{"id":"ac970d90-42b6-402f-958c-476eedbb79cc","arxiv_id":"2505.05012","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The nearby cycles of GKS kernels quantizing smoothed support-function Hamiltonians act on constructible sheaves by convolution with twisted polytope sheaves, matching the mirror Picard group action.","lead":"This paper shows that the mirror action of line bundles on toric varieties can be realized by quantized contact isotopies in the coherent-constructible correspondence, the sheaf-theoretic model of the A-side. The result provides a sheaf-theoretic counterpart to Hanlon's Floer-theoretic theorem and holds for arbitrary normal toric varieties.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the analytic estimates in Props. 4.7 and 4.11 are delicate but appear correct; only a face/facet notational slip.","rationale":"The reader's verdict is ACCEPT with moderate confidence and identifies the analytic derivative bounds in Propositions 4.7 and 4.11 as the weakest assumption. I agree that these are the most intricate and load-bearing analytic inputs: they control the singular support of the nearby cycles kernel in Proposition 4.13 and provide the properness needed in Theorem 4.18. However, on close inspection the estimates are internally consistent. The bound (4.9) is elementary and correct; Proposition 4.11 follows immediately. The limsup containment in Proposition 4.7 is valid because the ambient space is M_R × \\dot N_R, so the limiting covector is nonzero, and the face τ of the limit direction contains enough information to force dφ_{ε_n}(ξ_n) into χ_σ + τ^⊥. The only genuine defect is the printed convention that ≼ means 'facet' while every use requires 'face'; this is a minor notational slip that does not affect the argument once corrected. The remaining steps — Proposition 3.9's commuting of nearby cycles with group convolution, the microlocal uniqueness argument in Theorem 4.18, and the noncomplete reduction in Theorem 4.22 — are all supported by the text or by standard references. I therefore see no reason to alter the reader's verdict, and the recommended action is UNCHANGED.","tokens_in":22166,"tokens_out":37411,"duration_ms":365073,"concrete_test":"Independently re-derive Proposition 4.7 for a non-simplicial complete fan, e.g. the cone over a square in R^3, which has a codimension-2 face. Choose a piecewise-linear support function φ with distinct Cartier slopes on the maximal cones, take a sequence ξ_n in the relative interior of a maximal cone σ converging to a point on the codimension-2 face, and compute dφ_{ε_n}(ξ_n) with ε_n chosen so that the mollifier ball crosses the relevant fan walls. Verify explicitly that the limit lies in χ_σ + τ^⊥, where τ is the face containing the limiting direction, and that the containment (4.8) holds when ≼ is read as 'face'. This isolates the step where a missing term in (4.9) or an incorrect face/facet convention would first surface.","verdict_should_be":"UNCHANGED","load_bearing_attack":"After tracing the proof of Theorem 1.1, I cannot identify a load-bearing flaw. The most delicate input is the analytic control of the mollified support-function derivatives (Propositions 4.7 and 4.11), which supports the singular-support bound in Proposition 4.13 and the properness argument in Theorem 4.18. Re-examining these estimates: the bound on the non-convolution term in (4.9) is correct — the integrand reduces to |⟨χ_{σ_i}|εu⟩| on each maximal cone, giving |g_ε(ξ)| ≤ ε Σ|χ_{σ_i}|; Proposition 4.11 then follows from this bound together with df_ε(ξ̂) ∈ Conv({χ_σ}). The limsup argument in Proposition 4.7 also works: limits are taken in M_R × \\dot N_R, so the covector cannot degenerate to zero, and for a limiting direction in the relative interior of a face τ, the differences χ_{σ_i} − χ_σ lie in τ^⊥ for every contributing maximal cone σ_i. The only real issue is notational: the text prints 'τ ≼ σ indicates that τ is a facet of σ', but (4.8) and (4.15) require ≼ to mean 'face' (including τ = σ); with 'face' the statements and proofs are consistent, and the later union over facets in (4.15) is reconciled by closure of the strata. This is a presentation slip, not a mathematical gap.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves Theorem 1.1, which states that for a torus-invariant Cartier divisor D on a normal toric variety X_Σ, the nearby cycles kernel ψ(K_{D,I}) obtained from a family of quantized contact isotopies acts on sheaves by convolution with the twisted polytope sheaf p_!P(D). In the category of constructible sheaves realizing the coherent-constructible correspondence, this action is shown to coincide with the mirror action of the line bundle O(D). The proof combines the GKS quantization of Hamiltonian isotopies with a detailed asymptotic analysis of the derivatives of mollified support functions, proves singular support bounds for the action on shard sheaves, and passes to the full kernel via Čech resolutions. The non-complete fan case is reduced to the complete case by a fan refinement argument.","tokens_in":22404,"tokens_out":16153,"duration_ms":150935,"significance":"The result is a significant contribution to toric mirror symmetry: it provides a purely sheaf-theoretic, Floer-theory-independent realization of the mirror Picard group action as the ε→0 limit of quantized contact isotopies, and it holds for arbitrary normal toric varieties, going beyond the smooth projective hypotheses of prior Floer-theoretic results. The proof is well-structured and detailed: the analytic estimates in Propositions 4.1, 4.7, and 4.11 are explicit and checkable, the singular support bounds are applied carefully, and the passage from shard sheaves to the full kernel via the resolution of C_0 is legitimate. The paper is honest about the scope of its methods, including the remark that the toric stack case is not treated.","major_comments":[],"minor_comments":[{"comment":"The notation 'τ ≼ σ indicates that τ is a facet of σ' is inconsistent with the use of ≼ in (4.8), (4.15), and the proof of Proposition 4.7, where τ ranges over all faces of σ including τ = σ; under the literal definition, a point in the relative interior of σ would not be covered. The authors should change the definition to mean 'face' (including equality) and then reconcile the union in (4.15) by taking closures if needed.","section":"Section 4, after Proposition 4.1"},{"comment":"The step from the limsup bound on the singular support of ~K_{D,ε,1} ∘ C_{Int(σ^∨)} to the singular support bound (4.16) for the nearby cycles is justified by a reference to [NS20, Lem. 3.16]; since this is a delicate point, a sentence explaining why the hypotheses of that lemma hold would be helpful.","section":"Section 4, proof of Proposition 4.13"},{"comment":"The statement of Proposition 3.9 restricts to cohomologically constructible F for the full diagram, but the proof notes that the vertical isomorphisms are defined without any constructibility hypotheses; stating this explicitly in the proposition would avoid confusion about the scope of the theorem that uses it.","section":"Section 3, Proposition 3.9"},{"comment":"There are a few typographical errors that should be corrected: the abstract contains 'to ric mirror symmetry' instead of 'toric mirror symmetry', and in the proof of Proposition 3.2 the expression 'ad^*_{dϕ(ξ)}ξ' appears without a space in the subscript; these do not affect the mathematics.","section":"Throughout"}],"recommendation":"minor_revision","confidential_remarks":"The theorem is convincing and the proof appears correct; my recommendation of minor revision is driven by the notational slip in the definition of ≼ and a few places where the exposition is terse. I do not see any load-bearing error, and I expect the revision to be straightforward."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is the real thing. Bose and Williams prove that the mirror Picard group action in the coherent-constructible correspondence arises as the nearby-cycles limit of quantized contact isotopies — the sheaf-theoretic analog of Hanlon's Floer-theoretic result, and it covers arbitrary normal toric varieties, not just smooth projective ones. Theorem 1.1, the identification of ψ(K_{D,I}) ∘ F with (p_!P(D)) * F, is new; nobody had connected nearby cycles of GKS kernels to twisted polytope sheaves before.\n\nWhat the paper does well: the heavy lifting is in the analytic control of the mollified support-function derivatives (Propositions 4.1, 4.7, 4.11). I traced those estimates — the bounding of the non-convolution term in (4.9), the limsup argument with covectors restricted to the strata, and the uniform bound in Proposition 4.11 — and they are correct. The singular-support arguments in Proposition 4.13 are careful, and the passage from shard sheaves to the full sheaf via the Čech resolution is legitimate. The general machinery on GKS kernels for Lie groups in Section 3 (convolution description, nearby-cycles compatibility) is clean and likely to be independently useful.\n\nSoft spots, in proportion: they are minor. The notation 'τ ≼ σ' is defined as 'τ is a facet of σ', but (4.8) and (4.15) really need it to mean 'face' including τ = σ; the arguments work with the intended reading, so it is a presentation slip, not a gap. The proof of Theorem 4.17's identification of the induced maps on the resolution is slightly compressed — it leans on a stalkwise determination that is fine but could be spelled out. The paper is honest about two places it does not fully cover: Remark 4.20 admits cohomological constructibility is needed for the clean nearby-cycles-of-the-action formulation, only conjecturing the singular weakly-constructible case, and Remark 4.23 explicitly declines to extend to toric stacks. Neither undermines the stated theorems.\n\nThe citation pattern is clean: GKS, FLTZ, Kuwagaki, Zhou, Hanlon/Hicks, KS, NS20 are all used appropriately, with no self-citation leverage.\n\nWho this is for: anyone working on toric mirror symmetry, sheaf quantization, or the CCC. It deserves a serious referee — the analytic estimates in §4 take time but check out, and the referee should focus there. Send it to review; accept after the notation slip is fixed.","headline":"Solid, genuinely new proof that the mirror Picard action in the coherent-constructible correspondence is the nearby-cycles limit of quantized contact isotopies; deserves a serious referee.","tokens_in":22982,"tokens_out":11454,"would_cite":true,"duration_ms":94914,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14M25","14F08","53D10","53D37"],"pacs":[],"model":"deepseek-v4-flash","headline":"The mirror of tensoring by a toric line bundle is the zero-smoothing limit of quantized Hamiltonian flows.","keywords":["coherent-constructible correspondence","toric mirror symmetry","Picard group action","contact isotopy","sheaf quantization","twisted polytope sheaf","nearby cycles","constructible sheaves"],"falsifier":"Run the construction on $\\mathbb{P}^2$ with $D$ a hyperplane: compute both sides of Theorem 1.1 on the skyscraper sheaf at a point of $T^2$ and compare stalks; any mismatch refutes the claim. More surgically, for a fan whose adjacent maximal cones have Cartier data far apart in $\\mathbb{M}_\\mathbb{R}$, numerically sample $d\\varphi_\\varepsilon(\\xi)$ for small $\\varepsilon$ at points approaching the common wall: a value outside $B_{\\varepsilon R}(\\operatorname{Conv}\\{\\chi_\\sigma\\})$ would break Proposition 4.11, the analytic core of the proof.","tokens_in":21904,"feed_emoji":"🪞","tokens_out":14438,"duration_ms":128034,"temperature":0.7,"pith_summary":"This paper shows that the mirror Picard group action on the sheaf side of toric mirror symmetry is not merely an algebraic convolution: it is realized geometrically as the zero-smoothing limit of quantized Hamiltonian flows. Starting from a torus-invariant Cartier divisor $D$ on a normal toric variety $X_\\Sigma$, the authors mollify the piecewise-linear support function $\\varphi_D$, form the GKS kernels (sheaf quantizations of Hamiltonian isotopies) of the resulting homogeneous Hamiltonians on $T^*T^n$, and take their nearby-cycles limit. The main theorem states that this limiting kernel $\\psi(K_{D,I})$ acts on any sheaf $F$ exactly as convolution with the twisted polytope sheaf $p_!P(D)$, the established mirror of $\\mathcal{O}(D)$; on the category $\\mathrm{Sh}^c_{\\Lambda_\\Sigma}(T^n)$ this makes $\\psi(K_{D,I})\\circ F$ isomorphic to $A(\\mathcal{O}(D)\\otimes B(F))$. This provides a sheaf-theoretic counterpart to the Hamiltonian-flow description of the mirror Picard group known in the symplectic setting, and it holds for every normal toric variety, with no smoothness or projectivity assumption.","feed_headline":"Mirror line-bundle twists are limits of quantized Hamiltonian flows","feed_subtitle":"Nearby cycles of smoothed support-function flows reproduce the mirror Picard action for all normal toric varieties.","key_machinery":"The proof combines two mechanisms. The first is a general fact about GKS kernels on a Lie group $G$: a kernel quantizing a $G\\times G$-invariant Hamiltonian acts on sheaves by group convolution, $K\\circ F \\cong P*F$ with $P=K\\circ C_e$, and this action commutes with nearby cycles under a compactness bound on the derivatives of the Hamiltonian (Propositions 3.5 and 3.9). The second is a convex-geometric analysis of the homogenized smoothing $\\varphi_\\varepsilon(\\xi)=\\|\\xi\\|(\\eta_\\varepsilon*\\varphi_D)(\\hat{\\xi})$ of the support function: as $\\varepsilon\\to 0$ the derivatives $d\\varphi_\\varepsilon$ remain in an $\\varepsilon R$-neighborhood of $\\operatorname{Conv}(\\{\\chi_\\sigma\\})$, and their limit points over each cone face $\\tau\\preceq\\sigma$ lie in $(\\chi_\\sigma,0)+\\tau^\\perp\\times \\dot{\\tau}$ (Propositions 4.1, 4.7, 4.11). These bounds control the singular support of the nearby-cycles kernel, allowing a comparison (Proposition 4.13) between its action on the shard sheaves $\\mathbb{C}_{\\mathrm{Int}(\\sigma^\\vee)}$ and the \\v{C}ech summands $\\mathbb{C}_{\\mathrm{Int}(\\chi_\\sigma+\\sigma^\\vee)}$ of the twisted polytope sheaf $P(D)$; assembling these comparisons through the equivariant-to-nonequivariant reduction $p_!$ yields Theorem 1.1.","core_discovery":"Theorem 1.1 asserts that for every torus-invariant Cartier divisor $D$ on a normal toric variety $X_\\Sigma$, the nearby-cycles kernel $\\psi(K_{D,I})$ satisfies $\\psi(K_{D,I})\\circ F \\cong (p_!P(D))*F$ for all $F\\in \\mathrm{Sh}(T^n)$, where $P(D)$ is the twisted polytope sheaf mirror to $\\mathcal{O}(D)$ and $*$ is convolution on the torus. In the subcategory $\\mathrm{Sh}^c_{\\Lambda_\\Sigma}(T^n)$ identified by the coherent-constructible correspondence with $\\mathrm{Coh}(X_\\Sigma)$, this becomes the autoequivalence $F\\mapsto A(\\mathcal{O}(D)\\otimes B(F))$. The substance of the result is that the $\\varepsilon\\to 0$ limit of the quantized flows exists as a genuine sheaf even though the limiting Hamiltonian---the pullback of $\\varphi_D$---is not smooth and has no Hamiltonian flow of its own; that limit coincides with the convolution action of $p_!P(D)$. Thus the mirror of tensoring by a line bundle is a limiting contact isotopy, giving the Picard action a direct symplectic-geometric mechanism on the A-side.","pith_inferences":["The mollifier-independence suggested by the estimates points to a general construction: any piecewise-linear pullback Hamiltonian may acquire a canonical 'limiting flow' via nearby cycles of quantized smoothings, extending the method beyond the toric Cartier case.","Because the proof passes first through an equivariant statement on $\\mathbb{R}^n$, it should extend to the toric stacks covered by the nonequivariant CCC, as the authors anticipate in Remark 4.23, yielding the same geometric Picard action for stacky fans.","For compact Lie groups, the convolution description of bi-invariant-norm GKS kernels hints at explicit central convolution operators acting on equivariant sheaf categories, a direction the paper does not pursue."],"forward_implications":["The mirror action of $\\mathcal{O}(D)$ on constructible sheaves is a genuine limiting Hamiltonian flow, giving line-bundle twists a symplectic-geometric meaning rather than only a combinatorial convolution formula.","The isomorphism $\\psi(K_{D,I})\\circ F\\cong (p_!P(D))*F$ holds for every sheaf $F$ on $T^n$, so the geometric description is not confined to the constructible CCC category.","The theorem applies to arbitrary normal toric varieties, including singular and non-projective ones where the analogous Floer-theoretic statements are not available.","For a compact Lie group $G$, the same GKS mechanism expresses the action of a bi-invariant norm Hamiltonian as convolution with a kernel supported on the geodesic flow (Propositions 3.2 and 3.5), an interesting case in its own right."],"supporting_citations":[{"why":"Quantizes homogeneous Hamiltonian isotopies into the kernels $K_{D,I}$ and gives the singular-support control used across the proof.","marker":"[GKS12]"},{"why":"Establishes the equivariant CCC with its convolution–tensor compatibility, used to identify $\\psi(\\widetilde K_{D,I})\\circ F$ with $P(D)*F$.","marker":"[FLTZ11]"},{"why":"Supplies the nonequivariant CCC, the functors $A,B$, and the base-change lemmas that reduce the equivariant statement to $T^n$.","marker":"[Kuw20]"},{"why":"Introduces the twisted polytope sheaves $P(D)$ and their shard/\\v{C}ech resolutions, the target of the kernel action.","marker":"[Zho19]"},{"why":"Provides the microlocal foundations—singular supports, base change, pullback formulas—on which the GKS and bounds arguments rely.","marker":"[KS94]"},{"why":"Gives the singular-support bound for nearby cycles used to control $\\dot{\\mathrm{ss}}(\\psi(\\widetilde K_{D,I}\\circ \\mathbb{C}_{\\mathrm{Int}(\\sigma^\\vee)}))$.","marker":"[NS20]"},{"why":"Supplies the toric-geometry dictionary between Cartier divisors, Cartier data, and support functions used to define $\\varphi_D$ and $\\chi_\\sigma$.","marker":"[CLS11]"},{"why":"Shows $p_!$ intertwines the forgetful pullback with the nonequivariant CCC, entering through diagram (2.4) in the proof of Theorem 4.18.","marker":"[Tre10]"}],"fun_headline_variants":["Line-bundle twists as limits of quantized flows","Quantized flows realize mirror Picard action","Nearby cycles give sheaf-theoretic Picard action","Toric mirror symmetry via limiting contact isotopies"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the gradients of the smoothed support functions remain uniformly within a shrinking neighborhood of the convex hull of the Cartier data as the smoothing radius tends to zero, with limits confined to the expected cone strata; if that analytic control failed near a fan wall, the nearby-cycles kernel would stop agreeing with the twisted-polytope convolution.","fun_headline_variants_meta":{"raw":{"variants":["Line-bundle twists as limits of quantized flows","Quantized flows realize mirror Picard action","Nearby cycles give sheaf-theoretic Picard action","Toric mirror symmetry via limiting contact isotopies"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000365,"raw_usage":{"total_tokens":1999,"prompt_tokens":1011,"completion_tokens":988,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":627,"completion_tokens_details":{"reasoning_tokens":927}},"tokens_in":627,"tokens_out":988,"duration_ms":7520,"temperature":1.0,"reasoning_tokens":927,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T23:16:29.236530+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the construction on $\\mathbb{P}^2$ with $D$ a hyperplane: compute both sides of Theorem 1.1 on the skyscraper sheaf at a point of $T^2$ and compare stalks; any mismatch refutes the claim. More surgically, for a fan whose adjacent maximal cones have Cartier data far apart in $\\mathbb{M}_\\mathbb{R}$, numerically sample $d\\varphi_\\varepsilon(\\xi)$ for small $\\varepsilon$ at points approaching the common wall: a value outside $B_{\\varepsilon R}(\\operatorname{Conv}\\{\\chi_\\sigma\\})$ would break Proposition 4.11, the analytic core of the proof.","supporting_citations":[],"review_version":1}