{"id":"1507f2c9-21f6-4506-b90c-8e78308ffcbc","arxiv_id":"2607.20578","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The Fisher width and inverse-Fisher width of any compact set multiply to at least the square of the ordinary Gaussian width, and sparse recovery thresholds depend on Fisher curvature at the support.","lead":"This paper introduces a pair of geometric complexity measures built from the Fisher information matrix and proves that their product can never drop below the classical Gaussian-width product. It also shows that sparse recovery with inverse-Fisher measurements depends on which coordinates of the signal sit in high-curvature regions.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the flagged ALMT lower-bound estimate survives the proposed counterexample and is a standard citation; no load-bearing flaw found in the central claim.","rationale":"The reader's main concern is that Eq. (7), used for the two-sided recovery lower bound, is unproved and likely invalid, based on a Euclidean-norm sanity check. That check is arithmetically incorrect: the descent cone of the Euclidean norm at a unit vector is a halfspace with statistical dimension d−1/2, not 1/2, and the claimed RHS value d−2 satisfies the inequality. The paper cites Amelunxen et al. for the estimate; assuming the citation is accurate, the lower-bound argument in Theorem 4.6 is legitimate. The central product inequality (Theorem 5.3) is proved by a clean log-convexity argument, Section 3's local lower bound is internally consistent, and the recovery upper bound is standard. The numerical experiments are illustrative and consistent with the theory. No load-bearing defect was identified. Therefore the honest stress-test outcome is a non-finding; the reader's conditional verdict may be overcautious, but we do not have a concrete reason to overturn it, so the verdict is left unchanged.","tokens_in":20325,"tokens_out":44175,"duration_ms":415906,"concrete_test":"Verify Eq. (7) directly against the cited ALMT proposition (checking that the error term is 2 sup_{z∈∂f(x)}||z||_2 / f(x/||x||_2), not a variant without the denominator), and recompute the Euclidean-norm sanity check using δ({h1≤0}) = d−1/2 instead of 1/2. If the quoted estimate matches ALMT, the reason for a conditional verdict is removed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's weakest assumption targets Eq. (7), the ALMT descent-cone lower bound. The proposed sanity check does not refute it. For f(x)=||x||_2 at x=e1, the descent cone is the halfspace {h1≤0}, whose statistical dimension is δ = d−1/2 (in particular 1/2 for d=1), not 1/2 in general. The RHS of (7) is inf_τ E||g−τe1||² − 2 = d − 2, and d−1/2 ≥ d−2 holds for every d. So the stated estimate is consistent with the example. More fundamentally, since τ∂f_G(x) ⊂ cone(∂f_G(x)) = (descent cone)°, standard conic duality gives E dist²(g, τ∂f_G(x)) ≥ δ, so the subtracted error term in (7) is precisely a quantitative gap bound; the paper cites ALMT for this estimate and applies it in the homogeneous-norm setting to which it is attributed. The lower bound is non-vacuous only when the active-support Fisher mass is not too small, as Remark 4.7 explicitly discloses. I find no internal inconsistency in Theorem 5.3, Section 3, or the recovery upper bound, and the conditional verdict's main basis does not land.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a primal-inverse pair of Fisher widths, w_G(T) = w(G^{1/2}T) and w_{G^{-1}}(T) = w(G^{-1/2}T), and studies their roles in learning and in anisotropic Gaussian recovery. Section 3 proves a nonasymptotic local lower bound showing that the scale w_G(H_r)/√n is attained on small Fisher balls for Fisher-regular losses. Section 4 analyzes inverse-Fisher measurements, reduces unweighted ℓ1 recovery to a weighted-ℓ1 descent cone, and claims a two-sided statistical-dimension estimate in terms of the functional U_G(S). Section 5 establishes the sharp product inequality w_G(T)w_{G^{-1}}(T) ≥ w(T)^2, with a noncommuting geometric-mean extension. The paper closes with controlled numerical experiments on recovery transitions and width redistribution.","tokens_in":20668,"tokens_out":23229,"duration_ms":206119,"significance":"If the claims hold, the most valuable contribution is Theorem 5.3, an elegant and sharp relation showing that Fisher anisotropy cannot reduce both primal and inverse widths below the Euclidean width. Theorem 3.4 is also a solid finite-sample lower bound with explicit constants. The recovery analysis in Section 4 is potentially significant because it predicts support-location effects in sparse recovery under Fisher-induced covariance. The numerical experiments are reproducible in spirit and match the theoretical upper functional closely. However, the two-sided recovery claim rests on an external estimate that is neither derived nor precisely located, and the lower-bound half of Theorem 4.6 is therefore not currently established.","major_comments":[{"comment":"The lower bound in Theorem 4.6 depends on Eq. (7), cited as the 'descent-cone error estimate of Amelunxen et al. [2014]'. The standard ALMT descent-cone upper bound is δ(D(f,x)) ≤ inf_{τ≥0} E dist²(g, τ∂f(x)); the displayed inequality with the additional subtractive term 2 sup_{z∈∂f(x)}‖z‖₂/f(x/‖x‖₂) is not a standard consequence and is not derived. The reader's sanity check with f=‖x‖ does not refute it (the RHS is d−2 while δ is d−1/2), but the estimate remains unsupported. Since the 'two-sided' statistical-dimension estimate is a central claim of the paper, Equation (7) must either be proved in the manuscript or supplied with an exact, verifiable reference, including the hypotheses under which it holds. Without this, Theorem 4.6's lower bound is not established.","section":"§4.3, Eq. (7)"}],"minor_comments":[{"comment":"The introductory subsection contains duplicated paragraphs (the 'Fisher information matrix' paragraph and the 'Throughout this paper' paragraph appear twice).","section":"§1.1"},{"comment":"The displayed lower bound appears to be missing a division sign. The proof computes the error term as 2√(Tr(G)/Σ_{i∈S}γ_i), but the theorem statement as typeset reads like 2√(Tr(G)·Σ_{i∈S}γ_i), which is dimensionally inconsistent. Please correct the display.","section":"Theorem 4.6"},{"comment":"Even if Eq. (7) is a known result, the manuscript should give a precise equation/location in Amelunxen et al. [2014]; the current citation is too vague for a load-bearing estimate.","section":"Eq. (7)"}],"recommendation":"major_revision","confidential_remarks":"The product inequality (Theorem 5.3) is the strongest part of the paper and appears correct. The local lower bound (Theorem 3.4) is also carefully argued. The recovery section, however, makes a two-sided claim whose lower bound depends on an unverified external estimate. The reader's specific counterexample does not land, but the underlying concern about the status of Eq. (7) is legitimate. I would be willing to accept after the author either supplies a proof of Eq. (7) or gives an exact theorem-and-equation reference from ALMT; if no such result exists, the lower bound should be removed or substantially reworked."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is worth a referee's time. The one result I'd remember is Theorem 5.3: for any compact T and positive definite G, w_G(T)w_{G^{-1}}(T) ≥ w(T)^2, with equality for scalar G. The proof via log-convexity along the one-parameter group (Theorem 5.1) is genuinely new and clean. That alone justifies the paper.\n\nWhat else is solid: the local learning lower bound (Theorem 3.4) is carefully done—Paley-Zygmund plus a controlled quadratic remainder—and it genuinely matches the known upper scale. The recovery interpretation is less novel: UG(S) is the standard weighted-l1 distance-to-subdifferential functional, and the author says so. The support-ordering corollaries are elementary but useful, and the numerical experiments are well controlled. The observation that inverse-Fisher measurement can shift the empirical phase transition from roughly 22 to 178 measurements while holding sparsity fixed is a good illustration.\n\nWhere I'd push back: the two-sided estimate in Theorem 4.6 rests on Eq. (7), which is cited to Amelunxen et al. but is neither derived nor located in that reference. The reader's proposed counterexample does not actually refute it—the descent cone of the Euclidean norm at e1 is a halfspace with statistical dimension d−1/2, not 1/2, and d−1/2 is comfortably above the right-hand side. So I don't think the theorem is false; I just want the authors to say exactly where (7) comes from, or prove it inline. It is load-bearing for the lower half of the estimate. The extreme-sparsity case is also openly left as Conjecture 4.8, and the single configuration reported in Section 6.1 is only a consistency check.\n\nMinor but annoying: Section 1.1 contains duplicated paragraphs. That suggests the manuscript was assembled from pieces without a final pass. Not substantive, but it should be cleaned up.\n\nBottom line: the product inequality and the learning lower bound hold up; the recovery lower bound is plausible but under-derived. This is a compact theoretical contribution, appropriate for a specialist journal or conference. I'd engage with it, and I'd accept it for peer review with a request to fix Eq. (7) and tidy the text.","headline":"A clean and new width product inequality with a careful learning lower bound; the recovery side is plausible but leans on an under-derived citation.","tokens_in":21092,"tokens_out":2979,"would_cite":true,"duration_ms":29707,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60D05","62B10","94A12","52A20"],"pacs":[],"model":"deepseek-v4-flash","headline":"Fisher anisotropy can transfer Gaussian-width complexity between the Fisher and inverse-Fisher geometries, but it cannot make both widths smaller than the Euclidean Gaussian width.","keywords":["Fisher width","Gaussian width","Fisher information","statistical dimension","sparse recovery","weighted l1 minimization","descent cone","information geometry"],"falsifier":"For G = diag(1,10,10), support S={1,2}, and x* = (1,1,0) in d=3, compute δ(G^{-1/2}D(||·||_1,x*)) by Monte Carlo projection onto the weighted descent cone and compare it with the interval [U_G(S) − 2√(21/11), U_G(S)]. A value outside that interval would refute Theorem 4.6 directly; the paper's own conjecture asks for a uniform lower bound c U_G(S), which can be tested by shrinking γ on the active support and checking whether δ/U_G(S) collapses.","tokens_in":20226,"feed_emoji":"📐","tokens_out":12193,"duration_ms":121993,"temperature":0.7,"pith_summary":"This paper pairs two complementary geometries carried by the Fisher information matrix of a statistical model: the primal Fisher width w_G(T)=w(G^{1/2}T) and the inverse-Fisher width w_{G^{-1}}(T)=w(G^{-1/2}T), each a Gaussian width of a locally deformed parameter set. It proves a sharp universal relation between them: on every compact coordinate set T and every positive-definite G, w_G(T) w_{G^{-1}}(T) ≥ w(T)^2, so a metric that inflates one width can shrink the other only up to that barrier. On the learning side, it shows that for Fisher-regular losses the uniform empirical-risk fluctuation scale w_G(H_r)/√n is attained on sufficiently small Fisher balls, making the standard upper bound tight rather than loose. On the recovery side, it shows that sparse recovery from Gaussian measurements with covariance G^{-1} is governed by a weighted-ℓ1 descent cone whose statistical dimension has support-dependent bounds, so recovery thresholds depend on where the active coordinates sit in the Fisher spectrum and not merely on sparsity. The net effect is a sharper picture of when local geometry helps and when it hurts: Fisher anisotropy can transfer complexity between learning and recovery, but it cannot erase it.","feed_headline":"Fisher geometry can't beat Euclidean width in both directions","feed_subtitle":"Sharp inequality ties the two width functionals; recovery thresholds shift with support location in the Fisher spectrum.","key_machinery":"The key objects are the two width functionals obtained by deforming a common compact set T by G^{1/2} and G^{-1/2}. The product inequality follows from log-convexity of the map α ↦ w_{G^α}(T) along commuting powers of G, using Sudakov–Fernique comparison and the arithmetic–geometric mean inequality. On the recovery side, the load-bearing identity rewrites G^{-1/2}D(||·||_1,x*) as the descent cone of the weighted ℓ1 norm f_G(x)=||G^{1/2}x||_1, whose statistical dimension is then controlled by the standard distance-to-subdifferential functional U_G(S) and by U_G(S) minus an additive correction involving the Fisher mass on the support.","core_discovery":"The central claim is a sharp product inequality between the two widths induced by the Fisher metric and its inverse: for every nonempty compact parameter set T and every positive-definite Fisher information matrix G, w_G(T) w_{G^{-1}}(T) ≥ w(T)^2. Equality holds for isotropic G, where both widths are scaled Euclidean widths, and for antipodal two-point sets aligned with an eigenvector of G. The proof uses log-convexity of α ↦ w_{G^α}(T) for commuting powers of the metric, established through a Sudakov–Fernique comparison against a weighted sum of the two widths and a scalar arithmetic–geometric mean step. The same machinery yields a noncommutative geometric-mean bound for arbitrary positive-","pith_inferences":["A testable consequence of the support ordering is that intentionally placing a sparse signal on low-Fisher-curvature coordinates should lower basis-pursuit sample complexity; the paper's own low-support profile shows a several-fold reduction, which could be probed in larger randomized designs.","The product inequality suggests a conservation principle for local statistical geometry: the same Fisher matrix governs both parameter sensitivity and estimation noise, so any metric that flattens one geometry must steepen the other; this may inform preconditioning and natural-gradient-style optimization, though the paper does not develop that direction.","The two-sided recovery estimate is non-vacuous only when the active support carries a non-negligible fraction of total Fisher mass; proving the paper's Conjecture 4.8 would extend it to extreme sparsity, and that conjecture is a natural target for a direct cone-geometric argument.","If the descent-cone error bound used in the recovery proof turns out not to hold in its stated form, the product inequality and the learning-side lower bound would remain intact; the recovery lower half would reduce to a known upper functional plus a conjectured comparison."],"forward_implications":["For Fisher-regular losses, empirical-risk fluctuation on a small Fisher ball is, up to universal constants, exactly w_G(H_r)/√n, so the Fisher-geometric upper bound is achieved.","In inverse-Fisher sparse recovery, the measurement requirement for basis pursuit is governed by U_G(S), which depends on which coordinates are active; swapping active coordinates toward larger Fisher curvature raises the threshold.","Compensating for anisotropy by inverse-square-root weighting or finite-sample column normalization removes profile dependence but can raise the required measurements in profiles where the unweighted decoder already benefits from the geometry.","The product inequality implies that any preconditioning of a parameter set by G and by G^{-1} cannot make both widths smaller than the Euclidean Gaussian width; complexity can be redistributed but not destroyed.","Nested supports are ordered monotonically: adding active coordinates cannot decrease the recovery upper bound, and equal-cardinality supports can have very different recovery thresholds."],"fun_headline_variants":["Fisher width product can't beat Euclidean square","Shrink Fisher width, pay in inverse-Fisher width","Universal bound: product of Fisher widths ≥ Euclidean squared","Anisotropy can't reduce both Fisher widths below Euclidean"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The recovery-side lower bound rests on an unproved descent-cone error bound invoked in the proof of the two-sided statistical-dimension estimate; if that bound is not generally valid, the lower half of the recovery theorem and the support-ordering consequences lose their proof, while the width product inequality and learning-side result would still stand.","fun_headline_variants_meta":{"raw":{"variants":["Fisher width product can't beat Euclidean square","Shrink Fisher width, pay in inverse-Fisher width","Universal bound: product of Fisher widths ≥ Euclidean squared","Anisotropy can't reduce both Fisher widths below Euclidean"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000601,"raw_usage":{"total_tokens":2673,"prompt_tokens":799,"completion_tokens":1874,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":543,"completion_tokens_details":{"reasoning_tokens":1811}},"tokens_in":543,"tokens_out":1874,"duration_ms":14967,"temperature":1.0,"reasoning_tokens":1811,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T11:08:06.514567+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For G = diag(1,10,10), support S={1,2}, and x* = (1,1,0) in d=3, compute δ(G^{-1/2}D(||·||_1,x*)) by Monte Carlo projection onto the weighted descent cone and compare it with the interval [U_G(S) − 2√(21/11), U_G(S)]. A value outside that interval would refute Theorem 4.6 directly; the paper's own conjecture asks for a uniform lower bound c U_G(S), which can be tested by shrinking γ on the active support and checking whether δ/U_G(S) collapses.","supporting_citations":[],"review_version":1}