{"id":"97cf1211-0c22-4172-b46e-bad552335e14","arxiv_id":"2412.15062","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper refines Zhu's proof of the Pappas-Rapoport coherence conjecture from an equality of dimensions to isomorphisms of representations, with applications to affine Demazure modules.","lead":"This paper upgrades a known dimension equality between spaces of sections on two kinds of Schubert varieties to an isomorphism that also matches group actions. It is a refinement of the Pappas-Rapoport coherence conjecture proved by Zhu, and it leads to new relations among affine Demazure modules, core objects in representation theory.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The fiberwise T′_τ-isomorphism in Eq. (38) rests on an unproved flatness assertion for the direct image sheaf, which is not a consequence of the cited flatness of the total space.","rationale":"The reader's weakest assumption identifies the same load-bearing point: the unproved flatness of H^0(Gr^μ_{G,C}, L_ξ) and the resulting fiberwise T′_τ-module isomorphism (38). I agree that this is the most serious technical gap, since the entire refinement from dimension equality to representation isomorphism passes through it. The later G′_ϑ-upgrade in characteristic zero is not itself a separate obstruction: equality of T′_τ-characters does determine a finite-dimensional representation of a reductive group in characteristic zero, so that step is reasonable once a T′_τ-equivariant isomorphism exists. The reduction to non-simply-connected groups via an analogue of [HY, Lemma 5.16] is also asserted without proof, but it is secondary to the base-change gap. The paper is otherwise well structured, makes good use of known equivariant line-bundle techniques, and the Demazure-module applications are interesting; the concern is about a missing proof step, not about correctness of the overall strategy. Hence the existing CONDITIONAL verdict is appropriate, and I would not change it.","tokens_in":42910,"tokens_out":9908,"duration_ms":77531,"concrete_test":"Re-derive (38) via Grauert's cohomology and base change theorem for π: Gr^μ_{G,C}[ξ]→C and F = L_ξ. Verify H^i(Gr^μ_{G,a}, F|_a) = 0 for all i>0 and all k-points a, and confirm h^0(Gr^μ_{G,a}, F|_a) is independent of a. If these conditions hold—for instance by combining Frobenius splitting of global Schubert varieties with Zhu's dimension equality—then π_*F is locally free and (38) is justified. If they cannot be supplied from existing results, the proof of Theorem 3.5(2) is incomplete at its key base-change step.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equation (38) is the hinge of the proof of Theorem 3.5. It asserts an isomorphism of T′_τ-modules between H^0(A_Y(μ), L(Λ)_κ) and H^0(Gr_{G,μ}, L^c), and the paper derives it by saying that H^0(Gr^μ_{G,C}, L_ξ) is a T′_τ-module and flat over C. This flatness is stated without proof in §3.4, and it does not follow from Theorem 3.22 (flatness of the total space Gr^μ_{G,C} over C) together with Lemma 3.24 (T′_τ×C-equivariance of L_ξ). If the direct image were known to be flat, then its weight decomposition under T′_τ would be a direct sum of locally free sheaves on C, giving equal weight-space dimensions on all fibers and hence (38); but that is precisely the missing base-change input. The standard way to obtain it is cohomology and base change: one must show R^iπ_*L_ξ = 0 for i>0 and that h^0(Gr^μ_{G,a}, L_ξ|_a) is constant over C. Zhu's theorem supplies equality of total dimensions only between the special fiber and a generic fiber, and the paper does not cite vanishing of higher cohomology for all fibers. Without this step, the central upgrade from Zhu's dimension equality to an equivariant isomorphism is not established. The separate reduction to non-simply-connected G via an unproved analogue of [HY, Lemma 5.16] is a second, smaller gap.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper refines Zhu's proof of the Pappas--Rapoport coherence conjecture from a dimension equality to an isomorphism of representations. The main geometric comparison, Theorem 3.5, asserts that for a dominant coweight μ and an ample line bundle L(Λ)_κ on the relevant component of the partial affine flag variety, H⁰(A_Y(μ),L(Λ)_κ) is isomorphic as a T'τ-module to H⁰(Gr_{G,μ},L^c)⊗k_{-λ}, and, in characteristic zero, as a G'_ϑ-module. The proof is built on a parahoric Bruhat--Tits group scheme ℤ over the projective line ramified at 0, Faltings-style equivariant structures on line bundles, and transport of equivariance through the global affine Grassmannian. Section 4 claims a canonical L^+G-equivariant structure on any line bundle over a global Schubert variety, and Section 5 translates the comparison into an isomorphism of affine Demazure modules D(c,μ)⊗C_λ ≅ ∑_{w∈Wτ}D_τ(Λ,w(μ)), with explicit examples in the appendix.","tokens_in":43246,"tokens_out":9059,"duration_ms":83925,"significance":"The claimed result is a genuine strengthening of a theorem with substantial arithmetic impact. If Eq. (38) and the surrounding base-change assertions are justified, the upgrade from a dimension equality to an equivariant isomorphism is meaningful, and the Demazure-module consequence in Theorem 5.5 is new and clearly formulated, with helpful examples. The paper also contributes independent constructions: the equivariant realization of parahoric group schemes beyond the simply-connected case (Theorem 2.4), the canonical Levi subgroup description (Corollary 2.10), and the uniqueness theorem for L^+G-equivariant structures (Theorem 4.6). The latter result is presented with a clean fppf central-extension argument. However, the central proof currently relies on an unproved flatness assertion for a direct image sheaf, and the reduction to general G is supported by an unproved analogue of a lemma from the authors' earlier work. Those gaps are load-bearing rather than cosmetic.","major_comments":[{"comment":"The proof of Theorem 3.5 asserts, without proof, that H⁰(Gr^μ_{G,C}, L_ξ) is a T'τ-module and flat over C. This does not follow from Theorem 3.22, which only says that the total space Gr^μ_{G,C} is flat over C, together with Lemma 3.24, which supplies a T'τ × C-equivariant structure on L_ξ. Flatness of the direct image is a base-change statement: one needs, for example, vanishing of R^iπ_*L_ξ for i>0 and constancy of h⁰(Gr^μ_{G,a}, L_ξ|_a) on fibers. Complete reducibility of T'τ only converts such flatness into a fiberwise weight-space isomorphism; it cannot supply the missing base-change input. Since Eq. (38) is the hinge of the proof of Theorem 3.5, this gap must be repaired by a proof or a precise reference.","section":"§3.4, Eq. (38)"},{"comment":"When the proof passes from simply-connected G to a general simple group, it invokes the statement 'By a similar proof as in [HY, Lemma 5.16], there is an isomorphism Gr_{G',C} ≅ Gr_{G,C}[o]' but does not prove it or give a reference for the analogous lemma. This isomorphism is used to define L_ξ for non-simply-connected G, to identify restrictions to the fiber over 0, and to compare equivariant structures. The non-simply-connected case of Theorem 3.5 is therefore not supported unless this missing lemma is supplied.","section":"§3.4, reduction to general G"},{"comment":"In the injectivity part of Proposition 4.4, the proof states that since Gr_G is ind-projective over Ā¹ and each fiber has one-dimensional global sections, π_*(L) is a locally free sheaf of rank one on Ā¹. This is not immediate for an ind-scheme; one needs to work with the filtration by finite-type Schubert varieties and justify that the relevant direct image sheaf is defined and that base change to each fiber behaves as claimed. Since Proposition 4.4 is used in the uniqueness part of Theorem 4.6, this point needs clarification.","section":"§4.2, Proposition 4.4"}],"minor_comments":[{"comment":"The abstract contains a typo: 'coherence conjcture' should read 'coherence conjecture'.","section":"Abstract"},{"comment":"The subgroups U_{α,ℓ} are used in Definition 2.1 before the cases with 2α∈R are defined in the following paragraph; moving that explanation before Definition 2.1 would improve readability.","section":"§2.2, Definition 2.1"},{"comment":"The element n_μ defined in Eq. (19) depends a priori on a choice of lift μ of μ̄; the text should state explicitly that the construction is independent of this choice or fix a convention for the lift.","section":"§3.1, Eq. (19)"},{"comment":"The phrase 'flat over C' for H⁰(Gr^μ_{G,C},L_ξ) should specify the scheme structure and the direct image functor being used; as written it is not clear whether the claim is about a coherent sheaf on C or about a vector space with a T'τ-action.","section":"§3.4, proof of Theorem 3.5"},{"comment":"The examples in Appendix A refer to 'Figure 1a', 'Figure 2b', and similar diagrams that are not reproduced in the text; adding a short description of the weight diagrams would make the examples more self-contained.","section":"Appendix A"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely within scope for a journal in algebraic geometry. I agree with the reader's assessment: the main technical gap is the unproved flatness of the direct image sheaf used in Eq. (38), and the reduction to non-simply-connected G relies on an unproved analogue of [HY, Lemma 5.16]. Both issues appear repairable, possibly using the Frobenius-splitting techniques already cited in the paper, and the rest of the manuscript is plausible. I would not recommend rejection, but the current version is not publishable as is."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The short version: this paper upgrades Zhu's dimension equality in the coherence conjecture to an isomorphism of representations, and derives new identities among affine Demazure modules. If the main technical step holds, it's a substantial result. The authors also prove uniqueness of L^+G-equivariant structures on line bundles over global Schubert varieties, which is new and likely useful.\n\nThe paper is well organized and carefully written. The use of the parahoric Bruhat-Tits group scheme follows Zhu closely, and the extension to general G and twisted loop groups is genuinely attempted. The Demazure module consequences in Section 5 follow cleanly from the comparison theorem plus Kumar-Mathieu, and the appendix examples are helpful.\n\nThe soft spot is in Section 3.4, proof of Theorem 3.5. Equation (38) needs an isomorphism of T'_τ-modules between fibers of H^0(Gr^μ_{G,C}, L_ξ). The paper asserts that this space is flat over C and then concludes via complete reducibility of T'_τ. That flatness is stated without proof. It does not follow from Zhu's flatness of the total space (Theorem 3.22) plus equivariance of the line bundle. You need a base-change theorem: either vanishing of higher direct images and constancy of h^0, or something equivalent. Zhu only gives equality of total dimensions between special and generic fibers, which is weaker. This is not a nitpick; the missing step is the hinge that turns dimension equality into a representation isomorphism.\n\nThere is a second, smaller gap: the reduction to non-simply-connected G uses an analogue of [HY, Lemma 5.16] (an isomorphism Gr_{G',C} ≃ Gr_{G,C}[o]) that is left implicit. It is probably fixable, but it should be written down. The characteristic-zero upgrade from T'_τ-modules to G'_ϑ-modules is standard once the characters match, so that part is fine.\n\nI would not desk-reject this. The ideas are good, the statements are precise, and most of the argument is careful. I would send it to a referee with a specific request to check the flatness/base-change claim. If the authors can supply that missing argument, the paper is strong; as it stands, the central proof is incomplete. No fitting, no circularity, and no invented entities—just one gap that needs to be closed.","headline":"Real upgrade of Zhu's theorem, but the proof of Theorem 3.5 has an unproved flatness step that is load-bearing.","tokens_in":43746,"tokens_out":3170,"would_cite":false,"duration_ms":27792,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14M15","14L15","17B67"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper refines the coherence conjecture by upgrading its equality of dimensions to an isomorphism of representations of the canonical Levi subgroup, and derives an isomorphism of affine Demazure modules as a consequence.","keywords":["coherence conjecture","affine Grassmannian","partial affine flag variety","parahoric group schemes","affine Demazure modules","equivariant line bundles","global Schubert varieties","affine Lie algebras"],"falsifier":"A concrete check is to compute the $T^{\\tau'}$-weight multiplicities of both sides of the isomorphism in a small example, such as $G=SL_2$ with $Y=\\{1\\}$ and $\\mu=\\check{\\alpha}$, in a prime characteristic $p$ satisfying $p\\nmid a_Y$; if the dimensions coincide but the characters differ, the specialization isomorphism (Equation (38)) is false. Equivalently, computing $H^1$ of the line bundle over the global Schubert variety at the special fiber and finding a nonzero class would falsify the flatness of the direct image on which the module comparison rests.","tokens_in":42676,"feed_emoji":"🔁","tokens_out":16567,"duration_ms":114956,"temperature":0.7,"pith_summary":"The coherence conjecture asserts that the space of global sections of a line bundle over a Schubert variety in the affine Grassmannian has the same dimension as the space of sections of a related line bundle over a union of Schubert varieties in a partial affine flag variety. This paper upgrades that equality of numbers to an actual isomorphism of representations: in characteristic zero, the two section spaces are isomorphic as modules over the canonical Levi subgroup, and under mild characteristic assumptions the torus action is already equivariant. In representation-theoretic terms, the affine Demazure module $D(c,\\mu)$, twisted by the one-dimensional character $\\mathbb{C}_\\lambda$, is isomorphic to the sum $\\sum_{w\\in W^\\tau} D^\\tau(\\Lambda, w(\\mu))$ of twisted affine Demazure modules. The paper also shows that every line bundle on the global affine Grassmannian of the ramified parahoric group scheme carries a unique equivariant structure under the global jet group scheme, strengthening the foundations for the comparison.","feed_headline":"A dimension equality becomes an isomorphism of representations","feed_subtitle":"The coherence theorem is refined: the same section spaces match as modules over the Levi subgroup, not just dimensions.","key_machinery":"The central mechanism is a parahoric Bruhat–Tits group scheme over $\\mathbb{P}^1$ ramified at 0, whose global affine Grassmannian has generic fiber the ordinary affine Grassmannian and special fiber the partial affine flag variety. The paper constructs a line bundle on this global Grassmannian by descent from the moduli stack of $\\mathcal{G}$-bundles, so that it specializes to $\\mathcal{L}(\\Lambda)_\\kappa$ at 0 and to $\\mathcal{L}^c$ at a generic point, and then equips it with an equivariant structure under the negative loop group; restricting to the canonical Levi subgroup gives the same action on every fiber. The equivariant realization of parahoric group schemes as $\\sigma$-fixed points provides the canonical Levi subgroup and the subgroup $G_\\vartheta$, and a central-extension splitting criterion extends the equivariant structure from a generic open subset to the whole base curve. In characteristic zero, complete reducibility of the torus turns the fiberwise identifications into an isomorphism of representations of $G'_\\vartheta$.","core_discovery":"The paper's central claim is that the coherence conjecture's dimension identity is the shadow of an equivariant isomorphism: for a dominant coweight $\\mu$ and a line bundle $\\mathcal{L}(\\Lambda)_\\kappa$ of central charge $c$ on the relevant component of the partial affine flag variety, there is an isomorphism of $G'_\\vartheta$-representations $H^0(A_Y(\\mu), \\mathcal{L}(\\Lambda)_\\kappa) \\simeq H^0(Gr_{G,\\mu}, \\mathcal{L}^c) \\otimes k_{-\\lambda}$ in characteristic zero, and of $T^{\\tau'}$-modules under the mild assumptions $p \\nmid a_Y$ and $p \\nmid |X_*(T)/\\check{Q}|$, where $G'_\\vartheta$ is the preimage in the simply connected cover of the centralizer of a rational interior point $\\vartheta$ of the facet. Equivalently, the paper proves the affine Demazure module isomorphism $D(c,\\mu) \\otimes \\mathbb{C}_\\lambda \\simeq \\sum_{w\\in W^\\tau} D^\\tau(\\Lambda, w(\\mu))$ as modules over $\\mathfrak{g}_\\lambda$, the Lie algebra of $G_\\vartheta$. The equality of dimensions proved earlier is therefore upgraded to a genuine identification of representations, matching the Levi subgroup action on the two sides, and this is what produces the Demazure-module consequence.","pith_inferences":["A natural extension would be to ask whether the equivariant isomorphism lifts to a derived equivalence between the pushforwards of the two line bundles over the global Schubert variety; the paper's flatness premise would then be replaced by a full base-change theorem.","The uniqueness of the global jet group equivariant structure suggests that equivariant Picard groups and equivariant $K$-theory of partial affine flag varieties can be computed from the special fiber, paralleling standard localization formulas; the authors mention $K$-theory as future work.","One might test whether the Demazure-module isomorphism admits an integral or $q$-deformed version, since both sides carry natural standard monomial bases; the paper proves it in characteristic zero over algebraically closed fields."],"forward_implications":["In characteristic zero, $H^0(A_Y(\\mu), \\mathcal{L}(\\Lambda)_\\kappa) \\simeq H^0(Gr_{G,\\mu}, \\mathcal{L}^c) \\otimes k_{-\\lambda}$ as $G'_\\vartheta$-modules, and under mild characteristic assumptions as $T^{\\tau'}$-modules.","The affine Demazure modules satisfy $D(c,\\mu) \\otimes \\mathbb{C}_\\lambda \\simeq \\sum_{w\\in W^\\tau} D^\\tau(\\Lambda, w(\\mu))$ as $\\mathfrak{g}_\\lambda$-modules, so the coherence conjecture's character identity is a consequence of an actual module isomorphism.","Every line bundle on the global affine Grassmannian $\\mathrm{Gr}_{\\mathcal{G}}$ and its restriction to a global Schubert variety admits a unique equivariant structure under the global jet group scheme, so line bundles on $\\mathrm{Gr}_{\\mathcal{G}}$ are classified by their restrictions to the special fiber.","The irreducible components of $A_Y(\\mu)$ are exactly the Schubert varieties $\\mathrm{Fl}_{Y,\\eta}$ for $\\eta$ in the set $S_{Y,\\mu}$ of $\\breve{Y}$-dominant elements in the $W^\\tau$-orbit of $\\mu$.","The comparison is valid for non-simply-connected groups and for twisted loop groups, not only for the simply-connected untwisted case."],"supporting_citations":[{"why":"Supplies the earlier proof of the coherence conjecture and the flatness of the global Schubert variety that the comparison builds on.","marker":"[Zh1]"},{"why":"States the coherence conjecture and gives the Picard-group description of partial affine flag varieties used to parametrize line bundles.","marker":"[PR]"},{"why":"Provides the method for constructing unique opposite-parahoric and loop-group equivariant structures on line bundles.","marker":"[Fa]"},{"why":"Gives the equivariant construction of parahoric group schemes in the simply-connected case, extended here to general groups.","marker":"[DH1]"},{"why":"Prior work of the authors giving the special case of the global jet equivariance and the reduction lemma used for general groups.","marker":"[HY]"},{"why":"Identifies the dual of section spaces over Schubert varieties with affine Demazure modules.","marker":"[Ku1]"},{"why":"Companion identification for affine flag varieties used in the affine Demazure module statement.","marker":"[Ma]"},{"why":"Identifies the dual of sections over the union $A_Y(\\mu)$ with the sum of Demazure modules used in the reformulation.","marker":"[Li]"}],"fun_headline_variants":["Dimension equality becomes representation isomorphism","Coherence conjecture refined: dimension equality upgraded","Affine Demazure modules matched via equivariant isomorphism","Pappas-Rapoport coherence: from dimensions to modules"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the direct image of the line bundle over the global Schubert variety is flat over the base curve, with all higher direct images vanishing, so that the section spaces over the two fibers are isomorphic as modules under specialization; the paper asserts this flatness rather than proving the needed base-change theorem, and the reduction to non-simply-connected groups also assumes an analogue of a lemma from the authors' earlier work.","fun_headline_variants_meta":{"raw":{"variants":["Dimension equality becomes representation isomorphism","Coherence conjecture refined: dimension equality upgraded","Affine Demazure modules matched via equivariant isomorphism","Pappas-Rapoport coherence: from dimensions to modules"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000166,"raw_usage":{"total_tokens":1272,"prompt_tokens":981,"completion_tokens":291,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":597,"completion_tokens_details":{"reasoning_tokens":233}},"tokens_in":597,"tokens_out":291,"duration_ms":2981,"temperature":1.0,"reasoning_tokens":233,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T11:40:17.966142+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete check is to compute the $T^{\\tau'}$-weight multiplicities of both sides of the isomorphism in a small example, such as $G=SL_2$ with $Y=\\{1\\}$ and $\\mu=\\check{\\alpha}$, in a prime characteristic $p$ satisfying $p\\nmid a_Y$; if the dimensions coincide but the characters differ, the specialization isomorphism (Equation (38)) is false. Equivalently, computing $H^1$ of the line bundle over the global Schubert variety at the special fiber and finding a nonzero class would falsify the flatness of the direct image on which the module comparison rests.","supporting_citations":[],"review_version":1}