused by quantum projection noise33. Analyzing the frequency 136 difference between regions from 92 hours of data, we find a fractional frequency instability of 137 4.4×10-18/ ( is the averaging time in seconds), resulting in a fractional frequency uncertainty 138 of 7.6×10-21 for full measurement time, nearly two orders of magnitude lower than the previous 139 record. From this measurement we infer a single region instability of 3.1×10-18/ . Dividing the 6 140 fractional frequency difference by the spatial separation between each region’s center of mass 141 gives a frequency gradient of -1.30(18)×10-19/mm. Correcting for additional systematics as 142 before results in a gradient of -1.28(27)×10-19/mm, again fully consistent with the predicted 143 redshift. 144 In conclusion, we have established a new paradigm for atomic clocks. The vastly 145 improved atomic coherence and frequency homogeneity throughout our sample allow us to 146 resolve the gravitational redshift at the submillimeter scale, observing for the first time the 147 frequency gradient from gravity within a single sample. We demonstrate a synchronous clock 148 comparison between two uncorrelated regions with a fractional frequency uncertainty of 149 7.6×10-21, advancing precision by nearly two orders of magnitude. These results suggest that 150 there are no fundamental limitations to inter-clock comparisons reaching frequency 151 uncertainties at the 10-21 level, offering new opportunities for tests of fundamental physics. 7 152 153 Fig. 1: Experimental system and quantum state control. a, A millimeter length sample of 154 ~100,000 87Sr atoms are trapped in a 1D optical lattice formed within an in-vacuum cavity. The 155 longitudinal axis of the cavity, z, is oriented along gravity. We probe atoms along the 1S0 → 3P0 156 transition using a clock laser locked to an ultrastable crystalline silicon cavity6,34. b, Rabi 157 spectroscopy with a 3.1 s pulse time. Open purple circles indicate data with a corresponding 158 Rabi fit in green. c, Neighboring lattice sites are detuned by gravity, creating a Wannier-Stark 159 ladder. Clock spectroscopy probes the overlap of Wannier-Stark states between lattice sites 160 that are m sites away with Rabi frequency Ωm. d, Rabi spectroscopy probes Wannier-Stark 161 state transitions, revealing wavefunction delocalization of up to 5 lattice sites. The number of 162 lattice sites is indicated above each transition, with blue(red) denoting Wannier-Stark 163 transitions to higher(lower) lattice sites. 8 164 165 166 Fig. 2: Atomic coherence. We use Ramsey spectroscopy with a randomly sampled phase for 167 the second pulse to determine the coherence time of our system11. a, We measure the 168 excitation fraction across the cloud, shown in purple for a single measurement, and calculate 169 the average excitation fractions in regions p1 and p2, separated by 2 pixels. b, Parametric plots 170 of the excitation fraction of p1 versus p2 in purple for 6 s, 30 s and 50 s dark time demonstrate 171 a phase shift between the two regions and contrast decay. Using a maximum likelihood 172 estimator, we extract the phase and contrast for each dark time with the fit, shown in green. c, 173 Contrast decay as a function of time in green is fit with an exponential decay in gold, giving an 174 atomic coherence decay time of 36.5(0.7) s and a corresponding quality factor of 3.6×1016. 175 After accounting for the finite radiative decay contribution, we infer an additional decoherence 176 time constant of 52.8(1.5) s. 177 178 9 179 180 Figure 3: Evaluating frequency gradients. a, For each measurement we construct a 181 microscopic frequency map across the sample, with raw frequencies shown in green. The 2nd 182 order Zeeman correction is shown as a dashed gold line. Processed frequencies shown in 183 purple include both density shift corrections and 2nd order Zeeman corrections, with 184 uncertainties arising from the quadrature sum of statistical, density shift correction, and 2nd 185 order Zeeman correction uncertainties. To this we fit a linear function, shown in black. b, Over 186 the course of 10 days, we completed 14 measurements. For each measurement, we create a 187 corrected frequency map and fit a linear slope as in a. This slope is plotted for each 188 measurement, as well as a weighted mean (black) with associated statistical uncertainty 189 (dashed black) and total uncertainty as reported in Table 1 (dotted black). The expected 190 gravitational gradient is shown in red. All data is taken with Rabi spectroscopy using a 3.1 s π191 pulse time except for 08/13 which used a 3.0 s pulse time. The reduced chi-square statistic is 192 3.0, indicating a small underestimation of error variances entirely consistent with the additional 193 systematic uncertainties in Table 1. 194 195 196 10 197 198 Figure 4: In-situ synchronous clock comparison. a, The cloud is separated as in Fig. 2a. 199 The gravitational redshift leads to the higher clock(blue) ticking faster than the lower one(red). 200 The length scale is in millimeters. b, Allan deviation of the frequency difference between the 201 two regions in a over 92 hours. Purple points show fractional frequency instability fit by the 202 solid green line, with the quantum projection noise limit indicated by the dashed black line. We 203 attribute the excess instability of the measurement relative to QPN to detection noise. The 204 expected single atomic region instability is shown in gold. 205 206 207 208 209 11 210 Methods 211 In-Vacuum Cavity 212 Central to our system is an in-vacuum lattice buildup cavity oriented along gravity (Fig. 213 1a). Two mirrors with radius of curvature of 1 m are separated by ~15 cm, achieving a mode 214 waist of 260 μm. Our over-coupled cavity has a finesse at the lattice wavelength (813 nm) of 215 ~1300 and a power buildup factor of ~700 (ratio of circulating to input intensity). This enables 216 lattice depths in excess of 500 Erec (lattice photon recoil energy) using a diode-based laser 217 system. The dimensional stability of the cavity combined with the simplified diode laser system 218 enables robust operation compared with our previous Ti:Sapphire retro-reflected design10. The 219 cavity mirrors are anti-reflection coated at the clock wavelength of 698 nm. 220 One cavity mirror is mounted to a piezo for length stabilization while the other mirror is 221 rigidly mounted for phase reference for the clock laser. Grounded copper shields between 222 atoms and mirrors prevent DC Stark induced shifts due to charge buildup on the mirrors and 223 piezo35,36. Each shield (5 mm thick) has a centered hole of 6 mm diameter to accommodate 224 the optical lattice beam, with shielding performance verified through evaluation of the DC Stark 225 shift systematic. 226 Atomic sample preparation 227 87Sr atoms are cooled and loaded into a 300 Erec optical lattice using standard two stage 228 magneto-optical trapping techniques10. Once trapped, atoms are simultaneously nuclear spin 229 polarized, axially sideband cooled, and radially doppler cooled into a single nuclear spin state 230 at temperatures of 800 nK. The lattice is then adiabatically ramped to the operational trap 231 depth of 12 Erec, where a series of pulses addressing the clock transition prepares atoms into 12 232 . Clock spectroscopy is performed by interrogating the to 233 transition, the most magnetically insensitive 87Sr clock transition6. 234 Imaging 235 The clock excitation fraction is read out using standard fluorescence spectroscopy 236 techniques6,10,37. Photons are collected on both a photo-multiplier tube for global readout and 237 electron multiplying charge coupled device camera for an in-situ readout of clock frequency. 238 Camera readout is performed in full vertical binning mode, averaging the radial dimension of 239 the lattice. This provides 1D in-situ imaging for all synchronous evaluations. 240 We use a 25 μs fluorescence probe with an intensity of I/Isat ~ 20 (Isat being the 241 saturation intensity), ensuring uniform scattering across the atomic sample. Before imaging, 242 the optical lattice is ramped back to 300 Erec to decrease imaging aberration resulting from the 243 extended radial dimension at 12 Erec. 244 Analysis 245 Standard clock lock techniques and analysis are used6,11,33, with differences in 246 excitation fraction converted to frequency differences using Rabi lineshapes. Each dataset is 247 composed of a series of clock locks, tracking the center of mass frequency of the atomic 248 sample. A clock lock is four measurements probing alternating sides of the Rabi lineshape for 249 opposite nuclear spin transitions. Frequency corrections based on excitation fraction become 250 ambiguous when the excitation fraction measured is consistent with the Rabi lineshape at 251 multiple detunings. To avoid erroneous frequency corrections, we remove clock locks with 252 excitation fractions above (below) .903×C (.116×C), where C is the Rabi contrast. From each 13 253 clock lock, a pixel specific center frequency fi and frequency splitting between opposite mF 254 states Δi are calculated, creating an in-situ frequency map of the 1D atomic sample. This 255 allows rejection of vector shifts on a pixel-by-pixel basis and probes the magnetic field induced 256 splitting of mF transitions. The atom weighted mean frequency is subtracted from every lock 257 cycle to reject common mode laser noise. 258 For each dataset we approximate the atomic profile with a Gaussian fit, identifying a 259 center pixel and associated Gaussian width (σ). All analysis is performed within the central 260 region of ±1.5σ which demonstrates the lowest frequency instability. Identifying a center pixel 261 for data processing ensures rejection of any day-to-day drift in the position of the cloud due to 262 varying magnetic fields modifying MOT operation on the narrow line transition. The density 263 shift coefficient (see Density Shift section) is derived from the average center frequency per 264 pixel. Using this coefficient, we correct fi and Δi for the density shift. 2nd order Zeeman 265 corrections using these updated frequencies are then applied. 266 Gradient analysis is based on the processed center frequencies per pixel. A linear fit to 267 the frequencies as a function of pixel is performed using least squares, with uncertainty per 268 pixel arising from the quadrature sum of statistical frequency uncertainty, statistical 2nd order 269 Zeeman uncertainty, and density shift correction uncertainty. 270 For the two-clock comparison (Fig. 4b), all data from 8/14-8/22 was taken with the same 271 duty cycle and π-pulse time (3.1 s). Data was processed relative to a fit center pixel as 272 discussed and concatenated. Two equal regions extend from the center of the sample to a 273 width of ±1.5σ, with two empty pixels between regions to ensure uncorrelated samples. Each 274 region is processed for the atom weighted mean frequency, enabling a synchronous frequency 275 comparison between two independent clocks. 14 276 Atomic Coherence 277 We use a Ramsey sequence to measure the atomic coherence. We prepare a sample 278 in the state and apply a π/2 pulse along the to 279 transition. After waiting for a variable dark time, we apply a second π/2 pulse with a random 280 phase relative to the first. We then measure the excitation fraction. 281 Two regions, p1 and p2, are identified using the same technique as in the synchronous 282 instability measurement. For each experimental sequence, we find the average excitation 283 fraction in p1 and p2. A mean frequency shift across the sample primarily due to a magnetic 284 field gradient creates a differential phase as a function of time between p1 and p2. We create a 285 parametric plot of the average excitation in p1 and p2 for each dark time and use a maximum 286 likelihood estimator to fit an ellipse to each dataset, calculating phase and contrast15,38. To 287 estimate uncertainty in the contrast for each dark time a bootstrapping technique is used11. 288 Fitting the contrast as a function of dark time with a single exponential returns an effective 289 atomic coherence time. 290 Systematics 291 Imaging 292 We calibrate our pixel size using standard time of flight methods: we observe an atomic 293 sample in freefall for varying times to determine an effective pixel size along the direction of 294 gravity. Immediately after our 10-day data campaign we measured our effective pixel size to be 295 6.04 μm. Due to thermal drift of our system, the pixel size can vary by up to 0.5 μm /pixel over 296 months which we take as the calibration uncertainty. 15 297 Spatial correlations may limit imaging resolution. We measure these correlations by 298 placing atoms into a superposition of clock states. Any measured spatial correlation is due to 299 the imaging procedure. In our system we find no correlations between neighboring pixels11. 300 The optical resolution of our imaging lens is specified at 2 μm. 301 Lattice tilt from gravity will modify the measured gradient. We find the lattice tilt in the 302 imaging plane to be 0.11(0.06) degrees, providing an uncertainty orders of magnitude smaller 303 than the pixel size uncertainty. We are insensitive to lattice tilt out of the imaging plane. 304 Zeeman Shifts 305 First order Zeeman shifts are rejected by probing opposite nuclear spin states6. The 2nd 306 order Zeeman shift is given by ( ) 307 where is the splitting between opposite spin states and the corresponding 2nd order 308 Zeeman shift coefficient. For stretched spin state operation ( ), =-2.456(3)×10-7 Hz-1. 309 Using known atomic coefficients39 we find the 2nd order Zeeman coefficient for the 310 to transition to be =-1.23(8)×10-4 Hz, with the uncertainty arising from 311 limited knowledge of atomic coefficients. 312 The 2nd Order Zeeman corrections are made for every clock lock (analogous to the in- 313 situ density shift corrections). For a typical day (8/13) the average 2nd order Zeeman gradient 314 is -7.0×10-20/mm, corresponding to a splitting between opposite nuclear spin states of 12.7 315 mHz/mm (0.291 mG/mm). We include an error of 4×10-21/mm in Table 1 to account for the 316 atomic uncertainty in the shift coefficient. 16 317 DC Stark 318 Electric fields perturb the clock frequency via the DC Stark effect. We evaluate 319 gradients arising from this shift by using in-vacuum quadrant electrodes to apply bias electric 320 fields in all three dimensions. We find a DC Stark gradient of 3(2)×10-21/mm. 321 Black Body Radiation Shift 322 The dominant frequency perturbation to room temperature neutral atom clocks is black 323 body radiation (BBR). Similar to our previous work10, we homogenize this shift by carefully 324 controlling the thermal surroundings of our vacuum chamber. Attached to the vacuum chamber 325 are additional temperature control loops, with each vacuum viewport having a dedicated 326 temperature control system. This ensures our dominant BBR contribution – high emissivity 327 glass viewports – are all the same temperature to within 100 mK. 328 To bound possible BBR gradients, we introduce a 1 K gradient between the top and 329 bottom of the chamber along the cavity axis by raising either the top or bottom viewports by 1 330 K. We compare these two cases and find no statistically significant changes in the frequency 331 gradient across the entire sample. Accounting for uncertainty in linear frequency fits for each 332 case, we estimate an uncertainty of 3×10-21/mm. This finding is supported with a basic thermal 333 model of the vacuum chamber. 334 Density Shift 335 Atomic interactions during Rabi spectroscopy lead to clock frequency shifts as a 336 function of atomic density40. For each gradient measurement, we evaluate the density shift 337 coefficient by fitting the average frequency per pixel versus average camera counts per 338 pixel to an equation of the form 17 339 ( . 340 Here B is an arbitrary offset. Once is known, we remove the density shift at each pixel. 341 Residual density shift corrections may lead to error in our linear gradient. To bound this 342 effect, we compare the density shift coefficient and gradient from our data run with a separate 343 dataset at 8 Erec. With the trap depth at 8 Erec we found a linear gradient of s=-1.08×10-18/mm 344 and a density shift coefficient of =-1.39×10-6 Hz/count. During our data run we had an 345 average density shift coefficient of =-2.43×10-8 Hz/count. We bound the uncertainty in our 346 gradient from density shift as |=1.7×10-20/mm. 347 Lattice Light Shifts 348 Lattice light shifts arise from differential AC Stark shifts between the ground and excited 349 clock states. An approximate microscopic model of the lattice light shift ( ) in our system is 350 given by41 351 ( ( ) * + , 352 where is the trap depth in units of Erec , the differential electric dipole polarizability, 353 the differential multi-polarizability, and ( the detuning between lattice 354 frequency and effective magic frequency . Our model has no dependence on the 355 longitudinal vibrational quanta since we are in the ground vibrational band. We neglect higher 356 order corrections from hyperpolarizability due to our operation at depths <60 Erec. At our 357 temperatures thermal averaging of the trap depth is a higher order correction (<5%) that is also 358 neglected. 18 359 We model the linear differential lattice light shift across the atomic cloud as 360 ( ( [ ) ] , 361 where z is the coordinate corresponding to the axis of the cavity along gravity. To evaluate our 362 differential lattice light shift at our operational depth we need and . We modulate our 363 lattice between two trap depths (u1=14 Erec, u2=56 Erec) and find our detuning from scalar 364 magic frequency to be =7.4(0.6) MHz. To evaluate at our operational depth ( we 365 measure the linear gradient across the atomic cloud at + 250 MHz and - 250 MHz, the 366 difference given by 367 ( ) ( ) [ ] , 368 where = 500 MHz. We find 0.0383/mm, which when combined with =7.4 MHz, 369 gives us a fractional frequency gradient of -5×10-21/mm. Accounting for error in our lattice 370 detuning and linear gradient gives us an uncertainty of 1×10-21/mm. 371 Other Systematics 372 For a 3.1 s π-pulse the probe AC Stark shift7 is -3(2)×10-21. A frequency scan of the 373 to transition limits the variation of excitation fraction across the atomic 374 sample to 1% or below, bounding any possible probe AC Stark gradient across the sample to 375 <1×10-22. 376 Known Redshift 19 377 The gravitational acceleration (rounding to 4 digits) within our lab was evaluated by a 378 USGS survey42 to be a=-9.796 m/s2. 379 Systematic Budget Systematic Gradient (Fig. 3) BBR Density Lattice light shift DC Stark Pixel Calibration 2nd Order Zeeman Other Corrected Gradient Known Redshift 380 Slope (10-20/mm) -10.0 0 -0.5 0.3 0 0 -9.8 -10.9 Uncertainty (10-20/mm) 1.2 0.3 1.7 0.1 0.2 .8 .4 <.1 2.3 <.1 381 Table 1: Gradient Systematic Budget. Fractional frequency gradients and corresponding 382 uncertainties. Fractional frequencies denoted with ‘-‘ are corrected on a pixel-by-pixel basis 383 during initial data processing (Fig. 3a). The corrected gradient has known systematics 384 removed with uncertainty given by the quadrature sum of all correction uncertainties. 385 386 Acknowledgements 20 387 We acknowledge funding support from Defense Advanced Research Projects Agency, 388 National Science Foundation QLCI OMA-2016244, DOE Quantum System Accelerator, NIST, 389 NSF Phys-1734006, and Air Force Office for Scientific Research. We are grateful for theory 390 insight from Anjun Chu, Peiru He, and Ana María Rey. We acknowledge stimulating discussion 391 and technical contributions from John Zaris, James Uhrich, Josephine Meyer, Ross Hutson, 392 Christian Sanner, William Milner, Lindsay Sonderhouse, Lingfeng Yan, Maya Miklos, Yee Ming 393 Tso, and Shimon Kolkowitz. We thank James Thompson, Cindy Regal, John Hall, and Serge 394 Haroche for careful reading of the manuscript. 395 Authors' note: While performing the work described here, we became aware of 396 complementary work where high measurement precision was achieved for simultaneous 397 differential clock comparisons between multiple atomic ensembles in vertical 1D 398 lattices separated by centimeter scale distances using a hertz-linewidth clock laser43. 399 Author Contributions 400 All authors contributed to carrying out the experiments, interpreting the results, and writing the 401 manuscript. 402 *tobias.bothwell@colorado.edu, ye@jila.colorado.edu 403 Competing interests 404 The authors declare no competing interests. 405 Data and Code Availability 406 The experimental data and code analysis are available from the corresponding authors upon 407 reasonable request. 408 21 409 References 410 411 1. 412 Einstein, A. Grundgedanken der allgemeinen Relativitätstheorie und Anwendung dieser Theorie in der Astronomie. Preuss. Akad. der Wissenschaften, Sitzungsberichte 315, (1915). 413 2. Chou, C. W., Hume, D. B., Rosenband, T. & Wineland, D. J. Optical clocks and relativity. 414 Science 329, 1630–1633 (2010). 415 3. Herrmann, S. et al. Test of the Gravitational Redshift with Galileo Satellites in an Eccentric Orbit. 416 Phys. Rev. Lett. 121, 231102 (2018). 417 4. Delva, P. et al. Gravitational Redshift Test Using Eccentric Galileo Satellites. Phys. Rev. Lett. 418 121, 231101 (2018). 419 5. 420 421 6. 422 423 7. 424 Campbell, S. L. et al. A Fermi-degenerate three-dimensional optical lattice clock. Science 358, 90–94 (2017). Oelker, E. et al. Demonstration of 4.8×10−17 stability at 1 s for two independent optical clocks. Nat. Photonics 13, 714–719 (2019). Nicholson, T. et al. Systematic evaluation of an atomic clock at 2×10−18 total uncertainty. Nat. Commun. 6, (2015). 425 8. McGrew, W. F. et al. Atomic clock performance enabling geodesy below the centimetre level. 426 Nature 564, 87–90 (2018). 427 9. 428 Brewer, S. M. et al. 27Al+ Quantum-Logic Clock with a Systematic Uncertainty below 10-18. Phys. Rev. Lett. 123, 33201 (2019). 429 10. Bothwell, T. et al. JILA SrI optical lattice clock with uncertainty of 2.0×10-18. Metrologia 56, 430 (2019). 431 11. Marti, G. E. et al. Imaging Optical Frequencies with 100  μHz Precision and 1.1  μm Resolution. 432 Phys. Rev. Lett. 120, (2018). 433 12. Pedrozo-peñafiel, E. et al. Entanglement on an optical atomic-clock transition. Nature 588, 434 (2020). 435 13. Kaubruegger, R. et al. Variational Spin-Squeezing Algorithms on Programmable Quantum 436 Sensors. Phys. Rev. Lett. 123, 260505 (2019). 437 14. Kómár, P. et al. Quantum Network of Atom Clocks: A Possible Implementation with Neutral 438 Atoms. Phys. Rev. Lett. 117, 060506 (2016). 439 15. Young, A. W. et al. Half-minute-scale atomic coherence and high relative stability in a tweezer 440 clock. Nature 588, 408–413 (2020). 441 16. Safronova, M. S. et al. Search for new physics with atoms and molecules. Rev. Mod. Phys. 90, 442 25008 (2018). 443 17. Sanner, C. et al. Optical clock comparison for Lorentz symmetry testing. Nature 567, 204–208 444 (2019). 445 18. Kennedy, C. J. et al. Precision Metrology Meets Cosmology: Improved Constraints on Ultralight 22 446 Dark Matter from Atom-Cavity Frequency Comparisons. Phys. Rev. Lett. 125, 1–10 (2020). 447 19. Boulder Atomic Clock Optical Network. Frequency ratio measurements at 18-digit accuracy 448 using an optical clock network. 591, (2021). 449 20. Kolkowitz, S. et al. Gravitational wave detection with optical lattice atomic clocks. Phys. Rev. D 450 94, 124043 (2016). 451 21. Hafele, J. C. & Keating, R. E. Around-the-World Atomic Clocks: Science 177, 166 (1972). 452 22. Takamoto, M. et al. Test of general relativity by a pair of transportable optical lattice clocks. Nat. 453 Photonics 14, 411–415 (2020). 454 23. Laurent, P., Massonnet, D., Cacciapuoti, L. & Salomon, C. The ACES/PHARAO space mission. 455 Comptes Rendus Phys. 16, 540–552 (2015). 456 24. Tino, G. M. et al. SAGE : A proposal for a space atomic gravity explorer. Eur. Phys. J. D 73, 228 457 (2019). 458 25. Grotti, J. et al. Geodesy and metrology with a transportable optical clock. Nat. Phys. 14, 437–441 459 (2018). 460 26. Flechtner, Frank, Nico Sneeuw, and Wolf-Dieter Schuh. Observation of the system earth from 461 space: CHAMP, GRACE, GOCE and future missions. Springer (2014). 462 27. Kolkowitz, S. et al. Spin-orbit-coupled fermions in an optical lattice clock. Nature 542, 66–70 463 (2017). 464 28. Bromley, S. L. et al. Dynamics of interacting fermions under spin-orbit coupling in an optical 465 lattice clock. Nat. Phys. 14, 399–404 (2018). 466 29. Wilkinson, S. R., Bharucha, C. F., Madison, K. W., Niu, Q. & Raizen, M. G. Observation of 467 atomic wannier-stark ladders in an accelerating optical potential. Phys. Rev. Lett. 76, 4512–4515 468 (1996). 469 30. Lemonde, P. & Wolf, P. Optical lattice clock with atoms confined in a shallow trap. Phys. Rev. A 470 - At. Mol. Opt. Phys. 72, 1–8 (2005). 471 31. Aeppli, A. et al., In Preparation (2021). 472 32. Muniz, J. A., Young, D. J., Cline, J. R. K. & Thompson, J. K. Cavity-QED measurements of the 473 87Sr millihertz optical clock transition and determination of its natural linewidth . Phys. Rev. Res. 474 3, 023152 (2021). 475 33. Ludlow, A. D., Boyd, M. M., Ye, J., Peik, E. & Schmidt, P. O. Optical atomic clocks. Rev. Mod. 476 Phys. 87, 637–701 (2015). 477 34. Matei, D. G. et al. 1.5 µm Lasers with Sub-10 mHz Linewidth. Phys. Rev. Lett. 118, 263202 478 (2017). 479 480 References - Methods 481 35. Lemonde, P., Brusch, A., Targat, R. Le, Baillard, X. & Fouche, M. Hyperpolarizability Effects in 482 a Sr Optical Lattice Clock. Phys. Rev. Lett. 96, 103003 (2006). 23 483 36. Lodewyck, J., Zawada, M., Lorini, L., Gurov, M. & Lemonde, P. Observation and Cancellation of 484 a Perturbing dc Stark Shift in Strontium Optical Lattice Clocks. IEEE Transactions on 485 Ultrasonics, Ferroelectrics, and Frequency Control. 59, 411–415 (2012). 486 37. Leibfried, D., Blatt, R., Monroe, C. & Wineland, D. Quantum dynamics of single trapped ions. 487 Rev. Mod. Phys. 75, 281–324 (2003). 488 38. Marti, G. E. et al. Imaging Optical Frequencies with 100 μhz Precision and 1.1 μ m Resolution. 489 Phys. Rev. Lett. 120, 103201 (2018). 490 39. Boyd, M. M. et al. Nuclear spin effects in optical lattice clocks. Phys. Rev. A 76, 022510 (2007). 491 40. Martin, M. J. et al. A quantum many-body spin system in an optical lattice clock. Science 341, 492 632–636 (2013). 493 41. Ushijima, I. et al. Operational Magic Intensity for Sr Optical Lattice Clocks. Phys. Rev. Lett. 121, 494 263202 (2018). 495 42. van Westrum, D. NOAA Technical Memorandum NOS NGS-77. (2019). 496 43. Xin Zheng Jonathan Dolde, V. L. B. M. H. L. & Kolkowitz, S. High precision differential clock 497 comparisons with a multiplexed optical lattice clock (submitted 2021). 498 499 24 a Counts E lattice Si cavity b 1.0 0.8 Excitation Fraction 0.6 B g 0.4 0.2 z E clock c … |e,0 |e,-m … |e,+m Ω Ω-m … Ω+m … |g,0 |g,-m |g,+m 0.0 −2.0 −1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 2.0 Frequency (Hz) d 1.0 0.8 -1 +1 -2 +2 0.6 Excitation Fraction 0.4 0.2 -3 +3 -5 -4 0.0 −4 +4 +5 −2 Frequen0cyL(okrHemz) i2psum 4 Excitation Fraction Contrast p2 a 1.0 b 1 6s 30 s p1 p2 0.8 0.6 c 00 1.0 p1 1 0.8 0.4 0.6 50 s 0.4 0.2 0.2 τ=36.5(0.7) s 0.0 −0.5 0.0 0.5 z (mm) 0.0 0 20 40 60 Dark Time (s) a 1.00 0.75 0.50 1 0.25 0 0.00 −0.25 −0.50 −0.75 −1.00 −0.4 −0.2 0.0 0.2 0.4 z (mm) b3 Fractional Frequency (10-18) Fractional Frequency Gradient (10-19/mm) 000000088888880000000///////88888882122212///////152125112111117034698CBBBAAA Gravitational Acceleration (m/s2) 2 −1 −2 −3 20 10 0 −10 −20 −30 Measurement Date z (mm) Fractional Frequency Uncertainty a -0.5 0.0 0.5 b 10-18 10-19 4.4×10−18/√ τ 3.1×10−18/√ τ 2.6×10−18/√ τ 10-20 102 103 104 105 Time (s)