{"id":"12a8576e-f8dc-4745-9f3f-577cbe9217bd","arxiv_id":"2501.01499","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"A stacking statistic over galaxy-overdense pixels distinguishes simulated isotropic from AGN-traced anisotropic nanohertz gravitational-wave backgrounds.","lead":"The authors propose a stacking method that sums gravitational wave background signals from sky regions rich in AGN and galaxies to detect anisotropies in the nanohertz stochastic background. They test it on simulations and report that it can tell apart an isotropic background from one that traces galaxies, and argue it beats angular power spectrum methods.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central claim of superiority over angular power spectra is unsupported: no C_l comparison on same maps and quoted p-values are not null-hypothesis probabilities.","rationale":"The reader's verdict is CONDITIONAL and already flags the missing C_l comparison and the misuse of 'p-value' in its rationale; my independent reading converges on the same statistical gap as the most load-bearing issue, rather than on the AGN tracer fidelity that the reader listed as the weakest assumption. The tracer-fidelity point is real for application to actual data, but the method's internal demonstration does not hinge on it: even if the Schechter-function mass assignment were perfectly faithful, the paper still would not have shown that stacking outperforms angular power spectra, because no C_l-based statistic was computed on the same maps and no proper null-hypothesis false-alarm rate was reported. The self-consistency check (injecting anisotropy into the AGN map and recovering a positive stack) is useful but is not a detection claim. The concrete test I propose—ROC curves on identical realizations plus correctly defined null p-values—would settle whether the central 'surpassing' assertion survives. Since the requested changes are consistent with the reader's conditional verdict, no verdict adjustment is needed.","tokens_in":15675,"tokens_out":4037,"duration_ms":44221,"concrete_test":"On the same 1000 simulated realizations used for Figs. 6–10 (both Frac and isotropic cases), compute (1) the stacked statistic S; (2) a C_l-based statistic, e.g., χ² = Σ_l (C_l^obs − C_l^iso)^2 / Var(C_l^iso) or the cross-power with the AGN map; then build ROC curves by thresholding on the 95th/99th percentile of the isotropic realizations. If the anisotropic detection fraction for stacking is not higher than for the C_l statistic at equal false-alarm rate, the 'surpassing' wording must be dropped. Separately, re-express Table 1 p-values as the fraction of isotropic realizations with S ≥ the observed anisotropic median; those are the actual false-alarm probabilities and should be reported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline claim—that Multi-Tracer Correlated Stacking 'uniquely distinguishes' isotropic from anisotropic SGWB and 'surpasses' angular power spectrum methods—is not established by the reported statistics. The demonstration injects a fraction Frac of the GW sources into the same AGN map that is later used to select stacking pixels; a positive stacked value is therefore partly built in. More importantly, no C_l-based detection statistic is ever computed on the same realizations, so 'surpassing' is an assertion, not a result. The p-values in Table 1 and Fig. 10 are defined as the fraction of anisotropic realizations with stacked signal ≤ 0; that is a conditional non-detection probability, not the false-alarm probability under the isotropic null, which is what a detection significance requires. Because the null distribution of the stacked statistic is never used to threshold detections, the paper does not demonstrate that the method can discover anisotropy, nor that it does so better than C_l.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a method, Multi-Tracer Correlated Stacking, for detecting anisotropy in the nanohertz stochastic gravitational wave background (SGWB). The method selects sky pixels where a tracer (here an AGN catalog, with stellar masses assigned from a redshift-dependent Schechter function) has positive overdensity and sums the SGWB density fluctuation in those pixels. The authors simulate a supermassive black hole binary population, inject a fraction Frac of the SGWB from hosts in the AGN catalog and the rest from an isotropic distribution, and show that the stacked statistic is positive for the AGN-traced case and near zero for the isotropic case. They study dependence on Frac, astrophysical parameters, pulsar number and distribution, and tomographic redshift bins. The paper claims that this technique uniquely distinguishes isotropic from anisotropic SGWB and surpasses angular power spectrum methods.","tokens_in":15796,"tokens_out":3883,"duration_ms":43271,"significance":"If the comparative claim were established, this would be a useful addition to the SGWB anisotropy toolkit: the estimator is simple, parameter-free in its construction, and directly tied to a physical tracer of supermassive black hole hosts. The forward-model exercise is internally consistent and the qualitative separation between injected anisotropic and isotropic cases is a sensible injection-recovery test. However, the paper's headline claims—that the method 'uniquely distinguishes' and 'surpasses' angular power spectrum methods—are not supported by the quantitative results presented. The paper does not compute any C_l-based detection statistic on the same maps, and the reported p-values are not false-alarm probabilities under the isotropic null. The central methodological claim therefore needs additional work before the paper can be accepted as a demonstration of superiority.","major_comments":[{"comment":"The central claim that the stacking technique 'surpasses the capabilities of angular power spectrum-based methods' is not tested anywhere in the paper. No C_l-based detection statistic (e.g., a matched-filter S/N, a likelihood ratio, or a false-alarm rate from C_l) is computed on the same simulated SGWB maps. Figure 2 shows only the distribution of C_l coefficients and of pixel fluctuations; it does not quantify whether C_l fails to detect the injected anisotropic signal. To support the comparative claim, the authors should compute a C_l detection statistic on the same realizations and compare its detection efficiency/false-alarm rate with the stacked statistic. Without this, the superiority claim in the abstract and Sec. 2 is an assertion, not a result.","section":"Abstract; Sec. 2"},{"comment":"The quantity called 'p-value' is not a null-hypothesis probability. It is defined as the fraction of anisotropic realizations for which the stacked signal is less than or equal to zero. That is a conditional non-detection probability given that the signal is present, not the false-alarm probability under the isotropic null. Moreover, the noise realizations in Fig. 10 are centered at the median stacked signal of the anisotropic case, so the noise p-value measures only the tail of the noise distribution below zero given that the median is positive. To demonstrate that the method can 'discover' anisotropy, the authors need to compute the distribution of the stacked statistic under the isotropic null (with the same noise and the same pixel selection) and report the false-alarm probability or detection significance for a threshold. As written, the p-values in Fig. 10 and Table 1 do not establish a detection criterion.","section":"Sec. 5, Fig. 10, Table 1"},{"comment":"The positive stacked signal is partly built in by construction: a fraction Frac of the injected SGWB sources is placed in the same AGN overdensity map that is later used to select stacking pixels. This is a valid injection-recovery test of the estimator's response to a correlated signal, but it does not by itself show that the method can discover anisotropy when the tracer is an imperfect or incomplete representation of the SMBHB host population. The authors acknowledge catalog limitations in Sec. 6, but no simulation tests the effect of tracer incompleteness, tracer bias, or stochastic mismatch between the tracer and the true host distribution. A concrete test using, for example, a tracer map with a different selection function or with added noise would show whether the stacked statistic remains a reliable indicator of anisotropy beyond the idealized same-map injection.","section":"Sec. 3.2 and Sec. 5"},{"comment":"The claim that angular power spectra 'primarily capture the Gaussian component' and 'can misrepresent or underestimate' non-Gaussian features is not demonstrated quantitatively. The paper shows that the pixel-space distribution is skewed (Fig. 2b), but skewness alone does not imply that a C_l-based estimator is biased or less sensitive for the signals considered here. The authors should either provide a quantitative comparison (e.g., detection S/N of C_l versus the stacked statistic on the same anisotropic realizations) or soften the claim to state that C_l is not optimized for spatially localized, tracer-correlated fluctuations. As written, the non-Gaussianity argument is an unsupported premise for the central superiority claim.","section":"Sec. 2"}],"minor_comments":[{"comment":"The caption of Fig. 10 says 'Np = 400 and ℓmax = 24', while the text says the figure is for Np = 600 and ℓmax = 24. The numerical p-values in the figure match the Np = 600 row of Table 1, so the caption appears to contain a typo.","section":"Sec. 5, Fig. 10"},{"comment":"The phrase 'isotopic scenario' should read 'isotropic scenario'.","section":"Sec. 2"},{"comment":"The spherical harmonic coefficients a_lm are introduced for the SGWB fluctuation but the frequency dependence is suppressed. Since ΔΩ_gw is frequency dependent, the definition of C_l should specify the frequency at which it is evaluated, or state that the evaluation is per frequency bin.","section":"Sec. 2, Eq. (1)"},{"comment":"The parameter Frac is described both as the 'fraction of the SGWB power spectrum' contributed by AGNs and later as the fraction of Ω_iso_gw. These are not the same quantity; clarify that Frac is the fraction of the SGWB energy density assigned to AGN hosts in the simulation.","section":"Sec. 3.2"},{"comment":"The quantity called 'p-value' should be renamed, for example 'tail probability' or 'non-detection fraction', to avoid implying a test of a null hypothesis. The current terminology is misleading in a detection context.","section":"Sec. 5, Fig. 10 and Table 1"}],"recommendation":"major_revision","confidential_remarks":"The paper is a reasonable methods proposal, and the injection-recovery exercise is a sensible first step. The main issue is that the abstract and Sec. 2 make comparative and discovery claims that are not supported by the statistics actually computed. These claims are fixable within the manuscript's scope by adding a C_l comparison on the same maps, computing the null distribution of the stacked statistic, and testing tracer robustness. I do not see a need to reject the paper, but it should not be accepted until the quantitative support for the central claim is added or the claims are substantially weakened."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a simulation method paper that stacks SGWB fluctuations in AGN-overdense pixels. It works on injected signals, but the headline claim that it surpasses angular power spectrum methods is not supported by the analysis, and the quoted p-values are not false-alarm probabilities. Worth engaging, but it needs substantial revision.\n\nWhat's actually new: the specific estimator is a close relative of the galaxy–SGWB cross-correlation that the same group already demonstrated (Sah & Mukherjee 2024), but thresholded stacking of ΔΩ_GW in tracer-selected pixels is a distinct twist. It targets a real gap: angular power spectra are a sufficient statistic for Gaussian isotropic fields, and the nHz background from a sparse SMBHB population is expected to be shot-noise dominated and non-Gaussian. The forward modeling is solid in spirit: a realistic AGN catalog (Million Quasars), stellar masses assigned from a Schechter function, 1000 Monte Carlo realizations, and a demonstration that an isotropic source distribution gives a stacked signal centered at zero while an AGN-correlated one gives a positive signal. The redshift-tomography figure is a useful negative result, honestly showing the method cannot pick out the dominant redshift bin.\n\nSoft spots, in order. First, the central claim of superiority over C_l is not tested. No angular power spectrum detection statistic is computed on the same maps, so \"surpassing\" is an assertion. Second, the p-values in Table 1 and Fig. 10 are the fraction of anisotropic realizations with stacked signal ≤ 0 — a conditional probability given that the signal exists, not the false-alarm rate under the isotropic null. A proper detection significance needs the null distribution. Third, the \"multi-tracer\" language overstates what is demonstrated: only the AGN catalog is used, with the multi-tracer extension deferred to future work. Injecting a fraction Frac of the GW sources into the same AGN map that later selects stacking pixels is standard injection-recovery, not circular reasoning, but it does mean the positive stacked value is built in by construction. The noise model also uses a single value σ_IJ = 1 for all pulsar pairs.\n\nWho this is for: PTA analysts and people working on SGWB anisotropy estimators. The method may prove useful with future denser pulsar arrays and deeper tracer catalogs, but the current version doesn't demonstrate detection power against the standard alternative. I'd send it to peer review because the idea deserves a serious look, with the request that the authors add a direct C_l comparison on the same simulations and recompute p-values under the isotropic null.","headline":"A promising but overstated stacking estimator for nHz SGWB anisotropy: injections work, but the claimed superiority over C_l is untested and the p-values are not false-alarm rates.","tokens_in":16377,"tokens_out":3455,"would_cite":false,"duration_ms":37008,"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":"This paper proposes a stacking estimator that detects non-Gaussian anisotropy in the nano-hertz gravitational-wave background by adding signal only in tracer-overdense pixels, distinguishing AGN-traced from isotropic source populations…","keywords":["gravitational waves","stochastic gravitational wave background","pulsar timing arrays","anisotropy","supermassive black hole binaries","active galactic nuclei","stacking analysis","large-scale structure"],"falsifier":"Apply the stacking estimator to real pulsar-timing-array residuals using a complete AGN catalog at a resolution of $\\ell_{\\max} \\approx 20$ or higher: if the measured $\\hat{\\Omega}_{\\rm stacked}$ is consistent with zero in a regime where the simulation predicts a positive signal with a p-value near or below one percent, the tracer assumption or the simulated source population is wrong.","tokens_in":15399,"feed_emoji":"📡","tokens_out":7688,"duration_ms":68668,"temperature":0.7,"pith_summary":"This paper argues that the standard tool for measuring anisotropy in the nano-hertz gravitational-wave background, the angular power spectrum, is blind to the strongly non-Gaussian fluctuations expected when supermassive black hole binaries are sparse and trace massive galaxies. It introduces a new estimator, Multi-Tracer Correlated Stacking: select sky pixels where a tracer such as an active galactic nucleus is overdense, then sum the gravitational-wave fluctuations in those pixels. On simulations built from an AGN catalog out to redshift five, the stacked signal is strongly positive when the sources follow the tracer and centered at zero when the sources are isotropic, while the angular power spectrum cannot separate the two cases. The method's sensitivity grows with pulsar number and map resolution, and it can also probe what fraction of the background comes from AGN hosts. The authors position this as a route to discovering anisotropy in real pulsar-timing-array data with upcoming surveys.","feed_headline":"Stacking galaxy-rich pixels exposes gravitational wave anisotropy","feed_subtitle":"Technique beats angular power spectra at detecting non-Gaussian nano-hertz signals from black hole binaries.","key_machinery":"The central object is the Multi-Tracer Correlated Stacking estimator $\\hat{\\Omega}_{\\rm stacked} = \\sum_{i: \\delta_g^i > \\delta_{\\rm cut}} \\Delta\\Omega_{\\rm gw}^i$, where $\\delta_g$ is the fractional galaxy number-density fluctuation in a pixel and $\\Delta\\Omega_{\\rm gw}$ is the fractional stochastic-background energy-density fluctuation. The estimator threshold-selects pixels from a tracer map, here an AGN catalog, and coherently adds the gravitational-wave signal only in overdense regions, converting the spatial correlation between binaries and their host galaxies into a large positive sum. The paper contrasts this with the angular power spectrum $C_\\ell$, which measures variance and is shot-noise dominated for the sparse nano-hertz source population.","core_discovery":"The central claim is that stacking gravitational-wave background fluctuations in pixels selected by tracer overdensity is a direct, non-Gaussian-sensitive measurement of anisotropy, and it outperforms angular power spectra in the sparse-source regime of the nano-hertz background. Using Monte Carlo realizations of supermassive black hole binaries assigned to galaxies in an AGN catalog, the paper shows that the stacked estimator $\\hat{\\Omega}_{\\rm stacked} = \\sum_{i: \\delta_g^i > \\delta_{\\rm cut}} \\Delta\\Omega_{\\rm gw}^i$ is positive with high significance when sources trace the AGN distribution and consistent with zero for isotropic sources. At a resolution of $\\ell_{\\max}=20$ with full AGN contribution, the noise-only p-value is $0.001$, and the distinction persists across astrophysical parameters and under a realistic anisotropic pulsar distribution. The paper also finds that the method cannot identify which redshift bin dominates the signal, a task left to angular cross-correlation.","pith_inferences":["Applied to real data, a null result from stacking could set an upper bound on the AGN fraction of the nano-hertz background even before a full anisotropy map is available, since the simulated signal scales with that fraction.","The same estimator could be run with quasar, bright-galaxy, or other tracer catalogs; each tracer's stacked signal would measure how tightly that population tracks the supermassive-black-hole-binary hosts.","Because angular-power-spectrum upper limits are insensitive to non-Gaussian anisotropy, existing null anisotropy searches may be consistent with a strongly anisotropic background; stacking offers a direct, assumption-light test.","A combined pipeline using stacking for detection and cross-correlation for redshift tomography would recover both the amplitude and the redshift origin of the anisotropy."],"forward_implications":["At $\\ell_{\\max}=20$ with the AGN contributing the full background, the noise-only p-value for a positive stacked signal is $0.001$, so a positive measurement at that resolution would be strong evidence that the nano-hertz background is anisotropic.","The stacked signal grows with pulsar count and map resolution, so future arrays with many more pulsars will sharpen the test.","A realistic anisotropic pulsar distribution, such as that of current arrays, degrades precision but does not erase the distinction between isotropic and tracer-correlated sources.","The stacked signal increases monotonically with the AGN fraction $\\mathrm{Frac}$, which gives a route to estimating how much of the background is hosted by active galactic nuclei.","Stacking in redshift-selected tracer bins cannot locate the dominant redshift of the signal; recovering that information requires angular cross-correlation with galaxy surveys."],"supporting_citations":[{"why":"Supplies the Monte Carlo simulation framework for assigning supermassive black hole binaries to galaxies and the black-hole-mass-stellar-mass relation used to populate the catalog.","marker":"Sah et al. 2024"},{"why":"Provides the Million Quasars AGN catalog used as the tracer map in the stacking demonstration.","marker":"Flesch (2023)"},{"why":"Provides the redshift-dependent Schechter function used to assign stellar masses to AGNs lacking mass measurements.","marker":"McLeod et al. (2021)"},{"why":"Furnishes the background amplitude and spectral index fitted to set the isotropic reference gravitational-wave density.","marker":"Agazie et al. 2023a"},{"why":"Underlies the likelihood and Fisher-information framework used for the map-domain pixel noise.","marker":"Pol et al. 2022"},{"why":"Motivates the choice linking pulsar count to sky resolution.","marker":"Boyle & Pen 2012"},{"why":"Justifies the far-away point-source approximation for the Fisher matrix and pixel noise.","marker":"Romano & Cornish 2017"},{"why":"Provides the angular cross-correlation method that the stacking technique is compared against for redshift information.","marker":"Sah & Mukherjee 2024"}],"fun_headline_variants":["Tracer stacking uncovers nano-Hz gravitational wave anisotropy","Multi-tracer stacking beats power spectra for gravitational wave background","Stacking galaxy regions exposes gravitational wave anisotropy","New stacking technique reveals nano-Hz gravitational wave anisotropy","Tracer-correlated stacking detects non-Gaussian gravitational wave anisotropy"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The demonstration assumes the AGN catalog faithfully traces where supermassive black hole binaries actually live, and that the catalog's missing stellar masses are realistically modeled; if that tracer assumption is wrong, a positive stacked signal no longer proves the background is anisotropic.","fun_headline_variants_meta":{"raw":{"variants":["Tracer stacking uncovers nano-Hz gravitational wave anisotropy","Multi-tracer stacking beats power spectra for gravitational wave background","Stacking galaxy regions exposes gravitational wave anisotropy","New stacking technique reveals nano-Hz gravitational wave anisotropy","Tracer-correlated stacking detects non-Gaussian gravitational wave anisotropy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000292,"raw_usage":{"total_tokens":1758,"prompt_tokens":1055,"completion_tokens":703,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":671,"completion_tokens_details":{"reasoning_tokens":625}},"tokens_in":671,"tokens_out":703,"duration_ms":7491,"temperature":1.0,"reasoning_tokens":625,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:28:04.650636+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Apply the stacking estimator to real pulsar-timing-array residuals using a complete AGN catalog at a resolution of $\\ell_{\\max} \\approx 20$ or higher: if the measured $\\hat{\\Omega}_{\\rm stacked}$ is consistent with zero in a regime where the simulation predicts a positive signal with a p-value near or below one percent, the tracer assumption or the simulated source population is wrong.","supporting_citations":[{"cited_title":"2021, Monthly Notices of the Royal Astronomical Society, 503, 4413","cited_arxiv_id":null,"evidence_quote":"Provides the redshift-dependent Schechter function used to assign stellar masses to AGNs lacking mass measurements."},{"cited_title":"R., & Romano, J","cited_arxiv_id":null,"evidence_quote":"Underlies the likelihood and Fisher-information framework used for the map-domain pixel noise."},{"cited_title":"2012, Physical Review D—Particles, Fields, Gravitation, and Cosmology, 86, 124028","cited_arxiv_id":null,"evidence_quote":"Motivates the choice linking pulsar count to sky resolution."},{"cited_title":"D., & Cornish, N","cited_arxiv_id":null,"evidence_quote":"Justifies the far-away point-source approximation for the Fisher matrix and pixel noise."}],"review_version":1}