{"id":"4d52c4c8-4bb5-476e-88c4-4c82331870f2","arxiv_id":"2607.13279","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"If a heat-observability inequality with weight h holds from a set ω, then h(t) decays at least like exp(-κ L(ω)²/t) for every κ<1/2, yielding the sharp lower bound γ≥L(ω)²/2.","lead":"This mathematics paper proves a universal lower bound on how fast the weight in a heat-observability inequality must decay at small times, forcing the familiar e^{-γ/t} factor from geometry alone. It settles an open problem by showing the infinite-time constant is at least half the squared maximal distance to the observation set.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the proof is coherent, and the main risk—the pointwise Weyl law (A3)—is an explicit, verified hypothesis, not a hidden gap.","rationale":"The reader's weakest-assumption analysis correctly identifies (A3) as the load-bearing input: it is the only source of the lower bound in Lemma 2 and hence of the exponential scale r²/(2t). I agree with that diagnosis. However, identifying an assumption as load-bearing is not the same as finding a defect: the theorem is explicitly conditional, and the paper verifies (A3) in all four settings. The most delicate verification is the δ′-graph and the Grushin/sub-Riemannian case, but the cited pointwise Weyl laws are standard there. I checked the key inequalities in the proof (Lemma 1's transmutation identity, Lemma 2's spectral-interval estimate, Lemma 3's off-diagonal decay via wave-kernel support, Lemma 4's large-time bound, and the optimization over the base point in Theorem 1) and found no algebraic or logical error. The admissible-weight monotonicity near zero is used properly, the treatment of T=∞ via the (ker A)⊥ variant is justified, and the normalization of the observability constant is harmless. The paper also honestly flags its own limitations (no endpoint κ=1/2 with constant prefactor, no claim of a universal exact cost). Therefore the reader's ACCEPT verdict stands without modification.","tokens_in":38197,"tokens_out":47547,"duration_ms":479065,"concrete_test":"Independently re-derive (A3) for the least classical case, the δ′-coupled metric graph: at an interior point x of an edge, compute E_x(λ)=Σ_{λ_n≤λ}|φ_n(x)|² using the eigenfunction asymptotics of [9] and check that E_x(λ)/√λ → 1/π. If the ratio fails to converge to a positive constant, Lemma 2's lower bound would lose its input for that application and the theorem would not cover δ′-graphs as claimed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I read the paper as a conditional theorem: if (A1)–(A3) hold, the geometric barrier follows. The only place the exponential scale r²/(2t) is generated is Lemma 2, and that lower bound relies entirely on (A3) at a single base point. If (A3) failed on a positive-measure set, the lower bound would be lost and the barrier would not be established under these hypotheses. This is a genuine spectral input, not derivable from (A1)–(A2) alone. However, the paper states (A3) explicitly and verifies it in all four application classes (compact Riemannian, Schrödinger on R^d, sub-Riemannian/Grushin, δ′-graph). I found no internal inconsistency in those verifications and no step that smuggles in extra assumptions. The A2-to-finite-speed passage is a cited theorem [53] and is standard for the local operators considered. The kernel calculus in the appendix is intricate but appears rigorous; no circularity or fitted parameters were found. Thus the central claim holds as stated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves a universal small-time geometric barrier for weighted integrated observability of heat semigroups. Under hypotheses (A1) ultracontractivity, (A2) Davies–Gaffney estimates/finite propagation speed, and (A3) a pointwise local Weyl law on a doubling metric measure space, Theorem 1 shows that if an admissible weight h satisfies the integrated observability inequality (7) on a measurable set ω, then h(t) ≤ A_{T,κ} exp(−κ L(ω)^2/t) for every κ<1/2, equivalently limsup_{t↓0} t log h(t) ≤ −L(ω)^2/2. Corollary 1 gives the previously open infinite-time bound γ∞ ≥ L(ω)^2/2 for exponential weights; Corollary 2 recovers Miller's fast-control lower bound; Corollary 3 gives a corresponding high-frequency spectral constraint. The proof uses the spectral packet (cosh(r√A)−1)e^{−tA}, with a lower bound from the Weyl law at a base point and an upper bound on ω via a weak wave kernel and Kannai transmutation. Appendices develop pointwise spectral measures and L^∞_Δ kernel calculus without compact resolvent or kernel continuity. Applications include compact Riemannian Laplace-type operators, Schrödinger operators on R^d, sub-Riemannian/Grushin structures, and δ′-coupled metric graphs.","tokens_in":38436,"tokens_out":23938,"duration_ms":227922,"significance":"If the proof is correct — and I found it coherent — the result is significant: it shows that the ubiquitous e^{−γ/t} scale in heat observability is a geometric obstruction rather than an artifact of Carleman or spectral methods. The theorem is conditional on explicit, stated hypotheses, and the proof has no fitted parameters: the observability inequality is only the starting assumption, while the lower bound is fed by the Weyl law and the upper bound by finite-speed propagation, so there is no circularity. The framework is genuinely broader than earlier treatments, covering non-continuous kernels and operators without compact resolvent, and the four application classes give the result concrete reach. The paper is also honest about its limitations: Remark 3 explicitly disclaims the constant-prefactor endpoint κ=1/2, and (A3) is a real spectral input that is verified rather than derived. These features make the central claim credible and well-scoped.","major_comments":[],"minor_comments":[{"comment":"The paragraph after (1) uses C_T for both the observability constant in (1) and the usual L2 null-control cost, then states 'C_T = C_T^2'. This is confusing. Please rename one of the two constants (e.g., use K_T for the control cost) so the conversion between the squared-observability and control-norm normalizations is unambiguous.","section":"1.1"},{"comment":"Remark 3 correctly flags that the theorem proves the logarithmic endpoint and the quantitative family (8) for every κ<1/2, but not an endpoint estimate h(t) ≤ A e^{−L(ω)^2/(2t)} with a bounded prefactor. This is a genuine limitation and is appropriately stated; I mention it to confirm that the abstract's phrase 'sharp logarithmic endpoint' should not be overread as a constant-prefactor result.","section":"Remark 3"},{"comment":"For Schrödinger operators on R^d with bounded potential, the verification of (A3) is given by two arguments: Hörmander's local asymptotics and Feynman–Kac plus Karamata. Since Hörmander's theorem is usually stated for compact manifolds, the Karamata route is the more transparent justification in the noncompact setting; consider making it the primary proof and keeping the Hörmander remark as a comment.","section":"4.2"},{"comment":"The proof of Lemma B.1 is correct but compressed, especially in the Seeley-extension step after inequality (53). Adding one or two sentences explaining how the zeroth-order and 2m-th order extension bounds combine to yield (52) would improve readability. This is purely expository.","section":"B.1"}],"recommendation":"minor_revision","confidential_remarks":"The manuscript is well within the scope of math.AP and control theory, and I saw no citation or novelty concerns. The central theorem is sound as a conditional statement, and the appendix supplies the necessary functional-analytic machinery. My recommendation of minor revision is driven only by the local clarity items listed above; these are easily addressed and do not affect the mathematical content."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a strong paper. It answers the open question from Ervedoza–Zuazua and Zuazua by proving, under explicit assumptions, that any admissible weight in an integrated observability inequality must satisfy limsup_{t↓0} t log h(t) ≤ −L(ω)²/2; for h(t)=e^{−γ/t} with T=∞, this gives γ ≥ L(ω)²/2. The coefficient 1/2 is optimal as a universal threshold.\n\nWhat is actually new: the finite-time maximal-distance bound was Miller's, but the infinite-time constant was open. The paper also treats arbitrary admissible weights rather than only exponentials, and it works on doubling metric measure spaces without compact resolvent or kernel continuity. That last point is not decorative: the pointwise spectral measure and weak wave kernel constructions are genuinely needed for δ′-coupled graph Laplacians, where the heat kernel is discontinuous, and the four application classes are nontrivial. The proof strategy is clear: the test packet (cosh(r√A)−1)e^{−tA} is large at the base point by the local Weyl law and exponentially small on ω by finite speed of propagation; balancing the two gives the barrier.\n\nSoft spots, in proportion. The theorem is conditional on (A3), the pointwise local Weyl law, and that assumption does real work: it is the only source of the lower bound that fixes the r²/(2t) scale. The paper states this plainly and verifies (A3) in all four settings, so it is not a hidden gap, but it does mean the abstract result is narrower than the title alone might suggest. Minor points: the constants depend on the data, and there is no endpoint estimate at κ=1/2 with a bounded prefactor; the paper is explicit that only the logarithmic endpoint is asserted. The appendix is dense; I did not verify every line of the Seeley extension argument, but the main steps check out and I found no circularity and no fitted parameters.\n\nBottom line: for anyone working on heat controllability or spectral barrier arguments, this is worth careful reading and citing. I would send it to a serious referee and expect acceptance after normal minor revision.","headline":"Settles the infinite-time maximal-distance bound γ∞ ≥ L(ω)²/2 with a careful, honest proof; the main spectral assumption is explicit and verified in all claimed settings, so the paper deserves a serious referee.","tokens_in":38943,"tokens_out":3285,"would_cite":true,"duration_ms":45002,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35K05","35B60","93B07","35P20","58J35"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that any weight in a heat observability inequality must decay like e^{-d^2/(2t)}, where d is the essential maximal distance to the observation set.","keywords":["observability","heat semigroup","geometric barrier","essential maximal distance","Weyl law","finite speed of propagation","Kannai transmutation","metric measure space"],"falsifier":"Compute the optimal observability weight for the Dirichlet Laplacian on a dumbbell domain (two unit squares joined by a thin corridor) with observation set equal to one square: the theorem predicts limsup_{t→0} t log h(t) = -L(ω)^2/2, where L(ω) is the maximal distance from the far square. A numerical or spectral computation of the optimal weight that finds a strictly greater (less negative) value would contradict the logarithmic endpoint (9) and would indicate a hidden failure of one of the three assumptions, most likely the pointwise Weyl law at the maximal-distance point.","tokens_in":38051,"feed_emoji":"📏","tokens_out":9862,"duration_ms":84038,"temperature":0.7,"pith_summary":"Integrated observability inequalities for heat equations typically come with a small-time factor e^{-γ/t}. This paper proves that this exponential scale is not an artifact of Carleman or spectral estimates: it is imposed by the geometry of the observation set. On a doubling metric measure space, for any nonnegative self-adjoint operator satisfying ultracontractivity, Davies–Gaffney estimates and a pointwise Weyl law, any admissible weight h satisfying the inequality must obey h(t) ≤ A e^{-κ L(ω)^2/t} for every κ<1/2, where L(ω) is the essential maximal distance to the observation set. Equivalently, limsup_{t→0} t log h(t) ≤ -L(ω)^2/2. This yields the optimal lower bound γ∞ ≥ L(ω)^2/2 for the infinite-time observability constant, resolving an open problem, and recovers Miller's fast-control bound. The result covers Riemannian manifolds, Schrödinger operators, sub-Riemannian (Grushin) structures and δ'-coupled metric graphs.","feed_headline":"Observability weights must decay like e^{-d²/2t}","feed_subtitle":"The exponential barrier is set by the maximal distance to the observed set.","key_machinery":"One test object drives the proof: the spectral packet (cosh(r√A) − 1)e^{-tA} centered at x, with r below δ = d_ess(x,ω). A pointwise Plancherel identity expresses its L^2 norm through the pointwise spectral measure ν_x; the pointwise Weyl law makes this norm grow at least like e^{(1−η)r^2/(2t)}. Finite speed of propagation (Davies–Gaffney) and the Kannai transmutation formula make the packet exponentially small on ω, of order e^{-(δ^2−r^2)/(2t)}. Matching the two rates yields h(t) ≲ e^{-κ r^2/t}; optimizing r ↑ δ and x gives L(ω)^2/2.","core_discovery":"The central claim is that the e^{-γ/t} factor in heat observability is a geometric necessity, not a by-product of Carleman or spectral proofs. Under three assumptions — ultracontractivity, Davies–Gaffney estimates (finite wave propagation speed), and a pointwise local Weyl law — Theorem 1 says that if an integrated observability inequality holds with an admissible weight h on a measurable set ω, then for every κ<1/2, h(t) ≤ A_{T,κ} e^{-κ L(ω)^2/t} for 0