{"id":"7b1caa9c-4eb7-40ae-86f3-29a9e43039a4","arxiv_id":"2506.01898","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A coherent multiband Bayesian parameter estimation method with extrinsic-parameter marginalization extracts useful information from LISA observations of stellar-mass binary black holes down to LISA SNR 3, nearly doubling the expected multiband source count and sharply improving chirp-mass precision.","lead":"Researchers combined LISA data with future ground-based gravitational wave detectors in one coherent analysis, using a statistical trick to handle faint signals. Their method works down to LISA signal-to-noise ratio 3, nearly doubling the number of black hole mergers LISA can contribute to, and makes one year of LISA data sharpen chirp-mass measurements far beyond ground detectors alone.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The SNR-3 claim relies on a 2% LISA-waveform match over the 3G extrinsic posterior that is only demonstrated for non-precessing circular IMRPhenomD/HM waveforms with an analytical LISA orbit; outside that family the importance proposal can undercover and bias the marginalized likelihood.","rationale":"The reader's weakest assumption and this stress-test converge on the same load-bearing premise: the 3G extrinsic posterior must be an adequate importance proposal for the multiband likelihood, which is validated only by the 2% match claim for a restricted waveform family and an analytical orbit. This premise is not a side detail; it is what makes the method computationally efficient and what allows the SNR threshold to be lowered from 5 to 3. If the LISA waveform were more sensitive to extrinsic parameters in the presence of precession, eccentricity, higher modes, or a dynamical LISA orbit, the importance weights would become unstable and the marginalized likelihood would be biased. That would undermine both the specific SNR-3 examples and the population-scale credible-interval widths, since Fig. 14 aggregates posterior widths without coverage or effective-sample-size validation. The paper's own Sec. V acknowledges waveform systematics as a concern, and App. A notes the analytical-orbit limitation, so this is an explicit rather than manufactured gap. The concern is addressable with a targeted numerical experiment, which is why the CONDITIONAL verdict remains appropriate rather than a rejection. The paper has genuine strengths: a coherent joint likelihood with rotation-matrix frame unification, analytic phase marginalization, a public codebase, and population-scale validation on the restricted waveform family. These do not resolve the proposal-coverage question outside that family, but they make the central claim plausible enough to warrant the stated conditions.","tokens_in":38041,"tokens_out":6469,"duration_ms":74618,"concrete_test":"Run LISA SNR~3 injections with precessing (IMRPhenomXPHM) and eccentric waveforms using a numerical LISA orbit, recompute the Fig. 4 match distribution for LISA TDI A/E/T waveforms over the ET+2CE (tc, alpha, delta) posterior, and compute the effective sample size of the Eq. (24) importance weights for each injection. If the worst-case mismatch exceeds 2% or the effective sample size fraction drops below about 1% for any SNR~3 event, re-run full Bayesian parameter estimation both with the full importance marginalization and with (tc, alpha, delta) fixed to the 3G maximum-SNR point; a significant difference would show that the claimed SNR-3 extraction is proposal-dependent rather than robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is the use of the 3G posterior for (tc, alpha, delta) as the importance-sampling proposal for the multiband likelihood, justified in Sec. III.D and Fig. 4 by a LISA TDI waveform match within 2% over that posterior. The importance weights in Eq. (24) are only reliable if the proposal covers the LISA conditional posterior; if real LISA waveforms are more sensitive to extrinsic parameters, the weights will have large variance and the estimated marginalized likelihood will be biased. This premise is only checked for non-precessing, quasi-circular IMRPhenomD/HM waveforms with an analytical LISA orbit, as stated in Sec. IV.A and App. A, and the paper explicitly flags waveform systematics in Sec. V. The 'SNR 3 extraction' and the near-doubling of multiband events are precisely the regime where this flatness is used in place of a full marginalization, so the central claim is not established outside that restricted family. Moreover, the population-scale credible-interval widths in Fig. 14 are not accompanied by coverage or importance-weight diagnostics, so even a small number of unstable weights among the SNR~3 events could bias the reported improvements.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a coherent multiband Bayesian parameter-estimation method for stellar-mass and intermediate-mass binary black holes observed jointly by LISA and next-generation ground-based detectors (ET+2CE). The method evaluates a joint likelihood in a common geocentric frame using rotation matrices, marginalizes luminosity distance and orbital phase analytically/numerically, and uses importance sampling with a proposal derived from the 3G posterior for the extrinsic parameters (tc, alpha, delta). The key efficiency ingredient is the observation, illustrated in Fig. 4, that the LISA TDI waveform match varies by less than 2% over the 3G extrinsic posterior, so the 3G information can drive the marginalization. The authors validate the method on simulated GWTC-3-like populations, claim successful extraction of signals with LISA SNR as low as 3 (Sec. IV.C.5, Fig. 13), nearly doubled multiband event rates (Fig. 8), and population-scale parameter estimation with chirp-mass 90% credible intervals commonly below 1e-4 solar masses (Fig. 14). They also compare against a semi-coherent '3G posterior as LISA prior' approach and against LISA-only analysis.","tokens_in":38265,"tokens_out":7450,"duration_ms":81337,"significance":"If the central claims hold, this is an important step for multiband LISA+3G science: it would make population-scale Bayesian multiband inference practical, roughly double the number of usable multiband events by lowering the LISA SNR threshold from 5 to 3, and produce chirp-mass measurements far more precise than 3G-only estimation, even with one year of LISA data. The paper has concrete strengths: the code is public, the comparison with the semi-coherent approach is a useful check, the coordinate-system transformation is explicit, and the population-scale simulation campaign (20 random-seed datasets per observation scenario, plus 50 for IMBHBs) is substantially more ambitious than earlier single-event multiband studies. However, the strongest claims presently rest on a small number of example signals and on the flatness of the LISA response over the 3G extrinsic posterior, which is demonstrated only for a restricted waveform family. The significance is therefore real but conditional on additional validation and on narrowing or carefully qualifying the claims.","major_comments":[{"comment":"The central claim that the method 'successfully analyzes all multiband signals in simulated datasets with LISA SNR greater than 3' is supported by four randomly selected signals with rho_LISA about 3 and by qualitative inspection of their posteriors. There is no quantitative success criterion (for example, fraction of runs whose 90% credible interval contains the injected values, bias relative to injection, or HPD coverage), and no importance-weight diagnostics (effective sample size or weight variance) for these low-SNR runs. Because the paper itself notes that the chirp-mass posterior has 'multiple small modes' around the true peak, the presence of a narrow peak near the truth does not by itself establish that the marginalized likelihood is unbiased. I recommend defining a success metric and applying it to all events above threshold in the simulated datasets, or explicitly stating that the SNR-3 claim is example-based rather than a demonstrated population-level property.","section":"Sec. IV.C.5, Fig. 13; abstract; Sec. V"},{"comment":"The load-bearing flatness of the LISA TDI waveform over the 3G posterior of (tc, alpha, delta), quantified as a match within 2% in Fig. 4, is demonstrated only for non-precessing, quasi-circular IMRPhenomD/HM waveforms with an analytical LISA orbit. Precession, eccentricity, additional higher modes, or a realistic dynamical LISA orbit can make the LISA response more sensitive to extrinsic parameters; in that case the proposal q(tc, alpha, delta) from Eq. (23) will undercover the LISA conditional posterior, the importance weights in Eq. (24) will have large variance, and the marginalized likelihood at rho_LISA about 3 could be biased. The paper acknowledges waveform systematics in Sec. V, but the abstract and Sec. IV.C.5 state the SNR-3 capability unconditionally. I ask either to restrict the claim to the modeled waveform family or to add a robustness study on a subset with eccentricity or precession to demonstrate that the proposal coverage remains adequate.","section":"Sec. III.D, Fig. 4, Sec. IV.A, App. A"},{"comment":"The statement that in the marginalization over arrival time and sky location one can 'use (tc, alpha, delta) corresponding to maximum SNR in 3G samples for all the importance samples to accelerate' appears to replace per-sample LISA likelihood evaluations with a single evaluation at the 3G maximum-likelihood extrinsic point. If this is what is implemented, it is an additional approximation beyond standard importance sampling: the LISA contribution to each weight becomes a constant, so the LISA likelihood is not actually averaged over the proposal samples. The 2% match bound in Fig. 4 is computed for a time/phase-maximized match, whereas the joint likelihood uses a common reference frequency (5 Hz) and a fixed phase relationship; these two quantities need not have the same sensitivity to extrinsic-parameter errors. Please clarify whether per-sample LISA evaluations are performed, and if the max-point shortcut is used, validate it against the full evaluation for a subset of signals, especially near rho_LISA about 3.","section":"Sec. III.D.3, around Eq. (24)"},{"comment":"The population-scale credible-interval widths are presented without posterior coverage validation. For multimodal low-SNR posteriors, the width of a 90% credible interval is not, by itself, evidence of statistical validity; the fact that a narrow mode happens to contain the true value is anecdotal. I recommend adding a coverage or P-P diagnostic for the multiband runs, or at least reporting, binned by rho_LISA, the fraction of events whose 90% credible interval contains the injected chirp mass and effective spin. This is directly relevant to the headline claim that LISA improves chirp-mass precision by two to three orders of magnitude.","section":"Sec. IV.C.6, Fig. 14"}],"minor_comments":[{"comment":"The text says 'After we have chosen a specific frame, we can convert them into the same theta', but it is not immediately clear that this conversion is performed on the fly during likelihood evaluation using the rotation matrices in App. A; please make that connection explicit.","section":"Sec. III, after Eq. (4)"},{"comment":"The placement of the real-part operator and parentheses in Eq. (7) is ambiguous; please add brackets so the reader can see that the real part applies to the entire sum over detectors and polarizations.","section":"Eq. (7)"},{"comment":"The time resolution Delta_rho of the SNR time series is used in the discrete time-delay mapping but its value or the rule for choosing it is never stated; please specify it, as it controls the accuracy of the sky-position discretization and hence the proposal prior Pi(tau).","section":"Sec. III.D.3, Eqs. (20)-(21)"},{"comment":"The labels 'signal 108528', 'signal 131308', 'signal 341522', and 'signal 344896' are not defined in the text or caption; please state whether these are dataset event IDs and list their injected parameters, including the true rho_LISA values.","section":"Fig. 13"},{"comment":"The restriction to non-precessing, quasi-circular IMRPhenomD/HM waveforms and to an analytical LISA orbit is stated clearly, but the same caveat should appear in the abstract or introduction where the SNR-3 and near-doubling results are advertised.","section":"Sec. IV.A and App. A"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the journal's scope and the method is a worthwhile contribution. My main concern is that the strongest claims (SNR-3 extraction, doubling of the event rate, population-scale chirp-mass precision) are presented more strongly than the validation supports: only four example events, no coverage or importance-weight diagnostics, and a proposal-flatness argument demonstrated for a restricted waveform family. I do not see grounds for rejection; the issues are fixable by adding diagnostics and by narrowing or qualifying the claims. The authors should also clarify whether the maximum-SNR shortcut described in Sec. III.D.3 is actually used in the reported runs, since that changes the interpretation of the marginalization."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know: this paper does something real. It builds the first coherent joint LISA+3G likelihood for Bayesian parameter estimation, with rotation matrices to unify frames and importance sampling to marginalize over extrinsic parameters. The method is concrete, the code is public, and the population-scale simulations are a genuine step beyond the sequential '3G posterior as LISA prior' approach. If the method holds up, lowering the usable LISA SNR from 5 to 3 would roughly double the multiband event yield, and the chirp-mass precision gains look striking.\n\nWhat it does well: the frame transformations are carefully worked out, the marginalization scheme is sensible, and the comparison against the semi-coherent method on a specific signal shows a real improvement. The population simulations are extensive — 20 seeds per observing scenario, 120 datasets total — and the paper is honest about the restricted waveform family and the concerns about phase coherence and waveform systematics. That is good practice.\n\nThe soft spots are mostly about validation of the central claim. The paper says the method \"successfully analyzes all multiband signals\" with LISA SNR above 3, but that is supported by four example signals at SNR about 3, with no quantitative success criteria, no coverage or bias checks, and no importance-sampling diagnostics. The load-bearing premise is the 2% LISA waveform match over the 3G extrinsic posterior, which is only demonstrated for non-precessing, quasi-circular IMRPhenomD/HM waveforms with an analytical LISA orbit. Within that family the claim is plausible, but outside it — precession, eccentricity, higher modes, a realistic orbit — the proposal could undercover and bias the marginalized likelihood. The paper itself flags this, but it does not test it. The event-rate doubling also assumes analyzability rather than demonstrated detectability, which is a framing issue more than a fatal flaw.\n\nWho it is for: people working on LISA data analysis and multiband science. The method is worth engaging with, and the paper deserves a serious referee. I would send it to review, but with a clear request for posterior coverage checks, importance-weight variance diagnostics, and at least a test of the proposal's robustness to a broader waveform family before the SNR-3 claim is taken as established.","headline":"A genuinely new coherent multiband likelihood with extrinsic marginalization, but the SNR-3 claim rests on a thin validation set; worth refereeing with a demand for coverage diagnostics.","tokens_in":38877,"tokens_out":1358,"would_cite":true,"duration_ms":17247,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["04.30.-w","95.55.Ym"],"model":"deepseek-v4-flash","headline":"The paper claims that coherent phase-locked multiband analysis can extract reliable parameter estimates from LISA signals with SNR as low as 3, nearly doubling the available source count.","keywords":["multiband gravitational wave astronomy","LISA","third-generation ground-based detectors","Bayesian parameter estimation","importance sampling","extrinsic marginalization","stellar-mass black hole binaries","chirp mass measurement"],"falsifier":"Generate a mock multiband event with LISA SNR near 3 whose waveform includes precession or eccentricity, run the coherent method with the same ground-based importance proposal, and check whether the normalized importance weights remain bounded and the chirp-mass posterior stays unbiased; a collapse of the weights would show that the 2-percent waveform-match assumption is the load-bearing limit.","tokens_in":37776,"feed_emoji":"🔭","tokens_out":5606,"duration_ms":59719,"temperature":0.7,"pith_summary":"The paper claims that by evaluating a joint likelihood over LISA and next-generation ground-based data, rather than feeding ground-based posteriors into LISA as priors, one can do reliable multiband parameter estimation for stellar-mass black hole binaries with LISA signal-to-noise ratios as low as 3. This matters because population models predict most multiband sources have LISA SNR below 5, the effective limit of previous sequential methods; lowering the threshold to 3 roughly doubles the available multiband events. The efficiency comes from marginalizing extrinsic parameters, including distance, orbital phase, arrival time, and sky location, while preserving phase coherence at a common reference frequency. On simulated populations, the method yields detector-frame chirp-mass 90% credible intervals below $10^{-4}\\,\\mathrm{M}_\\odot$ for most events even with one year of LISA data, better by one to three orders of magnitude than the ground-based network alone.","feed_headline":"Lowering LISA's SNR threshold to 3 doubles multiband sources","feed_subtitle":"A coherent LISA-plus-ground analysis measures chirp mass to 0.0001 solar masses, even with one year of LISA data.","key_machinery":"The central object is the coherent multiband likelihood, which sums the log-likelihoods of the LISA time-delay-interferometry channels and all ground-based detectors for the same physical waveform, expressed in a common geocentric frame through rotation-matrix transformations. Extrinsic parameters are marginalized to reduce the sampled parameter space: luminosity distance by a precomputed inner-product look-up table, orbital phase analytically through a modified Bessel function, and arrival time plus sky location by importance sampling using the ground-based SNR time series as the proposal distribution. Because the LISA waveform match varies by less than 2% over the ground-based posterior for the restricted waveform family used, the same extrinsic samples serve the LISA part of the likelihood, and the remaining sampling is over intrinsic parameters only.","core_discovery":"The central claim is that the low-SNR barrier in multiband analysis is not fundamental: a coherent likelihood that keeps LISA and ground-based waveform phases consistent, combined with importance-sampling marginalization over extrinsic parameters, can extract information from signals with LISA SNR as low as 3, where noise-only matched-filter peaks are comparable to the signal peak. The paper demonstrates this on population scale, analyzing all simulated multiband signals with LISA SNR above 3 and showing that chirp mass, mass ratio, and spin estimates improve substantially over ground-based-only results. It also shows that even a single year of LISA data, analyzed this way, gives most multiband sources a detector-frame chirp-mass 90% credible interval below $10^{-4}\\,\\mathrm{M}_\\odot$, exceeding the best ET+2CE-only measurements by at least an order of magnitude.","pith_inferences":["If the 2%-match assumption fails for precessing, eccentric, or higher-mode waveforms, the SNR-3 threshold would likely rise, or the method would need a proposal distribution built iteratively from the multiband posterior itself.","The same extrinsic-marginalization machinery could be applied to LISA-only or LISA-plus-current-ground-based early-warning searches, where the proposal distribution is weaker and the practical SNR threshold would be set by proposal quality rather than by the likelihood alone.","The population-scale posterior widths already trace the underlying mass distribution, so a natural next step is hierarchical population inference that treats the multiband chirp-mass measurements as data, potentially sharpening merger-rate and formation-channel estimates.","The claim that one LISA year is enough revises earlier mission-planning expectations and could be tested against LISA Data Challenge-style datasets that include overlapping signals and realistic confusion noise."],"forward_implications":["Multiband analyses can work at LISA SNR roughly 3, nearly doubling the number of usable sources relative to the previous threshold of 5.","Population-scale multiband Bayesian parameter estimation becomes computationally practical, with individual runs taking days on a small CPU cluster rather than being prohibitive.","Even one year of LISA data yields detector-frame chirp-mass 90% credible intervals below $10^{-4}\\,\\mathrm{M}_\\odot$ for most multiband sources, beating ET+2CE-only measurements by at least an order of magnitude.","LISA's main multiband contribution is precision on intrinsic phase-sensitive parameters such as chirp mass, mass ratio, and spins, while sky localization improves only marginally because the ground-based network already constrains it well.","Lowering the effective LISA SNR threshold from 5 to 3 shifts the multiband event-rate bottleneck from data-analysis capability to the intrinsic LISA sensitivity and mission duration."],"supporting_citations":[{"why":"Defines the sequential 3G-posterior-as-LISA-prior method whose LISA SNR around 5 limit this paper lowers to roughly 3.","marker":"[12]"},{"why":"Supplies the population model, including the power-law-plus-peak mass distribution, used to generate the mock multiband event sets.","marker":"[6]"},{"why":"Establishes the analytic LISA orbit and coordinate-frame conventions that the rotation-matrix transformations build on.","marker":"[48]"},{"why":"Provides the LISA waveform and time-delay-interferometry response model used for injections and likelihood evaluations.","marker":"[66]"},{"why":"Supplies the fast importance-sampling marginalization over arrival time and sky location that the proposal distribution is based on.","marker":"[69]"},{"why":"Gives the Rayleigh-distributed noise-SNR behavior and the mock-data generation recipe used to build and validate the population datasets.","marker":"[71]"}],"fun_headline_variants":["LISA SNR 3 doubles multiband source yield","Coherent LISA+ground analysis works at SNR 3","Chirp mass to 10^-4 solar masses from one-year LISA","Importance sampling enables LISA analysis at SNR 3","Precise chirp mass from low-SNR LISA and ground"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method assumes that, over the arrival-time and sky-location region allowed by the ground-based posterior, the LISA waveform changes by less than about 2 percent, so ground-based samples can serve as the proposal for the LISA part of the likelihood; the paper demonstrates this only for non-precessing, quasi-circular, aligned-spin waveforms with an analytic LISA orbit.","fun_headline_variants_meta":{"raw":{"variants":["LISA SNR 3 doubles multiband source yield","Coherent LISA+ground analysis works at SNR 3","Chirp mass to 10^-4 solar masses from one-year LISA","Importance sampling enables LISA analysis at SNR 3","Precise chirp mass from low-SNR LISA and ground"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001289,"raw_usage":{"total_tokens":5313,"prompt_tokens":1043,"completion_tokens":4270,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":659,"completion_tokens_details":{"reasoning_tokens":4180}},"tokens_in":659,"tokens_out":4270,"duration_ms":29912,"temperature":1.0,"reasoning_tokens":4180,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T11:31:52.533870+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Generate a mock multiband event with LISA SNR near 3 whose waveform includes precession or eccentricity, run the coherent method with the same ground-based importance proposal, and check whether the normalized importance weights remain bounded and the chirp-mass posterior stays unbiased; a collapse of the weights would show that the 2-percent waveform-match assumption is the load-bearing limit.","supporting_citations":[],"review_version":1}