{"id":"fc4ffaf2-840f-46fa-95b4-9aeaa2fbac74","arxiv_id":"2501.12240","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A trapped on-chip Rubidium 87 Ramsey clock is characterized with a complete budget of its known noise sources, achieving about 4e-12 single-shot fractional frequency stability and about 1.4e-12 after empirical corrections.","lead":"This paper reports the realization and full noise budget of an on-chip cold atom Rubidium 87 clock based on a Ramsey interferometer with 600 ms interrogation, operated without magnetic shielding and with large temperature drift. It is a reference characterization for compact cold atom sensors, showing single-shot stability near 4e-12 and a plausible path to a few 1e-13.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Density-correction slope k is fit to the same data and varies by 20% between datasets, while the noise budget uses only the 1.8% intra-dataset fit error, so the headline stability may be an artifact of overfitting.","rationale":"The reader's weakest assumption identifies the density-correction slope k as the load-bearing element: the protocol assumes k removes only the true collisional shift, yet k varies by 20% between datasets, implying unmodeled correlations and a ~5e-12 systematic. My reading agrees and sharpens the point by noting the quantitative inconsistency inside the paper: Section II.E reports the 20% inter-dataset variation, while Section IV.E / Table I use only δk = 1.8% from the in-sample fit. This understates the density-correction noise by roughly a factor of four to five and leaves the headline stability dependent on a correction that is not demonstrated to be transferable. The paper is transparent about the 20% variation and even provides a partial cross-validation for the polynomial correction in Fig. 8(b), which is commendable. However, that cross-validation shows a ~35% degradation, and the analogous test for the linear correction used in the main result is absent. Because the central claim is specifically that the noise budget agrees with the measured Allan deviation, and because the measured values themselves are produced by the in-sample correction, this is the single most load-bearing concern. A conditional acceptance with a request for this cross-check is the right verdict; the paper's own disclosures already point to the required test, so no new objection beyond the reader's is raised.","tokens_in":21021,"tokens_out":4852,"duration_ms":48913,"concrete_test":"Recompute both Allan deviations in Fig. 3(d) using the k value fitted to the other dataset (apply the 15 Nov 2016 slope to the 22 Nov 2016 data and vice versa) instead of the per-dataset fitted k, and add the 20%-level systematic uncertainty in k in quadrature to the Table I total. If the cross-applied one-shot σ_y exceeds the reported 3.6–5.0e-12 range by more than ~40%, or if the revised total noise exceeds the measured value while the cross-applied σ_y also degrades, the headline stability is an artifact of the in-sample fit and the central agreement claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that a measured single-shot stability of 3.6–5.0e-12 is reproduced by a noise budget. This relies on the density correction in Eq. (11), δf_cor = δf − k·n_est, with k fitted to the same data in Fig. 5(a). Section II.E discloses that k differs by 20% between the two datasets behind Fig. 3(d), stating that this implies \"other experimental correlations between the density n and the frequency displacement δf than the collisional frequency shift,\" with a systematic effect up to ~5e-12. Yet Table I's density-correction contribution (1.13e-12) is computed from Eq. (45) using δk = 1.8% of k, which is the intra-dataset fit uncertainty, not the inter-dataset reproducibility. The known 20% variation is therefore omitted from the budget. More importantly, because k is fitted in-sample, any noise source correlated with n_est—temperature, magnetic field via the cooling correlations, or detection noise in O_max and N_tot—is partially absorbed into the correction, artificially lowering the reported σ_y. The cross-validation in Fig. 8(b) for the higher-order polynomial correction shows a ~35% degradation when coefficients are transferred between datasets (1.3–1.4e-12 to 1.8–1.9e-12), indicating that much of the improvement is dataset-specific; the analogous cross-check is not shown for the linear k correction that produces the headline 3.6–5.0e-12. Without addressing the k instability and the in-sample fitting, the agreement between the noise budget and the measured Allan deviation is not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a compact cold-atom rubidium-87 clock on an atom chip, based on a Ramsey interferometer with ~600 ms interrogation in a magnetic trap, operated without magnetic shielding and under large laboratory temperature drift. The authors measure single-shot stabilities of 3.6e-12 and 5.0e-12 on two datasets, construct a noise budget from magnetic-field and temperature instabilities, quantum projection noise, detection noise, Rabi-pulse noise, density-correction noise, atomic-loss noise, and the Dick effect, and show that the total budget is close to the measured Allan deviation. They further apply a polynomial correction in atom number and temperature that reduces the one-shot stability to 1.3-1.4e-12, with a cross-dataset degradation to 1.8-1.9e-12, and discuss routes toward a few times 1e-13 stability.","tokens_in":21331,"tokens_out":2612,"duration_ms":27699,"significance":"A detailed, self-contained noise budget for an on-chip cold-atom clock is of genuine value to the atom-chip metrology community, particularly because the paper goes beyond the usual headline stability number and documents the experimental sources of noise, including the image-denoising protocol based on reference-image eigenfaces and the density-correction procedure. The manuscript is also useful as a reference for the expected noise contributions in chip-based accelerometers and gyroscopes that share the same clock architecture. The paper's strengths include a relatively complete derivation of the magnetic-field and temperature sensitivity, a transparent account of the detection noise, and an explicit cross-dataset check for the higher-order correction in Fig. 8(b). However, the central claim that the noise budget accounts for the measured stability is weakened by in-sample fitting of the key parameters sigma_B, sigma_T, and the density-correction slope k, and by the paper's own disclosure that k varies by 20% between datasets.","major_comments":[{"comment":"The density correction is the load-bearing element of the reported stability: applying Eq. (11) with a slope k fitted to the same data reduces the fluctuations seen in Fig. 5(b) to Fig. 5(c). Section II.E discloses that k differs by 20% between the two datasets of Fig. 3(d), and states that this implies 'other experimental correlations between the density n and the frequency displacement δf than the collisional frequency shift', with a possible systematic effect of about 5e-12. Yet the noise-budget entry 'Density correction' in Table I is computed from Eq. (45) using δk = 1.8% of k, which is the intra-dataset fit uncertainty, not the inter-dataset reproducibility. The paper therefore omits from the budget the largest known uncertainty in the correction that produces the headline 3.6-5.0e-12 stability. The authors should either propagate the 20% k variation into σ_DensCorr and into the total budget, or justify why the 1.8% value is the appropriate uncertainty for the correction.","section":"II.E, Eq. (11), Table I"},{"comment":"The magnetic-field instability σ_B = 0.791 ± 0.055 mG and temperature instability σ_T = 28.3 ± 1.2 nK are obtained by fitting Eq. (28) to the measured Allan deviations of Fig. 6(a). These same fitted values are then used in Section IV.A to predict the magnetic and thermal contributions to the noise budget, and in the third column of Table I they are refit to the same data while applying the density correction. This is an in-sample fit, so the 'agreement' between the budget total and the measured Allan deviation is partly built into the parameters rather than being an independent test of the model. To support the claim that the noise budget explains the measured stability, the authors should provide an out-of-sample check: for example, use the σ_B and σ_T values obtained from one dataset or from an independent measurement (e.g., magnetometer or current-noise calibration) to predict the Allan deviation of the other dataset, or report the covariance and fitting residuals explicitly.","section":"III.D and IV.A, Eq. (28), Eq. (30)-(31), Table I"},{"comment":"The Allan deviation data are shown without error bars or confidence intervals. With only two datasets, and with the measured one-shot values (3.6e-12 and 5.0e-12) differing by 40% while the budget total is 5.1e-12 (uncorrected) or 3.8e-12 (corrected), the reader cannot judge whether the discrepancies are statistically significant. The paper should provide uncertainty estimates on the Allan-deviation points, for example using the effective number of independent samples or bootstrap/overlapping-sample confidence intervals, at least for the first few points that carry the headline stability claim.","section":"Fig. 3(d), Fig. 6, Fig. 8"},{"comment":"The cross-dataset check for the polynomial correction in Fig. 8(b) is commendable, but it shows a degradation from 1.3-1.4e-12 to 1.8-1.9e-12 when the 10 fitted coefficients are transferred between datasets, indicating that a substantial part of the improvement is dataset-specific. The analogous cross-validation is not shown for the linear density correction k that produces the main result of Section II.E. Since the paper's conclusion emphasizes both the 3.6-5.0e-12 budget agreement and the 1.4e-12 corrected stability, the authors should apply the same transfer test to the linear k correction, or at least state whether the 20% k variation between datasets already provides that test and how it affects the interpretation of Fig. 3(d).","section":"V.A, Eq. (55), Fig. 8"}],"minor_comments":[{"comment":"The density estimator n_est is defined up to a proportionality factor, but the text does not discuss how sensitive the correction is to the assumed scaling n ∝ (O_max)^(3/2) sqrt(N_tot); a brief justification or reference for this functional form would improve readability.","section":"II.E, Eq. (6)"},{"comment":"The table header 'Sumary' should be 'Summary'.","section":"Table I"},{"comment":"The phrase 'lead to two biaises' should be 'lead to two biases'.","section":"II.D.1"},{"comment":"The sentence 'It remains to improve σ_DensCorr which is link to the improvement...' contains a typo ('link' should be 'linked') and should be reworded.","section":"V.B.g"},{"comment":"The text says 'the first point of the Allan deviation as a function of the magnetic field is shown in figure 6.a', but the units and the fitted curve in Fig. 6(a) are not clearly defined in the caption; adding a caption note that the y-axis is σ_y(τ0) and that the line is the fit to Eq. (28) would help.","section":"III.D.d"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid experimental-noise-budget study, but the central quantitative claim rests on parameters fitted in-sample to the same data that are then claimed to be explained. The authors are unusually candid about the 20% variation of k and the 5e-12 possible offset, yet they do not propagate this into the budget. The requested changes are substantial but feasible within the manuscript's scope: propagate the inter-dataset k uncertainty, provide an out-of-sample test for σ_B and σ_T, and add uncertainty estimates to the Allan-deviation points. If the authors can supply those, the paper would be a valuable reference for chip-clock noise budgeting."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Honest take: this is a careful noise-budget study of one specific unshielded on-chip Rb-87 clock, and it is exactly what it says on the tin. It is not a record clock (they compare with ref [25] at 2e-13 one-shot) and not a new universal method, but it is a thorough, transparent account of all known noise sources, and that is genuinely useful for anyone building a compact atom-chip sensor.\n\nWhat the paper does well: it walks through every known noise – magnetic, thermal, QPN, detection, Rabi, Dick, atom losses, and the density correction – with formulas and measurements, and it shows how the measured Allan deviation can be reproduced when the magnetic and temperature noise parameters are re-fitted with the density correction applied. The eigenface-style imaging denoising is a nice practical trick, and the authors are unusually candid about limitations: the data are from 2016, the density-correction slope k varies by 20% between datasets, and they give a systematic uncertainty estimate of about 5e-12 from that variation. That kind of honesty is rare and worth respecting.\n\nThe soft spots are real and mostly acknowledged. The k in Eq. (11) is fitted to the same data it corrects, and Table I uses an intra-dataset fit error of 1.8% rather than the 20% inter-dataset reproducibility the paper itself reports. The higher-order polynomial correction in Eq. (55) has ten coefficients fitted to the same data, and the cross-validation they do show (Fig. 8b) degrades from 1.3–1.4e-12 to 1.8–1.9e-12, which is a clear sign of overfitting. No analogous cross-check is shown for the linear k correction that produces the headline 3.6–5.0e-12. The agreement in Table I's third column is also partly enforced because sigma_B and sigma_T are re-fit to the same Allan deviation data. So the central claim that the noise budget accounts for the measured stability is plausible, but it is not rigorously established.\n\nThat said, these are proportionate concerns for an engineering-oriented characterization paper. The authors do not hide the flaws, and the paper is a solid reference for noise formulas and for how to handle a density correction in a compact clock. It deserves a serious referee, but the referee should ask for error bars on the Allan deviation points, a cross-check of the linear k correction, and a propagation of the inter-dataset k variation through the final budget.\n\nFor researchers building compact atom sensors, this is a useful case study. I would cite this as a noise-budget reference for atom-chip clocks, not for a record stability. For a reading group, it is a good example of how in-sample corrections can flatter an Allan deviation, but it is not a must-read.","headline":"A transparent, useful noise-budget case study for an unshielded on-chip Rb-87 clock, but the agreement claim rests on parameters fitted to the same data.","tokens_in":21926,"tokens_out":2979,"would_cite":true,"duration_ms":28324,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper establishes that the residual frequency noise of an on-chip rubidium-87 Ramsey clock is accounted for by a measured budget of known noises, with single-shot stability between 3.6e-12 and 5e-12.","keywords":["cold atom clock","atom chip","Ramsey interferometry","noise budget","Allan deviation","rubidium-87","collisional frequency shift","magnetic field noise"],"falsifier":"Repeat the Allan-deviation measurement while actively suppressing magnetic-field noise, with all other settings unchanged: if the density-correction slope $k$ shifts or the residual scatter is no longer linear in $n_{\\rm est}$, the correction is absorbing magnetic or thermal correlations rather than only collisions. A second independent check is to measure $k$ on two long datasets of the same run; the paper already reports a 20% difference, so an experiment that explains that difference would settle whether the correction is unbiased.","tokens_in":20766,"feed_emoji":"⚛️","tokens_out":10915,"duration_ms":61184,"temperature":0.7,"pith_summary":"The paper reports a working cold-atom clock on an atom chip and, more importantly, a complete accounting of why its frequency fluctuates. The central claim is that every known noise source—magnetic field, temperature, atomic collisions, detection, quantum projection, Rabi-pulse parameters, atomic loss, and local-oscillator aliasing (the Dick effect)—has been modelled and measured, and that the sum agrees with the measured Allan deviation. The clock gives 3.6e-12 to 5e-12 relative stability in a single shot, integrates to about 1e-12 after 1000 s, and improves to 1.3–1.4e-12 at one shot after correcting for atom-number and temperature correlations. The authors frame the work not as the best on-chip clock but as the first step toward a noise budget for the atom-chip accelerometers and gyroscopes that share the same trap and Ramsey sequence.","feed_headline":"On-chip atom clock noise is fully identified at 3.6-5e-12","feed_subtitle":"A rubidium-87 Ramsey clock on a chip now has a measured budget for every known noise source.","key_machinery":"The central object is a Ramsey interferometer in a dimple magnetic trap created by currents on an atom chip. The transition frequency is written $f=f_0+\\delta f_B+\\delta f_{col}$, where the magnetic shift is quadratic in the field offset, $\\delta f_B=b(B-B_0^m)^2$, and the collisional mean-field shift is $\\delta f_{col}=-2\\hbar(a_{aa}-a_{bb})/(mN)\\int n^2\\,dV$. The density correction uses the estimator $n_{\\rm est}=(O_{\\max}^b+O_{\\max}^a)^{3/2}(N_{\\rm tot})^{1/2}$ with a linear slope $k$ fitted to measured frequency shifts. The noise budget combines measured magnetic and temperature fluctuations ($\\sigma_B=0.791$ mG, $\\sigma_T=28.3$ nK) with the Dick-effect sensitivity function $g(t)$, the quantum projection noise formula, and a camera-photon-noise model for detection.","core_discovery":"Operating between the two magnetically trappable rubidium-87 states $|F=1, m_F=-1\\rangle$ and $|F=2, m_F=1\\rangle$ near the 3.229 G magic field, with two-photon microwave/RF $\\pi/2$ pulses and a Ramsey time of 603 ms, the clock reaches a single-shot relative stability of 3.6–5e-12. The claim is that this stability is quantitatively understood: the measured noise budget totals 5.1e-12 at one shot, or 3.8e-12 when the density correction is included in the magnetic and thermal estimates, matching the measured values. The dominant contribution is magnetic-field noise (4.69e-12, from unshielded current supplies), followed by the density-correction noise (1.13e-12), thermal noise (1.28e-12, or 0.45e-12 after density correction), the Dick effect (0.72e-12), technical detection (0.56e-12), and quantum projection noise (0.50e-12). Applying an empirical third-order polynomial correction in the measured atom number and temperature reduces the single-shot stability to 1.3–1.4e-12, and the authors argue that better supplies, a longer Ramsey time, and lower detection noise could bring it to a few times 1e-13.","pith_inferences":["The reported 20% run-to-run variation in the density-correction slope $k$ implies that the density estimator tracks something beyond the true collisional shift, so a physical model of that correlation would tighten the accuracy budget.","A testable consequence of the budget is that a version of this clock with magnetic shielding and quieter current sources should have its one-shot stability set by the sum of density-correction, detection, and quantum projection noises, around the low 1e-12 range.","The cross-calibration stability (1.8–1.9e-12) is the safer estimate of what the empirical correction achieves, since it leaves no freedom to fit the noise being corrected.","The budget's separation into atomic, local-oscillator, and technical terms can be transferred to other atom-chip clock geometries or species once the density estimator is recalibrated for each trap."],"forward_implications":["If the budget is correct, the single highest-leverage improvement is to cut magnetic-field noise: the measured 0.791 mG field noise contributes 4.7e-12 at one shot.","With the empirical $N_{tot}$-and-$T$ correction, one-shot stability falls to 1.3–1.4e-12, and the cross-calibrated version reaches 1.8–1.9e-12.","Extending the Ramsey time to 6 s, lengthening the trap lifetime, and reducing the cycle time to 7.3 s would bring the projected total noise to about 3.2e-13, dominated by the density-correction term.","Because the same trap and Ramsey sequence underlie the atom-chip accelerometer and gyroscope configurations, the clock noise budget sets the floor for those inertial sensors as well."],"supporting_citations":[{"why":"Introduces the two trapped clock states and the magic-field transition used by this interferometer.","marker":"[24]"},{"why":"Previous atom-chip clock used as stability baseline and as the source of the Dick-effect, density-correction, and atom-loss formulas.","marker":"[25]"},{"why":"Supplies the quadratic magnetic-field coefficient $b\\approx 431$ Hz/G$^2$ used in the field-dependent frequency shift.","marker":"[52]"},{"why":"Gives the two-photon Rabi formula used to calibrate the $\\pi/2$ pulses and extract the light shift.","marker":"[53]"},{"why":"Breit-Rabi formula used to convert measured frequencies into magnetic fields and to compute the theoretical transition frequency.","marker":"[54]"},{"why":"Provides the rubidium-87 $s$-wave scattering lengths that set the collisional mean-field shift.","marker":"[66]"},{"why":"Gives the quantum projection noise formula $\\sigma_P=1/(2\\sqrt{N})$.","marker":"[70]"},{"why":"Provides the clock sensitivity function $g(t)$ used to compute the Dick-effect aliasing.","marker":"[75]"},{"why":"Camera photon-noise model used to estimate technical detection noise and the path to reduce it.","marker":"[79]"}],"fun_headline_variants":["Full noise budget for on-chip Rb-87 clock matches measured stability","Chip atom clock's every known noise source quantified, magnetic leads","On-chip cold atom clock: noise budget complete, stability 3.6e-12","Rb-87 chip clock: correction drops noise to 1.3e-12","Noise decomposition of chip-scale Rb-87 clock reveals magnetic limit"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The density-correction protocol assumes that the image-derived quantity $n_{\\rm est}$ stays proportional to the collisional mean-field density across runs and that the single fitted slope $k$ removes only the collisional shift.","fun_headline_variants_meta":{"raw":{"variants":["Full noise budget for on-chip Rb-87 clock matches measured stability","Chip atom clock's every known noise source quantified, magnetic leads","On-chip cold atom clock: noise budget complete, stability 3.6e-12","Rb-87 chip clock: correction drops noise to 1.3e-12","Noise decomposition of chip-scale Rb-87 clock reveals magnetic limit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000781,"raw_usage":{"total_tokens":3429,"prompt_tokens":903,"completion_tokens":2526,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":519,"completion_tokens_details":{"reasoning_tokens":2426}},"tokens_in":519,"tokens_out":2526,"duration_ms":18542,"temperature":1.0,"reasoning_tokens":2426,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T17:22:03.063130+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the Allan-deviation measurement while actively suppressing magnetic-field noise, with all other settings unchanged: if the density-correction slope $k$ shifts or the residual scatter is no longer linear in $n_{\\rm est}$, the correction is absorbing magnetic or thermal correlations rather than only collisions. A second independent check is to measure $k$ on two long datasets of the same run; the paper already reports a 20% difference, so an experiment that explains that difference would settle whether the correction is unbiased.","supporting_citations":[{"cited_title":"Treutlein, P","cited_arxiv_id":null,"evidence_quote":"Introduces the two trapped clock states and the magic-field transition used by this interferometer."},{"cited_title":"Dupont-Nivet, C","cited_arxiv_id":null,"evidence_quote":"Previous atom-chip clock used as stability baseline and as the source of the Dick-effect, density-correction, and atom-loss formulas."},{"cited_title":"Rosenbusch, Appl","cited_arxiv_id":null,"evidence_quote":"Supplies the quadratic magnetic-field coefficient $b\\approx 431$ Hz/G$^2$ used in the field-dependent frequency shift."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the two-photon Rabi formula used to calibrate the $\\pi/2$ pulses and extract the light shift."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the rubidium-87 $s$-wave scattering lengths that set the collisional mean-field shift."},{"cited_title":"Egorov, B","cited_arxiv_id":null,"evidence_quote":"Gives the quantum projection noise formula $\\sigma_P=1/(2\\sqrt{N})$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the clock sensitivity function $g(t)$ used to compute the Dick-effect aliasing."},{"cited_title":"Santarelli, C","cited_arxiv_id":null,"evidence_quote":"Camera photon-noise model used to estimate technical detection noise and the path to reduce it."}],"review_version":1}