{"id":"f82f5d22-0cda-463d-9ead-9410cdca8d37","arxiv_id":"1908.03026","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":7,"one_line_summary":"Using 31 years of combined public timing data for PSR J1939+2134, the paper reports a nearly sinusoidal timing variation with a period close to the data span, which it interprets as either a moon-sized planet at about 11 AU or slow precession.","lead":"A single astronomer stitched together 31 years of public radio-telescope data on the millisecond pulsar PSR J1939+2134, producing one long timing record. It reports a remarkably stable pulsar clock and finds that a moon-sized planet or a slow wobble could explain the pulsar's 31-year timing wiggle.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The planet claim rests on a ~31-year sinusoid visible in one cycle of data, but the paper never tests whether a red-noise-only model fits the same timing residuals as well as the Keplerian orbit; absent that comparison, the fitted orbit parameters and derived moon mass are not established.","rationale":"The paper does substantial and valuable data-archaeology work: combining 36 sub-systems, deriving a consistent DM curve, estimating instrumental offsets, and obtaining timing residuals with rms near 0.5 microsecond. The clock-stability and ISM spectral-index results are plausible and do not depend on the planetary interpretation. The load-bearing step is the interpretation of the residual curvature as a coherent 31-year periodicity. The author explicitly recognizes the one-cycle limitation in Section 4 and the period scatter in Section 5.2, which is already a serious caveat. But the decisive missing test is a quantitative model comparison against a red-noise-only description. Because the data span equals the claimed period, a steep red-noise process can easily produce apparent quasi-sinusoidal curvature; a low-frequency Fourier component of red noise will look periodic over one cycle. The paper's robustness check in Figure 10 and Table 4 only shows that the fitted period does not wildly vary when early data are excised; it does not test whether a red-noise process would behave similarly. The quoted formal errors are not reliable given the large reduced chi-square values and the unmodeled red-noise uncertainty. A Bayes factor or likelihood-ratio test against a power-law red-noise model with identical white-noise parameters would settle the question. If such a test favors a red-noise-only model, or is inconclusive, the moon-mass companion claim should be downgraded to 'not ruled out' at best. This is exactly the condition attached by the reader, so the CONDITIONAL verdict remains appropriate and no adjustment is needed.","tokens_in":19861,"tokens_out":4108,"duration_ms":46405,"concrete_test":"Fit the same combined timing residuals under two models: (M1) a Keplerian orbit plus per-sub-system white noise (EFAC/EQUAD), and (M2) a power-law red-noise process plus the same white-noise parameters, using nested sampling or an equivalent marginal-likelihood estimate. Repeat the analysis with the MJD 51200–53200 segment both included and excluded. If the log Bayes factor for M1 over M2 is not strongly positive (for example, below about 5), or if M2 is preferred, the planetary-companion claim is unsupported; additionally report the posterior distribution of the orbital period to check whether it is bounded away from the total data span.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The decisive assumption is that the quasi-sinusoidal curvature in the timing residuals is a deterministic periodic signal rather than a realization of red timing noise. The paper's own numbers expose the fragility: Table 1 gives P = 11105–13068 d for four fits with formal errors of only a few days, and Section 5.2 concedes the true period uncertainty is about 2.1 yr; Section 4 concedes only one cycle is available. The robustness test in Figure 10 and Table 4 is not a model comparison: excising 0–8 yr of initial data changes P from 13068 to 13370 d, which is plausibly how a red-noise realization would respond, and the fit still excludes MJD 51200–53200. No likelihood ratio, Bayes factor, or false-alarm probability is computed against a steep-spectrum red-noise-only process (e.g., a power-law or harmonic model) with the same white-noise treatment. The reduced chi-square values of 24–39 in Table 1 also show that the quoted parameter uncertainties are not statistically meaningful under the assumed noise model. Therefore the inference from A, e, and P to a 3.5e-8 solar-mass companion at 11 AU is underdetermined.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents a uniform reanalysis of 31 years of public IPTA and NANOGrav timing data for the millisecond pulsar PSR J1939+2134 (B1937+21). It derives a dispersion-measure curve, applies consistent data selection and instrumental-offset corrections across 36 sub-systems, and models red-noise correlations with both the IPTA and NANOGrav prescriptions. The resulting timing residuals are reported to be very close to a sinusoid with period about 31 years. The paper fits this timing noise with a Keplerian planetary-companion model, obtaining a projected semi-major axis A of about 137 microseconds, eccentricity e about 0.215, and period P about 11296 days, which it interprets as a Moon-sized companion at about 11 AU; a precession model is fitted as an alternative. Additional results include a clock stability of almost one part in 10^15, an interstellar electron-density spectral index beta = 3.86 +/- 0.04, and excess achromatic timing noise of about 8 microseconds amplitude during epochs of steep DM gradient.","tokens_in":20159,"tokens_out":3941,"duration_ms":41888,"significance":"If the quasi-sinusoidal timing residual is a coherent, deterministic signal, then the paper reports a notable result for millisecond-pulsar timing and for planet formation around recycled pulsars. The manuscript has real strengths: the careful merging of heterogeneous public data, the explicit handling of 36 instrumental sub-systems, the cross-check between two independent red-noise treatments, and the candid reporting of large reduced chi-square values and the one-cycle limitation. However, the central claim that the timing noise is a true periodicity rather than a realization of steep-spectrum red noise is not yet supported by a model comparison, and the fitted period is not robust across the four analyses. The significance of the paper is therefore conditional: the data combination and the descriptive fits are valuable, but the companion and precession interpretations require substantially stronger statistical evidence.","major_comments":[{"comment":"The paper fits deterministic sinusoidal (Keplerian and precession) models to the timing residuals but never compares them against a red-noise-only model, such as a power-law or harmonic process with the same white-noise and T2 treatment. With only one cycle of the putative 31-year period in the data, a realization of steep-spectrum red noise can plausibly produce the same apparent curvature. A likelihood ratio, Bayes factor, or false-alarm probability against a red-noise-only process is needed before the claim that the timing noise is 'very close to a sinusoid' can be accepted.","section":"Section 3.2, Figures 6 and 7"},{"comment":"The orbital period is not robust: the four analyses give P = 11105, 13068, 11403, and 11381 days, a spread of about 2.1 years against formal errors of a few days. The paper itself acknowledges this in Section 5.2 and advises using the larger uncertainty, but the companion mass of about 3.5 x 10^-8 solar masses is quoted without propagating this systematic period uncertainty. The derived companion mass and semi-major axis are therefore not established to the accuracy implied by the text.","section":"Section 3.2.1, Table 1 and Section 5.2"},{"comment":"The fit explicitly ignores the excess timing noise between MJD 51200 and 53200, which spans about 5.5 years of the 31-year baseline. Excluding a large contiguous block of data from a fit to a long-period sinusoid can bias the estimated period and amplitude, especially when the excluded interval coincides with a large DM-gradient epoch. The paper should either model this excess noise simultaneously, include it with an additional variance term, or demonstrate explicitly that its exclusion does not change the inferred period and amplitude beyond the quoted uncertainties.","section":"Section 3.2.1, Figure 6 and Section 3.3"},{"comment":"The preference for the planetary-companion model over the precession model is based on the rms and reduced chi-square of the best fits, but the reduced chi-square values in both tables are 24-39 and 104-226 respectively, far above unity. This indicates that the formal parameter uncertainties are not statistically meaningful under the assumed noise model, so the comparison of the two models is not a valid model selection without a common treatment of the noise and a proper goodness-of-fit or Bayesian evidence calculation.","section":"Section 3.2.2, Tables 1 and 2"}],"minor_comments":[{"comment":"The notation 'sin 2 (omega_p (t - t_0))' in Equation (1) is ambiguous: it should be written as sin(2 omega_p (t - t_0)) or with an explicit square, as the context indicates a second harmonic rather than a squared sine.","section":"Equation (1)"},{"comment":"Several axis labels contain corrupted characters, for example 'T m ng Res duals' and 'Calenda( Yea(' in Figure 1, and 'Epoch in Calenda( Yea(' in Figure 5. These should be fixed in the final version.","section":"Figures 1, 2, and 5"},{"comment":"The value beta = 3.86 +/- 0.04 is derived from a spline fit to the DM data, but the spline smoothing and the absence of error bars on the earlier digitized DM points are not fully propagated into this uncertainty. The quoted error likely underestimates the systematic contribution from the spline model choice.","section":"Section 3.1"},{"comment":"The statement that there is 'no correlation' of the derived period with the length of data used is not quantitatively supported. Table 4 shows P increasing monotonically from 13068 to 13370 days as the first 0 to 8 years are excised; a correlation coefficient or a significance test should be reported.","section":"Section 5.2, Table 4"},{"comment":"The digitized DM curve from Ramachandran et al. (2006) and Demorest (2007) is used without a description of the digitization uncertainty beyond the alignment rms of 3 x 10^-4 pc cm^-3. A brief discussion of how the digitization error enters the DM curve and the subsequent timing analysis would improve reproducibility.","section":"Section 2.2"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is an honest, clearly written single-author analysis of a difficult data-combination problem, and the descriptive results (DM curve, data selection, instrumental offsets) are useful. The main weakness is statistical: the periodic-signal claim and the companion inference are not tested against a red-noise-only alternative, and the large period spread across the four analyses shows that the systematic uncertainty is far larger than the formal errors. Major revision with a model-comparison analysis is appropriate; I do not see the issues as beyond repair, because the claims can be reframed as tentative and supported by the required statistical tests."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things. First, this is the first analysis to properly combine the IPTA and NANOGrav public data sets for J1939+2134, and the effort put into handling frequency-dependent profile evolution, DM estimation, and inter-sub-system delays is real. Second, the headline result—a moon-mass companion at 11 AU—is not something I would put money on. The paper is honest about its own limits, but those limits are close to the center of the claim.\n\nThe genuinely useful parts: a 31-year clock stability estimate at the one-part-in-10^15 level, a clean measurement of the ISM spectral index (beta = 3.86 ± 0.04), a careful DM curve, and a documented excess timing noise episode that coincides with steep DM gradient. These are worth having, and the data-selection scheme for making NANOGrav wide-band data compatible with IPTA narrow-band data is thoughtful. The four-way fits with MCMC error estimation are also more thorough than what the original IPTA or NANOGrav papers did for this pulsar.\n\nThe soft spot is the planetary interpretation. The residuals do show a roughly sinusoidal term with a period near the data span, but there is only one cycle in 31 years. The paper's own Table 1 shows the fitted period bouncing between 11105 and 13068 days across four reasonable analyses, a spread of about 2.1 years against formal errors of a few days. The paper acknowledges this and says the larger uncertainty should be used, which is good, but it does not fix the problem. More importantly, there is no formal test against a steep-spectrum red-noise process. The robustness test in Figure 10 excises the first 2–8 years and shows the period drifts from 13068 to 13370 days; the paper reads that as stability, but it is exactly the kind of drift you would expect from a red-noise realization. The reduced chi-square values of 24–39 also tell you the quoted uncertainties are not statistically meaningful under the assumed noise model.\n\nThe paper also excludes the MJD 51200–53200 data from the orbit fit because of excess noise whose cause is unknown. That is a significant chunk of the already-short baseline, and it raises the question of what else in the residuals is being absorbed by the model. The precession alternative is presented as a worse fit, which is fair, but neither model is tested against a \"no periodic signal\" hypothesis.\n\nMy take: the paper deserves a serious referee. An editor should not desk-reject it, because the data combination and the timing-noise characterization are useful, and the planet claim, while underdetermined, is not absurd. But the referee should push for a formal likelihood comparison against red-noise-only models, a model of the excluded MJD segment, and a demonstration that the period is stable when fit with a proper noise model. I would cite the DM/clock work, not the planet. And yes, I would bring it to a reading group, mainly to argue about the red-noise analysis.","headline":"A careful, valuable combination of 31 years of public timing data for J1939+2134, but the moon-mass planet claim rests on one cycle of a quasi-sinusoid and never gets tested against a red-noise-only model.","tokens_in":20740,"tokens_out":2200,"would_cite":true,"duration_ms":23842,"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":"After merging 31 years of public timing data, millisecond pulsar J1939+2134 shows a near-sinusoidal timing signal that a Moon-sized companion in an 11-AU eccentric orbit could explain.","keywords":["pulsar timing","millisecond pulsar","PSR B1937+21","timing noise","planetary companion","neutron star precession","dispersion measure","interstellar medium"],"falsifier":"Continue timing J1939 for another ~15–30 years to see whether the sinusoid repeats at the same phase, period, and amplitude; if the residuals do not trace the same curve in the second cycle, both the Moon-mass companion and the precession model are falsified. A shorter-term test is to compute a periodogram of the merged residuals with a red-noise false-alarm threshold: unless the ~31-year peak is significant against that threshold, it should not be assigned to any physical mechanism.","tokens_in":19603,"feed_emoji":"🪐","tokens_out":15401,"duration_ms":145499,"temperature":0.7,"pith_summary":"The paper tries to establish that 31 years of combined public timing data for the millisecond pulsar J1939+2134 contain a near-sinusoidal timing signal with a period of about 31 years. It argues that the most natural physical source is a Moon-sized planetary companion in an eccentric orbit at about 11 astronomical units, and that a torque-driven precession of the neutron star is a less good but still viable alternative. It further reports that the pulsar's rotation clock is stable to almost one part in $10^{15}$ over three decades, and that the interstellar electron-density fluctuation spectrum toward the pulsar has a power-law index $3.86 \\pm 0.04$. If the companion interpretation holds, the planet's inferred mass, separation, and eccentricity would sit between those of the planets of PSR B1257+12 and PSR B0329+54, tightening constraints on how planets form around recycled pulsars.","feed_headline":"Millisecond pulsar's 31-year wobble hints at a Moon-sized planet","feed_subtitle":"Combining 31 years of public timing data shows a near-sinusoidal signal a lunar-mass companion at 11 AU could explain.","key_machinery":"The load-bearing object is the merged, uniformly processed 31-year set of pulse arrival times for J1939. The procedure keeps the entire first public timing data set, prunes the wide-band observations of the second down to a narrow frequency slice around the first data set's frequencies (removing the need for frequency-dependent 'FD' delay parameters), estimates the dispersion measure epoch by epoch with a dual-frequency method and a digitized early DM curve, aligns the 36 instrument sub-systems with fixed clock-offset ('JUMP') terms, and models red-noise (low-frequency wander) correlations using either a spectral function or a power-law harmonic approach in the standard timing-modeling software. The fit that carries the planetary claim is a full Keplerian orbit, with no linearized approximation: the projected semi-major axis $A\\approx137$ µs and period $P\\approx11{,}300$ days relate through Kepler's third law to a Moon-mass companion at about 11 AU. The competing precession mechanism is the formula $f(t)=k+a_1\\sin(\\omega_p(t-t_0))-a_2\\sin 2(\\omega_p(t-t_0))$, whose first and second harmonic amplitudes carry the oblateness and wobble-angle information.","core_discovery":"The central discovery is that the timing residuals of PSR J1939+2134, after merging the two public long-term timing data sets with one uniform selection and dispersion-measure treatment, are dominated by a near-sinusoidal signal with period $P \\approx 11{,}300$ days ($\\approx 31$ yr), projected semi-major axis $A \\approx 137$ µs, and eccentricity $e \\approx 0.21$. A full Keplerian fit gives a companion mass of about $3.5\\times 10^{-8}$ solar masses, close to the Moon's mass, in a relative orbit of semi-major axis $\\approx 11$ AU. A torque-driven precession model with oblateness $\\epsilon \\approx 1.6\\times 10^{-12}$ and $\\theta \\tan\\chi \\approx 0.4$ also fits the data, but with residual scatter about 2.5 times larger than the planetary fit. The paper also reports an unexplained, frequency-independent timing excess of about 8 µs concentrated during the epoch of steepest DM gradient, and a clock-stability level of almost one part in $10^{15}$ over about 31 years.","pith_inferences":["Because the four fits in the paper give periods spread from 11,105 to 13,068 days, a fair reader should treat the nominal '31-year' period as known only to about ±2 years until a second cycle is observed.","A stronger test would fit a Keplerian companion and a red-noise process simultaneously in one Bayesian model; the paper fits the sinusoid separately from the noise model, so it cannot fully exclude the possibility that the sinusoid is absorbed noise.","If the excess noise during the steepest DM gradient is interstellar, it should appear in other pulsars behind the same scattering screen; if it does not, the excess likely originates in the pulsar's own magnetosphere.","Should the timing baseline reach a second cycle, a stable phase and period would favor the companion, while a drifting period or growing second harmonic would favor torque-driven precession."],"forward_implications":["If the sinusoid is real, J1939's rotation is a clock stable to almost one part in $10^{15}$ over 31 years, making it one of the most stable celestial clocks known.","A Moon-mass companion at about 11 AU with $e\\approx0.2$ would place J1939's planet between the close, nearly circular planets of PSR B1257+12 and the wider, eccentric planet of PSR B0329+54, so any formation scenario must explain both extremes.","If torque-driven precession is instead correct, the inferred oblateness $\\epsilon\\approx1.6\\times10^{-12}$ implies a nearly unstrained, largely decoupled neutron-star crust, consistent with the complete absence of glitches in J1939.","The power-law index $\\beta=3.86\\pm0.04$ for electron-density fluctuations indicates the interstellar scattering environment along this line of sight has been statistically stable across three decades.","The unexplained ~8 µs excess coinciding with the steepest DM gradient marks a new phenomenon that future multi-frequency monitoring can target directly."],"supporting_citations":[{"why":"Supplies the original timing method, the DM estimation equation, and the first stability estimate for J1939.","marker":"Kaspi, Taylor & Ryba (1994)"},{"why":"Provides the first of the two public long-term timing data sets and its spectral red-noise treatment.","marker":"Verbiest et al. (2016)"},{"why":"Provides the second public long-term timing data set, with its wide-band and power-law noise treatment.","marker":"Arzoumanian et al. (2018)"},{"why":"Supplies the method for including red-noise correlations in weighted least-squares timing fits.","marker":"Coles et al. (2011)"},{"why":"Supplies the intermediate DM measurements that are digitized to bridge the public-data gap.","marker":"Ramachandran et al. (2006)"},{"why":"Provides the torque-precession model, oblateness formulas, and crust-superfluid coupling discussion.","marker":"Link & Epstein (2001)"},{"why":"Supplies an alternative precession formula and comparison oblateness values for other pulsars.","marker":"Jones & Andersson (2001)"},{"why":"Supplies the second-harmonic precession timing formula used for the curve fit.","marker":"Akgun et al. (2006)"},{"why":"Earlier 26-year timing-noise study whose asteroid-belt model is the direct precedent the paper's periodic fit goes beyond.","marker":"Shannon et al. (2013)"}],"fun_headline_variants":["Moon-sized planet may explain pulsar's 31-year wobble","Pulsar's 31-year signal hints at Moon-mass companion","Millisecond pulsar's timing reveals possible exomoon","31-year pulsar wobble points to Moon-sized world","Exomoon around pulsar? New analysis of J1939+2134"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the near-sinusoidal ~31-year wobble is a real coherent signal and not a chance realization of slowly wandering timing noise; only one full cycle lies inside the dataset, and the fitted period moves by about two years depending on how the noise is treated.","fun_headline_variants_meta":{"raw":{"variants":["Moon-sized planet may explain pulsar's 31-year wobble","Pulsar's 31-year signal hints at Moon-mass companion","Millisecond pulsar's timing reveals possible exomoon","31-year pulsar wobble points to Moon-sized world","Exomoon around pulsar? New analysis of J1939+2134"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000887,"raw_usage":{"total_tokens":3921,"prompt_tokens":1128,"completion_tokens":2793,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":744,"completion_tokens_details":{"reasoning_tokens":2703}},"tokens_in":744,"tokens_out":2793,"duration_ms":20556,"temperature":1.0,"reasoning_tokens":2703,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:28:33.915633+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Continue timing J1939 for another ~15–30 years to see whether the sinusoid repeats at the same phase, period, and amplitude; if the residuals do not trace the same curve in the second cycle, both the Moon-mass companion and the precession model are falsified. A shorter-term test is to compute a periodogram of the merged residuals with a red-noise false-alarm threshold: unless the ~31-year peak is significant against that threshold, it should not be assigned to any physical mechanism.","supporting_citations":[{"cited_title":"& Ryba, M.F","cited_arxiv_id":null,"evidence_quote":"Supplies the original timing method, the DM estimation equation, and the first stability estimate for J1939."},{"cited_title":"2018, ApJS , 235, 37 Backer D.C., Kulkarni S.R., Heiles C., Davis M.M., Goss W.M","cited_arxiv_id":null,"evidence_quote":"Provides the second public long-term timing data set, with its wide-band and power-law noise treatment."},{"cited_title":"Hobbs, G., Champion, D.J., Manchester, R.N., & J","cited_arxiv_id":null,"evidence_quote":"Supplies the method for including red-noise correlations in weighted least-squares timing fits."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the intermediate DM measurements that are digitized to bridge the public-data gap."},{"cited_title":"2001, ApJ, 556, 392","cited_arxiv_id":null,"evidence_quote":"Provides the torque-precession model, oblateness formulas, and crust-superfluid coupling discussion."},{"cited_title":"2006, MNRAS, 365, 653","cited_arxiv_id":null,"evidence_quote":"Supplies the second-harmonic precession timing formula used for the curve fit."},{"cited_title":"et al 2013, ApJ, 766, 5","cited_arxiv_id":null,"evidence_quote":"Earlier 26-year timing-noise study whose asteroid-belt model is the direct precedent the paper's periodic fit goes beyond."}],"review_version":1}