{"id":"a9b04b60-a322-41dd-a9a8-035ecb2c46dd","arxiv_id":"2607.17393","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"In d=1,2, C(s,t) ≍ min{s,t}^{1-d/2} max{s,t}^{1-d/2-d/κ} as s,t→0 for the white-noise Schrödinger trace under two-sided power-law potential growth.","lead":"For random Schrödinger operators with white noise in one and two dimensions, this paper proves sharp small-time asymptotics for the covariance of the exponential trace and shows they are optimal for power-law confining potentials. The result quantifies trace decorrelation and yields a quantitative semigroup hyperuniformity rate.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 4.2's dominated-convergence step is invalid for d=2: the Young bound (u(t-u))^{-d/2}(v(s-v))^{-d/2} is non-integrable, leaving Lemma 4.1 and the d=2 bounds unsupported as written; fixable via the exact bound p_{ε+z} ≤ (2π z)^{-d/2}.","rationale":"The reader's weakest_assumption focused on the d=2 trace construction and its dependence on imported exponential moment bounds. That is a legitimate structural concern, but the most concrete load-bearing defect in the manuscript is the invalid dominated-convergence step in Proposition 4.2: the printed Young-type bound is non-integrable for d=2, and Proposition 4.2 underpins Lemma 4.1, which is used in both the upper and lower proofs of the main theorem. This is not fatal because the same proof contains the exact Gaussian-convolution identity that supplies an integrable majorant, so the result is likely correct and the gap is easily repaired. Since the reader already assigned CONDITIONAL and this correction does not change that assessment, I leave the verdict unchanged. I mark agreement as partial because my primary concern is the internal proof gap in Prop 4.2 rather than the d=2 trace construction, though both point to the need for revision before the d=2 claims are taken as fully rigorous.","tokens_in":30452,"tokens_out":36826,"duration_ms":355242,"concrete_test":"Correct the domination step in Proposition 4.2 by replacing the Young inequality with the exact bound E[p_ε(B^{x,x}_t(u)-B^{y,y}_s(v))] = p_{ε+z}(x-y) ≤ (2π z)^{-d/2}, z = u(t-u)/t + v(s-v)/s. Verify that ∫_0^t∫_0^s z^{-1} du dv < ∞ for d=2 (and z^{-1/2} for d=1). If this corrected domination is integrable, Lemma 4.1 and the d=2 part of Theorem 1.4 are restored; if it is not, the d=2 upper and lower bounds fail.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Section 4.6, proof of Proposition 4.2, the paper justifies passing ε→0 inside the (u,v) integral by Young's convolution inequality, obtaining the bound C^d [t/(u(t-u))]^{d/2}[s/(v(s-v))]^{d/2}. For d=2 this behaves like (u v)^{-1} near the boundaries of [0,t]×[0,s], so it is not integrable; the stated dominated-convergence argument fails exactly in the new d=2 case. Proposition 4.2 supplies the first-moment formula for the mutual intersection local time used in Lemma 4.1, which is the engine for both the upper bound (4.5)/(4.8) and the optimality lower bound (4.30)–(4.31). Thus, as written, the proof of Theorem 1.4 is incomplete in d=2. The gap is repairable: earlier in the same proof one has the exact identity E[p_ε(B^{x,x}_t(u)-B^{y,y}_s(v))] = p_{ε+z}(x-y) with z = u(t-u)/t + v(s-v)/s, and |p_{ε+z}(x-y)| ≤ (2π z)^{-d/2}, which is integrable for d=1,2. So this is a real but fixable flaw, not a refutation of the theorem.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the covariance C(s,t) = Cov(Tr[e^{-sH}], Tr[e^{-tH}]) for Schr\\\"odinger operators with Gaussian white noise in dimensions d=1,2, under a deterministic potential V with power-law growth (1.2). The main result, Theorem 1.4, states an upper bound C(s,t) ≲ min{s,t}^{1-d/2} max{s,t}^{1-d/2-d/κ} as (s,t)→0, and if the additional upper-growth condition (1.4) holds, a matching lower bound, giving optimal asymptotics. The proof uses Feynman-Kac representations for the trace and its covariance, reducing the problem to estimates of mutual and self-intersection local times of Brownian bridges. The central technical estimate is Lemma 4.1, proved via explicit first- and second-moment formulas (Propositions 4.2 and 4.3). Applications include a semigroup hyperuniformity ratio R_s(t) ≍ t^{2-d/2} (Corollary 1.10) and a correlation estimate ρ(s,t) ≍ (min{s,t}/max{s,t})^{d/(2κ)} (Corollary 1.12). In dimension 2, the trace observable is defined as an L^2 renormalized limit of regularized traces, since the operator H itself is not constructed.","tokens_in":30804,"tokens_out":14652,"duration_ms":128646,"significance":"If the result holds, it represents a substantial improvement over previous one-dimensional bounds and the first two-dimensional covariance estimates for general power-law potentials. The matching upper and lower bounds establish that the exponents are optimal, and the Brownian-bridge local-time method is elegant and detailed. The paper ships explicit, parameter-free asymptotic exponents in terms of d and κ, and the applications (Corollaries 1.10 and 1.12) are concrete falsifiable predictions. The proof of Theorem 1.4 is largely written out, with the key scaling estimates in Lemma 4.1 checked in detail; the d=2 construction of the trace relies on imported results from [GLP26] and [Mat22], which is acknowledged. The main caveats are a genuine but locally fixable gap in the proof of Proposition 4.2 for d=2, and the fact that the d=2 theorem concerns the renormalized trace observable T(t) rather than an operator H constructed independently of the regularization.","major_comments":[{"comment":"The dominated-convergence step uses Young's convolution inequality to dominate E[p_ε(B_t(u)-B_s(v))] by C^d [t/(u(t-u))]^{d/2} [s/(v(s-v))]^{d/2}. For d=2 this bound is non-integrable on [0,t]×[0,s] (near u=0 or v=0 it behaves like (uv)^{-1}), so the passage ε→0 inside the double integral is not justified as written. Since Proposition 4.2 feeds Lemma 4.1, which yields both the upper bound (4.8) and the lower bound (4.30)–(4.31), the d=2 proof of Theorem 1.4 is incomplete as written. The gap is local and fixable: earlier in the same proof one has the exact identity E[p_ε(B_t(u)-B_s(v))] = p_{ε+z}(x-y) with z=u(t-u)/t+v(s-v)/s, and |p_{ε+z}(x-y)| ≤ (2π z)^{-d/2}, which is integrable for d=1,2. Please replace the Young bound by this estimate.","section":"Section 4.6, proof of Proposition 4.2"},{"comment":"In d=2, Tr[e^{-tH}] is defined as an L^2 limit of Tr[e^{-t(H_ε+c_ε)}] for t<ϑ; H itself is not constructed (Remark 1.2). Theorem 1.4 is therefore a statement about the renormalized trace observable T(t). The proof of Proposition 2.4 establishes convergence for Gaussian mollifiers with the specific c_ε=(1/2π)log(1/ε), but no independence of the limit under other regularizations is shown. The theorem statement should make this conditioning explicit (e.g., by phrasing the d=2 half in terms of T(t)) and discuss the uniqueness question, so that the object whose covariance is estimated is unambiguous.","section":"Section 2.2 / Definition 2.5"}],"minor_comments":[{"comment":"There is a typo: 'ind“1,2' should read 'in d=1,2'.","section":"Abstract"},{"comment":"The expression 'Ct,spx, yqdxdyq' contains a stray 'q'; it should be 'Ct,spx, yqdxdy'.","section":"Section 4.3, Eq. (4.31)"},{"comment":"The symbol '—' for asymptotic equivalence is nonstandard; consider using '≍' consistently with the abstract.","section":"Definition 1.3"},{"comment":"In the proof of Proposition 4.2, both bridge densities are denoted Φ_B; use different symbols (e.g., Φ_1, Φ_2) for clarity.","section":"Section 4.6"},{"comment":"The use of Minkowski's determinant inequality to get det(K0) ≥ t^2 det(A(a)) + s^2 det(A(b)) is terse; adding a one-line explanation via the matrix square root would help the reader.","section":"Section 4.7"}],"recommendation":"major_revision","confidential_remarks":"The paper relies heavily on the authors' prior works (GL21, LGL20, GL25, GLP26) for the d=2 construction, exponential bounds, and moment formulas. Given that the d=2 theorem is conditional on the existence of the renormalized trace T(t), the editor may want the authors to either supply a self-contained construction or explicitly mark the d=2 result as conditional on [GLP26]. The regularization-dependence issue deserves airing. The Proposition 4.2 gap is real but is easily fixed by using the exact p_{ε+z} bound instead of the Young inequality, so rejection is not warranted on that basis."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main thing to know: this paper proves matching upper and lower bounds for the covariance of Tr e^{-sH} and Tr e^{-tH} as s,t→0 in d=1,2, improving the d=1 exponent from 1/4 to 1/2 and giving the first d=2 bounds for power-law potentials. I believe the theorem is true, but the proof as written has a fixable gap in Section 4.6.\n\nThe genuinely new content is the two-time asymptotics with a matching lower bound. The scaling exponents are consistent, and the Feynman-Kac framework with mutual and self-intersection local times is the right tool. The paper is also honest about a delicate point: in d=2 it constructs a renormalized trace observable as an L2 limit rather than the operator itself (Remark 1.2), and it imports exponential moment bounds from earlier published work. I don't see circularity there; the cited results are published and independently checkable.\n\nThe soft spot is Proposition 4.2. To pass ε→0 inside the double integral, the paper invokes Young's convolution inequality and dominates the integrand by C^d [t/(u(t-u))]^{d/2} [s/(v(s-v))]^{d/2}. For d=2 that behaves like (u v)^{-1} near the boundaries of [0,t]×[0,s], so it is not integrable and the stated dominated-convergence argument fails exactly in the new d=2 case. This is a real flaw in the write-up, but it is repairable: earlier in the same proof one has the exact identity E[p_ε(B^{x,x}_t(u)-B^{y,y}_s(v))] = p_{ε+z}(x-y) with z = u(t-u)/t + v(s-v)/s, and |p_{ε+z}(x-y)| ≤ (2π z)^{-d/2}, which is integrable for d=1,2. Replacing the Young bound with this pointwise bound fixes the proof of Lemma 4.1 and the d=2 bounds.\n\nWho this is for: people working on random Schrödinger operators, Feynman-Kac estimates, and rigidity/hyperuniformity questions for point processes. It deserves a serious referee. Send it to peer review and ask the authors to replace the dominated-convergence argument in Proposition 4.2 and to state the integrability of the exact density bound explicitly. That is a revision, not a rejection.","headline":"Sharp covariance asymptotics for white-noise Schrödinger traces in d=1,2: the result looks right, but Proposition 4.2's dominated-convergence step is non-integrable in d=2 and needs a small fix.","tokens_in":31376,"tokens_out":3306,"would_cite":true,"duration_ms":29781,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60H25","60J55","82B44"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper establishes optimal asymptotics for the covariance of the Schrödinger semigroup trace in dimensions one and two: as s,t→0, the covariance of Tr[e^{-sH}] and Tr[e^{-tH}] is bounded above by min{s,t}^{1-d/2} max{s,t}^{1-d/2-d/κ}, a","keywords":["random Schrödinger operator","white noise","semigroup trace","covariance asymptotics","Feynman–Kac formula","Brownian bridge local times","intersection local times","hyperuniformity"],"falsifier":"Compute the L^2 limit in Proposition 2.4 with two different mollifiers (e.g., Gaussian vs. compactly supported) for the same white noise; if the limiting random variable T(t) differs, the d=2 claim collapses. Alternatively, for d=1 with κ=1, numerically evaluate C(s,t) for very small s,t and check whether it follows min{s,t}^{1/2} max{s,t}^{−1/2}; a mismatch would disprove the claimed optimal exponent.","tokens_in":30279,"feed_emoji":"🎲","tokens_out":6579,"duration_ms":56930,"temperature":0.7,"pith_summary":"This paper gives the first matching upper and lower bounds on the covariance between Tr[e^{-sH}] and Tr[e^{-tH}], where H is the random Schrödinger operator −½Δ+V+white noise in d=1,2 and V grows like |x|^κ. The result is that as s,t→0 the covariance is exactly of order min{s,t}^{1−d/2} max{s,t}^{1−d/2−d/κ}, up to constants, once V also satisfies a matching upper growth bound; without that upper bound only the upper estimate is claimed. The proof uses Feynman–Kac formulas to write the covariance entirely in terms of Brownian-bridge self-intersection and mutual intersection local times, then controls those local times by scaling and exact Gaussian moment formulas. Because the bounds match, the paper obtains quantitative consequences: a semigroup notion of hyperuniformity with rate t^{2−d/2}, and a decorrelation rate (min/max)^{d/(2κ)} between eigenvalues sampled at different temperature scales.","feed_headline":"Exact covariance rates for white-noise Schrödinger traces in d=1,2","feed_subtitle":"Matching upper and lower bounds from Brownian-bridge local times also prove a semigroup hyperuniformity law.","key_machinery":"The central object is the Brownian-bridge intersection local time: the self-intersection local time β_t (how often a bridge crosses itself) and the mutual intersection local time α_{s,t} (how often two independent bridges cross each other). The paper derives exact Feynman–Kac formulas: E[Tr e^{-tH}] and Cov(Tr e^{-sH}, Tr e^{-tH}) are Gaussian integrals over x,y of expectations of exponentials of these local times. The estimates then reduce to moment asymptotics for α, proved through explicit Gaussian-density integral representations (Propositions 4.2, 4.3) and scaling arguments. In d=2 the intersection local times must be renormalized by subtracting a divergent mean, and the trace is define","core_discovery":"The paper proves that for d=1,2, under a lower power-law growth condition V(x) ≥ a|x|^κ − b, the covariance C(s,t) = Cov(Tr[e^{-sH}], Tr[e^{-tH}]) obeys C(s,t) ≲ min{s,t}^{1−d/2} max{s,t}^{1−d/2−d/κ} as (s,t)→0. When V also satisfies an upper growth condition V(x) ≤ c|x|^κ + f, the reverse bound holds, giving C(s,t) ≍ min{s,t}^{1−d/2} max{s,t}^{1−d/2−d/κ}. In d=1 this sharpens earlier upper bounds for s≠t; in d=2 it is the first covariance estimate for general power-law potentials. The d=2 trace is defined not through a full operator construction but as an L^2 limit of renormalized smoothed traces, which the paper constructs for small times using Feynman–Kac formulas and exponential moment b","pith_inferences":["Editorial inference: If the d=2 renormalized trace construction can be shown independent of the mollifier, the same Feynman–Kac machinery should extend to other self-intersection-local-time functionals, such as higher moments of the trace, yielding a full fluctuation theory.","Editorial inference: The hyperuniformity exponent 2−d/2 suggests a transition at d=4, where the ratio would fail to vanish; this is consistent with the known criticality of white-noise Schrödinger operators in d≥4 and could be tested numerically in d=3.","Editorial inference: The decorrelation rate (min/max)^{d/(2κ)} gives a concrete prediction for eigenvalue counting-function correlations (Conjecture 1.18) if the Abelian/Tauberian heuristic holds; simulating the one-dimensional model with V(x)=|x|^κ and comparing counting-function correlations to this rate is a direct test.","Editorial inference: The method of splitting the covariance integral at a cutoff {α>c} and showing the tail is polynomially small is a template for other observables expressible through intersection local times with subexponential moment bounds, and may yield similar two-scale asymptotics in related singular SPDE models."],"forward_implications":["The semigroup hyperuniformity ratio Var[Tr e^{-tH}] / E[Tr e^{-tH}] decays like t^{2−d/2} as t→0 under (1.2)+(1.4), so the eigenvalue point process is hyperuniform in the semigroup sense, with a rate depending only on dimension.","The correlation between Tr[e^{-sH}] and Tr[e^{-tH}] decays like (min{s,t}/max{s,t})^{d/(2κ)}, quantifying how eigenfunctions share the same noise.","In d=1 the upper bound C(s,t) ≲ min{s,t}^{1/2} max{s,t}^{1/2−1/κ} improves earlier bounds for s≠t, e.g., replacing the (min/max)^{1/4} bound in the κ=1 case.","In d=2 the result gives the first covariance estimates for general power-law potentials; formally setting κ=∞ recovers the flat-potential bound C(t,t)=O(1).","The matching lower bound shows the exponent 1−d/2−d/κ is intrinsic under the upper growth condition (1.4), so the upper bound cannot be improved in general."],"fun_headline_variants":["Optimal covariance bounds for white-noise Schrödinger traces in d=1,2","Sharp trace covariance rates for 1D and 2D random Schrödinger","Matching upper/lower bounds for Schrödinger semigroup traces","Hyperuniformity and decorrelation in white-noise Schrödinger traces","First optimal covariance estimates for d=2 Schrödinger with noise"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the d=2 renormalized trace, defined as the L^2 limit of Tr[e^{-t(H_ε+c_ε)}], exists and is independent of regularization for small t; this depends on uniform exponential moment bounds for renormalized intersection local times imported from earlier work.","fun_headline_variants_meta":{"raw":{"variants":["Optimal covariance bounds for white-noise Schrödinger traces in d=1,2","Sharp trace covariance rates for 1D and 2D random Schrödinger","Matching upper/lower bounds for Schrödinger semigroup traces","Hyperuniformity and decorrelation in white-noise Schrödinger traces","First optimal covariance estimates for d=2 Schrödinger with noise"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000272,"raw_usage":{"total_tokens":1497,"prompt_tokens":803,"completion_tokens":694,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":547,"completion_tokens_details":{"reasoning_tokens":601}},"tokens_in":547,"tokens_out":694,"duration_ms":7330,"temperature":1.0,"reasoning_tokens":601,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T18:09:14.585178+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the L^2 limit in Proposition 2.4 with two different mollifiers (e.g., Gaussian vs. compactly supported) for the same white noise; if the limiting random variable T(t) differs, the d=2 claim collapses. Alternatively, for d=1 with κ=1, numerically evaluate C(s,t) for very small s,t and check whether it follows min{s,t}^{1/2} max{s,t}^{−1/2}; a mismatch would disprove the claimed optimal exponent.","supporting_citations":[],"review_version":1}