{"id":"d8974075-059e-4daf-a38d-635a856c3850","arxiv_id":"2608.09781","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"PTA anisotropy sensitivity scales as Npsr^0.8, delta-t^-0.08, and angular-scale exponents 1.6 to 2.1, and a new multi-resolution pixel basis is demonstrated.","lead":"This paper computes how sensitive pulsar timing arrays are to maps of gravitational wave background anisotropy, using simulations and Fisher information. It finds scaling laws with pulsar count, timing noise, frequency, and angular scale, and introduces a new multi-resolution sky pixel basis for searches.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The scaling exponents in §3.2 are fit to diag(M) alone, but §5 reports full pixel-basis Fisher matrices with condition numbers 10^14–10^18; if the omitted off-diagonal structure changes with Npsr, δt, or ℓ, the headline scaling laws may be systematically biased.","rationale":"The reader's weakest assumption correctly identifies the diagonal-only approximation. Our stress test finds no additional load-bearing concern beyond this: the paper honestly labels the multi-resolution basis as preliminary, the f_GW scaling is explicitly limited to the Fisher matrix, and the forecasting acknowledges bin-matching issues. The key gap is that the scaling exponents are never re-derived from the full inverse Fisher matrix, despite Section 5 showing the matrix is near-singular. This makes the central claim conditional on an untested proxy. The proposed concrete check directly settles whether the proxy biases the exponents, so the reader's CONDITIONAL verdict remains appropriate; no change is needed.","tokens_in":11,"tokens_out":7543,"duration_ms":473035,"concrete_test":"Recompute the scaling-law fits of Section 3.2 using the diagonal of the full inverse Fisher matrix, diag(M^{-1}), instead of diag(M), for a representative subset of the same simulated datasets (e.g., pixel basis at Nside=8 and spherical harmonic basis at lmax=8, using SVD-regularized pseudo-inversion to handle the near-singular M). Apply the identical median-of-medians fitting procedure and compare the resulting exponents with Eqs. (13)–(15) and the Figure 4 slope. If any exponent shifts by more than its quoted uncertainty, the diagonal-only approximation is load-bearing and the scaling laws need qualification or correction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central quantitative claim—Eqs. (13)–(15) and the ΔΩ exponent in Figure 4—is that PTA sensitivity to anisotropy scales as Npsr^0.8, δt^-0.08, and ΔΩ^1.6–2.1. These exponents are fit to sqrt(diag(M)), the diagonal of the Fisher matrix, with the covariance between pixels/spherical harmonics explicitly neglected (Section 2.1). Section 5 reports that the full pixel-basis Fisher matrix has condition numbers O(10^14) for uniform pulsars and O(10^18) for the NANOGrav distribution, and that zeroing the off-diagonal elements reduces the condition number to O(1). This demonstrates that the off-diagonal structure dominates M and that individual pixels are nearly degenerate. The true per-pixel uncertainty is sqrt(diag(M^{-1})), which can be much larger than 1/sqrt(diag(M)) when M is ill-conditioned; the two coincide only for diagonal M. Nothing in the paper checks whether the fitted exponents are robust to using diag(M^{-1}) instead of diag(M). If the degeneracy structure changes with Npsr, δt, or ℓ—for example, if adding pulsars shrinks the smallest eigenvalues faster than it grows the diagonal—then the headline exponents may not describe the actual sensitivity that a search would achieve. This is not an external objection; it is a testable internal consequence of the paper's own condition-number analysis.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper develops empirical scaling relations for PTA sensitivity to anisotropy in the gravitational-wave background, using the per-frequency optimal statistic (PFOS) to build Fisher information matrices in both a HEALPix pixel basis and a spherical harmonic basis. The authors simulate datasets based on the NANOGrav 15-year data and report that their sensitivity measure, sqrt(diag(M)), scales approximately as Npsr^0.8, delta-t^-0.08, f_GW^-0.12, and Delta-Omega^1.58-2.1 depending on the ranges of ell and m. They also simulate a 30-year NANOGrav-like array and find that increasing the observing baseline alone has negligible effect on sensitivity except at the lowest frequencies. Finally, they introduce a multi-resolution HEALPix pixel basis and demonstrate it with analytical injection-recovery tests.","tokens_in":23327,"tokens_out":11817,"duration_ms":99312,"significance":"If the reported scaling relations are robust, they provide practical guidance for PTA array design and for forecasting sensitivity to anisotropic backgrounds, complementing earlier work by Siemens et al. (2013), Pol et al. (2022), and Depta et al. (2024). The paper's strengths are its realistic simulation setup, explicit comparison with hasasia in the isotropic limit, discussion of pair covariance, and public availability of the analysis code as a fork of Defiant. The multi-resolution pixel basis is a promising idea, though it is presented as a proof of concept rather than a complete method. The main limitation is that the headline exponents are derived from a diagonal-only approximation of the Fisher matrix, which the paper's own condition-number analysis shows to be a very strong simplification; this needs to be addressed before the scaling laws can be taken as general statements about PTA anisotropy sensitivity.","major_comments":[{"comment":"The central scaling laws in Eqs. (13)-(15) and Figure 4 are based on sqrt(diag(M)) (equivalently 1/sqrt(diag(M)) under the diagonal approximation), which neglects covariances between basis functions. Section 5 reports that the full pixel-basis Fisher matrix has condition numbers O(10^14) for uniform pulsars and O(10^18) for the NANOGrav distribution, and that zeroing the off-diagonal elements reduces this to O(1). This shows that the off-diagonal structure dominates the Fisher matrix, so the true marginal uncertainty sqrt(diag(M^{-1})) can be very different from 1/sqrt(diag(M)). The authors should recompute the scaling exponents using sqrt(diag(M^{-1})) with an appropriate regularization or pseudo-inverse, or otherwise demonstrate that the exponents are robust to including the off-diagonal covariances. If the off-diagonal structure changes with Npsr, delta-t, or ell, the reported scaling may describe only the diagonal radiometer statistic and not the sensitivity of a full anisotropy search. At minimum, the abstract and conclusions should explicitly qualify that the scaling laws apply to the diagonal-only approximation.","section":"Section 2.1 / Section 5"},{"comment":"The angular scaling exponent is derived from a non-representative subset of spherical harmonic coefficients: the headline slope 1.58 uses only m=0 with 8<=ell<=20, while the right panel of Figure 4 shows a strong dependence on m, with exponents ranging roughly from 1.5 to 2.2. Because m=0 corresponds to spherical harmonics concentrated near the poles, where the NANOGrav pulsar density is low, a single power law fitted to m=0 may not be a meaningful summary of the angular scaling. The authors should provide an explicit aggregation rule over m and ell, and report the goodness of fit of the power-law model over the chosen ell range. The abstract's range \"Delta-Omega^1.6-Delta-Omega^2.1\" should be tied directly to the chosen statistic and to the scatter shown in Figure 4.","section":"Section 3.2 / Figure 4"},{"comment":"There is an inconsistency in the text about which quantity is used for sensitivity curves versus scaling-law fits. Section 2.2 states that uncertainties are \"the square root of the diagonal elements of M^{-1}\", while Section 3.2 says the scaling laws are fit to sqrt(diag(M)). These two quantities are equal only if M is diagonal, which Section 5 argues is not the case. The authors should clarify the exact quantity used for each figure and analysis, and ensure that the sensitivity curves (e.g., Figure 1) and the scaling-law fits are based on the same definition. This clarification is important for interpreting the reported exponents and for comparing them with prior scaling relations.","section":"Section 2.2 / Section 3.2"}],"minor_comments":[{"comment":"Please state the HEALPix Nside used for the pixel-basis sensitivity curves and for the scaling-law fits in Section 3.2; this information is needed to reproduce the results and to assess the condition-number discussion in Section 5.","section":"Section 3.1 / Figure 1"},{"comment":"The abstract reports \"Delta-Omega^1.6-Delta-Omega^2.1\" while the text reports 1.58 +/- 0.03 for m=0 and a range of roughly 1.5-2.2 for fixed m in Figure 4; please reconcile these numbers and state the precise range and selection criteria in the abstract.","section":"Abstract / Section 3.2"},{"comment":"The text says the uncertainty on the scaling exponent is \"1 standard deviation\" but the fits use asymmetric error bars based on the 25th-75th percentiles; please clarify how the standard deviation is obtained from the percentile-based errors or use consistent error reporting throughout.","section":"Section 3.2"},{"comment":"The y-axis of Figure 5 is restricted and the text notes that some changes extend beyond the range; please consider showing the full distribution or stating explicitly the fraction of cases that fall outside the plotted range.","section":"Figure 5"},{"comment":"The final row labeled \"All (GWB)\" should be explicitly identified in the table caption as the injected GWB parameters rather than a pulsar-specific noise parameter.","section":"Table 1"},{"comment":"The notation \"p 1/diag(M)\" is nonstandard and could be confused with the power vector P; please write 1/sqrt(diag(M)) or sqrt(diag(M^{-1})) explicitly, depending on which quantity is intended.","section":"Section 2.1"}],"recommendation":"major_revision","confidential_remarks":"This is a well-structured and useful simulation study from the NANOGrav collaboration, with reproducible code and honest caveats. The main issue is that the headline scaling exponents rest on a diagonal-only Fisher approximation that the paper itself shows to be a poor representation of the full Fisher matrix. This is fixable by recomputing the exponents with the full covariance or by clearly limiting the claims to the diagonal radiometer statistic. I would not reject; the work is likely to be valuable to the PTA community after revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Plainly: this is a useful, honest methods paper. The new bits are the empirical scaling laws for PTA anisotropy sensitivity (Npsr^0.8, δt^-0.08, f_GW^-0.12, and ΔΩ^1.6-2.1), the forecast that extending Tspan alone does little except at the lowest frequency bins, and the multi-resolution pixel basis. The simulations are detailed, the code is forked on GitHub, and validation against hasasia in the isotropic limit gives me confidence the pipeline is sound. The pair-covariance factor-of-two effect is also worth knowing.\n\nThe soft spot is real and it is the one the reader flags: the headline exponents are fit to sqrt(diag(M)), not sqrt(diag(M^-1)). The paper itself reports condition numbers of 1e14-1e18 for the full pixel-basis Fisher matrix, which means off-diagonal covariances dominate and diag(M) is not a reliable proxy for per-pixel uncertainty. The true uncertainty can be much larger, and there is no check that the fitted exponents survive if you use diag(M^-1). This matters because the scalings are presented as guidance for array design. I also notice an internal inconsistency: Section 2.2 says the pixel-basis sensitivity maps use diag(M^-1), while Section 3.2 fits scalings to diag(M). That should be reconciled.\n\nThe angular scaling is more nuanced than the abstract suggests: the 1.58 exponent comes from l=8-20 and m=0, and the m-dependence in the right panel of Figure 4 spans 1.5-2.2. They do say this, so it is not hidden, but the abstract's single range understates the orientation dependence.\n\nThe Tspan result is plausible and well explained by the number of pulsars whose baselines reach the lowest bins. The multi-resolution basis is clever but explicitly preliminary; the authors limit the demonstration to a few pixels because of the same ill-conditioning. That honesty is to their credit.\n\nOverall the paper deserves a serious referee. It is not a paradigm changer, but it gives the PTA community quantitative scaling laws and a new tool, and the main weakness is addressable. I would ask the authors to redo the scaling fits using diag(M^-1) (or at least check robustness with regularization) and to clarify which quantity is used where.","headline":"Solid empirical scaling laws for PTA anisotropy sensitivity, but the headline exponents are fit to the Fisher diagonal while the paper's own condition-number analysis shows off-diagonal terms dominate.","tokens_in":24420,"tokens_out":3340,"would_cite":true,"duration_ms":30586,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A pulsar timing array's sensitivity to directional structure in the gravitational wave background scales roughly as the number of pulsars to the 0.8 power, barely moves with timing precision, and gains little from extending a 16-year run…","keywords":["gravitational wave background","pulsar timing arrays","anisotropy","Fisher information matrix","sensitivity scaling","HEALPix","multi-resolution pixel basis"],"falsifier":"Recompute the scaling-law fits for $N_\\mathrm{psr}$, $\\delta t$, and $\\Delta\\Omega$ using the full inverse Fisher matrix $\\mathbf{M}^{-1}$ (with SVD regularization, since the paper reports condition numbers $\\sim10^{14}$ to $10^{18}$) instead of diagonal-only inversion; if the fitted exponents move by more than the quoted uncertainties, the reported scaling relations are artifacts of the diagonal approximation.","tokens_in":2127,"feed_emoji":"📡","tokens_out":4818,"duration_ms":138603,"temperature":0.7,"pith_summary":"This paper asks how a pulsar timing array's ability to map direction-dependent structure in the gravitational wave background depends on the array's size, timing precision, observing frequency, and angular resolution. Using simulations built from a real 15-year dataset, it finds that sensitivity to anisotropy scales roughly as the number of pulsars to the 0.8 power, only weakly with per-pulse timing error, and steeply with the angular size of the anisotropy. It also forecasts that extending observations from 16 to 30 years, by itself, barely improves anisotropy sensitivity except at the lowest frequencies, where newly matured pulsar baselines add weight. These scaling laws give array designers a concrete target: adding pulsars and filling sky coverage buys far more anisotropy sensitivity than waiting longer. The paper closes by introducing a multi-resolution pixel basis that places smaller pixels where the array is more sensitive, and demonstrates that injected sky maps are recovered accurately with it.","feed_headline":"Doubling pulsars boosts gravitational-wave sky-map sensitivity ~74%","feed_subtitle":"Adding pulsars beats waiting: longer runs barely improve anisotropy sensitivity except at low frequencies.","key_machinery":"The central object is the cross-correlation Fisher information matrix $\\mathbf{M}=\\mathbf{R}^T\\mathbf{C}^{-1}\\mathbf{R}$, where $\\mathbf{R}$ maps sky power to pulsar-pair correlations and $\\mathbf{C}$ is the covariance matrix of those correlations, including, in the pair-covariant per-frequency optimal statistic, the covariance induced by an isotropic gravitational wave background. The sensitivity measure is $\\sqrt{\\mathrm{diag}(\\mathbf{M})}$, i.e. the reciprocal of the diagonal-approximated uncertainty $\\sqrt{1/\\mathrm{diag}(\\mathbf{M})}$, which deliberately neglects covariances between pixels or spherical harmonics. The angular-scale scaling is driven by the effective number of pulsars that fall inside the characteristic beam of each spherical harmonic, which is why the exponent depends on orientation $m$ and on sky location. The multi-resolution basis replaces the single-resolution HEALPix normalization $1/N_\\mathrm{pix}$ with $\\Delta\\Omega_k/4\\pi$, so pixel area can vary across the sky and resolution can be spent where pulsar density is high.","core_discovery":"The paper's central claim is that, for cross-correlation-based anisotropy searches, PTA sensitivity—quantified by $\\sqrt{\\mathrm{diag}(\\mathbf{M})}$, the reciprocal of the diagonal-approximated per-pixel uncertainty—follows empirical power laws: $\\sqrt{\\mathrm{diag}(\\mathbf{M})}\\propto N_\\mathrm{psr}^{0.8\\pm0.1}$, $\\delta t^{-0.08\\pm0.01}$, and $f_\\mathrm{GW}^{-0.12\\pm0.04}$, with spherical-harmonic sensitivity scaling as $\\Delta\\Omega^{1.58\\pm0.03}$ for angular scales corresponding to $8\\le\\ell\\le20$ at $m=0$ and roughly $\\Delta\\Omega^{1.5}$ to $\\Delta\\Omega^{2.1}$ depending on orientation. The paper also claims that increasing observation duration from 16 to 30 years has negligible median effect on anisotropy sensitivity except in the lowest frequency bins, because the dominant growth there comes from the number of pulsars whose baselines are long enough to reach those bins. Finally, it claims that a multi-resolution HEALPix pixel basis, with response matrix $R_{ab}^k=(\\Delta\\Omega_k/4\\pi)R_{ab}(\\hat{\\Omega}_k)$, recovers injected isotropic and anisotropic backgrounds to numerical precision in analytic tests, allowing resolution to be concentrated where pulsar density is high.","pith_inferences":["Because the paper's sensitivity measure drops off-diagonal Fisher terms, and because the paper itself reports condition numbers $\\sim10^{14}$ to $10^{18}$, a direct follow-up would be to repeat the scaling-law fits with a regularized full inverse Fisher matrix; the absolute sensitivities and possibly the fitted exponents could shift.","The factor-of-2–3 sky-location dependence suggests a concrete design rule: new pulsars placed in low-density sky regions improve angular resolution at that location more than anywhere else, so sky coverage and pulsar count should be optimized jointly rather than separately.","The multi-resolution basis's ability to trade localization for uncertainty points toward a data-driven mesh-selection scheme in which pixel sizes are chosen from the local sensitivity map, extending the paper's planned reversible-jump MCMC approach."],"forward_implications":["Adding pulsars is the strongest lever: doubling $N_\\mathrm{psr}$ raises anisotropy sensitivity by roughly $2^{0.8}\\approx1.7$.","Reducing per-pulse TOA errors by a factor of 10 improves anisotropy sensitivity by only about 20%, so spending on more pulsars usually beats spending on pushing $\\delta t$ down.","Extending from 16 to 30 years without adding pulsars gives negligible median sensitivity gain except at the lowest frequency bins, where pulsars whose baselines have matured first enter the analysis.","Sensitivity to an anisotropy of angular scale $\\Delta\\Omega$ grows roughly as $\\Delta\\Omega^{1.6}$ to $\\Delta\\Omega^{2.1}$, so smaller-scale features are substantially harder to detect and detections will be sky-position dependent by a factor of 2–3.","At the lowest frequencies, neglecting the covariance between pulsar-pair correlations overestimates sensitivity by about a factor of two, so forecasts built on the covariance-free approximation are optimistic there."],"supporting_citations":[{"why":"Supplies the per-frequency optimal statistic and the pulsar-pair covariance matrix $\\mathbf{C}$ that defines the Fisher information matrix used throughout.","marker":"K. A. Gersbach et al. 2025"},{"why":"Provides the earlier cross-correlation-based anisotropy search framework that this paper's Fisher construction extends.","marker":"N. Pol et al. 2022"},{"why":"Defines the Hellings-Downs correlation pattern that anchors the isotropic null hypothesis and the response construction.","marker":"R. W. Hellings & G. S. Downs (1983)"},{"why":"Gives the optimal-statistic SNR scaling laws in pulsar number and TOA uncertainty that this paper compares its fitted exponents against.","marker":"X. Siemens et al. (2013)"},{"why":"Supplies the isotropic sensitivity-curve formalism used to validate the map-averaged anisotropy sensitivity curves.","marker":"J. S. Hazboun et al. (2019)"},{"why":"Provides the spherical-harmonic basis formulation for PTA anisotropy searches that the harmonic-basis analysis uses.","marker":"S. R. Taylor & J. R. Gair (2013)"},{"why":"Derives expected angular power spectra and the spherical-harmonic response used for decomposing the background power.","marker":"C. M. F. Mingarelli et al. (2013)"},{"why":"Introduces the multi-resolution HEALPix tessellation concepts that the new pixel basis adapts.","marker":"L. P. Singer & L. R. Price (2016)"},{"why":"Provides the multi-resolution HEALPix software implementation used to build and test the new basis.","marker":"I. Martinez-Castellanos et al. (2024)"}],"fun_headline_variants":["PTA sky-map sensitivity scales as N_psr^0.8, not time","Longer runs don't boost PTA anisotropy maps, more pulsars do","Multi-resolution pixels sharpen PTA gravitational-wave sky maps","Pulsar count, not observation time, drives anisotropy sensitivity"],"cache_read_input_tokens":25856,"weakest_assumption_plain":"The load-bearing premise is that a pulsar array's sensitivity to anisotropy can be quoted from the diagonal of its Fisher information matrix alone, ignoring correlations between neighboring directions on the sky; the paper itself reports near-singular matrices from those correlations, so including them could lower the absolute sensitivities and shift the fitted scaling exponents.","fun_headline_variants_meta":{"raw":{"variants":["PTA sky-map sensitivity scales as N_psr^0.8, not time","Longer runs don't boost PTA anisotropy maps, more pulsars do","Multi-resolution pixels sharpen PTA gravitational-wave sky maps","Pulsar count, not observation time, drives anisotropy sensitivity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000201,"raw_usage":{"total_tokens":1440,"prompt_tokens":1071,"completion_tokens":369,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":687,"completion_tokens_details":{"reasoning_tokens":291}},"tokens_in":687,"tokens_out":369,"duration_ms":3970,"temperature":1.0,"reasoning_tokens":291,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:53:46.081029+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the scaling-law fits for $N_\\mathrm{psr}$, $\\delta t$, and $\\Delta\\Omega$ using the full inverse Fisher matrix $\\mathbf{M}^{-1}$ (with SVD regularization, since the paper reports condition numbers $\\sim10^{14}$ to $10^{18}$) instead of diagonal-only inversion; if the fitted exponents move by more than the quoted uncertainties, the reported scaling relations are artifacts of the diagonal approximation.","supporting_citations":[{"cited_title":"P., Burns, E., et al","cited_arxiv_id":null,"evidence_quote":"Provides the multi-resolution HEALPix software implementation used to build and test the new basis."}],"review_version":1}