{"id":"e3876af4-f6be-47ec-bec3-2eed7ef3bce7","arxiv_id":"2411.13648","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A fourth-order moment extension of the FPFS shear estimator achieves sub-percent bias accuracy and, when combined with the second-order estimator, reduces shape noise by about 35 percent for isolated galaxies in simulations.","lead":"This paper describes a new method for measuring the tiny image distortions (shear) caused by weak gravitational lensing, combining standard ellipticities with higher-order shape features of galaxies. In simulations, combining the two estimators reduces the shape noise that limits cosmic shear measurements by about 35 percent for isolated galaxies.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The renoising noise-bias correction (Eq. 20) is extended to fourth-order moments without a dedicated proof or isolated validation; if the 90-degree-rotated noise layer does not cancel spin-2 anisotropies in the M42 response, the |m|<3e-3 accuracy claim is not yet established.","rationale":"The paper's central claim has two parts: the accuracy of the fourth-order estimator and the precision gain from combining it with the second-order estimator. I focus on the accuracy part because it is the foundation: if the noise-bias correction is biased, both the multiplicative-bias result and the shape-noise reduction inherit that bias. The reader's weakest assumption identifies the same step, and I agree that it is the most load-bearing unverified component. The in-sample tuning of C(2), C(4), and mu, and the 30% versus 35% discrepancy, are real concerns, but they affect the quantitative size of the precision improvement rather than the existence of fourth-order shear information. I do not claim the renoising method fails; rather, the paper neither proves its applicability to fourth-order moments nor isolates it with an independent noise-bias correction. The large simulation suites and public code are genuine strengths, and the end-to-end accuracy tests provide supporting evidence, but a dedicated cross-check is needed to close the gap. Since this matches the condition the reader already attached to acceptance, the verdict remains conditional.","tokens_in":19516,"tokens_out":12871,"duration_ms":983262,"concrete_test":"On a subset of the isolated HSC-like simulations (e.g., 500 of the 4000 subfields), re-measure the fourth-order multiplicative bias m1 using the Hessian-based noise-bias correction of Li et al. (2024b) in place of the renoising correction of Eq. (20), keeping detection, selection, and C(4)=10 identical. If the difference between the two m1 values exceeds the statistical error on that subset, the renoising cancellation is incomplete for fourth-order moments and the |m|<3e-3 claim requires revision. As a complementary check, for a small set of noiseless galaxy images, average the renoised estimator over many independent noise realizations and verify that it converges to the noiseless estimate within the target 10^-3 tolerance.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is the renoising noise-bias correction (Section 2.5, Eq. 20) applied to the fourth-order ellipticity e = M42/(M00+C). The method adds a 90-degree-rotated noise layer and relies on cancellation of spin-2 anisotropies after PSF deconvolution. For the fourth-order estimator, the shear response of M42 couples to M20, M60, and the spin-4 mode M64 (Eqs. 17-18), and the noise covariance of these modes does not transform trivially under a 90-degree rotation when the PSF is anisotropic. The paper adopts the renoising correction by reference to Li et al. (2024a) and reports no derivation or dedicated numerical check specifically for fourth-order moments. Because the headline |m|<3x10^-3 result is obtained using this same correction, an incomplete cancellation would directly bias the central accuracy claim. The end-to-end simulations do provide a real test of the correction, but they cannot separate a residual renoising bias from other modeling choices unless an independent noise-bias correction is applied for comparison.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper extends the FPFS/AnaCal shear estimator to use fourth-order polar shapelet moments (M42) to define a spin-2 ellipticity, with analytic corrections for detection/selection bias and noise bias via the renoising method of Li et al. (2024a). Using HSC-like and LSST-like image simulations of isolated and blended galaxies, the authors report multiplicative shear bias |m| below 3e-3 at the claimed 99.7% confidence and additive biases consistent with zero, while combining the second- and fourth-order estimators reduces shape noise by about 35% for isolated galaxies and by only about 2% in effective number density for blended galaxies.","tokens_in":19714,"tokens_out":6711,"duration_ms":60142,"significance":"If the claims hold, the paper demonstrates that fourth-order moments carry usable complementary shear information that can be calibrated analytically, improving precision by roughly 30-35% in isolated-galaxy samples without exceeding LSST multiplicative-bias requirements. The work is backed by extensive simulation tests (4000 isolated and 5000 blended subfields, ring tests, two survey setups) and public code, which are notable strengths. The main reservations are that the renoising noise-bias correction is applied to higher-order moments without a dedicated derivation or isolated validation, and that the headline shape-noise improvement is measured after in-sample parameter optimization, so the reported gain may be optimistic.","major_comments":[{"comment":"The renoising noise-bias correction is adopted by reference to Li et al. (2024a), but the paper does not derive or validate it for the fourth-order ellipticity e = M42/(M00+C(4)). The shear response of M42 couples to M20, M60, and M64 (Eqs. 17-18), and the effect of the 90-degree-rotated added noise layer on the covariance of these modes after PSF deconvolution is not demonstrated in the manuscript. Because the headline |m|<3e-3 result in Sections 4.1.4 and 4.2.2 is obtained with this correction, the end-to-end simulations cannot separate a residual renoising bias from other modeling choices. Please provide a dedicated test isolating the renoising correction for fourth-order moments, for example by comparing Eq. (20) with an independent noise-bias correction (Hessian-based AnaCal or a numerical method) on identical simulations, or an explicit analytic check of the spin-2 cancellation for the M42 response.","section":"Section 2.5, Eq. (20)"},{"comment":"The values C(2), C(4), and the combination weight mu are selected by minimizing the measured variance on the same 100-subfield simulations that are then used to report the shape-noise reduction. This in-sample optimization makes the reported 30-35% improvement an optimistic estimate, and no cross-validation or bootstrap uncertainty on the gain is given. Please report the improvement with a split-sample or cross-validated procedure, or at minimum provide an uncertainty on the variance reduction.","section":"Section 4.1.1, Fig. 3 and Eqs. (22)-(24)"},{"comment":"The paper quotes a \"~30%\" reduction in Section 4.1.1 and Section 4.1.2, while the abstract and conclusion quote \"~35%\". If these numbers refer to different sample definitions or different comparison baselines (e.g., with or without detection/selection cuts, or versus reGauss), that should be stated explicitly; as written, the headline improvement is internally inconsistent.","section":"Section 4.1.1 and Abstract/Conclusion"}],"minor_comments":[{"comment":"The phrase \"the fourth-order shear estimator, which is independent of the second-order\" overstates the case; Eq. (23) explicitly includes the covariance rho between the two estimators, so they are correlated. Suggest replacing \"independent\" with \"complementary\" or \"partially independent\".","section":"Section 1"},{"comment":"The sentence beginning \"In this work, use the same detection and selection\" is missing the subject \"we\"; please correct the grammar.","section":"Section 2.4"},{"comment":"The claim that \"A 30 per cent reduction in shape noise is equivalent to the increase in sample size that would be achieved by expanding the survey area by 70 per cent\" appears arithmetically inconsistent with the usual scaling sigma proportional to 1/sqrt(area); please check whether the reduction refers to sigma or to the variance and correct the area-equivalence statement.","section":"Section 4.1.1"},{"comment":"The sentence \"The n_eff ~ 15 arcmin^-2 for HSC setup is 35% smaller than the n_eff ~ 20 arcmin^-2\" is inconsistent: 15 is 25% smaller than 20. Please correct the percentage or the quoted values.","section":"Section 4.2.1"},{"comment":"The statement that |m| is below 3e-3 at the 99.7% confidence interval would be clearer if the authors specified whether this is a one-sided upper limit or a two-sided interval, and how the confidence bound is derived from the displayed 1-sigma and 3-sigma error bars.","section":"Section 4.1.4"}],"recommendation":"major_revision","confidential_remarks":"The paper is a solid extension of the FPFS/AnaCal program with extensive simulation tests and publicly available code. My main reservations are (i) the renoising correction is applied to fourth-order moments without dedicated validation, and (ii) the headline shape-noise reduction is computed in-sample. Both are addressable with additional analysis. The inconsistency between the 30% and 35% improvement figures should also be resolved. I would support publication after a major revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper does something genuinely new: it builds a fourth-order shapelet shear estimator within the FPFS/AnaCal framework, derives its shear response, corrects detection/selection and noise bias analytically, and combines it with the second-order estimator. The bias testing is extensive: 4000-5000 subfields, ring tests, HSC and LSST configurations, and the fourth-order estimator keeps |m| below 3e-3 within 3 sigma for both isolated and blended galaxies. That matches the LSST requirement and is the main accuracy claim. The code is public, which is a plus.\n\nThe precision gain is plausible and important: for isolated galaxies, combining second and fourth order cuts the shear uncertainty by roughly a third. But the paper quotes ~30% in Section 4.1.1 and ~35% in the abstract and conclusion. That needs to be reconciled. More substantively, the gain is measured on 100 subfields, and the parameters C(2), C(4), and mu are chosen from those same subfields. There is no cross-validation or error bar on the improvement. This is the softest spot. It does not sink the claim, because the bias tests use separate subfields and the improvement is consistent across seeing and noise levels in Fig. 6, but a referee should ask for an independent check.\n\nThe renoising noise-bias correction is imported from Li et al. 2024a and applied to fourth-order moments without a dedicated derivation. The underlying method is general for spin-2 observables, and the end-to-end simulations provide a real test, so I do not think this is a load-bearing flaw. But a sentence explaining why the 90-degree rotation cancels spin-2 anisotropies for M42 specifically would close the gap.\n\nThe paper is honest about limitations: blending only gives ~2% improvement, and the homogeneous-noise assumption is flagged as potentially inadequate for Stage IV surveys. That is the right level of caution.\n\nWho is this for? People working on shear estimation and analytic calibration. It deserves a serious referee. I would recommend acceptance after a minor revision that reconciles the 30/35 numbers and cross-validates the precision gain.","headline":"A solid extension of FPFS to fourth-order moments with credible bias tests; the headline shape-noise gain is real but quoted inconsistently and measured in-sample.","tokens_in":20302,"tokens_out":2200,"would_cite":true,"duration_ms":21528,"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":"The paper shows that fourth-order galaxy shape moments, corrected analytically, reach shear bias below 0.003 and cut shape noise by about 35% when combined with the standard estimator.","keywords":["weak gravitational lensing","cosmic shear","shapelet moments","shear estimation","analytic calibration","noise bias","shape noise","galaxy blending"],"falsifier":"A decisive test would be to run the fourth-order estimator on simulated images whose noise has anisotropic pixel-to-pixel correlations and check whether the residual multiplicative and additive bias stays within $|m|<3\\times10^{-3}$ after renoising; if the added rotated noise layer leaves a nonzero spin-2 correlation after PSF deconvolution, the cancellation assumed by Eq. (20) is incomplete and the analytic correction is biased for higher-order moments.","tokens_in":19249,"feed_emoji":"🔭","tokens_out":12939,"duration_ms":123828,"temperature":0.7,"pith_summary":"Weak lensing shear is usually measured from the second-order moments of galaxy light, but this paper asks whether fourth-order moments contain additional, usable information about the shear. Using the FPFS shapelet framework and the AnaCal analytic calibration, the authors construct a fourth-order ellipticity from the $M_{42}$ moment and correct its detection, selection, and noise biases without external simulation-based calibration. In HSC- and LSST-like image simulations, the corrected estimator has multiplicative bias $|m|$ below $3\\times10^{-3}$ at the $3\\sigma$ level for both isolated and blended galaxies. When combined with the second-order estimator, it reduces per-galaxy shape noise by about 35% for isolated galaxies, roughly equivalent to the precision gain from a 70% larger survey area. If the claim holds, higher-order moments become a practical resource for tightening cosmic shear constraints while staying inside the LSST systematic error budget.","feed_headline":"Fourth-order galaxy moments cut shear shape noise by ~35 percent.","feed_subtitle":"Added fourth-order shapelet moments keep multiplicative bias below 0.003 in simulated galaxies.","key_machinery":"The central object is the fourth-order spin-2 polar shapelet moment $M_{42}$ and its linear shear response, which couples $M_{42}$ to $M_{20}$, $M_{60}$, and $M_{64}$ (Eqs. 17–18). Dividing by $M_{00}+C^{(4)}$ turns this mode into a dimensionless fourth-order ellipticity, and the smooth selection weights of the FPFS/AnaCal framework give its detection and selection responses analytically. Noise bias is removed by the renoising procedure: an extra noise layer with the same statistics, rotated by 90 degrees, is added to the image so that the spin-2 anisotropies cancel after PSF deconvolution, making the estimator noise-bias-free to second order in shear without computing noisy high-order derivatives. Finally, the second- and fourth-order estimators are combined with the variance-minimizing weight $\\mu$ derived from their covariance, so that independent information from each order is used where it is most constraining.","core_discovery":"The central claim is that fourth-order polar shapelet moments carry shear information that is partly independent of the second-order moments, and that this information can be extracted at sub-percent accuracy through the analytic AnaCal correction scheme. The shear estimator built from the normalized fourth-order spin-2 moment $M_{42}/(M_{00}+C^{(4)})$ has multiplicative bias $|m| < 3\\times10^{-3}$ with $99.7\\%$ confidence after analytic correction for detection, selection, and noise bias, in isolated and blended galaxy simulations under HSC and LSST observing conditions. Combining this fourth-order estimator with the existing second-order estimator through a variance-minimizing weight $\\mu$ reduces per-galaxy shape noise by roughly 35% for isolated galaxies, matching the precision gain of about a 70% larger survey area, while for blended galaxies the gain is only about 2% because the fourth-order moment amplifies the effects of blending and image noise. The paper therefore claims that the two estimators are complementary rather than redundant in high-SNR, isolated, or space-based regimes.","pith_inferences":["Editorial inference: if the ~35% isolated-galaxy gain carries over to space-based surveys where blending is negligible, the fourth-order combination could raise effective number density by tens of percent, a testable prediction for Euclid- or Roman-like simulations.","Editorial inference: the optimal weighting parameters $C^{(2)}=7$ and $C^{(4)}=10$ were chosen on the same simulations used for validation, so an out-of-sample optimization would measure how much of the reported precision gain is tuning rather than information.","Editorial inference: the fourth-order moment's distinct sensitivity to small radii could make the ratio of second- to fourth-order shear estimates a practical diagnostic for PSF modeling error; the paper recommends but does not perform this test.","Editorial inference: a direct extension would inject correlated anisotropic noise into the renoising step and verify that the 90-degree rotation still cancels spin-2 correlations for fourth-order modes; this would distinguish the method's generic validity from the homogeneous-noise approximation used here."],"forward_implications":["For isolated, high-SNR galaxy samples, combining second- and fourth-order estimators reduces shape noise by about 35%, equivalent to increasing survey area by roughly 70% at fixed statistical precision.","The fourth-order estimator alone satisfies the LSST ten-year multiplicative-bias requirement $|m|<3\\times10^{-3}$ in HSC-like and LSST-like simulations for both isolated and blended galaxies.","Because the two estimators weight different spatial scales, their combination provides an internal cross-check that can expose PSF leakage and modeling errors that affect one order more than the other.","In blended ground-based images the fourth-order estimator contributes only about 2% to effective number density, concentrating the practical gain in high-SNR, isolated, or space-based observations.","The same analytic calibration machinery can be applied to further moment orders or to redshift-dependent shear without rerunning external image-calibration simulations."],"supporting_citations":[{"why":"Defines the second-order FPFS shear estimator, detection modes, and smooth selection weights that this work extends to fourth-order moments.","marker":"Li & Mandelbaum 2023"},{"why":"Supplies the analytical renoising noise-bias correction that makes the fourth-order estimator bias-free without noisy derivative computations.","marker":"Li et al. 2024a"},{"why":"Establishes the AnaCal framework's analytic shear calibration and the blended-galaxy test setup reused here.","marker":"Li et al. 2024b"},{"why":"Provides the original idea of adding a noise layer with 90-degree-rotated statistics to cancel noise bias.","marker":"Sheldon & Huff 2017"},{"why":"Provides the descwl-shear-sims blended galaxy simulations and the benchmark used for the blended accuracy tests.","marker":"Sheldon et al. 2023"},{"why":"Supplies the polar shapelet formalism and the shear responses of shapelet modes used to write the fourth-order shear-response equations.","marker":"Massey & Refregier 2005"},{"why":"Introduces the FPFS shapelet-mode construction and the normalization scheme used in the ellipticity definition.","marker":"Li et al. 2018"},{"why":"Supplies the COSMOS HST parametric galaxy catalog used to generate the isolated galaxy image simulations.","marker":"Mandelbaum et al. 2019"},{"why":"Defines the ten-year LSST systematic error requirement on multiplicative shear bias that the paper uses as its accuracy target.","marker":"The LSST Dark Energy Science Collaboration et al. 2018"}],"fun_headline_variants":["Fourth-order moments slash shear shape noise by 35%","Higher-order shapelets: shear bias below 0.003, noise cut 35%","Fourth-order moments: shear noise down 35%, bias sub-0.003","Shape noise cut 35% with fourth-order shear estimator, bias <0.003","Sub-percent bias, 35% less noise: fourth-order shear wins"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The result rests on the renoising noise-bias correction: the added 90-degree-rotated noise layer must cancel the relevant spin-2 anisotropies after PSF deconvolution for fourth-order moments; if that cancellation is incomplete, the analytic corrections are biased and the $|m|<3\\times10^{-3}$ claim fails.","fun_headline_variants_meta":{"raw":{"variants":["Fourth-order moments slash shear shape noise by 35%","Higher-order shapelets: shear bias below 0.003, noise cut 35%","Fourth-order moments: shear noise down 35%, bias sub-0.003","Shape noise cut 35% with fourth-order shear estimator, bias <0.003","Sub-percent bias, 35% less noise: fourth-order shear wins"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00101,"raw_usage":{"total_tokens":4295,"prompt_tokens":998,"completion_tokens":3297,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":614,"completion_tokens_details":{"reasoning_tokens":3208}},"tokens_in":614,"tokens_out":3297,"duration_ms":25018,"temperature":1.0,"reasoning_tokens":3208,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T16:01:40.181373+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A decisive test would be to run the fourth-order estimator on simulated images whose noise has anisotropic pixel-to-pixel correlations and check whether the residual multiplicative and additive bias stays within $|m|<3\\times10^{-3}$ after renoising; if the added rotated noise layer leaves a nonzero spin-2 correlation after PSF deconvolution, the cancellation assumed by Eq. (20) is incomplete and the analytic correction is biased for higher-order moments.","supporting_citations":[],"review_version":1}