{"id":"6556002e-3a12-4075-9f83-21460798098c","arxiv_id":"2411.19899","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A new catalog gives the first directly observed mass loss rates for 12 Milky Way globular clusters, inferred from Gaia stellar stream densities.","lead":"Astronomers measured how fast 12 globular clusters are losing mass by comparing the density of their tidal streams in Gaia data with computer-generated mock streams. The resulting catalog of mass loss rates, the first of its kind, matches theoretical predictions based on cluster mass and orbital frequency.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The inferred Mdot is proportional to the assumed completeness inside detected stream segments; fsel_1 is set to 1 by normalizing the mock to the observed maximum, so an unmeasured constant incompleteness factor would bias the entire catalog and could distort the scaling.","rationale":"The paper's central claim is a catalog of mass-loss rates and a scaling relation. For that claim to hold, the conversion from observed stream counts to Mdot must be robust. The conversion has two large multiplicative corrections: the missing-mass weight w_i from the MF and the selection function fsel. The MF correction is explicitly tested and its alpha-sensitivity is quantified in Appendix B (0.1 dex per 0.3 in alpha). The selection-function correction is not similarly calibrated: fsel_1 is a 0/1 mask constructed by amplitude-matching the mock to the observed density, so any constant incompleteness in the 'complete' segments is absorbed into the overall normalization and misattributed to a lower Mdot. Since Eq. (9) is linear in 1/fsel, a stream-dependent factor of, say, 0.5 would lower all rates by about 0.3 dex and change the relative ranking used for the M-Omega correlation. The Pal 5 deep-photometry check is an excellent validation, but it calibrates only one stream; it cannot certify fsel = 1 for the other 11. I therefore agree partially with the reader: the MF assumption is important, but the selection-function normalization is more central and even less constrained. The appropriate verdict remains CONDITIONAL: the concern is addressable by the injection test or by deep photometry for a second stream, and it does not by itself invalidate the method, only the current error budget and the word 'directly observed.' Hence verdict_should_be is UNCHANGED.","tokens_in":16248,"tokens_out":10109,"duration_ms":96125,"concrete_test":"Injection-recovery test: generate C25 mock streams with known Mdot and a known cluster MF, inject them as synthetic Gaia DR3 sources for representative cases (e.g., NGC 3201-like disk-crossing stream, omega Cen-like wide stream, and a Pal 5-like stream), run the I24 STREAMFINDER pipeline, reconstruct fsel with the §2.5 procedure, and recover Mdot via Eq. (9). Repeat with artificial completeness levels c = 0.5, 0.7, and 1.0 inside the true stream segments. If the recovered Mdot deviates from the input by more than the quoted roughly 0.1 dex errors for any c < 1, the binary fsel = 1 assumption is the dominant systematic and the Table 1 rates and Eq. (12) slopes need revision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing step is the reconstruction of the spatial selection function in §2.5, not the MF correction. In Eq. (9), Mdot = Ndot_tracer * Msel / sum_j fsel(phi_j), so any multiplicative error in fsel enters Mdot linearly. The paper sets fsel_1 = 1 in segments where the observed 1D density is within 50% of the mock after 'normaliz[ing] the mock stream to match the observed maximum' (§2.5). That normalization absorbs any constant completeness factor c < 1 inside a segment: the shape comparison cannot distinguish c*P_mock from P_mock, so the reconstructed fsel is 1 even if only half the stream stars are detected. The same degeneracy applies to smoothly varying incompleteness along the stream. Thus fsel = 1 is an assumption of full STREAMFINDER completeness at the density peak, not a measurement. For streams crossing the disk (NGC 3201) or wide streams (omega Cen), c could be well below 1. The Pal 5 CFHT check (Appendix D) validates c = 1 for Pal 5 only; it does not calibrate the other 11 streams. Because this normalization error is not included in the quoted Poisson and MF uncertainties, the 'directly observed' mass-loss rates in Table 1 may be systematically low by a stream-dependent factor, and the fitted slopes a and b in Eq. (12) could be biased. The MF issue identified by the reader is real, but its sensitivity is quantified in Appendix B (about 0.1 dex per 0.3 in alpha); the fsel normalization is unquantified and at least as large.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a method to infer orbit-averaged mass loss rates of Galactic globular clusters by comparing the stellar density of observed streams (from the Ibata et al. 2024 Gaia DR3 catalog) with mock streams generated by the Chen et al. (2025) particle spray algorithm. The mass density of a stream is corrected for stars below the detection limit using a power-law stellar mass function, and the spatial selection function of the stream finder is reconstructed along and perpendicular to the stream track. Applying this to 12 of 16 globular cluster streams yields mass loss rates from 0.5 to 200 M_sun/Myr, and power-law fits give Mdot proportional to M_GC^0.66 Omega^0.54 r_h^0.12, consistent with N-body expectations. The paper includes an unbiasedness proof for the mass correction (Appendix C) and an external validation of the correction using deeper CFHT photometry of Pal 5 (Appendix D).","tokens_in":16641,"tokens_out":2328,"duration_ms":24704,"significance":"If the derived rates are accurate, this would be the first direct observational calibration of globular cluster mass loss rates and their scaling with cluster properties, providing a valuable constraint for models of cluster disruption and for initial cluster mass function evolution. The methodological core is attractive: Eq. (9) is a clean importance-sampling estimator, the unbiasedness proof in Appendix C is a useful formal step, the mock-stream comparison is carried out with a modern particle spray algorithm, and the code is publicly available. The external Pal 5 check against independent CFHT data is a genuine strength that partially validates the completeness correction. However, the central result depends on two unquantified assumptions: the normalization of the reconstructed selection function and the equality of the stream and cluster stellar mass functions. Both enter the quoted rates almost linearly, so the catalog values and the reported scaling exponents should be treated as provisional until these assumptions are tested.","major_comments":[{"comment":"The normalization of the reconstructed spatial selection function is degenerate with a constant completeness factor. In §2.5 the mock stream is normalized to match the observed maximum density, and fsel,1 is set to unity wherever the observed density is within 50% of the mock. Any constant incompleteness c < 1 inside a segment is absorbed by this normalization, so fsel = 1 is an assumption of full STREAMFINDER completeness at the stream density peak, not a measurement. Since Mdot in Eq. (9) is inversely proportional to the sum of fsel over mock particles, an unmeasured constant incompleteness would bias every catalog entry by a stream-dependent factor and could distort the fitted slopes a and b in Eq. (12). The Pal 5 CFHT test in Appendix D validates fsel = 1 only for Pal 5, not for the other 11 streams. Please quantify this normalization uncertainty, for example by injecting synthetic streams with a known completeness into the STREAMFINDER-like selection or by propagating a conservative prior on the peak completeness into the Mdot uncertainties.","section":"§2.5, Eq. (9)"},{"comment":"The mass correction assumes that the stellar mass function of the stripped stream is the same power law as that of the progenitor cluster, with slope alpha taken from Baumgardt et al. (2023). The correction weights w_i range from 2 to 200, so the inferred Mdot scales almost linearly with the assumed mass fraction below the Gaia detection limit. The paper acknowledges that low-mass stars are ejected preferentially (Webb & Bovy 2021) and that the correction is sensitive to the MF slope (Appendix B), but the quoted uncertainties include only Poisson noise and a 0.1 dex MF-slope error. A systematic difference between the stream MF and the cluster MF would change the catalog values by a large factor and could bias the comparison with N-body scalings. Please estimate the magnitude of this effect, e.g., by testing a mass-dependent ejection model or by marginalizing over a plausible range of stream MFs.","section":"§2.4, Eq. (6), Appendix B"},{"comment":"The analysis excludes four of the sixteen streams associated with GCs in the I24 catalog: three whose morphology cannot be reproduced with MWPotential2014 (NGC 1261, NGC 5466, NGC 7099) and one with too few stars (NGC 7089). If the excluded streams are systematically different in mass loss rate, orbital frequency, or stream age, the power-law fits in Eq. (12) will be biased even if each individual measurement is unbiased. The paper should at least report the properties of the excluded clusters and discuss whether the sample selection is likely to correlate with Mdot; a quantitative selection function for inclusion would be stronger.","section":"§3, Table 1"}],"minor_comments":[{"comment":"The sentence 'The two models produce similar density distributions, indicating fast mixing of stars in the stream.' appears twice verbatim; one copy should be removed.","section":"§2.2"},{"comment":"There is a typo in the affiliation: 'Tsinghua Univeristy' should be 'Tsinghua University'.","section":"Author affiliations"},{"comment":"The distance relation d(phi1) used to assign stellar masses is fitted from the mock stream, which couples the mass correction to the assumed orbital model; a brief discussion of how a mis-modeled track affects m_min(d_i) would help the reader assess this step.","section":"§2.4"},{"comment":"The left and right panels of Figure 3 are described with nearly identical captions in the text; please clarify in the caption which panel shows the mass loss rate and which shows the mean correction factor.","section":"Figure 3"}],"recommendation":"major_revision","confidential_remarks":"This is a promising and potentially influential paper, but the two main systematic uncertainties—the fsel normalization and the stream-versus-cluster mass function—are not yet quantified, and both affect the headline catalog values and the scaling exponents. The external Pal 5 test is a good step but does not calibrate the other eleven streams. I would not reject the paper; rather, I would ask for an explicit treatment of these systematics before publication. The sample selection (four of sixteen streams excluded) should also be discussed explicitly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read Chen, Li & Gnedin on stellar streams as mass-loss probes. Punchline: this is the first catalog of directly observed mass-loss rates for GCs (12 of them, 0.5–200 Msun/Myr), built by comparing I24 stream densities with C25 particle-spray mocks, and the method is mostly sound. If correct, it gives the first empirical handle on the Mdot–M_GC and Mdot–Omega scalings that N-body models have been arguing about for years.\n\nWhat’s good: The estimator (Eq. 9) is clean, and the unbiasedness proof for the mass correction (Appendix C) works. They ship the particle-spray code on Zenodo, and the method rests on an independently calibrated algorithm (C25) rather than something invented here. The Pal 5 CFHT check is a genuine external validation—deeper photometry, uniform selection, same answer—which is more than most papers at this level do. The paper is also careful in places: it names the three streams it cannot reproduce and one it drops for too few stars, and it quantifies the MF-slope sensitivity at about 0.1 dex per 0.3 in alpha.\n\nThe soft spot is the spatial selection function reconstruction in §2.5. They set fsel_1 = 1 in segments where observed density agrees with the mock after normalizing the mock to the observed maximum. That normalization absorbs any constant incompleteness factor c < 1 inside a segment. A shape comparison cannot distinguish c*P_mock from P_mock, so fsel = 1 is an assumption of full STREAMFINDER completeness at the density peak, not a measurement. The Pal 5 CFHT check calibrates c = 1 for Pal 5 only; it does not validate the other 11 streams. Since Mdot is proportional to fsel⁻¹, a stream-dependent constant incompleteness of, say, 2 would shift the catalog by that factor and could distort the fitted slopes. This is unquantified in the quoted uncertainties, which only include Poisson and MF errors. I think this is a larger issue than the MF correction, which is at least bounded.\n\nThe exclusion of four of 16 streams after morphology matching is worth noting but not damning; the paper states it plainly. The true uncertainties also almost certainly dominate over the quoted error bars because orbit and potential choices are not propagated. That said, the central idea is right, and the paper is honest about its limitations—it explicitly lists what it does not capture (dark remnants, binary kicks).\n\nWho for: anyone working on GC evolution, stream detection, or initial cluster mass functions. It deserves a serious referee. I would recommend major revision, with the selection-function normalization made explicit and, ideally, an external check on at least one more stream (e.g., omega Cen) with deeper data.\n\nMy two cents: do not desk reject; send to review with a request to quantify the fsel normalization degeneracy.","headline":"A genuinely new stream-density-based mass-loss catalog for 12 GCs, with an honest method section; the main weakness is the unquantified normalization in the spatial selection function, which can bias the whole catalog.","tokens_in":17162,"tokens_out":2166,"would_cite":true,"duration_ms":18891,"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":"Matching Gaia stream densities to mock streams yields the first direct mass-loss rates for 12 globular clusters, from 0.5 to 200 solar masses per Myr, with a scaling that rises with cluster mass and orbital frequency.","keywords":["stellar streams","globular clusters","mass loss rates","Gaia DR3","tidal stripping","stellar mass function","particle spray models","Milky Way"],"falsifier":"Take a stream such as NGC 6397 and observe it with a wide-field telescope reaching roughly $r=23$, comparable to the CFHT Pal 5 data, count stars below the main-sequence turnover, and derive the mass-loss rate with a correction factor near 6; if the resulting rate disagrees with the Gaia-based value by more than the quoted uncertainties, the assumed stream mass function is wrong.","tokens_in":16062,"feed_emoji":"🌌","tokens_out":8869,"duration_ms":71630,"temperature":0.7,"pith_summary":"This paper sets out to turn stellar streams, the tidal debris shed by globular clusters, into a direct observational measure of how fast the clusters are losing mass. Using Gaia DR3 stream detections and mock streams generated by a particle spray model, it derives orbit-averaged mass-loss rates for 12 Milky Way globular clusters, the first catalog of its kind, spanning 0.5 to 200 solar masses per million years. The rates correlate positively with cluster mass and orbital frequency, and the fitted power-law scaling is consistent within uncertainties with N-body predictions for clusters without black holes. If the method holds, stream density becomes a working observational probe of cluster disruption that can be extended to far more clusters with deeper surveys.","feed_headline":"First direct mass-loss rates from 12 globular cluster streams","feed_subtitle":"Gaia stream densities plus mock streams yield rates from 0.5 to 200 solar masses per Myr.","key_machinery":"The carrying mechanism is a comparison between observed and mock stream densities built on three components. First, a particle spray algorithm generates tracer particles from a Gaussian phase-space distribution calibrated on N-body simulations of disrupting clusters, reproducing stream width and length to better than ten percent while keeping morphology independent of the mass-loss rate. Second, the observed stellar mass above the Gaia detection limit is corrected to a total mass using per-star weights, the ratio of the integrated stellar mass function above the hydrogen-burning limit to that above each star's minimum detectable mass; a proof in Appendix C shows the resulting estimate is unbiased. Third, a spatial selection function, estimated separately along and across the stream track, down-weights mock particles in regions the stream-finding algorithm would miss. These pieces combine into an importance-sampling estimator that yields the orbit-averaged mass-loss rate from the observed star masses, the correction weights, and the mock tracer particles.","core_discovery":"The paper's central claim is that the surface density of a globular cluster's stellar stream encodes the cluster's orbit-averaged mass-loss rate, once the stream's width, length, and selection effects are modeled. By fitting Gaia DR3 stream density profiles with mock streams produced by a particle spray algorithm, the authors measure $|\\dot{M}|$ for 12 Galactic globular clusters, with rates spanning roughly 0.5 to 200 $M_\\odot\\,\\mathrm{Myr}^{-1}$. A multivariate power-law fit gives $\\dot{M}\\propto M^{0.66}\\,\\Omega^{0.54}\\,r_h^{0.12}$, with the half-mass radius slope consistent with zero; the mass and orbital-frequency slopes are positive and agree at the one-to-two $\\sigma$ level with the $(1/3,1,0)$ no-black-hole scaling from N-body simulations. The authors also show their unbiased estimator recovers the mass-loss rate of Pal 5 from independent CFHT photometry, where the correction for missing faint stars is much smaller than for Gaia.","pith_inferences":["Editorial inference: If the true stream mass function is depleted or enriched in low-mass stars relative to the parent cluster, as preferential tidal ejection would produce, the inferred rates shift by roughly the same large factor as the correction weights, so the catalog values should be read as conditional on the adopted mass-function slopes.","Editorial inference: Because the method requires known progenitor positions and velocities, it cannot yet see clusters that have already fully dissolved; combining the same density-matching idea with full six-dimensional phase space for orphan streams would make disruption a measurable endpoint rather than an inferred one.","Editorial inference: The orbital-frequency dependence is positive but noisy, and the clusters the paper had to exclude are precisely those on highly eccentric orbits near the Galactic center; a dedicated central Galactic potential model would let those clusters enter the sample and provide the sharpest test of the $\\Omega$ scaling."],"forward_implications":["Stream density becomes a direct observable for cluster mass loss, breaking the previous degeneracy between stream width, length, and mass-loss rate.","The positive mass scaling $\\dot{M}\\sim M^{0.66}$ supports the slope $a<1$ that theoretical models require to convert an abundant low-mass initial cluster population into today's roughly log-normal globular cluster mass function.","At current rates, most of the 12 clusters would dissolve within 10 to 100 billion years, while Pal 5 and NGC 2298 have dissolution times below 10 billion years.","The measured scaling is consistent with N-body predictions for clusters without black holes, and the Pal 5 rate agrees with the earlier black-hole-free estimate; the comparison with a black-hole-rich model is left unresolved because the mock algorithm is not calibrated for such clusters.","Deeper wide-field surveys and spectroscopic velocities could extend the same procedure to roughly a thousand streams and to streams of fully disrupted clusters."],"supporting_citations":[{"why":"Particle spray algorithm whose Gaussian initial conditions reproduce stream width and length and make morphology independent of the mass-loss rate; it generates all mock streams.","marker":"C25"},{"why":"Gaia DR3 stream catalog that supplies the 12 observed streams and the member-star list whose incompleteness the pipeline corrects.","marker":"I24"},{"why":"Provides the present-day stellar mass function slopes used in the per-star correction weights of Eq. (6).","marker":"Baumgardt et al. (2023)"},{"why":"N-body scaling relation $(a,b,c)=(1/3,1,0)$ that the fitted mass-loss rate slopes are compared against.","marker":"Gieles & Gnedin (2023)"},{"why":"Supplies carefully selected phase-space coordinates for Pal 5 used to match the observed stream track.","marker":"Erkal et al. (2017)"},{"why":"CFHT photometry of Pal 5 used as an independent dataset with a deep detection limit and simple selection function to validate the method.","marker":"Ibata et al. (2016)"},{"why":"Shows that low-mass stars are ejected preferentially, the physical mechanism that could break the assumption that stream and cluster mass functions coincide.","marker":"Webb & Bovy (2021)"},{"why":"Catalog that supplies cluster masses, half-mass radii, positions, velocities, and mass-function slopes for the sample.","marker":"Hilker et al. (2019)"}],"fun_headline_variants":["Globular cluster streams yield first direct mass-loss rates","Stream densities reveal mass-loss rates for 12 globular clusters","Mass loss of 12 globular clusters measured via stellar streams","Stellar streams expose globular cluster mass loss from 0.5 to 200","New catalog: mass-loss rates for 12 globular clusters from Gaia"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the faint stars missed by Gaia have the same mass distribution as the cluster's present-day stars, with that distribution fixed by one measured slope, so the correction for unseen stars, a factor of 2 to 200, scales the final mass-loss rate almost linearly.","fun_headline_variants_meta":{"raw":{"variants":["Globular cluster streams yield first direct mass-loss rates","Stream densities reveal mass-loss rates for 12 globular clusters","Mass loss of 12 globular clusters measured via stellar streams","Stellar streams expose globular cluster mass loss from 0.5 to 200","New catalog: mass-loss rates for 12 globular clusters from Gaia"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000698,"raw_usage":{"total_tokens":3127,"prompt_tokens":891,"completion_tokens":2236,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":507,"completion_tokens_details":{"reasoning_tokens":2144}},"tokens_in":507,"tokens_out":2236,"duration_ms":12844,"temperature":1.0,"reasoning_tokens":2144,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T05:42:17.947829+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a stream such as NGC 6397 and observe it with a wide-field telescope reaching roughly $r=23$, comparable to the CFHT Pal 5 data, count stars below the main-sequence turnover, and derive the mass-loss rate with a correction factor near 6; if the resulting rate disagrees with the Gaia-based value by more than the quoted uncertainties, the assumed stream mass function is wrong.","supporting_citations":[],"review_version":1}