{"id":"9e3d88f2-f91d-4dd1-956e-daff51ccfdeb","arxiv_id":"2509.08675","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Hyperuniform continuous random fields with tunable anisotropy can be generated in a single FFT pass by prescribing a superellipse spectral mask.","lead":"This paper generates 2D and 3D random fields with tuned hyperuniform spectra by filtering white noise in Fourier space with superellipse-shaped masks. The method is single-pass and FFT-based, so it is far faster than iterative reconstruction, though the speedup claim lacks a direct benchmark.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Sec. II B's derivation is internally inconsistent: i.i.d. complex Fourier weights do not guarantee a real field; Hermitian symmetry is required but not stated.","rationale":"The reader's weakest assumption (unsupported speedup baseline) is valid, but I see it as secondary: even a slow method that correctly generates real hyperuniform fields would be useful, whereas a method that cannot generate real fields as described fails the core objective. The realness error is an internal inconsistency in the derivation (Sec. II B), not a disagreement with external consensus. It is easily fixed by specifying Hermitian symmetry or by generating real-space white noise, so conditional acceptance remains appropriate: the authors must clarify the construction and correct the false inference. I therefore set verdict_should_be=CONDITIONAL and mark agreement_with_reader=disagree.","tokens_in":12627,"tokens_out":8909,"duration_ms":88109,"concrete_test":"Implement Eqs. (8)-(9) exactly as written: generate independent complex Gaussian variables F_wn(k) for every k on a 1024^2 reciprocal lattice, set A(k)=sqrt(χ(k)) from Eq. (13) with α=2, apply the inverse FFT, and compute max|Im(f)|. If nonzero, the claim 'f is strictly real valued' is false under the stated algorithm. Then verify that enforcing F(-k)=conj(F(k)) yields a real field whose empirical power spectrum matches χ(k), confirming the necessary fix.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"In Sec. II B, the authors define {F_wn(k)} as i.i.d. complex Gaussian variables on the full reciprocal lattice and claim 'Because χ_K is real and symmetric, f is strictly real valued.' This inference is false. For the inverse FFT f(x) to be real, the Fourier coefficients must satisfy F_filt(-k) = conj(F_filt(k)). The symmetry of χ(k) alone implies nothing about the phases of F_wn; with independent weights, F_filt(-k) and F_filt(k) are independent, so f is generically complex. The subsequent steps (variance normalization, thresholding, power-spectrum estimation) all assume a real scalar field. If a reader implements Eqs. (8)-(9) literally, the output is complex; the desired real field requires an additional Hermitian-symmetry constraint (or taking the real part), which is not part of the stated algorithm. Moreover, Sec. II B later mentions 'two FFTs (one forward on white noise and one inverse on the filtered spectrum)', which implies real-space white noise and hence Hermitian symmetry—contradicting the earlier i.i.d. complex Gaussian description. This inconsistency is load-bearing because the central claim is reconstruction of real continuous random fields with the prescribed spectrum.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a single-shot FFT-based method for generating Gaussian random fields with prescribed superellipse-shaped power spectral densities, targeting continuous and thresholded two-phase hyperuniform microstructures in 2D and 3D. The spectral density is defined analytically in Eq. (13) with parameters α (low-wavenumber exponent), σ (bandwidth), p (superellipse shape), and a/b (anisotropy). After filtering complex white noise by the square root of this mask and inverse-Fourier-transforming, the fields are demeaned and rescaled; thresholding is then explored as a route to binary morphologies. The authors report O(N log N) complexity, absolute generation times of about 0.03 s for 1024² and 1.3 s for 256³ grids, and claim orders-of-magnitude speedups over iterative reconstruction methods.","tokens_in":12979,"tokens_out":6716,"duration_ms":74117,"significance":"If validated, the proposed spectral mask family is practically useful: it provides explicit analytic control over hyperuniform exponent, angular shell shape, and anisotropy in a non-iterative generation pipeline, with straightforward extension to 3D. The parameterization in Eqs. (12)-(13) is transparent and easily reimplemented, which is a genuine strength. However, the core spectral-representation technique is standard, and the paper's added value rests on the mask design and on the validation and performance claims. At present, the hyperuniform-class verification is not independent of the construction, and the speedup comparison is unsupported by any measured baseline. These issues substantially affect the strength of the central claims.","major_comments":[{"comment":"The assertion that 'Because χ_K is real and symmetric, f is strictly real valued' is incorrect. For the inverse discrete Fourier transform in Eq. (9) to produce a real field, the Fourier coefficients must obey F_filt(-k) = conj(F_filt(k)). Under the stated i.i.d. complex Gaussian definition of F_wn(k), F_filt(k) and F_filt(-k) are independent, so f is generically complex. The later statement that the pipeline uses 'two FFTs (one forward on white noise and one inverse on the filtered spectrum)' implies real-space white noise and hence Hermitian symmetry, contradicting the earlier definition. This is load-bearing because variance rescaling, thresholding, and all spectral analyses in Sec. III presuppose real scalar fields. Please specify Hermitian-symmetric weights explicitly (e.g., start from real-space white noise and FFT it, or enforce conjugate symmetry).","section":"Sec. II B, Eqs. (8)-(9)"},{"comment":"The hyperuniform-class claims are not independently verified. Equation (13) prescribes the target spectrum and Eq. (8) multiplies white noise by sqrt(χ), so the empirical power spectra displayed in Figs. 1-5 necessarily reproduce the input mask; the 'confirmation' of spectral shell shapes is tautological. The paper never computes the real-space variance σ_F^2(R) or its scaling per Eqs. (3) and (6), nor any independent estimate of the low-k exponent. This matters because Eq. (14) is a continuous-space asymptotic statement; on a finite periodic grid, especially for large α (e.g., α=50,100), the low-k region contains very few modes and the realized class is not guaranteed by the mask formula alone. Please add real-space variance-scaling tests for representative (α,p,a/b) combinations, or otherwise provide a non-tautological validation of the hyperuniform class.","section":"Sec. III A/III B/IV, Figs. 1-5"},{"comment":"The claimed 'orders-of-magnitude speedup compared to existing approaches' is not supported by any measured baseline. The text reports absolute timings (≈0.03 s for a 1024² grid, ≈1.3 s for a 256³ grid) but gives no runtime for the methods it claims to outperform, nor does it state the hardware/implementation details. Since refs. [42,43] already describe non-iterative spectral-shaping methods, the relevant baseline is unclear. Please report benchmark timings for at least one representative iterative method (e.g., simulated annealing) and, ideally, for fast spectral methods; otherwise revise the speedup claim to describe the absolute cost of the single-shot FFT pipeline.","section":"Sec. III A and IV"}],"minor_comments":[{"comment":"For anisotropic masks (a≠b), the small-wavenumber expansion is χ(k) ~ C |k|^α multiplied by a direction-dependent factor (e.g., (1/a)^α along kx and (1/b)^α along ky). The exponent α is unchanged, so the hyperuniform class is unaffected, but Eq. (14) as written should be qualified.","section":"Eq. (14)"},{"comment":"Refs. [42,43] are cited among representation schemes but are not discussed as prior fast, non-iterative spectral-shaping methods. A brief comparison would strengthen the positioning of the contribution and clarify what is genuinely new.","section":"Introduction and Related Work"},{"comment":"The statement 'The codes and data are available upon request' is weak for a methods paper. Consider providing a public repository with the generation code and parameters used for the figures.","section":"Data Availability"},{"comment":"The paper uses 'reconstruction' for what is essentially forward generation from a prescribed spectrum. If the method is intended for microstructure reconstruction from correlation data, the connection should be made explicit; otherwise 'generation' avoids overclaiming.","section":"Terminology"}],"recommendation":"major_revision","confidential_remarks":"The core idea is simple and likely correct once the Hermitian-symmetry implementation is clarified; the main risks are insufficient validation of the hyperuniform-class claims and an unsupported speedup comparison. For this journal, the novelty may be viewed as incremental if the authors do not benchmark against existing fast spectral-shaping methods (refs. [42,43]) and release code. I would not reject on the current evidence, but these load-bearing points require substantive revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new thing here is the superellipse spectral mask in Eqs. (12)-(13). It gives you a clean, low-dimensional parameterization that separates hyperuniform exponent, bandwidth, angular shape, and anisotropy, and it plugs directly into the classical spectral-representation method. That is a legitimate, practically useful extension, and the 2D/3D demonstrations do show the expected qualitative morphologies. The pipeline is correctly O(N log N) for Gaussian fields if implemented properly, and the exposition is mostly clear. Credit where due: the mask family is new, and the paper is honest about thresholding not preserving exact hyperuniformity.\n\nNow the soft spots. First, the stress-test note is right: Sec. II B states that i.i.d. complex Gaussian Fourier weights give a strictly real field because the mask is real and symmetric. That is false. Realness requires Hermitian symmetry of the filtered coefficients, which i.i.d. complex weights do not satisfy. The later mention of two FFTs (one forward on real-space white noise) implicitly fixes this, but the stated algorithm in Eqs. (8)-(9) is inconsistent. This is load-bearing for the central claim, though easily repaired by specifying Hermitian-symmetric weights.\n\nSecond, the 'orders-of-magnitude speedup' headline has no baseline. They report 0.03 s for 1024^2 and 1.3 s for 256^3, but no runtime for Yeong-Torquato or the already-fast spectral shaping methods in refs. [42,43]. Since the paper itself cites those fast methods, the speedup claim needs a direct comparison or it should be dropped.\n\nThird, no code or data are released despite 'available upon request,' and stochastic results are shown without error bars. For a generative method, a single realization per parameter set tells you about visual trends, not statistical reproducibility. These are addressable, not fatal, but they matter for a paper that sells itself as a practical tool.\n\nThe reader's conditional verdict is about right. The math is sound once the Hermitian symmetry is fixed, and the tool is likely useful for people who need quick, tunable hyperuniform continuous fields for large-scale simulations or microstructure-design pipelines. It does not resolve an open physical question, but it is a solid methods contribution.\n\nRecommendation: this deserves refereeing, not a desk reject. The Hermitian slip must be fixed, the speedup benchmarked against the right baseline, and ideally code released. If those are done, it is a reasonable publication.\n\nGiven the gaps, I wouldn't cite it in my own work until the code and benchmarks appear, but I would bring it to a reading group to discuss the mask parameterization.","headline":"Useful parameterized mask family for spectral filtering, but the Hermitian-symmetry slip, unbenchmarked speedup, and missing code make it a conditional accept at best.","tokens_in":13392,"tokens_out":1440,"would_cite":false,"duration_ms":18246,"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":"One analytic mask generates 2D and 3D hyperuniform random fields in two FFT passes.","keywords":["hyperuniformity","spectral density","random fields","spectral filtering","superellipse","microstructure reconstruction","anisotropy","fast Fourier transform"],"falsifier":"Measure wall-clock times for the paper's 1024^2 and 256^3 examples against a Yeong-Torquato decoder and against the non-iterative spectrally shaped disorder methods cited as [42,43]; the orders-of-magnitude runtime claim stands only if the measured gap is large. Also compute the spectral density at k=0 for thresholded maps: the paper predicts a finite value, not a vanishing one.","tokens_in":1378,"feed_emoji":"🌀","tokens_out":1688,"duration_ms":49967,"temperature":0.7,"pith_summary":"Hyperuniform random fields suppress long-wavelength fluctuations while keeping local disorder, a combination prized for photonic, thermal, and mechanical materials. This paper claims that such fields, isotropic or anisotropic, can be generated at scale by one analytic Fourier-domain mask rather than by iterative reconstruction. The mask is a generalized superellipse inserted into a log-Gaussian spectral density, letting four parameters independently set hyperuniformity class, radial bandwidth, angular shape, and anisotropy. The authors report 1024x1024 realizations in about 0.03 seconds and 256^3 volumes in about 1.3 seconds, and show that thresholding the continuous field yields binary microstructures whose morphology tracks the mask shape. If the claim holds, large-scale simulation and inverse design of disordered hyperuniform materials becomes a single-shot FFT operation.","feed_headline":"One mask makes 2D and 3D hyperuniform fields at FFT speed","feed_subtitle":"An analytic superellipse filter sets class, bandwidth, shape, and anisotropy in a single shot.","key_machinery":"The key object is the generalized superellipse norm K_{p,a,b}(k) = [(|k_x|/a)^p + (|k_y|/b)^p]^(1/p), which interpolates between diamond, circular, and square contours in Fourier space and carries the anisotropy ratio a/b. It is inserted into the analytic spectral density exp[alpha ln K - K^2/(2 sigma^2)], where alpha sets the power-law suppression near k = 0, sigma sets the band width, and p and a/b shape the angular envelope. This single expression replaces iterative matching of target spectral densities with a direct, parameterized mask.","core_discovery":"The central claim is that the spectral density defined by Eq. (13), built from the generalized superellipse norm of Eq. (12), completely prescribes a continuous hyperuniform random field's second-order statistics, so the field can be reconstructed by filtering Gaussian white noise in Fourier space and taking one inverse FFT. Because the mask is analytic, alpha controls the low-wavenumber power-law and hence the hyperuniformity class, sigma controls the radial width, the exponent p controls angular shape, and a/b controls anisotropy. The authors argue this single-shot pipeline reproduces spectral features that iterative schemes target, extends directly to 3D, and leaves the field Gaussian so","pith_inferences":["A natural extension is inverse design: since the mask is differentiable in alpha, sigma, p, and a/b, one could gradient-optimize these parameters against target effective properties or scattering responses.","The paper's speedup claim would be sharpened by a direct wall-clock comparison against the non-iterative spectral-shaping methods it cites, since those methods may be the relevant baseline rather than simulated annealing.","The thresholding step is identified as the point where exact hyperuniformity is lost; a future variant could replace simple sign thresholding with a hole-preserving binarization to keep the zero-wavenumber density near zero.","Because the fields are Gaussian by construction, higher-order correlation functions are constrained by the two-point spectrum, which may limit how far the generated microstructures can deviate from Gaussian statistics in real-space textures."],"forward_implications":["Four independent controls let users tune class, bandwidth, angular shape, and anisotropy without disturbing the other features.","Thresholding a generated continuous field to a +1/-1 binary map preserves the visual shell morphology but refills the zero-wavenumber density, so strict hyperuniformity is lost under binarization.","Because the whole pipeline is linear and FFT-based, additional functional maps can be applied to the raw field with no additional spectral filtering cost.","The same parameter trends observed in 2D carry over to 3D, giving a dimension-agnostic way to tailor pore architectures.","The analytic mask makes the reconstruction fast enough for high-resolution parametric sweeps over microstructure families."],"supporting_citations":[{"why":"Supplies the spectral representation method used to filter white-noise coefficients with a prescribed spectral density.","marker":"[99]"},{"why":"Provides the circulant-embedding and FFT technique for sampling stationary Gaussian fields on periodic grids.","marker":"[100]"},{"why":"Defines hyperuniform random fields, the variance-decay criterion, and the power-law classification that alpha is designed to control.","marker":"[61]"},{"why":"The Yeong-Torquato iterative reconstruction method is the main baseline that the single-shot pipeline claims to outperform.","marker":"[25]"},{"why":"An existing non-iterative spectral-shaping method that represents a close competing fast approach for runtime comparison.","marker":"[42]"},{"why":"Supports the statement that thresholding a hyperuniform continuous field generally disturbs exact hyperuniform scaling.","marker":"[101]"}],"fun_headline_variants":["Single FFT yields anisotropic hyperuniform fields in 2D and 3D","Analytic mask sets hyperuniform class and shape in one pass","One inverse FFT designs hyperuniform fields with full spectral control","Superellipse spectral filter makes hyperuniform fields at FFT speed"],"cache_read_input_tokens":15232,"weakest_assumption_plain":"The claimed orders-of-magnitude speedup assumes the proper comparison is to iterative decoders such as Yeong-Torquato, but the paper measures no runtime for any competing method, so if the real baseline is already a fast spectral-shaping method the speedup could be small.","fun_headline_variants_meta":{"raw":{"variants":["Single FFT yields anisotropic hyperuniform fields in 2D and 3D","Analytic mask sets hyperuniform class and shape in one pass","One inverse FFT designs hyperuniform fields with full spectral control","Superellipse spectral filter makes hyperuniform fields at FFT speed"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000543,"raw_usage":{"total_tokens":2427,"prompt_tokens":726,"completion_tokens":1701,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":470,"completion_tokens_details":{"reasoning_tokens":1625}},"tokens_in":470,"tokens_out":1701,"duration_ms":11954,"temperature":1.0,"reasoning_tokens":1625,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T20:16:06.600674+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure wall-clock times for the paper's 1024^2 and 256^3 examples against a Yeong-Torquato decoder and against the non-iterative spectrally shaped disorder methods cited as [42,43]; the orders-of-magnitude runtime claim stands only if the measured gap is large. Also compute the spectral density at k=0 for thresholded maps: the paper predicts a finite value, not a vanishing one.","supporting_citations":[],"review_version":1}