{"id":"6555dbdd-6d07-46a3-be6c-4742606e6b91","arxiv_id":"2412.10951","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Two template banks for sub-solar-mass compact binary searches in LIGO-Virgo-KAGRA's fourth observing run were generated with a modified manifold algorithm, and simulation tests show sufficient match coverage for masses down to 0.2 solar masses.","lead":"This paper presents template banks for searching gravitational-wave signals from sub-solar-mass compact binary mergers in the fourth LIGO-Virgo-KAGRA observing run, covering component masses down to 0.2 solar masses, with two algorithmic fixes to the manifold bank generator. The banks are validated with simulated signals and will be used by the GstLAL pipeline for both low-latency and offline searches.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The bank validation does not directly test the 128 s duration-truncated templates; the raised flow cutoffs in Table II proxy for the cutoff, and the paper concedes the SNR loss is not reflected, so the lowest-mass corner's stated efficacy rests on an untested equivalence.","rationale":"The paper's central deliverable is two usable banks and evidence they cover their target parameter spaces. The strongest evidence is the SBank simulation campaign: about 293k injections across three mass/spin regions, plus BNS/NSBH checks. The manifold modifications are concrete algorithmic contributions, and the boundary-padding and neighborhood-metric methods are validated indirectly by the same simulations. The most load-bearing assumption is that the duration-truncated templates used in production behave in the search as the untruncated templates used in the validation, with the raised flow cutoffs capturing the truncation. The paper is unusually explicit that this is not directly tested, stating that the bank simulation test cannot reflect the SNR loss due to the duration cutoff. This is not an internal inconsistency, and for the bulk of the parameter space the proxy is conservative: for masses where the duration cutoff is inactive, raising flow removes low-frequency SNR and reduces the fitting factor. But at the specific low-mass corner the paper worked to improve via boundary padding, the proxy is fixed per region while the real truncation frequency depends on chirp mass, and the paper's own 16% SNR-loss estimate for 0.2+0.2 implies a match well below the quoted 90th-percentile values. Whether that matters for 'sufficient efficacy' depends on the sensitivity the search is willing to sacrifice at the lowest masses. The concern does not warrant rejection: the bank is a real product, the code and methods are concrete, the simulations are large, and a direct check is straightforward. A conditional acceptance with a required verification is the right disposition, matching the reader's verdict.","tokens_in":21241,"tokens_out":11880,"duration_ms":104171,"concrete_test":"Generate about 1000 simulated signals with component masses log-uniformly drawn from [0.2, 0.5] Msun and spins in [-0.1, 0.1], all starting at 45 Hz. For each, compute the fitting factor against the actual archive bank templates as generated by manifold with the 128 s maximum-duration cutoff, either by patching SBank to support the duration option or by loading the bank's template waveforms and computing match with GstLAL's match routine, using the O4a PSD of Table I. Compare the resulting match distribution, and in particular the 90th percentile and the minimum, with Fig. 4 and with the corresponding match obtained using the paper's flow=75 proxy.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the offline bank has sufficient efficacy rests on the SBank simulations of Section V. But the templates actually used by the search are generated by manifold with a 128 s maximum-duration truncation, while SBank cannot apply this truncation (Section V). The paper compensates by raising the signal low-frequency cutoff to 75/55/45 Hz by mass region (Table II), and it explicitly states that 'the bank simulation test below cannot reflect the SNR loss due to this duration cutoff in our templates.' For m1=m2=0.2 Msun, the paper itself reports up to 16% SNR loss from the cutoff (Section III); if that is a loss in recovered SNR, the corresponding match between a normalized full-band signal and the truncated template is only about 84%, far below the 96.5% target, and the low-low 90th-percentile match of 97.77% would not describe the lowest-mass corner. The raised flow choice is a proxy, not a direct test: it is held fixed per region (75 Hz for low-low) while the actual truncation frequency varies continuously with chirp mass (about 73 Hz for 0.2+0.2, about 55 Hz for 0.5+0.2, and below 45 Hz for 0.5+0.5). For masses where the duration cutoff is inactive, using flow=75 or 55 makes the simulation conservative; for masses where the cutoff is active, a region-averaged flow can either over- or under-state the true template match. Because the stated simulation cannot reflect the duration loss, the bank's 'sufficient efficacy' is not directly established in the lowest-mass corner it was specifically padded to cover.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper describes the construction and validation of two template banks for GstLAL's O4 subsolar-mass compact-binary searches. The offline bank covers m1 ∈ [0.2, 10] M⊙, m2 ∈ [0.2, 1] M⊙, mass ratio up to 10, spin-aligned, with a low-frequency cutoff of 45 Hz, a maximum waveform duration of 128 s, and 3,452,006 templates; the low-latency bank covers m1 ∈ [0.5, 10] M⊙, m2 ∈ [0.5, 1] M⊙, |χeff| < 0.3, and 1,069,651 templates. Methodologically, the paper introduces two modifications to the manifold geometric placement algorithm: neighborhood metric estimation to handle metric failures at very low masses, and boundary padding to improve coverage near the low-mass boundary. Efficacy is assessed with SBank simulations in three mass/spin regions and with BNS/NSBH injection sets. The headline results are 90th-percentile matches of 97.77%, 98.08%, and 96.79% in the low-low, high-low, and high-high regions, respectively, for the offline bank.","tokens_in":21653,"tokens_out":10788,"duration_ms":97393,"significance":"The banks are a practical deliverable for an ongoing search, and the paper's methodological additions (neighborhood metric estimation and boundary padding) address genuine numerical problems in applying manifold to low masses. The validation uses large, independent SBank simulation sets and standard fitting-factor definitions, which is a strength. However, the claimed 'sufficient efficacy' for the offline bank's lowest-mass corner is not directly supported: the simulations cannot apply the 128 s duration cutoff, and the paper's proxy (raised flow cutoffs) does not measure the SNR loss that the cutoff imposes on real signals. If that gap can be closed by direct simulation or by an analytic correction, the paper would be a solid instrument paper; as it stands, the central claim is only partially demonstrated.","major_comments":[{"comment":"The central efficacy claim for the offline bank is not directly established for the low-low mass region because the SBank simulations do not apply the 128 s maximum-duration cutoff that manifold uses in template placement. The paper concedes this: 'the bank simulation test below cannot reflect the SNR loss due to this duration cutoff in our templates.' Raising the flow cutoff to 75/55/45 Hz per region (Table II) does not fix the problem, because the simulated signals themselves are generated with the raised cutoff; the test therefore measures the bank's coverage of signals that begin at 75/55/45 Hz, not the recovery of full 45 Hz signals by the duration-truncated templates actually used in the search. This is quantitatively important: for m_i = 0.2 M⊙, Section III reports up to 16% optimal-SNR loss from the cutoff, implying that the overlap between a full-band signal and the truncated template at the same parameters is at most about 0.84, well below the 96.5% minimum match and the quoted 97.77% low-low 90th percentile. I recommend either (1) running SBank or an equivalent code with the actual 128 s-truncated templates against full-band flow = 45 Hz injections for the low-low region, or (2) presenting a separate calculation that multiplies the geometric match by the duration-cutoff overlap and then restating the efficacy claim for that corner of parameter space.","section":"Section V, Table II and Section III"}],"minor_comments":[{"comment":"The spin range stated in the conclusion ('−0.9 to 0.9 in the larger component and −0.05 to 0.05 in the smaller component') is inconsistent with Table I and Section III, which specify |s_i,z| < 0.1 for masses ≤ 0.5 M⊙ and |s_i,z| < 0.9 for masses > 0.5 M⊙; please correct.","section":"Section VI"},{"comment":"The boundary-padding width is effectively a free parameter tuned on simulation results ('we stopped widening once a significant majority of simulated signals near the original boundary achieved match values above 95%'); the paper should report the final padding width and discuss the possible optimistic bias from tuning on the same type of validation used in Section V.","section":"Section IV C"},{"comment":"There are typographical errors that should be corrected: 'independet' and 'power spectral densitys' in Section II, 'dipassive' in the Introduction, 'Paramter' in Table IV, and 'culster' in the Acknowledgments.","section":"Introduction and Section II"},{"comment":"For the BNS/NSBH simulations, the paper reports 90th-percentile matches as low as 0.70 and 0% for NSBH, which is expected because these lie outside the target space; however, the text should more clearly state that these numbers are not efficacy claims for the bank's design region but rather characterize a serendipitous detection capability, to avoid misinterpretation.","section":"Section V.2 and Appendix B"},{"comment":"The statement that a minimum match of 96.5% 'ensures that no more than approximately 10% of astrophysical signals can be missed' is imprecise: the minimum match is a per-template worst-case design value, and the fraction of signals below a given fitting factor depends on the signal population and the match distribution; consider rephrasing.","section":"Section III"}],"recommendation":"major_revision","confidential_remarks":"The main obstacle is the duration-cutoff validation gap. If the authors can implement a direct test with the actual truncated templates or provide a quantitative correction that propagates the 16% SNR loss into the match statistics, I would be willing to accept a revised version. The spin inconsistency and typos are easy to fix. I would also encourage the editor to ask whether the bank parameter files will be made publicly available, since reproducibility is an important consideration for a template-bank paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe short version: this paper is a solid, incremental contribution to the GstLAL template-bank toolkit. It delivers two new banks for sub-solar-mass searches in O4 and extends the manifold algorithm down to 0.2 solar masses with two new tricks. The validation is extensive but has a real gap at the lowest masses, and the authors mostly but not fully own up to it.\n\nWhat's new: first manifold-based bank for component masses down to 0.2 Msun, plus two modifications to the geometric placement code: neighborhood metric estimation to handle numerical failures in the low-mass region, and boundary padding to improve recovery at the low-mass edge. The banks are concrete products, and the simulations use SBank, an independent program, with about 300k injections. That's real evidence, and the paper credits the prior manifold work rather than overselling.\n\nThe main soft spot is the duration cutoff. SBank can't apply the 128-second truncation that manifold applies, so the validation uses raised low-frequency cutoffs (75/55/45 Hz per mass region) as a proxy. That proxy doesn't capture the SNR loss from the truncation. For a 0.2+0.2 system, their own Section III reports nearly 16% SNR loss due to the cutoff. That implies the actual fitting factor for the lowest-mass corner is about 0.84, not the 97%+ reported for the low-low region. So the 'sufficient efficacy' claim overstates what is directly tested. The paper does disclose this in Section V, so it's not hidden, but the abstract and conclusion don't carry the caveat.\n\nThere's also a mild circularity in the boundary padding: the padding width was chosen by iterating until the bank's own simulations showed good recovery. That doesn't invalidate the result, but it means the quoted matches are partially tuned. Finally, the conclusion misstates the spin bounds compared to Table I — it says the smaller component has |s|<=0.05, but Table I uses |s|<=0.1 for the low-mass regime and |s|<=0.9 for the high-mass regime. Minor.\n\nBottom line: worth reading if you work on template banks or SSM searches, and worth citing for the algorithmic extensions. It deserves peer review, with a request to either directly validate the truncated templates or soften the efficacy claim for the lowest masses. I'd send it out.","headline":"Solid, workmanlike template-bank paper with a real validation gap at the lowest masses; worth a peer review.","tokens_in":22347,"tokens_out":6027,"would_cite":true,"duration_ms":53347,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["04.30.-w","04.80.Nn"],"model":"deepseek-v4-flash","headline":"A 3.45-million-template bank covers sub-solar-mass mergers across LIGO-Virgo-KAGRA's fourth observing run.","keywords":["gravitational waves","template bank","sub-solar mass compact binaries","manifold algorithm","matched filtering","primordial black holes","LIGO-Virgo-KAGRA","low-latency search"],"falsifier":"Generate simulated 0.2–0.4 $M_\\odot$ signals and match them against templates constructed with the same 128-second duration truncation used in the bank, computing fitting factors directly rather than via the raised-cutoff proxy used by SBank; if the 90th-percentile match in the low-low region falls below the quoted 97.77% by more than the statistical uncertainty, the claimed efficacy is not reproduced.","tokens_in":21062,"feed_emoji":"🔭","tokens_out":7804,"duration_ms":65935,"temperature":0.7,"pith_summary":"The paper claims that the GstLAL pipeline now has template banks capable of searching for gravitational waves from compact binaries with a component below one solar mass throughout the fourth observing run, both in low latency and offline. To build them, the authors extend the geometric manifold placement algorithm down to component masses of 0.2 solar masses with two fixes: a neighborhood-based metric estimation for the unstable low-mass region and a boundary-padding step that adds a marginal bank along the low-mass edge. Simulated-signal tests (SBank) show that 90% of test signals spanning the archive bank's design space have a match of 97.42% or higher, and that the low-latency bank recovers 90% of its primary target signals at match 0.99. If these results hold, the search can place new constraints on primordial black holes and other exotic sub-solar-mass objects, and can issue low-latency alerts for potentially electromagnetically-bright low-mass mergers.","feed_headline":"3.45M-template bank hunts sub-solar black holes in O4 data","feed_subtitle":"Simulations find 90% of test signals match at 97.4% or better; the offline bank reaches 0.2 solar masses.","key_machinery":"The load-bearing machinery is the manifold treebank algorithm, a geometric placement method that tiles the intrinsic parameter space into hyperrectangular regions, each containing one template at its center, using a Fisher-information mismatch metric to decide when a region is too large. Two additions make it work at low masses: neighborhood metric estimation, which replaces a failed pointwise metric evaluation with the metric at a random nearby point inside a mismatch hyperellipsoid (up to a maximum number of attempts); and boundary padding, which unions the main bank with a smaller marginal bank extending past the low-mass boundary to raise fitting factors for extremal signals. The bank's efficacy is then tested with SBank, which computes the fitting factor of simulated signals against the bank.","core_discovery":"On its own terms, the paper establishes that two new template banks deliver sufficient coverage for sub-solar-mass (SSM) searches in O4. The archive bank contains 3,452,006 templates covering primary masses 0.2–10 $M_\\odot$, secondary masses 0.2–1 $M_\\odot$, mass ratios 1–10, and aligned spins up to 0.9 (above 0.5 $M_\\odot$) or 0.1 (below), built with IMRPhenomD waveforms, a 45 Hz lower cutoff, a 128-second maximum duration, and a 96.5% minimum match; the low-latency bank uses 1,069,651 templates with component masses 0.5–10 and 0.5–1 $M_\\odot$ and spins up to 0.3. In bank simulations, 90% of signals in the low-low, high-low, and high-high regions reach matches of 97.77%, 98.08%, and 96.79% respectively, and 90% of BNS-like signals reach about 95.8% for chirp masses below roughly 2.4 $M_\\odot$, while the bank does not recover NSBH signals above that chirp mass. The paper also claims two methodological advances: mismatch-neighborhood metric estimation, which samples nearby points when the Fisher metric has negative eigenvalues at very low masses, and boundary padding, which merges a low-mass-edge marginal bank with the main bank at the cost of up to roughly 30% more templates.","pith_inferences":["The 128-second duration cutoff can cost up to 16% optimal SNR for 0.2 $M_\\odot$ systems, and the SBank tests use raised low-frequency cutoffs rather than the truncation itself; a dedicated injection study with truncated templates would likely show lower real-world sensitivity at the very lowest masses than the quoted matches suggest.","The same neighborhood-estimation and boundary-padding recipe should transfer to other long-duration, low-chirp-mass searches, including eccentric binaries or ultralight subsolar objects, whenever the waveform approximant's smoothness degrades.","If O4 produces no SSM detections, the population model attached to the low-latency bank (built with an uninformative prior so all templates carry equal weight) can feed joint constraints with the O3 SSM upper limits into primordial black hole abundance limits."],"forward_implications":["The offline bank lets GstLAL search O4 archive data for mergers with component masses down to 0.2 $M_\\odot$, covering the same SSM parameter space as the previous O3 search and enabling updated rate upper limits on primordial black hole and dissipative dark matter binaries.","The low-latency bank makes GstLAL's SSM search real-time for the first time, so a detected low-mass merger can trigger electromagnetic follow-up while the signal is still fresh.","The two manifold enhancements remove the low-mass barrier for geometric bank placement, so future banks tuned to new noise curves or mass ranges can be regenerated in minutes rather than weeks.","Combining the offline SSM bank with the BNS/NSBH/BBH bank raises the 90th-percentile match for high-high edge signals from 96.79% to 97.14%, so events at the boundary between search classes are not lost.","A 90th-percentile match of at least 96.5% (the minimum-match design choice) implies that template spacing alone misses no more than roughly 10% of signals in the designed space, assuming the duration-cutoff caveat is accounted for."],"supporting_citations":[{"why":"Defines the manifold treebank placement algorithm whose low-mass behavior this paper extends.","marker":"[1]"},{"why":"The software implementation of the treebank algorithm used to generate both banks.","marker":"[2]"},{"why":"Describes the O4 supersolar bank and the SBank-based efficacy assessment and duration-cutoff feature this paper relies on.","marker":"[28]"},{"why":"Defines the O3 SSM search parameter space and upper-limit methodology that the offline bank is designed to match.","marker":"[46]"},{"why":"Reports the most recent LVK SSM search upper limits that the O4 searches will extend.","marker":"[47]"},{"why":"Supplies the GstLAL low-latency template bank design and the spin restriction used for the low-latency bank.","marker":"[73]"},{"why":"Establishes the match/mismatch metric formalism and minimum-match concept used to set bank density.","marker":"[32]"},{"why":"Describes the SBank simulation program used to measure fitting factors in the efficacy tests.","marker":"[76]"}],"fun_headline_variants":["O4 sub-solar hunt: 3.45M waveforms down to 0.2 M☉","New template bank expands sub-solar search to 3.45M signals","Sub-solar mergers: 3.45M-template bank ready for O4","3.45M templates cover sub-solar binaries in O4 data","Modified algorithm yields 3.45M-template sub-solar bank"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The claimed match values assume that the SBank simulations, which cannot apply the 128-second maximum waveform duration that actually truncates the bank's templates, are a faithful proxy for the templates' true coverage; if raising the low-frequency cutoff (to 75/55/45 Hz by mass region) does not fully compensate for the up-to-16% SNR loss at 0.2 $M_\\odot$, the quoted matches overstate the bank's effectiveness at the lowest masses.","fun_headline_variants_meta":{"raw":{"variants":["O4 sub-solar hunt: 3.45M waveforms down to 0.2 M☉","New template bank expands sub-solar search to 3.45M signals","Sub-solar mergers: 3.45M-template bank ready for O4","3.45M templates cover sub-solar binaries in O4 data","Modified algorithm yields 3.45M-template sub-solar bank"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001036,"raw_usage":{"total_tokens":4482,"prompt_tokens":1190,"completion_tokens":3292,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":806,"completion_tokens_details":{"reasoning_tokens":3186}},"tokens_in":806,"tokens_out":3292,"duration_ms":21489,"temperature":1.0,"reasoning_tokens":3186,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T15:27:07.510626+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Generate simulated 0.2–0.4 $M_\\odot$ signals and match them against templates constructed with the same 128-second duration truncation used in the bank, computing fitting factors directly rather than via the raised-cutoff proxy used by SBank; if the 90th-percentile match in the low-low region falls below the quoted 97.77% by more than the statistical uncertainty, the claimed efficacy is not reproduced.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Describes the O4 supersolar bank and the SBank-based efficacy assessment and duration-cutoff feature this paper relies on."}],"review_version":1}