{"id":"a6665483-4d54-4518-8ab4-253b7fa46d38","arxiv_id":"1908.03320","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A matched-filter search statistic for dark matter transients in GPS atomic clock networks is developed, with analytic sensitivity formulas and simulation-based detection thresholds.","lead":"This paper develops a signal-detection method, called matched filtering, for finding dark matter objects that sweep through networks of atomic clocks such as those on GPS satellites. The authors derive sensitivity formulas and demonstrate on simulated GPS clock data that the method can detect very weak transient dark matter signals.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Calibration rests on simulated GPS noise; real-data validation is missing, so the thresholds and 95% detection sensitivity may not carry over to archival search.","rationale":"The matched-filter SNR statistic is only calibrated if the covariance matrix used in Eq. (20) equals the true noise covariance. The paper validates this calibration on simulated data built from assumed PSDs and a white common reference; it never demonstrates that these assumptions hold for the actual JPL clock-bias streams. Since the paper's stated goal is a frequentist search of archival GPS data, the unvalidated noise model is the weakest link between the Monte Carlo results and the claimed method. The perturbative inversion concern is closely related: for the 2010 network xi ~ 3, and Table III shows the first-order expansion can be unreliable at comparable xi, so even the simulated calibration could differ from exact-inverse predictions; however, because the thresholds and detection probabilities are empirically re-measured from the simulations, the Monte Carlo detection claim is internally consistent. The missing real-data null test is what would settle the calibration transfer. This is a condition for acceptance, not a rejection: the analytic derivations and simulation study are coherent, and the paper clearly labels its simplifications. Therefore the verdict remains CONDITIONAL.","tokens_in":24377,"tokens_out":17404,"duration_ms":185907,"concrete_test":"Take 30 randomly selected, non-overlapping 24-hour segments of real JPL GPS clock-bias data from the 2010 constellation. Run the full search pipeline (differencing, covariance estimate, M=1024 template bank, perturbative inverse) with no injected signals. Compute for each segment the empirical standard deviation of template-specific SNR and the rate at which the max-over-template statistic exceeds the Table II 10-f.p./yr threshold (rho*=4.95). Compare the pooled sigma_rho to the simulated value 0.86 (Fig. 2) and the observed false-positive rate to the nominal 10 per year. A deviation in sigma_rho larger than ~10% or a false-positive rate significantly above nominal would show the simulated noise model does not reproduce real GPS data, invalidating the claimed calibration.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central promise is a calibrated frequentist search of archival GPS data, but every threshold (Table II) and detection-sensitivity number (Fig. 4, h95%=0.045 ns for 2010) is produced by Monte Carlo simulations whose noise is generated from prior PSDs plus a white-noise common reference clock with sigma_x=0.006 ns (Sec. V A). The matched-filter statistic's calibration depends on E in Eq. (20) being the true noise covariance; if the real differenced clock noise is non-stationary (clock changes, satellite maneuvers) or has correlated environmental contributions beyond the common reference, the Gaussian unit-variance property (Sec. VII, Eq. (25)) and the false-positive thresholds in Table II need not hold. The paper does not apply the pipeline to any real GPS segment to verify the null distribution of template-specific SNR or the false-positive rate of the max statistic. Its own Table III shows the perturbative inverse (A17) can fail badly once the reference is not very quiet, and the 2010 operating point (sigma~0.02, sigma_x~0.006, ND=33) gives xi=ND(sigma_x/sigma)^2 ~ 3, where the first-order expansion is not obviously small; this is not tested. Without real-data null validation, the step from Monte Carlo demonstration to a calibrated archival search is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops a frequentist matched-filter search for transient dark-matter signals, specifically thin domain walls, in networks of precision sensors with a common reference clock, using GPS atomic clock data as the target application. The authors define a template-specific SNR (Eq. 20), construct thin-wall signal templates with an importance-sampled template bank, and derive analytic Gaussian statistics for an idealized network of identical white-noise sensors with a common white-noise reference (mean in Eq. 24, unit variance in Eq. 25). They prove the unit-variance property in Appendix B.1, compute false-positive thresholds for four GPS network configurations (Table II), and use Monte Carlo simulations to estimate 95% detection probabilities (for example, h95%=0.045 ns for the 2010 network) and parameter-extraction resolutions. The paper closes with projected exclusion limits on DM couplings. The calibration is carried out entirely on simulated GPS noise; the pipeline is not applied to archival GPS data.","tokens_in":24576,"tokens_out":17442,"duration_ms":168201,"significance":"The analytic derivation is clean, and the unit-variance proof in Appendix B.1 is a genuinely useful contribution that extends matched-filter treatments to networks with cross-node correlated noise. The Monte Carlo checks, including the 94.6% versus 95% detection-probability agreement in Sec. IX B, are consistent with the Gaussian predictions. If the calibration survives contact with real GPS noise, the method offers a calibrated frequentist complement to the Bayesian search of Ref. [14] and gives concrete projected sensitivity estimates. The main open question is not the internal mathematics but whether the simulated-noise assumptions support the stated false-positive thresholds and detection probabilities for archival use.","major_comments":[{"comment":"As written, Section VIII states that the simulations use the perturbative inverse of the covariance matrix, and Appendix A2 states that the von Neumann expansion converges only when the relevant eigenvalues are smaller than unity. For the 2010 operating point quoted in Sec. XI B (sigma ~ 0.02 ns, sigma_x ~ 0.006 ns, ND = 33), the expansion parameter is xi = ND (sigma_x/sigma)^2 ~ 3, so the stated convergence condition is violated. Table III shows the approximation already produces sigma_rho = 8.96 at sigma_x/sigma = 0.5 (xi ~ 7.5 for ND = 30), and the paper does not test the 2010 value sigma_x/sigma ~ 0.3. If the simulations indeed used the perturbative inverse, the Table II thresholds and the Fig. 4 h95% = 0.045 ns value for 2010 are not established; the authors should recompute them with the exact inverse or demonstrate that the perturbative inverse is valid for their heterogeneous 2010 network. If the simulations instead used exact Cholesky inversion, the text should say so explicitly.","section":"Sec. VIII A and Appendix A2, Table III"},{"comment":"The null calibration and detection thresholds are obtained entirely from simulated noise generated from prior power spectral densities plus a white common reference clock (Sec. V A), with no validation on archival GPS data. The paper presents thresholds for one false positive in 20 years (Table II) and reports h95% for the 2010 network (Sec. IX B), but it never applies the pipeline to a real, signal-free GPS segment to check that template-specific SNRs are Gaussian with the simulated sigma_rho values or that the max-statistic false-positive rate matches Table II. Real GPS noise is nonstationary (clock swaps, satellite maneuvers, environmental correlations), so the calibration of the proposed archival search is unsupported. The authors should either add null-data validation or explicitly rescope the claims to simulated data only.","section":"Sec. VIII and Sec. IX"},{"comment":"The false-positive probability per epoch treats the M template-specific SNRs as independent. Because templates are drawn from continuous priors over velocity, direction, and arrival time, neighboring templates will produce correlated SNRs, so the distribution of the maximum statistic defined in Eq. (21) need not follow Eq. (27). The paper verifies only the marginal distribution of individual template SNRs in Fig. 2, not the false-positive rate of the actual detection statistic. Without a Monte Carlo characterization of the max statistic under H0, the thresholds in Table II are not fully calibrated even within the simulated-noise model.","section":"Eq. (27), Sec. VIII"}],"minor_comments":[{"comment":"The word 'whitten' in 'we \"whitten\" the data' appears to be a typo for 'whiten'.","section":"Sec. V"},{"comment":"The sentence 'geophysicists were been able to identify' contains a grammatical error and should read 'were able to identify'.","section":"Sec. II"},{"comment":"The main text should note explicitly that Eq. (24) drops the fractional-degeneracy factor lambda that appears in the exact expression (B8), and should comment on the size of this approximation for GPS.","section":"Eq. (24) and Eq. (B8)"},{"comment":"Reference [45] has an incomplete arXiv identifier ('arXiv preprint arXiv: . . . 10, 582 (2013)') and should be completed.","section":"Reference [45]"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear X,\n\nQuick take: this is a real contribution, not a repackaging. The matched-filter statistic itself is textbook, but the analytic treatment of cross-node correlated noise from a common reference clock is new and clean. Eq. (24) for the SNR mean and the Appendix B.1 proof that the SNR has unit variance under the true covariance are the core results, and they hold up. The simulation work is internally consistent — the 94.6% detection probability at the claimed 95% threshold is a nice check. The paper also does something useful that the earlier GPS.DM papers didn't: it gives a template-bank search, a false-positive threshold formula, and parameter estimation with believable resolutions.\n\nThe soft spots are real but mostly about what the paper promises versus what it shows. Every threshold and sensitivity number is generated from simulated noise: PSDs from prior GPS analyses plus a white-noise common reference clock with sigma_x=0.006 ns. The stress-test concern is correct — if the differenced clock noise is non-stationary in the archival data, or if there are correlated environmental contributions beyond the reference, the unit-variance property and the Table II thresholds don't necessarily carry over. The paper explicitly acknowledges the perturbative covariance inversion fails for modern clocks (Table III), and the 2010 operating point sits at xi ~ 3, where the first-order expansion is not obviously small. That's a limitation, not a fatal flaw, because the paper is framed as a method study, but the abstract's talk of 'utility and performance' on 'simulated GPS data' is accurate only up to that boundary.\n\nOther, smaller things: no code or data shipped, so the Monte Carlo curves are hard to audit; the curves in Fig. 4 lack error bars, though 128 trials a point is small enough that this is a genuine omission; and the final mean expression in Eq. (B8) drops the fractional degeneracy lambda in the approximation step, which is fine for an analytic estimate but should be stated more carefully. None of these undercut the central derivation — the variance proof is general and short, and it's correct.\n\nWho is this for? Someone designing or planning a transient search with a GPS/GNSS clock network, or any network with a common reference. It's a method paper, and a serious referee should engage with it. I'd recommend accepting with the expectation of major revision: the revision should either add real-data null validation on a segment of archival GPS data, or reframe the claims explicitly as simulation-only and ship code. As is, it earns a place in the literature but falls short of its own 'calibrated archival search' framing.\n\nMy vote: worth a serious referee, conditionally acceptable with revision.","headline":"A solid method paper that derives the right analytic SNR for common-reference clock networks and demonstrates it in simulation, but the promised calibrated archival search depends on noise assumptions not yet tested on real GPS data.","tokens_in":25130,"tokens_out":1781,"would_cite":true,"duration_ms":19059,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A matched-filter search over GPS atomic clock data can detect thin domain-wall dark matter transients at 95 percent probability for clock jumps as small as 0.045 nanoseconds.","keywords":["matched-filter technique","dark matter transients","thin domain walls","topological defects","atomic clock networks","GPS constellation","quantum sensors","signal-to-noise ratio"],"falsifier":"Histogram the template-specific SNR values on long, event-free stretches of the real archival GPS data using the paper's covariance inversion: the calibration claims a standard deviation of exactly 1 (Eq. 25), so an observed spread noticeably different from 1, the signature of non-stationary or extra correlated noise, would void the Table II thresholds. A complementary check is to splice synthetic thin-wall jumps of known height $h$ into real data streams and verify that the mean SNR follows $\\mu_{\\rho} = (h\\sqrt{N_D}/\\sigma)\\sqrt{(1+\\eta^2+\\xi)/(1+\\xi)}$; failure of that scaling would show the analytic mean formula omits some real network behavior.","tokens_in":2077,"feed_emoji":"🛰️","tokens_out":5246,"duration_ms":156020,"temperature":0.7,"pith_summary":"The paper claims that a standard template-matching technique, the matched filter, can be turned into a calibrated dark-matter search using the existing network of atomic clocks aboard GPS satellites, whose timing data is publicly archived for nearly two decades. The target signal is a 'thin' domain wall, a sheet of self-interacting ultralight dark matter, which would sweep through the satellite constellation at galactic velocity and momentarily shift each clock's frequency, leaving a correlated pattern of jumps across the network. The central technical result is that this search statistic stays Gaussian with unit variance even though every satellite clock is measured against one shared reference clock, whose noise corrupts all channels alike; that fact makes the false-alarm rate predictable and the threshold a matter of a formula rather than guesswork. If the method performs on real archival data as it does in the simulations, the GPS timing record becomes a nearly free dark-matter detector with projected sensitivity competitive with dedicated optical clock experiments.","feed_headline":"Matched filter can find dark-matter walls in GPS clock data","feed_subtitle":"The calibrated statistic could mine two decades of GPS timing records for compact dark matter objects.","key_machinery":"The carrying mechanism is the matched-filter SNR built from the inverse network covariance matrix. Each trial signal is a thin-domain-wall template (Eq. 22): in the differenced clock stream it is a single spike of height $h_a$ at the epoch the wall hits a satellite clock, and a spike of opposite sign $-h_R=-\\eta h_a$ at the later epoch it hits the shared reference clock. The statistic $\\rho = d^{T}E^{-1}s/\\sqrt{s^{T}E^{-1}s}$ compares every such template against the data, and the covariance matrix $E$, which encodes both individual clock noise and the common reference clock's contribution to every channel, is what makes the search nontrivial: for the idealized network its inverse is the closed form $(\\mathrm{E}^{-1})^{ab}_{jl} = \\sigma^{-2}\\delta_{jl}(\\delta^{ab} - \\xi/(1+\\xi)/N_D)$, while realistic GPS clocks use a perturbative inversion valid when the reference clock is quieter than the satellites. The decisive identity is the proof in Appendix B.1 that $\\mathrm{Var}\\{n^{T}E^{-1}s\\} = s^{T}E^{-1}s$, forcing the SNR variance to exactly one regardless of noise color or correlation; this single fact fixes the false-positive rate and converts thresholds into pure functions of the template-bank size and the desired false-alarm budget.","core_discovery":"On its own terms, the paper establishes a frequentist detection recipe for dark-matter transients in networks of precision clocks. For an idealized network of $N_D$ identical white-noise sensors sharing a common reference clock, the matched-filter SNR $\\rho = d^{T}E^{-1}s/\\sqrt{s^{T}E^{-1}s}$ is Gaussian with unit variance and mean $\\mu_{\\rho} = (h\\sqrt{N_D}/\\sigma)\\sqrt{(1+\\eta^2+\\xi)/(1+\\xi)}$, where $h$ is the strength of the dark-matter-induced clock jump, $\\eta$ the ratio of the signal on the reference clock to that on the satellites, and $\\xi = N_D\\sigma_{\\times}^2/\\sigma^2$ the relative noise power of the shared reference (Eqs. 24-25). The unit variance is proven for arbitrary covariance structure, so the statistic is self-calibrating even for colored noise and unknown cross-node correlations. In Monte Carlo simulations against realistic GPS constellations from 2000, 2005, 2010, and 2015, the pipeline detects injected thin-domain-wall signals with 95% probability at $h\\approx 0.045$ ns on the 2010 network, recovers the event's velocity, direction, and arrival time, and projects discovery reach for dark-matter couplings comparable to laboratory optical-clock limits.","pith_inferences":["A natural extension the paper leaves implicit: the same SNR statistic and unit-variance calibration apply to any sensor network whose channels share a common reference or common environmental noise, so ground-based clock networks and magnetometer arrays could reuse the pipeline as-is instead of deriving a new covariance model.","The sharpest untested premise is stationarity, so the single most informative follow-up is empirical: compute $\\sigma_\\rho$ on rolling segments of the real archive; if it wanders, per-segment recalibration of the thresholds would be needed before any claim of a 20-year false-alarm rate.","Because the SNR is an odd function of the signal strength, the same pipeline natively covers both signs of the dark-matter coupling (clocks speeding up or slowing down); a positive detection would let the mixed Rb/Cs/H-maser network disentangle which fundamental constant shifts, through the distinct coefficient combinations the paper tabulates.","The analytic mean neglects the fractional degeneracy $\\lambda$ of two satellites being hit within the same 30-second epoch; as future constellations grow denser, $\\lambda$ will rise, and the Gaussian calibration of the statistic would need to be revisited at that point."],"forward_implications":["The two decades of archival GPS timing records become searchable as a calibrated frequentist dark-matter experiment, with explicit SNR thresholds for one false positive per day, ten per year, or one per 20 years for each network generation (Table II).","The matched filter reaches 95% detection probability for thin-wall clock jumps around 0.045 ns on the 2010 constellation, a reach that extends the earlier GPS domain-wall analysis, which was limited to signals well above the instrument noise, by orders of magnitude.","Sensitivity scales as $\\sqrt{N_D}$ with the number of clocks and as $\\sqrt{N_E}$ with the number of independent encounters stacked in the search window, so the growing multi-constellation GNSS clock fleet directly strengthens the search.","A detected event carries enough information to reconstruct its velocity ($\\pm 27$ km/s), direction ($\\pm 0.05\\pi$ rad), and time ($\\pm 5$ s), enabling a standard-halo-model consistency veto; a null result instead converts into exclusion limits on dark-matter couplings to electrons, quarks, and the fine-structure constant at energy scales comparable to optical-clock constraints."],"supporting_citations":[{"why":"The preceding GPS domain-wall search; supplies the archival data analysis and the relation between signal strength and dark-matter coupling that this method extends.","marker":"[13]"},{"why":"Supplies the GPS clock noise characterization, auto-correlation functions, simulation procedures, and the Bayesian technique against which the matched-filter sensitivity is benchmarked.","marker":"[14]"},{"why":"The matched-filter formalism and optimality criteria that the network SNR statistic generalizes to correlated multi-sensor data.","marker":"[16]"},{"why":"The frequency-domain SNR for quantum-sensor networks searching for dark-matter waves, the counterpart this paper contrasts with its time-domain transient statistic.","marker":"[21]"},{"why":"The standard halo model velocity distribution used to set the prior templates for event speed and direction.","marker":"[22]"},{"why":"The public archival GPS clock-bias and orbit data set that the search procedure is designed to mine.","marker":"[41]"},{"why":"The proposal that distributed atomic-clock networks can serve as dark-matter detectors, which this paper operationalizes with matched filtering.","marker":"[12]"},{"why":"Laboratory optical-clock constraints used as the benchmark for the projected discovery reach of the GPS search.","marker":"[23]"}],"fun_headline_variants":["Matched filter finds dark-matter walls in GPS clock networks","Quantum sensor networks hunt dark matter with matched filter","GPS clocks as dark matter detectors via matched-filter analysis","Matched-filter technique reveals dark matter transients in GPS data","Self-calibrating statistic spots dark-matter walls in clock data"],"cache_read_input_tokens":27264,"weakest_assumption_plain":"The load-bearing premise is that the clock noise is completely and stably described by the covariance matrix built from prior GPS data, with the shared reference clock as the only source of correlation between satellites; if the differenced noise drifts over the archive or hides additional environmental correlations, the unit-variance calibration and every tabulated threshold in Table II cease to hold.","fun_headline_variants_meta":{"raw":{"variants":["Matched filter finds dark-matter walls in GPS clock networks","Quantum sensor networks hunt dark matter with matched filter","GPS clocks as dark matter detectors via matched-filter analysis","Matched-filter technique reveals dark matter transients in GPS data","Self-calibrating statistic spots dark-matter walls in clock data"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00031,"raw_usage":{"total_tokens":1768,"prompt_tokens":943,"completion_tokens":825,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":559,"completion_tokens_details":{"reasoning_tokens":743}},"tokens_in":559,"tokens_out":825,"duration_ms":8619,"temperature":1.0,"reasoning_tokens":743,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:17:39.384031+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Histogram the template-specific SNR values on long, event-free stretches of the real archival GPS data using the paper's covariance inversion: the calibration claims a standard deviation of exactly 1 (Eq. 25), so an observed spread noticeably different from 1, the signature of non-stationary or extra correlated noise, would void the Table II thresholds. A complementary check is to splice synthetic thin-wall jumps of known height $h$ into real data streams and verify that the mean SNR follows $\\mu_{\\rho} = (h\\sqrt{N_D}/\\sigma)\\sqrt{(1+\\eta^2+\\xi)/(1+\\xi)}$; failure of that scaling would show the analytic mean formula omits some real network behavior.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The frequency-domain SNR for quantum-sensor networks searching for dark-matter waves, the counterpart this paper contrasts with its time-domain transient statistic."},{"cited_title":"Shaerer, J","cited_arxiv_id":null,"evidence_quote":"The standard halo model velocity distribution used to set the prior templates for event speed and direction."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The proposal that distributed atomic-clock networks can serve as dark-matter detectors, which this paper operationalizes with matched filtering."},{"cited_title":"Auger, D","cited_arxiv_id":null,"evidence_quote":"Laboratory optical-clock constraints used as the benchmark for the projected discovery reach of the GPS search."}],"review_version":1}