{"id":"54ac31a7-07c7-4875-a7fd-67901006b76a","arxiv_id":"2502.09275","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"low","formal_verification":"none","parameter_count":5,"one_line_summary":"Reported 95% upper limits on microhertz gravitational-wave strain at 1 μHz are 1.26×10⁻¹² sky-averaged and 4.77×10⁻¹³ toward the pulsar.","lead":"Using 1.7 years of FAST radio telescope data on the millisecond pulsar PSR J1713+0747, this paper sets new upper limits on gravitational-wave strain from supermassive black-hole binaries in the microhertz band. The limits are a proof of concept for the Chinese Pulsar Timing Array and are comparable to, though somewhat weaker than, earlier European limits.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The headline 'sky-average 95% upper limit' is an average of per-cell upper limits over an unspecified sky region, not a 95% confidence limit for an unknown source direction; the quoted 1.26e-12 is not uniquely defined.","rationale":"Good-faith reading: the paper does not claim a detection and presents a standard single-pulsar Bayesian upper-limit analysis; the noise modeling is thorough, with model selection and a varying-noise check, and the comparison with the CRLB is a reasonable cross-check. The most load-bearing issue is not the noise model (a generic and acknowledged caveat) but the meaning of the headline 'sky-average' number. The text explicitly says the sky-average is an average of per-cell upper limits, and it depends on an unspecified exclusion around the anti-pulsar direction; the paper even concedes that the excluded area in Perera et al. is unknown. A reader cannot reproduce the quoted 1.26e-12 from the paper alone, and the number does not have the coverage properties implied by '95% confidence level' for an unknown source location. The pulsar-direction limit (4.77e-13) is better defined and is not affected by this concern; it could be the primary result if the sky-average statement is clarified. The reader's noise-model caveat is valid but secondary: if unmodeled red noise were present the limits would be optimistic, yet the model selection and the fact that the relevant band lies above the fitted DM band reduce this risk. For these reasons I retain the CONDITIONAL verdict: the science is likely sound, but the headline needs a precise statistical definition or a change of quantity before the result is treated as final.","tokens_in":13479,"tokens_out":16619,"duration_ms":176113,"concrete_test":"At 1 μHz, take the fixed-noise per-cell posterior distributions used for the 40x40 sky map (or the 20x20 grid) and compute: (A) the arithmetic mean of the per-cell 95% upper limits; (B) the 95% quantile of the equally weighted mixture of the per-cell posteriors, i.e., p(h) = (1/N) sum_i p_i(h), with the same sky prior over the region actually used. If (B) exceeds (A) by a factor much larger than unity (as expected from the anti-pulsar cells), the quoted 'sky-average 95% upper limit' is not the 95% upper limit for an isotropically distributed source; the paper should report (B) or explicitly define the sky-average as a sensitivity curve rather than a confidence limit. Also state the exact excluded sky region in both this paper and Perera et al. so the comparison is reproducible.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's headline quantity, the 'sky-average continuous source upper limit' of 1.26e-12 at 1 μHz, is not a well-defined 95% confidence limit. Section 4.2 says the sky-average is obtained 'by averaging the upper limits derived in 400 sky cells' on a 20x20 RA-cos(DEC) grid, and Section 4 says an area around the anti-pulsar direction is excluded when computing the all-sky upper limit. Averaging per-cell 95% upper limits does not yield the 95% upper limit for a source at an unknown sky location: with a uniform prior over sky, the amplitude posterior is a mixture of the per-cell posteriors, and its 95% quantile is generally much larger than the arithmetic mean of the per-cell quantiles. This matters because single-pulsar sensitivity vanishes in the anti-pulsar direction: the paper quotes 1.41e-10 for the anti-pulsar half-sky versus 7.01e-13 for the pulsar half. The reported 1.26e-12 therefore depends on the arbitrary grid and on the unspecified exclusion region; the paper itself notes that the excluded area in Perera et al. is not mentioned, so the 'sky-average' comparison is uncontrolled. If the intended quantity is instead a sky-averaged sensitivity (e.g., the inverse-square average of per-cell limits), that is a different, non-coverage statement and must be labeled as such; the abstract currently attaches '95% confidence level' to a number without a defined frequentist or Bayesian coverage for an unknown source direction.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript analyzes 1.67 years of FAST/CPTA timing data for PSR J1713+0747, including a dense September 2020 observing campaign, to set upper limits on the strain amplitude of monochromatic gravitational waves from circular supermassive black-hole binaries in the 10^-7 to 3e-6 Hz band. The noise analysis uses TEMPONEST with Bayes-factor model selection and adopts an EFAC+EQUAD+ECORR+DM noise model, with data after the April 2021 profile change excluded. The authors then derive per-sky-cell upper limits using both a Cramer-Rao lower bound (with fixed maximum-likelihood noise parameters) and a Bayesian fixed-frequency analysis, from which they construct sky-average and best-sky limits. The headline results are a sky-average 95% upper limit of 1.26e-12 at 1 μHz and a best-sky limit of 4.77e-13.","tokens_in":13819,"tokens_out":8619,"duration_ms":85441,"significance":"The paper is potentially valuable: it demonstrates that a single high-precision millisecond pulsar observed with FAST can probe the relatively unexplored μHz gravitational-wave band, and the agreement between the Bayesian and Cramer-Rao analyses provides a useful cross-check. The explicit disclosure of the data cutoff after the 2021 profile change and the use of Bayes factors for noise model selection are commendable. However, the statistical interpretation of the headline sky-average limit needs to be corrected or clarified before the quoted numbers can be used as 95% confidence upper limits for a source of unknown sky direction; the current construction is an average of per-cell limits and is not uniquely defined because of the unspecified anti-pulsar exclusion region.","major_comments":[{"comment":"The quantity called 'sky-average continuous source upper limit at the 95% confidence level' is not a well-defined 95% upper limit for a source at an unknown sky location. Section 4.2 states that the sky-average is obtained 'by averaging the upper limits derived in 400 sky cells' on a 20×20 grid, and Section 4 states that an area around the anti-pulsar direction is excluded when computing the all-sky limit. For a uniform prior over the sky, the amplitude posterior is a mixture of the per-cell posteriors, and its 95% quantile is generally larger than the arithmetic mean of the per-cell quantiles. Moreover, no value is given for the excluded anti-pulsar area, so the quoted 1.26e-12 is not uniquely defined. The authors should either label the reported number as a sky-averaged sensitivity (a non-coverage statement) and remove the '95% confidence level' claim attached to the average, or compute a proper all-sky upper limit, for example by marginalizing the Bayesian posterior over sky position or by quoting the worst-cell 95% limit. This issue is load-bearing because the abstract's headline result uses this quantity.","section":"Abstract and Section 4.2"},{"comment":"The comparison with Perera et al. (2018) is uncontrolled. The paper itself notes that the excluded area around the anti-pulsar direction in Perera et al. 'is not mentioned', so the two sky-averaged quantities need not be defined over the same sky region; given the orders-of-magnitude variation in sensitivity across the sky, the grid and exclusion choices can change the average substantially. The claim that the sky-average limit is ~3.5 times higher than the Perera et al. value should be recomputed using the same sky coverage and the same averaging prescription, or the comparison should be restricted to sensitivity curves that do not depend on the arbitrary sky-average definition.","section":"Section 5"}],"minor_comments":[{"comment":"The pulsar is named PSR J1713+5307 in the abstract but PSR J1713+0747 in the title and the rest of the paper; the catalog name should be made consistent.","section":"Abstract"},{"comment":"The statement that the Bayesian-inference result 'must be worse than the CRLB in any case' is not generally correct, since informative priors or boundary effects can make a posterior interval narrower than the Cramer-Rao lower bound; the intended comparison is with an unbiased frequentist estimator and should be phrased accordingly.","section":"Section 4.1"},{"comment":"The comparison uses twice the CRLB as a proxy for the 95% Bayesian upper limit, but for a one-sided 95% upper limit on a Gaussian parameter the appropriate multiplier is approximately 1.65 rather than 2; the authors should state which convention is intended so the agreement is meaningful.","section":"Fig. 4"},{"comment":"The sentence describing the grid, 'a 20×20 grid in equally large in RA and cos(DEC)', is grammatically unclear and should be rewritten to state explicitly that the grid cells are equal in RA and in cos(DEC) and are not equal-area on the sky.","section":"Section 4.2"},{"comment":"The sentence 'the sky-average sky limit is ~3.5 times higher, and the best-sky limit ~2.2 times higher' is ambiguous about which frequencies and which averaging convention are used, especially because the simulated-curve comparison is also quoted with a factor of 2.2.","section":"Section 5"},{"comment":"There are several minor grammatical slips, e.g., 'the current paper also places upper limit using single-pulsar data' should read 'places upper limits'; a careful copyedit is recommended.","section":"Introduction"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of the journal and addresses an interesting and timely topic. The main technical concern is the statistical definition of the headline sky-average upper limit; if the authors re-frame it as a sky-averaged sensitivity or compute a proper all-sky 95% upper limit, the paper would be acceptable. The comparison with Perera et al. also needs to be recomputed under a common definition. I do not see an issue of circularity; the noise parameters are estimated from the same data as the upper limits, but that is standard practice and the target quantity is not used to set those parameters."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"New numbers, old story: this is the first μHz single-source upper limit from FAST on PSR J1713+0747, and it is honestly weaker than Perera et al. (2018). The analysis is competent and cross-checked—Bayesian and CRLB agree, noise model selection is transparent, and the post-2021 profile-change cutoff is disclosed. The paper is a useful CPTA pipeline demonstration, not a result that changes the μHz landscape.\n\nThe actual limits at 1 μHz are 1.26×10⁻¹² (sky-average) and 4.77×10⁻¹³ (best-sky). The best-sky number is well defined. The sky-average is not. It is the arithmetic mean of 400 per-cell 95% upper limits on a 20×20 grid, with an unspecified region around the anti-pulsar direction excluded. Averaging per-cell quantiles does not produce a 95% quantile for an unknown source direction; the abstract's '95% confidence level' attached to 1.26×10⁻¹² is therefore misleading. The paper discloses the procedure, but the headline number is grid- and exclusion-dependent, and the comparison to Perera et al. is uncontrolled because Perera's excluded region is not stated. Fix: report a proper sky-averaged sensitivity (e.g., inverse-square average) labeled as a sensitivity, or give per-cell limits with the exclusion mask. This is the one issue that should hold up acceptance.\n\nMinor: the abstract misspells the pulsar as J1713+5307; Section 5 has an incomplete sentence ('provided the best SSGW strain amplitude limits at frequencies') that appears to contradict the paper's own comparison; 'stain amplitude' in the intro. All easy fixes.\n\nThe noise model (white + DM, no red) is defended with Bayes factors, and the possibility of unmodeled red noise masking a signal is acknowledged. The CRLB cross-check is internal but appropriate. No circularity.\n\nThis is a solid, incremental contribution. It deserves a serious referee—the data are new, the pipeline is real, and the statistical definition issue is fixable. If I were the editor, I would send it to review with a request to clarify the sky-average statement.","headline":"Solid but incremental μHz limits from FAST; the 'sky-average' headline number needs a statistical fix before it's citable.","tokens_in":14437,"tokens_out":3274,"would_cite":false,"duration_ms":32287,"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":"The Chinese Pulsar Timing Array sets microhertz gravitational-wave strain upper limits from one FAST pulsar: $1.26\\times10^{-12}$ sky-averaged, $4.77\\times10^{-13}$ toward the pulsar at 1 μHz.","keywords":["pulsar timing arrays","gravitational waves","microhertz frequencies","supermassive black hole binaries","PSR J1713+0747","FAST telescope","Bayesian upper limits","dispersion measure noise"],"falsifier":"Re-run the Bayesian upper-limit calculation on the same arrival times with an extra power-law red-noise component in the $10^{-7}$ to $3\\times10^{-6}$ Hz band; if the 95% sky-averaged limit at 1 μHz rises substantially above $1.26\\times10^{-12}$, the quoted bound depends on the absence of red noise. An independent check is to inject a known continuous-wave signal at a strain just below the claimed limit and confirm that it is recovered at the nominal rate.","tokens_in":13289,"feed_emoji":"🔭","tokens_out":14775,"duration_ms":125428,"temperature":0.7,"pith_summary":"The paper seeks to show that high-precision timing of a single millisecond pulsar can probe the microhertz gravitational-wave band, a frequency range above the traditional pulsar-timing-array band and less explored. Using FAST observations of PSR J1713+0747 from August 2019 to April 2021, including a dense campaign in September 2020, it reports 95% confidence upper limits on the strain of monochromatic continuous waves from circular supermassive-black-hole binaries: $1.26\\times10^{-12}$ sky-averaged and $4.77\\times10^{-13}$ in the pulsar's direction at 1 μHz. These are new single-source upper limits from CPTA/FAST data, and they reach the level of published EPTA limits that used roughly 4.3 years of data, thanks to the pulsar's high timing precision. The result matters because it maps the capability of a single-pulsar detector and indicates that combining the full CPTA pulsar set could push single-source limits toward $10^{-14}$ in the microhertz band.","feed_headline":"One pulsar's FAST data yield new microhertz gravitational-wave limits","feed_subtitle":"Sky-averaged 95% bound is $1.26\\times10^{-12}$ at 1 μHz, matching earlier EPTA limits with a single pulsar.","key_machinery":"The load-bearing machinery is the standard Gaussian pulsar-timing likelihood together with a covariance matrix built from the selected noise model (white noise plus power-law dispersion-measure variations, with no red-noise component). On top of this sits the seven-parameter monochromatic continuous-wave signal model for a circular supermassive-black-hole binary, whose Earth term and pulsar term both enter the residuals. The upper limits come from Bayesian posterior inference on the strain amplitude with uniform priors over the other six signal parameters, and they are cross-checked with the Cramér-Rao lower bound, the theoretical best variance any unbiased estimator can achieve, computed from the same covariance matrix.","core_discovery":"The central claim is that FAST timing data of PSR J1713+0747 set the current microhertz single-source gravitational-wave strain upper limits from the Chinese Pulsar Timing Array. In the Bayesian analysis, the 95% confidence sky-averaged upper limit at 1 μHz is $1.26\\times10^{-12}$, and the limit in the sky direction of the pulsar is $4.77\\times10^{-13}$; with the noise fixed to maximum-likelihood values the corresponding numbers are $1.10\\times10^{-12}$ and $1.01\\times10^{-13}$. Only circular orbits and monochromatic waves are considered, and the seven-parameter signal model includes the Earth and pulsar terms. Because a single pulsar has a dipolar response, the sky-averaged limit is computed over 400 sky cells with the anti-pulsar region excluded, and a fine 1600-cell sky map is produced at 1 μHz. The paper also shows via simulations that the gap to earlier EPTA limits is mostly explained by the shorter time span rather than by data quality.","pith_inferences":["The no-red-noise assumption is the place to stress-test the result: if a time-correlated red process exists between $10^{-7}$ and $3\\times10^{-6}$ Hz, the reported limits would be optimistic, so re-running the analysis with a red-noise term would show how much the quoted numbers depend on that choice.","The September 2020 dense campaign is what opens the microhertz window; applying the same short-cadence strategy to other high-precision pulsars could extend microhertz coverage much sooner than waiting for decade-long baselines.","Sky-averaged limits from different single-pulsar studies are not strictly comparable unless the same anti-pulsar exclusion region is used, so publishing the per-cell limit maps, as this paper does, is the better basis for future comparisons."],"forward_implications":["The 1 μHz sky-averaged and pulsar-direction limits are new single-source gravitational-wave bounds from CPTA/FAST data, and the best-sky value is comparable to earlier EPTA results despite a much shorter time span.","With a single pulsar the sensitivity pattern peaks in the pulsar direction and almost vanishes in the opposite direction, so the sky-averaged limit is effectively a pulsar-hemisphere constraint, not a uniform all-sky bound.","Simulations in the paper attribute most of the remaining gap to EPTA limits to the 1.67-year time span; extending to 4.3 years would improve the microhertz sensitivity by about a factor of 2.2.","Combining the rest of the CPTA pulsars over 3.4, 10, and 20 years is projected to lower the single-source strain upper limit to roughly $10^{-14}$, potentially making pulsar timing the most sensitive microhertz gravitational-wave detector."],"supporting_citations":[{"why":"Supplies the seven-parameter monochromatic continuous-wave model and the Earth-term/pulsar-term signal formalism used for the upper-limit calculation.","marker":"Lee et al. 2011"},{"why":"Previous single-pulsar microhertz continuous-wave upper limits from EPTA data; the main comparison baseline for the new limits.","marker":"Perera et al. 2018"},{"why":"Establishes the Bayesian noise-model selection methodology and the way red-noise and dispersion-measure components are handled in pulsar timing arrays.","marker":"Lentati et al. 2016"},{"why":"Provides the Gaussian pulsar-timing likelihood with timing parameters marginalized, the basis for both noise and signal analyses.","marker":"van Haasteren et al. 2009"},{"why":"Supplies the FORTYTWO analysis software and the Cramér-Rao lower-bound approach used to cross-check the Bayesian limits.","marker":"Caballero et al. 2016"},{"why":"Context for the CPTA dataset and the stochastic-background evidence that motivates single-source microhertz searches.","marker":"Xu et al. 2023"}],"fun_headline_variants":["FAST timing of one pulsar sets microhertz GW limits","Chinese PTA single-pulsar data set muHz gravitational-wave bounds","Microhertz GW upper limits from FAST pulsar data","PSR J1713+0747 FAST data probe microhertz GWs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The analysis assumes that the timing residuals are fully described by white noise plus dispersion-measure variations; if any additional time-correlated red noise exists in the microhertz band, the reported upper limits would be optimistically biased.","fun_headline_variants_meta":{"raw":{"variants":["FAST timing of one pulsar sets microhertz GW limits","Chinese PTA single-pulsar data set muHz gravitational-wave bounds","Microhertz GW upper limits from FAST pulsar data","PSR J1713+0747 FAST data probe microhertz GWs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000679,"raw_usage":{"total_tokens":3103,"prompt_tokens":981,"completion_tokens":2122,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":597,"completion_tokens_details":{"reasoning_tokens":2057}},"tokens_in":597,"tokens_out":2122,"duration_ms":16726,"temperature":1.0,"reasoning_tokens":2057,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T22:03:09.824455+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the Bayesian upper-limit calculation on the same arrival times with an extra power-law red-noise component in the $10^{-7}$ to $3\\times10^{-6}$ Hz band; if the 95% sky-averaged limit at 1 μHz rises substantially above $1.26\\times10^{-12}$, the quoted bound depends on the absence of red noise. An independent check is to inject a known continuous-wave signal at a strain just below the claimed limit and confirm that it is recovered at the nominal rate.","supporting_citations":[{"cited_title":"J., Wex, N., Kramer, M., et al","cited_arxiv_id":null,"evidence_quote":"Supplies the seven-parameter monochromatic continuous-wave model and the Earth-term/pulsar-term signal formalism used for the upper-limit calculation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Previous single-pulsar microhertz continuous-wave upper limits from EPTA data; the main comparison baseline for the new limits."},{"cited_title":"M., Coles, W","cited_arxiv_id":null,"evidence_quote":"Establishes the Bayesian noise-model selection methodology and the way red-noise and dispersion-measure components are handled in pulsar timing arrays."},{"cited_title":"N., Lee, K","cited_arxiv_id":null,"evidence_quote":"Supplies the FORTYTWO analysis software and the Cramér-Rao lower-bound approach used to cross-check the Bayesian limits."}],"review_version":1}