{"id":"cb277191-7136-4dc5-92c0-3022433f0c6d","arxiv_id":"2412.16372","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"For loud anisotropic gravitational wave backgrounds in LISA, Bayesian spherical harmonic searches reach their best angular resolution at ℓmax=16, limited by computation rather than by signal properties.","lead":"This paper uses simulations to measure how sharply the LISA space observatory can pinpoint concentrated sources of gravitational wave background noise on the sky. It finds the resolution is governed by the number of spherical harmonic terms in the search, up to the limits of available computing power.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The ℓmax=16 resolution claim may be injection-limited: all high-ℓmax injections are band-limited at ℓamax,inj=16, so the recovered FWHM at ℓmax=16 cannot be separated from the injected source's intrinsic width.","rationale":"The reader's conditional verdict is appropriate, and the weakest assumption identified is exactly the most load-bearing one. The paper's abstract and conclusions claim that the Bayesian search's angular resolution reaches ℓmax=16 and exceeds other LISA map-making techniques. That claim rests entirely on simulations whose injected point sources are themselves band-limited at ℓamax,inj=16, the same as the highest analysis cutoff. Because a band-limited 'point' has a finite width, the recovered FWHM at ℓmax=16 has a floor set by the injection, not necessarily by the search's resolving power. Without a control with ℓamax,inj>16, the observed saturation or agreement with the heuristic could be an artifact of the injection model. This is not an internal inconsistency or a question of consensus; it is a missing control in an otherwise clean and reproducible simulation study. The paper's strengths—open-source code, released posterior samples, a broad parameter grid, and clear FWHM heuristics—support the lower-ℓmax trends, but they do not establish the headline comparison at ℓmax=16. I therefore agree with the reader's assessment and see no reason to change the CONDITIONAL verdict; the concrete test above would upgrade or sharpen it.","tokens_in":16444,"tokens_out":8046,"duration_ms":70201,"concrete_test":"Compute the SP metric of the injected band-limited skymap itself for ℓamax,inj=16 (e.g., a delta expanded to ℓ=16 on a Healpy nside=32 map). If the recovered SP at ℓmax=16 in Fig. 2 equals this injection SP within the 95% credible interval, the ℓmax=16 point is injection-limited rather than resolution-limited. As a secondary check, compare sim 1 (ℓamax,inj=16) with sim 33 (ℓamax,inj=4) at ℓmax=4; if the SP differs measurably, injection width already matters at lower ℓmax. To fully settle the headline claim, run at least one single-point simulation with ℓamax,inj=32 (or a direct delta injection) and analyze at ℓmax=16; if the recovered SP stays the same, the search is limited by ℓmax; if it shrinks, the current endpoint underestimates the search's resolution.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline result is that for high-amplitude ASGWBs the angular resolution of the BLIP search is limited by ℓmax up to ℓmax=16, and that this exceeds other LISA map-making techniques. The data supporting the ℓmax=16 endpoint (single-point simulations 1–28, Table II; Fig. 2) all use injections with ℓamax,inj=16, the same value as the largest analysis cutoff. A truncated delta at ℓ=16 is not a point; it is a band-limited kernel with an intrinsic FWHM of roughly 0.25 rad (SP fraction ≈ 0.004–0.005). If the recovered SP metric at ℓmax=16 is consistent with this intrinsic width, the observed agreement with the Eq. (7) heuristic is expected regardless of whether the search can resolve finer structure. The trend from ℓmax=4 to 16 may then reflect the analysis model progressively matching the injection's band limit rather than the search's true resolution. No simulation with ℓamax,inj>16 is run, so the central comparison to Contaldi et al. is not established. The same table does contain a partial control at ℓmax=4 (sim 1 vs sim 33, injection 16 vs 4), but it is not used to quantify injection-width effects, and no analogous control exists at ℓmax=16.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper uses the Bayesian spherical-harmonic ASGWB search in the open-source BLIP package to study how well LISA can localize anisotropic stochastic gravitational-wave backgrounds. The authors simulate single- and two-point-source ASGWBs with power-law spectra over a grid of amplitudes Ωref, observing times Tobs, and spherical-harmonic cutoffs ℓamax, and they quantify angular resolution with FWHM-based 'spot size' metrics for single sources (SP) and separation-to-spot-size ratios for pairs (TP). The main reported trend is that the recovered spot size decreases with increasing analysis cutoff ℓamax, following an approximate heuristic SPmin≈sin²(1.1809π/(4ℓamax)), with no strong dependence on amplitude or observing time above an SNR threshold; the trend is reported to continue up to ℓamax=16, where computational cost becomes prohibitive. The authors also report that two-point resolution improves with ℓamax and separation, and that odd-ℓamax modes are insensitive in the reconstructed aℓm. They conclude that current Bayesian ASGWB searches are limited by the chosen ℓamax rather than by LISA's intrinsic response, and that their ℓamax=16 result exceeds the ℓamax≲15 limit estimated for frequentist map-making methods.","tokens_in":16723,"tokens_out":9547,"duration_ms":85188,"significance":"If the ℓamax=16 result holds, the paper would provide the first quantitative characterization of angular resolution for a Bayesian spherical-harmonic ASGWB search in LISA, which is directly relevant to planned global-fit analyses. The study is also useful as a systematic simulation campaign: it covers a broad grid in amplitude, observing time, and source separation; uses an open-source package; and makes posterior samples available on Zenodo, which supports reproducibility. The main caveat is that the headline endpoint at ℓamax=16 may be limited by the injected source's own spherical-harmonic truncation rather than by the search's resolving power, so the comparison to frequentist limits in Sec. IV is not yet established. The trends for ℓamax up to 14 and the explicit SNR-threshold discussion are valuable even if the final endpoint needs stronger support.","major_comments":[{"comment":"All single-point injections used for the ℓmax trend (simulations 1–28) have ℓamax,inj=16, equal to the largest analysis cutoff; because the footnote in §II B restricts analyses to ℓamax≤ℓamax,inj, the ℓamax=16 point is exactly the case where the analysis cutoff equals the injection cutoff. A delta function expanded to ℓ=16 has a finite intrinsic FWHM of order π/16 rad, which is comparable to the SP_min heuristic of Eq. (7) at ℓ=16, so the recovered FWHM at ℓmax=16 may be set by the injected source's band limit rather than by the search's resolving power. I recommend adding simulations with ℓamax,inj=24 or 32 (or otherwise narrower injections) at least for the highest-amplitude cases, and reporting the intrinsic FWHM of the truncated-delta injection alongside the recovered SP metric. The existing control at ℓmax=4 (simulation 1 vs simulation 33) should also be reported explicitly, because it is the only direct evidence on how injection bandwidth affects the recovered FWHM.","section":"§II B, Table II, Fig. 2"},{"comment":"The SP and TP heuristics are derived from the same spherical-harmonic truncation used in the reconstruction model, and the metrics are evaluated on the reconstructed band-limited skymap, so the observed improvement with ℓmax is partly a property of the parameterization rather than of LISA's angular response. This does not make the trend meaningless, but it means the paper's central claim that the search is 'currently limited at high amplitudes by the choice of ℓamax' needs to be separated from the trivial statement that a band-limited basis cannot represent features smaller than ~π/ℓmax. A concrete way to do this is to compare the recovered skymap to the best-fit band-limited projection of the true injection and to report the excess (the offset from Eq. (7)) as the actual resolution loss. Without such a comparison, the Sec. IV statement that ℓamax=16 'exceeds' the frequentist limit of Contaldi et al. is not a comparison of equivalent quantities.","section":"§II C, Eq. (7); §IV"}],"minor_comments":[{"comment":"The text says the angular separation grid spans (π/5, π) at a step of π/5, but Table III lists separations 2π/5, 4π/5, 6π/5, 8π/5, and π; please reconcile the stated range and step.","section":"§II B"},{"comment":"The sentence 'The recovered TP metric is compared to the TP heuristic in Fig. 2' should reference Fig. 6, where the dashed TPmax curve actually appears.","section":"§III B.2"},{"comment":"The phrase 'open souce Python package' contains a typo; it should be 'open-source Python package'.","section":"§I B"},{"comment":"The claim of amplitude/observing-time independence is conditioned on successful recovery; the low-SNR simulations that failed are excluded in §II B. Please state this explicitly in the Fig. 3 caption and in the abstract, since the current wording 'across a large grid in ... amplitude' could be read as unconditional.","section":"§III A.2 and Fig. 3"},{"comment":"The color-bar labels (e.g., '1.06893e-22 1.42898e-11') lack units and a description of the plotted quantity; please specify that these are ΩGW at f=1 mHz and state the units in the caption.","section":"Figs. 1 and 4"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for a gr-qc journal and the main computational campaign is a useful contribution. The unresolved point is the injection band limit at ℓmax=16: the headline comparison to Contaldi et al. rests on a single data point where the injection cutoff equals the analysis cutoff. This is fixable with additional simulations (or a clear injection-width comparison), so I do not see a reason for rejection; however, the abstract's claim that the result 'exceeds' other techniques should not be accepted until that control is provided."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The useful new thing here is a first quantitative map of how a Bayesian spherical-harmonic ASGWB search behaves as you push the analysis cutoff. The SP metric shrinks with ℓmax following the π/ℓ heuristic, the two-source separability results are a genuine addition, and the SNR threshold below which the spatial recovery fails is a practically important observation. They also ship code and posterior samples, which makes the whole grid reproducible. That is real value.\n\nThe main soft spot is exactly what the stress-test note flags. All single-point injections at high ℓmax use ℓamax,inj=16, the same value as the largest analysis cutoff. A band-limited delta at ℓ=16 is not a point; it has an intrinsic width, and the recovered FWHM at ℓmax=16 is consistent with that width. So the claim that resolution \"continues up to ℓmax=16\" is not cleanly supported. The trend from 4 to 14 is fine, but the ℓmax=16 data point, which is the one they use to say they beat Contaldi's ℓ~15, is confounded. They do have a control at ℓmax=4 (sim 1 vs sim 33) that suggests injection width doesn't matter at low cutoff, but they don't apply the same logic at ℓmax=16. Without a simulation with ℓamax,inj>16, the headline comparison to frequentist methods is not established.\n\nAlso worth noting: the frequentist comparison is not a matched baseline. Different metric, different assumptions, different noise treatment. The abstract overstates the result by omitting the high-amplitude qualifier and the injection caveat. And I would have liked some convergence diagnostics for the nested sampling runs, especially at ℓmax=16 where the parameter space is large.\n\nNone of this sinks the paper. The ℓmax 4–14 trend, the two-source behavior, and the SNR threshold are all useful and mostly robust. The paper deserves a serious referee. The referee should ask for an injection with ℓamax,inj>16, a softened abstract, and a more careful comparison to the frequentist limits. With those changes it would be a solid contribution to the LISA data-analysis literature.","headline":"First quantitative Bayesian resolution study for LISA ASGWB searches, with a solid ℓmax trend up to 14, but the ℓmax=16 point is plausibly injection-limited and the headline comparison to frequentist limits overstates the case.","tokens_in":17264,"tokens_out":2259,"would_cite":true,"duration_ms":21140,"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":"A Bayesian spherical-harmonic search can localize anisotropic stochastic gravitational-wave backgrounds in LISA down to the angular scale set by the spherical-harmonic cutoff $\\ell_{\\mathrm{max}}$, with resolution improving up to…","keywords":["gravitational waves","LISA","stochastic gravitational wave background","anisotropy","spherical harmonics","Bayesian inference","angular resolution","full-width half-maximum"],"falsifier":"Re-run the single-source simulations with injection cutoff $\\ell_{\\mathrm{max,inj}} > 16$ (for example 24) and recover at $\\ell_{\\mathrm{max}}=16$. If the recovered full-width-half-maximum is unchanged, the resolution is truly limited by the analysis model; if it shrinks, the reported $\\ell_{\\mathrm{max}}=16$ result was an artifact of the truncated injection.","tokens_in":16207,"feed_emoji":"🛰️","tokens_out":8968,"duration_ms":71824,"temperature":0.7,"pith_summary":"LISA will observe stochastic gravitational-wave backgrounds that are anisotropic on the sky, most prominently from tens of millions of unresolved white-dwarf binaries in the Milky Way. This paper asks how finely a Bayesian spherical-harmonic search can map such a background, and finds that the search's angular resolution is set by the spherical-harmonic cutoff $\\ell_{\\mathrm{max}}$ used to model the sky, not by the signal's amplitude or total observing time once the signal is loud enough. Resolution improves steadily as $\\ell_{\\mathrm{max}}$ grows from 4 to 16, and then a computational wall appears because the detector-response matrix and the sampling cost scale steeply with $\\ell_{\\mathrm{max}}$. This ceiling exceeds the $\\ell_{\\mathrm{max}}\\lesssim 15$ angular resolution estimated for frequentist map-making techniques, so a Bayesian search can in principle produce sharper maps of LISA's anisotropic backgrounds. The result matters because LISA data analysis is expected to be Bayesian and global-fit based, making the achievable sky resolution a key planning input.","feed_headline":"LISA Bayesian search resolves sky maps down to ℓ=16","feed_subtitle":"Higher spherical-harmonic cutoff sharpens LISA's maps until compute cost stops the search.","key_machinery":"The machinery is the spherical-harmonic expansion of the square root of the sky power, $S(n) = \\sqrt{P(n)} = \\sum_{\\ell m} b_{\\ell m} Y_{\\ell m}(n)$, which guarantees that any inferred sky map is real and non-negative; the usual $a_{\\ell m}$ coefficients are recovered through a Clebsch-Gordan convolution $b \\otimes b$. The analysis cutoff $\\ell_{\\mathrm{max}}$ (with $\\ell^a_{\\mathrm{max}} = 2\\ell^b_{\\mathrm{max}}$) is the load-bearing dial: it directly sets the number of fitted coefficients, the cost of the LISA detector-response model, and the smallest angular feature the map can represent. Resolution is measured with full-width-half-max contoured skymaps, using the fraction of sky inside the half-max spot for single sources and the separation-to-spot-size ratio for two sources.","core_discovery":"The central discovery is that, for loud enough anisotropic stochastic gravitational-wave backgrounds ($\\Omega_{\\rm ref} > 10^{-9}$ at 25 Hz for one year of observing), the angular resolution of a Bayesian spherical-harmonic search in LISA is limited by the analysis model's spherical-harmonic cutoff $\\ell_{\\mathrm{max}}$, equivalently by the number of fitted sky coefficients, rather than by the signal amplitude or observing time. For single point sources the full-width-half-maximum spot shrinks steadily as $\\ell_{\\mathrm{max}}$ increases from 4 to 16, tracking the angular-scale heuristic $\\theta \\sim \\pi/\\ell_{\\mathrm{max}}$; amplitude and observing time mainly reduce the posterior uncertainty, not the spot size. Two-source separation shows the same $\\ell_{\\mathrm{max}}$ dependence, with sources becoming resolvable at $\\ell_{\\mathrm{max}}\\ge 6$ and the separation-to-spot-size ratio growing into the tens. The paper concludes that the current resolution ceiling is a computational artifact, and that $\\ell_{\\mathrm{max}}=16$ already exceeds the $\\ell_{\\mathrm{max}}\\lesssim 15$ resolution ceiling estimated for frequentist angular power-spectrum analyses of LISA. It also confirms LISA's insensitivity to odd-$\\ell$ modes in the reconstructed $a_{\\ell m}$ distribution.","pith_inferences":["The computational ceiling is likely a moving target: as nested-sampling and response-matrix performance improve, $\\ell_{\\mathrm{max}}$ values above 16 should become feasible, potentially pushing resolution below the current ~11-degree floor for high-SNR backgrounds.","For realistic faint extragalactic backgrounds, the practical resolution may be set by confusion with the Galactic foreground rather than by $\\ell_{\\mathrm{max}}$, so extending these simulations to include a foreground and low-amplitude injections would map the actual noise floor.","The odd-$\\ell$ insensitivity means that $a_{\\ell m}$-based power-spectrum estimators discard roughly half the available modes; fitting directly in the $b_{\\ell m}$ basis, as done here, may extract more spatial information than traditional $\\ell$-summaries.","The FWHM metrics used here assume compact, roughly Gaussian sources; adapting them to extended or structured backgrounds like the Galactic disk would tell whether Bayesian maps can separate a diffuse foreground from localized point-like contributors."],"forward_implications":["LISA Bayesian searches can in principle resolve loud anisotropic backgrounds down to angular scales of roughly $\\pi/16$ radians (about 11 degrees), better than the $\\sim\\pi/15$ radians ceiling of frequentist angular power-spectrum methods.","The spherical-harmonic cutoff is the primary design lever for sky-map resolution: pushing $\\ell_{\\mathrm{max}}$ higher sharpens the map, so future analysis pipelines should raise $\\ell_{\\mathrm{max}}$ as computational resources allow.","Below a signal-to-noise threshold the search becomes information-limited, so low-amplitude backgrounds will not benefit from a high $\\ell_{\\mathrm{max}}$ unless longer observations or denoising raise the SNR.","Two-source separation requires at least six months of LISA data; shorter observations fail to resolve sources because the constellation's orbital motion has not yet broken the response degeneracies."],"supporting_citations":[{"why":"Defines the Bayesian spherical-harmonic formalism, the square-root sky model, and the LISA response computation used for all simulations and analyses.","marker":"[38]"},{"why":"Supplies the frequentist angular-resolution ceiling ($\\ell_{\\mathrm{max}} \\lesssim 15$) that this paper's Bayesian search is compared against at $\\ell_{\\mathrm{max}}=16$.","marker":"[44]"},{"why":"Provides the earlier estimate of LISA's anisotropic-background sensitivity and the prediction of odd-$\\ell$ insensitivity that the paper confirms in its $a_{\\ell m}$ posteriors.","marker":"[19]"},{"why":"Establishes the earlier $\\ell_{\\mathrm{max}} \\le 4$ resolution limit for LISA that the current search far exceeds.","marker":"[42]"},{"why":"Introduces the two-source separation-to-spot-size metric used to quantify angular resolution in the two-point simulations.","marker":"[20]"},{"why":"The nested-sampling algorithm used to generate all posterior distributions, and the source of the computational sampling-cost limit at high $\\ell_{\\mathrm{max}}$.","marker":"[50]"},{"why":"Provides the pixelized skymaps used to compute full-width-half-max contours and evaluate the recovered spatial distributions.","marker":"[51]"}],"fun_headline_variants":["LISA's Bayesian sky maps hit resolution ceiling at ℓ=16","ℓ=16 sets LISA's Bayesian resolution record","LISA's Bayesian search outresolves frequentist methods at ℓ=16","LISA's map resolution limited by ℓmax, not data","Bayesian LISA search caps resolution at ℓmax=16"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The injected point sources are simulated with spherical-harmonic expansions truncated at $\\ell_{\\mathrm{max}}=16$, the same value as the largest recovery cutoff, so the measured resolution at $\\ell_{\\mathrm{max}}=16$ could be capped by the finite width of the injected source rather than by the search's own resolving power; no simulation with a larger injection cutoff is run.","fun_headline_variants_meta":{"raw":{"variants":["LISA's Bayesian sky maps hit resolution ceiling at ℓ=16","ℓ=16 sets LISA's Bayesian resolution record","LISA's Bayesian search outresolves frequentist methods at ℓ=16","LISA's map resolution limited by ℓmax, not data","Bayesian LISA search caps resolution at ℓmax=16"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001225,"raw_usage":{"total_tokens":5092,"prompt_tokens":1061,"completion_tokens":4031,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":677,"completion_tokens_details":{"reasoning_tokens":3939}},"tokens_in":677,"tokens_out":4031,"duration_ms":27357,"temperature":1.0,"reasoning_tokens":3939,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:39:23.677450+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the single-source simulations with injection cutoff $\\ell_{\\mathrm{max,inj}} > 16$ (for example 24) and recover at $\\ell_{\\mathrm{max}}=16$. If the recovered full-width-half-maximum is unchanged, the resolution is truly limited by the analysis model; if it shrinks, the reported $\\ell_{\\mathrm{max}}=16$ result was an artifact of the truncated injection.","supporting_citations":[{"cited_title":"Cutler, Physical Review D 57, 7089 (1998)","cited_arxiv_id":null,"evidence_quote":"Supplies the frequentist angular-resolution ceiling ($\\ell_{\\mathrm{max}} \\lesssim 15$) that this paper's Bayesian search is compared against at $\\ell_{\\mathrm{max}}=16$."},{"cited_title":"Banagiri, A","cited_arxiv_id":null,"evidence_quote":"Establishes the earlier $\\ell_{\\mathrm{max}} \\le 4$ resolution limit for LISA that the current search far exceeds."},{"cited_title":"Breivik, C","cited_arxiv_id":null,"evidence_quote":"Introduces the two-source separation-to-spot-size metric used to quantify angular resolution in the two-point simulations."}],"review_version":1}