{"id":"d6ce7e2c-e28b-4870-929a-e7715c62872c","arxiv_id":"2607.20329","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For n×n Gaussian matrices over R, C, or H, the normalized permanent has density at most O(n^{(β+2)/4}), giving small-ball probability O(n^{(β+2)/4} ε^β) and settling the complex PACC.","lead":"This paper proves a sharp, dimension-dependent bound on how often a random Gaussian matrix's permanent lands close to any fixed value, resolving the Aaronson–Arkhipov permanent anticoncentration conjecture for complex matrices. The result extends to real and quaternionic Gaussian entries and to arbitrary independent perturbations of the matrix.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified","rationale":"The reader identified Proposition 3.5 as the load-bearing technical step; I agree that this is the crux. However, after close inspection, the proof of Proposition 3.5 is correct for all three fields, including the quaternionic case. The noncommutativity of H is handled because the permanental cofactor expansion is real-bilinear in the two distinguished columns, and the real Gaussian interpolation machinery does not require commutativity. The quaternionic lower bound works because the last-row entries appear as right-hand factors, giving independent centered real Gaussians with the stated variances. The induction in Theorem 3.6 is the standard Laplace-to-Fourier-to-Laplace cycle, and its base case is correct. The only notable imperfection is the mislabeling of (·,·)_R as the ordinary Euclidean inner product, but the subsequent arguments only use the fact that z ↦ Re(z w) ranges over all real linear functionals and that the Gaussian law is invariant under the associated orthogonal reflection; hence no substantive gap results. The density bound and the small-ball estimate follow as claimed. Since I cannot identify a load-bearing concern, I do not recommend changing the reader's ACCEPT verdict.","tokens_in":18576,"tokens_out":52078,"duration_ms":399816,"concrete_test":"Numerically verify Proposition 3.5 for K=H at n=3 and n=4: generate many samples of the permanental cofactor vector C (e.g., 10^5 samples), and for a random grid of z (e.g., 100 vectors with varied directions and norms), estimate φ_n(z) and φ_n(||z||e1) by Monte Carlo. Confirm that 0 ≤ φ_n(z) ≤ φ_n(||z||e1) holds within, say, three standard errors. This directly tests the one step on which the whole induction depends; a violation would signal a hidden flaw in the quaternionic two-coordinate compression.","verdict_should_be":"UNCHANGED","load_bearing_attack":"After a full pass over the argument, I do not find a load-bearing flaw. The central claim rests on Proposition 3.5 (Eq. 13), the Fourier coordinate compression for permanental cofactor vectors. I checked the quaternionic case in detail: the two-column expansion (14) is indeed real-bilinear in the two distinguished columns, so the real matrices T_R and L_R exist; the conditional Gaussian integrations and the interpolation Lemma 3.2 apply after rescaling by √β. The lower bound (nonnegativity) follows from conditioning on the last Gaussian row, and for H the row-ordered permanent still gives coefficients D_j such that Re(D_j h_j) are independent centered Gaussians. The induction in Theorem 3.6 is sound: scalar Laplace comparison at level n gives Fourier compression at level n+1, which via Lemma 3.4 yields the matrix Laplace comparison and then the next scalar comparison. The notation issue that (·,·)_R is called the Euclidean inner product is a harmless misstatement: the map z ↦ Re(z w) is bijective on the real dual space for each field, and the standard Gaussian is isotropic, so all Fourier/Laplace identities used remain valid. The density bound follows rigorously from the Laplace comparison through the approximate-identity limit in (25), and the final small-ball estimate is correct. I therefore see no significant objection to the central theorem.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves a Gaussian permanent anticoncentration theorem for the three Ginibre ensembles over R, C, and H. For a standard K-Ginibre matrix G_n^K with β=dim_R K, the normalized row-ordered permanent W_n^K=per_K G_n^K/√(n!) is shown to have a radial density p_n^K with sup norm O(n^{(β+2)/4}), yielding the uniform small-ball estimate sup_z P(|W_n^K-z|≤ε) = O(n^{(β+2)/4} ε^β). In the complex case this gives sup_z P(|per G_n-z|≤ε√(n!)) ≤ min{1, nε²}, resolving the Aaronson–Arkhipov Permanent Anticoncentration Conjecture and, via the perturbation result in Section 4, a conjecture of Bouland et al. The proof is built on a Laplace-transform comparison between the squared permanent and a gamma benchmark arising from the determinant (or Study determinant for H). The main technical innovation is a Fourier coordinate compression for permanental cofactor vectors (Proposition 3.5), which is fed into an induction (Theorem 3.6) that yields the matrix and scalar Laplace comparisons. The density bound follows from an approximate-identity argument using the explicit Laplace transform of the gamma benchmark.","tokens_in":18927,"tokens_out":46615,"duration_ms":362250,"significance":"If correct, this is a substantial resolution of a well-known open problem. The proof is explicit and structurally informative: it identifies the exact gamma benchmark, proves the comparison in Laplace-transform order, and then extracts a uniform density bound rather than only a centered lower-tail estimate. It also extends the result to real and quaternionic Ginibre ensembles and to arbitrary deterministic additive perturbations, giving a stronger form of the Bouland–Datta–Fefferman–Hernández conjecture. The paper ships no fitted parameters, and the main computations are checkable by hand. The appendices honestly delineate the limits of the method, showing that stronger stochastic/convex orders fail.","major_comments":[],"minor_comments":[{"comment":"The sentence that (·,·)_R is the ordinary Euclidean inner product on R^{βm} is false for K=C and K=H. For example, with K=C, Re(zw)=Re(z\\bar{w}) fails when w is not real. The subsequent Fourier and Laplace identities are still valid because the map z↦Re(z w) is a nondegenerate real bilinear form; after an orthogonal change of coordinates it becomes the standard real inner product. Please correct this statement and add a remark where Lemma 3.4 is applied in Theorem 3.6, so that the translation to the standard Fourier inversion formula is explicit.","section":"§1.2"},{"comment":"In the lower-bound paragraph, the expansion along the last Gaussian row is written as ∑_j D_j h_j. Since the last row is the final factor in the row-ordered permanent, the correct expression is ∑_j h_j D_j; the real parts coincide because Re(qr)=Re(rq), so the argument is unaffected. A one-sentence clarification would avoid confusion.","section":"§3.2, proof of Prop. 3.5"},{"comment":"The quaternionic case of Theorem 4.1 is deferred to Subsection C.2, but the proof of (29) for K∈{R,C} is written before the reader is told that the H case is handled later. Add a forward pointer at the start of the proof, as is done for Theorem C.1.","section":"§4.1"},{"comment":"The table of Wick-averaged values A,B,C is terse. A short sentence indicating how the quaternionic entries are computed (e.g., by writing each quaternionic entry as a 4-vector and using the real Gaussian Wick formula) would improve reproducibility, especially because the Q_Δ entry C=336 differs from the permanent's C=366.","section":"Appendix A"},{"comment":"The projection condition γ_{1,β} ≤_{Lt} |R_i c|²/(σ_i²∥c∥²) should explicitly exclude c=0. This is harmless but should be stated.","section":"§C.1, Eq. (35)"}],"recommendation":"minor_revision","confidential_remarks":"I agree with the reader's assessment: the central chain—Fourier compression, cofactor induction, determinant/Study comparisons, and the density bound—is internally consistent, and I do not see a load-bearing flaw. The main issue is a presentational one: the pairing (·,·)_R is not the Euclidean inner product for C and H, and the proof of the Fourier-to-Laplace step would be clearer if this were acknowledged with an orthogonal change of coordinates. The other comments are local clarifications. The result is significant and well within the scope of the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline is real: this paper proves the permanent anticoncentration conjecture for complex Ginibre matrices, with the sharp nε² small-ball bound, and also covers the real and quaternionic cases plus a perturbed-permanent conjecture. That is a major result, and the proof is more than just a clever trick. The central mechanism—Fourier coordinate compression for permanental cofactor vectors, feeding a Laplace-transform comparison against a gamma benchmark—is genuinely new as far as I can tell. The chain from Proposition 3.5 through the cofactor induction to the density bound is explicit and internally consistent. The reader's strongest claim is correct, and the stress-test note's verification of the quaternionic two-column expansion makes sense: the real-bilinear structure rescues the argument where noncommutativity might have broken it.\n\nWhat the paper does especially well is that it gives a usable, non-asymptotic bound with no fitted parameters. The benchmark is an explicit gamma product, the comparison is proved from Gaussian structure rather than assumed, and the final small-ball estimate follows from a monotone limit. The appendix showing that the Laplace comparison cannot be strengthened to stochastic or convex order is also a good sign: it demonstrates genuine understanding of the order's limits instead of overselling it.\n\nSoft spots are minor relative to the main result. I did not machine-check the quaternionic algebra, and the paper leaves a few degeneracies implicit (e.g., C=0 in the two-coordinate step). But these are technical gaps, not conceptual ones; the stress-test note addresses the main worry about Proposition 3.5, and the argument holds together on reading. The notation calling (·,·)_R the Euclidean inner product is harmless, as the note says. The self-citations to [KL26] are for applications and definitions, not load-bearing, so they don't bother me.\n\nThe only real limitation I see is that the paper is about Gaussian permanents, not Bernoulli or other discrete models. That is the hard case for BosonSampling, so it is a substantive gap relative to the full AA program, but it is not a flaw in this paper—the authors are explicit about what they prove. The typical-scale results of Tao–Vu and the exponential anticoncentration of Hunter–Kwan–Sauermann remain separate, and this paper complements rather than replaces them.\n\nWho should read it: anyone working on anticoncentration, random matrix theory, or quantum complexity lower bounds. It deserves a serious referee. I would send it out and expect acceptance after a check of the quaternionic computations and some cleanup of the edge cases.","headline":"A convincing resolution of the Aaronson–Arkhipov permanent anticoncentration conjecture, built on a genuinely new Laplace-transform comparison that holds up under scrutiny.","tokens_in":19319,"tokens_out":1061,"would_cite":true,"duration_ms":12163,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60B20","60E15","15A15"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that the normalized Gaussian permanent has a radial density bounded by O(n^{(β+2)/4}) over the reals, complexes, and quaternions, and in the complex case obtains sup_z P(|per G_n − z| ≤ ε√n!) ≤ nε², resolving the Permanent","keywords":["permanent","anticoncentration","Ginibre ensemble","Laplace-transform order","small-ball probability","quaternionic permanent","Study determinant","random matrix"],"falsifier":"A direct numerical check of Proposition 3.5 for K = H at n = 3: generate many standard quaternionic Gaussian (2×3) matrices B, compute the permanental cofactor entries C_j = per_H(B_{−j}), and estimate E e^{i(z,C)_R} versus E e^{i(||z||₂ e₁,C)_R} for a test vector z with two nonzero equal-weight coefficients; if the former exceeds the latter by more than sampling error, the compression inequality fails and the density bound loses its support.","tokens_in":18528,"feed_emoji":"🎲","tokens_out":6605,"duration_ms":49362,"temperature":0.7,"pith_summary":"The paper proves that the normalized Gaussian permanent—with row-ordered multiplication for real, complex, and quaternionic entries—has a density whose maximum grows only polynomially in n, as n^{(β+2)/4} where β is the real dimension of the field. In the complex case this gives sup_z P(|per G_n − z| ≤ ε√n!) ≤ min{1, nε²}, which resolves the Permanent Anticoncentration Conjecture for BosonSampling. The core is a comparison of the squared permanent with the squared determinant (or Study determinant for quaternions) in Laplace-transform order, showing the permanent is at least as spread out as the determinant at all scales. A separate stability result extends the same polynomial bound to arbitrary independent Gaussian perturbations of the matrix, proving a stronger form of a previously conjectured gentle-perturbation bound.","feed_headline":"Permanent anticoncentration proved: nε² bound for complex Ginibre","feed_subtitle":"The proof compares permanent with determinant in Laplace-transform order, yielding real and quaternionic bounds too.","key_machinery":"The load-bearing object is the Laplace-transform order of nonnegative random variables (U ⪯_Lt V when E e^{-sU} ≥ E e^{-sV} for all s ≥ 0) and its matrix-variate extension for outer products of random vectors. The main technical tool is the Fourier coordinate-compression inequality (Proposition 3.5): for the permanental cofactor vector C determined by an (n−1)×n Gaussian matrix, the characteristic function φ_n(z) = E e^{i(z,C)_R} is always nonnegative and is no larger than its value at the one-coordinate vector ||z||₂ e₁. This one-sided replacement for the determinant's exact spherical invariance lets the permanent's cofactor vector inherit the determinant's factorial-scale spreading, transf","core_discovery":"The central discovery is that permanent anticoncentration can be reduced to a one-sided comparison between the permanent and determinant cofactor vectors in Laplace-transform order. For an (n−1)×n Gaussian matrix B, the signed determinant cofactor vector is spherically invariant and obeys an exact recursion governed by a product of independent gamma variables; the paper shows, through a new Fourier coordinate-compression inequality, that the permanental cofactor vector is at least as well spread out in Laplace order. From this, the squared permanent Laplace-dominates the squared determinant, and the density of the normalized permanent at zero is bounded by that of a known gamma benchmark, yi","pith_inferences":["Editorial inference: The Fourier compression inequality is a one-sided analog of spherical invariance; if it holds for other row distributions satisfying the projection condition of Appendix C, the same Laplace-order method would transfer anticoncentration to non-Gaussian row models at the exact second-moment scale.","Editorial inference: The explicit nε² constant in the complex case is likely not optimal at small ε: the proof bounds the permanent's density by a gamma benchmark from the determinant, and comparing with the permanent's own moments might improve the linear n factor.","Editorial inference: Uniformity over centers z means the result controls not just |per G_n| but the full small-ball behavior over the plane, which is the property needed for zero-counting and continuity estimates in average-case complexity arguments.","Editorial inference: The method suggests a general recipe for anticoncentrating any symmetric multilinear polynomial of Gaussian entries: find a spherically invariant benchmark and prove Fourier compression for its cofactor vector."],"forward_implications":["For K = C the small-ball probability satisfies sup_z P(|per G_n − z| ≤ ε√n!) ≤ min{1, nε²}, settling the Permanent Anticoncentration Conjecture.","For every K ∈ {R,C,H}, the density of the normalized permanent at zero is bounded by κ_{n,β} ≍ n^{(β+2)/4}, giving polynomial small-ball bounds with exponent (β+2)/4 in all three fields.","For every deterministic A ∈ R^{n×n}, the perturbed permanent per(G_n^R + A)/√n! has density bounded by n^{3/4}; hence sup_w P(|per(G_n^R + A) − w| ≤ ε√n!) ≲ n^{3/4}ε, proving a stronger version of the gentle-perturbation conjecture.","The logarithmic asymptotics log|per_K G_n^K| = 1/2 log(n!) + O_P(log n) follow, refining the known typical-scale result to logarithmic error.","The comparison also gives Sdet(G_H^n)^2 ⪯_Lt |per_H G_H^n|² for the quaternionic Study determinant, extending anticoncentration to the symplectic Ginibre ensemble."],"fun_headline_variants":["Permanent anticoncentration proved for Ginibre ensembles","Aaronson-Archipov conjecture settled via Laplace-order comparison","Gaussian permanent spreads out: new nε² anticoncentration bound","Permanent vs determinant: Laplace-order trick for anticoncentration","Complex Ginibre permanent has flat density: conjecture proved"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire argument rests on the Fourier coordinate-compression inequality—that for the random cofactor vector of a Gaussian matrix, every linear test statistic has characteristic function bounded between zero and the value for a single coordinate of equal Euclidean norm—and this must hold without obstruction in the noncommutative quaternionic case.","fun_headline_variants_meta":{"raw":{"variants":["Permanent anticoncentration proved for Ginibre ensembles","Aaronson-Archipov conjecture settled via Laplace-order comparison","Gaussian permanent spreads out: new nε² anticoncentration bound","Permanent vs determinant: Laplace-order trick for anticoncentration","Complex Ginibre permanent has flat density: conjecture proved"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001362,"raw_usage":{"total_tokens":5354,"prompt_tokens":731,"completion_tokens":4623,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":475,"completion_tokens_details":{"reasoning_tokens":4552}},"tokens_in":475,"tokens_out":4623,"duration_ms":26805,"temperature":1.0,"reasoning_tokens":4552,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T10:08:20.174756+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct numerical check of Proposition 3.5 for K = H at n = 3: generate many standard quaternionic Gaussian (2×3) matrices B, compute the permanental cofactor entries C_j = per_H(B_{−j}), and estimate E e^{i(z,C)_R} versus E e^{i(||z||₂ e₁,C)_R} for a test vector z with two nonzero equal-weight coefficients; if the former exceeds the latter by more than sampling error, the compression inequality fails and the density bound loses its support.","supporting_citations":[],"review_version":1}