{"id":"505b95d2-1836-4241-ac91-a636965488ec","arxiv_id":"2505.10724","paper_version":1,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For GL_n over unramified p-adic fields, the paper proves the strongly generic part of the categorical local Langlands conjecture with F_l coefficients and derives Harris-Viehmann type identities and torsion vanishing for type A Shimura varieties.","lead":"This paper proves a torsion-coefficient version of the categorical local Langlands conjecture for GL_n over p-adic fields in the strongly generic case, and uses it to describe the cohomology of local Shimura varieties and prove new torsion vanishing results. It extends the author's earlier l-adic work to mod l coefficients.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The F_ℓ shift in Theorem 5.2 is fixed only by an unproved torsion-freeness assertion in Proposition 5.4; without it the modular categorical equivalence is not established.","rationale":"The reader's weakest_assumption identifies exactly the same step: the modular supercuspidal argument in Section 5.2 requires a compact type-A Shimura variety for which the localized Z_ℓ-cohomology is concentrated and torsion free, and the proof of Proposition 5.4 does not establish this. I agree that this is the most load-bearing point, because it is the only place where the unknown shift in Theorem 5.2 is determined, and Theorem 5.2 is the base case for the induction in Proposition 7.1 and hence for Theorem 7.6. I am not moving the verdict because the paper is already CONDITIONAL and the gap is potentially repairable by supplying the missing torsion-freeness argument or by a different globalization. The rest of the paper is structurally coherent, and I found no internal contradiction or circular use of the main conjecture. The agreement is complete rather than partial because the reader's stated weakest assumption is essentially my concern, though I would emphasize even more strongly that the proof does not currently contain the torsion-freeness argument at all, and that the shift cannot be recovered from the two characteristic-specific computations alone.","tokens_in":46357,"tokens_out":10148,"duration_ms":111756,"concrete_test":"Run the universal-coefficient spectral sequence for RΓ_c(Sh_Kp, Z_ℓ)[φ] in the smallest non-trivial case, n = 2, ℓ = 3, p = 5, using Mantovan's formula from [HL23, Theorem 1.12] and the characteristic-zero computation of [Shi12, Theorem 6.7]. Explicitly test whether the Tor_1^{Z_ℓ}(H^{i+1}(RΓ_c(Sh_Kp, Z_ℓ)[φ]), F_ℓ) terms vanish for i = h-2, h-1, h. If any is nonzero, then mod-ℓ concentration can coexist with integral torsion and the lift to characteristic 0 used to fix the shift in Theorem 5.2 is invalid; if they all vanish, the omitted argument should be written out and the existing proof completed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The decisive step is the transfer from Q_ℓ to F_ℓ for supercuspidal parameters, specifically the determination of the unknown shift in Theorem 5.2(2). Proposition 5.4 claims that RΓ_c(Sh_Kp, F_ℓ)[φ] is concentrated in degree h = dim Sh_Kp and then concludes that RΓ_c(Sh_Kp, Z_ℓ)[φ] is concentrated and torsion free, so that the mod-ℓ complex is the reduction of a characteristic-zero object whose degree is known. The proof only shows concentration in one degree up to the yet-unknown shift, and the torsion-freeness of the Z_ℓ-complex is asserted without an argument. Concentration of the mod-ℓ cohomology and knowledge of the Q_ℓ-cohomology do not by themselves determine the F_ℓ shift: one needs a universal-coefficient spectral sequence controlling the Tor_1 contribution from integral torsion, and no such control is provided. This is exactly the point where the unknown shift in equation (6) of Section 5.1 is fixed. If Proposition 5.4 fails, Theorem 7.6 for F_ℓ coefficients is not proved for the supercuspidal case, and the subsequent torsion applications in Section 8 lose their foundation. The paper is honest about many transfers from [Ngu23a], but this particular gap is not flagged.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript extends the author's earlier work [Ngu23a] from Q_ℓ to F_ℓ coefficients for the categorical local Langlands conjecture for G = Res_{F/Q_p} GL_{n,F}. For an L-parameter φ whose corresponding WF-representation is a direct sum of pairwise disjoint irreducibles with no unramified twist identifications (condition (A1) in Section 6.1), it describes the connected component [C_φ] of the stack of L-parameters as a quotient stack, constructs a block decomposition of D_lis(Bun_G, Λ)^ω, and proves an equivalence IndPerf^{bqc}_{Coh}([C_φ]) ≃ D_lis^{[C_φ]}(Bun_G,Λ)^ω together with an explicit spectral action formula. The paper also derives a Harris–Viehmann statement (Theorem 6.7) and torsion vanishing results for type-A Shimura varieties (Theorems 8.1, 8.6, and Corollary 8.10).","tokens_in":46581,"tokens_out":8621,"duration_ms":77518,"significance":"If the proof is completed, Theorem 7.6 would be a significant step: it verifies the strongly generic case of the Fargues–Scholze categorical local Langlands conjecture for GL_n with F_ℓ and Q_ℓ coefficients and would give new torsion vanishing results for Shimura varieties. The paper's precise formulation of condition (A1), the explicit description of the spectral component [C_φ] in Proposition 3.2, and the construction of the Hecke eigensheaf G_φ in Theorem 6.6 are valuable and clearly stated. However, the central modular shift is currently fixed by an incomplete argument in Proposition 5.4, and several key reductions are delegated to an unpublished preprint; these issues must be resolved before the main theorem can be regarded as established.","major_comments":[{"comment":"This is a comment.","section":"§5.2.2, Proposition 5.4"},{"comment":"This is a comment.","section":"§7.1, Theorem 7.6 and Proposition 7.4"},{"comment":"This is a comment.","section":"§6.3, Theorem 6.5 and §7.1, Proposition 7.1"}],"minor_comments":[{"comment":"This is a minor typo.","section":"§5.2.2"},{"comment":"This is a minor typo.","section":"§6.4.2"},{"comment":"This is a clarity issue.","section":"§8.2, Proposition 8.4"},{"comment":"This is a reference consistency issue.","section":"§5.3 (Mantovan's formula)"},{"comment":"This is a presentation issue.","section":"§5.1, Step 4"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the main theorem for F_ℓ coefficients hinges on Proposition 5.4, whose proof appears incomplete, and on a sequence of 'same arguments as in [Ngu23a]' claims. Since [Ngu23a] is an unpublished preprint, it would be prudent to request that the authors either supply full proofs for the F_ℓ adaptations or state Theorem 7.6 and its applications as conditional on [Ngu23a]. The manuscript is otherwise carefully written, and the results, if valid, are well within the journal's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the short version: this is a serious, readable paper that extends the author's Q_ℓ categorical local Langlands work to F_ℓ for strongly generic parameters and gets torsion vanishing results for type A Shimura varieties as applications. The novelty is real—the F_ℓ supercuspidal step needs a global argument, not just a translation—and the paper is honest about what is taken from [Ngu23a] and where it overlaps with Zou [Zou25]. I found no circularity.\n\nThe main soft spot is Proposition 5.4. It is supposed to pin down the unknown shift in Theorem 5.2 by analyzing the supercuspidal part of the cohomology of a Kottwitz–Harris–Taylor Shimura variety. The proof shows the F_ℓ complex is concentrated in one degree, but then immediately asserts the Z_ℓ complex is concentrated and torsion free. That does not follow without a universal coefficient argument. The mod ℓ cohomology being a single degree plus the Q_ℓ cohomology being a single degree does not tell you the integral complex is torsion free, and it does not tell you the common degree is h. The paper needs to either prove the torsion-freeness directly (or from known results on these Shimura varieties, citing them) or add a spectral sequence controlling Tor_1. Since this step is the only place the shift is fixed, the F_ℓ categorical equivalence for supercuspidals, and everything downstream in Section 8, is conditional on it.\n\nThe other soft spots are minor: several load-bearing results are delegated to [Ngu23a] with 'the same arguments still work,' which makes the paper hard to check without the companion preprint; and Remark 8.9 leaves open whether the π_ξ are pairwise non-isomorphic, which the author flags.\n\nBottom line: if the torsion-freeness assertion in Proposition 5.4 can be justified, this is a major result. As it stands, the paper should be sent to referees who can check the arithmetic geometry and the lifting step. This is a conditional accept, not a reject.","headline":"Serious and mostly coherent extension of the Q_ℓ categorical Langlands work to F_ℓ, but the shift-fixing step rests on an unproved torsion-freeness assertion and needs a referee who can check it.","tokens_in":47180,"tokens_out":4423,"would_cite":true,"duration_ms":44973,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11S37"],"pacs":[],"model":"deepseek-v4-flash","headline":"For L-parameters whose irreducible pieces are pairwise disjoint up to unramified twist, the paper proves a torsion-coefficient version of the categorical local Langlands conjecture for GL_n, computing the spectral action on the Whittaker…","keywords":["categorical local Langlands","spectral action","torsion coefficients","Shimura varieties","Harris-Viehmann conjecture","modular representations of GL_n","L-parameters","Bun_G"],"falsifier":"Take a small explicit type-A Shimura variety at a non-banal prime $\\ell$ (for instance a $\\mathrm{GU}(1,n-1)$ case), localize its mod-$\\ell$ cohomology at an $L$-parameter $\\varphi$ satisfying (A1), and check whether classes appear outside the asserted degrees: outside degree $d$ when the variety is compact or $\\varphi$ is irreducible, or outside the range $[0,d]$ in general. A single such class would falsify Theorem 8.1. Alternatively, test the identity $C_{\\chi_i}\\star F_\\chi \\simeq F_{\\chi\\otimes\\chi_i}$ from Theorem 6.5 directly by computing the restriction of $C_{\\chi_i}\\star F_\\chi$ to the stratum of $b_{\\chi\\otimes\\chi_i}$ and checking that all other stratum restrictions vanish.","tokens_in":46098,"feed_emoji":"📐","tokens_out":13503,"duration_ms":111667,"temperature":0.7,"pith_summary":"This paper establishes a large explicit piece of the categorical local Langlands program for $\\mathrm{GL}_n$ over a $p$-adic field with coefficients in $\\mathbb{F}_\\ell$ or $\\mathbb{Q}_\\ell$. For an $L$-parameter $\\varphi$ whose irreducible constituents are pairwise disjoint up to unramified twist (condition (A1)), it describes the spectral action on the component $[C_\\varphi]$ of the stack of $L$-parameters: the relevant category $D_{\\mathrm{lis}}^{[C_\\varphi]}(\\mathrm{Bun}_G,\\Lambda)^\\omega$ decomposes orthogonally over characters of the centralizer torus $S_\\varphi$, and each summand is the derived category of modules over an explicit algebra $A_\\varphi$. This proves the strongly generic part of the categorical local Langlands conjecture for $\\mathrm{GL}_n$ with torsion coefficients. The proof combines modular representation theory of inner forms of $\\mathrm{GL}_n$ with a global argument: cohomology of type-A Shimura varieties is shown, via a product formula and torsion-freeness, to be concentrated and torsion-free in the relevant block, which fixes the otherwise unknown shifts. Consequences include an analogue of the Harris-Viehmann conjecture and new vanishing results for the mod-$\\ell$ cohomology of type-A Shimura varieties.","feed_headline":"Spectral action computes torsion cohomology of GL_n Shimura varieties","feed_subtitle":"Verifies a strongly generic case of categorical local Langlands and yields torsion vanishing in type-A Shimura varieties.","key_machinery":"The argument is carried by four interlocking objects. (1) The spectral action: line bundles $C_\\chi$ on the $L$-parameter component $[C_\\varphi]$ act on $D_{\\mathrm{lis}}(\\mathrm{Bun}_G,\\Lambda)$, and Hecke operators are computed by first decomposing a representation $V$ into $S_\\varphi$-isotypic pieces as $V \\simeq \\bigoplus_\\chi \\chi\\boxtimes\\sigma_\\chi$ and then applying $\\bigoplus_\\chi C_\\chi\\star(-)\\boxtimes\\sigma_\\chi$. (2) The renormalized pushforward $i^{\\mathrm{ren}}_{b!} = i_{b!}(\\delta_b^{-1/2}\\otimes -)[-d_b]$, which embeds blocks of inner forms into sheaves on $\\mathrm{Bun}_G$. (3) Modular representation theory: the classification of smooth $\\mathbb{F}_\\ell$-representations of $\\mathrm{GL}_n$ and its inner forms by supercuspidal multisegments, together with a compact pro-generator of each block, identifies each block with the category of $A_\\varphi$-modules, where $A_\\varphi \\simeq \\mathbb{F}_\\ell[X_1^{\\pm1},\\dots,X_r^{\\pm1},Y_1,\\dots,Y_r]/(Y_i^{\\ell^{u_i}})$. (4) The global step: a product formula expressing the localized cohomology of a type-A Shimura variety in terms of local Shimura varieties and Igusa varieties; affineness of Igusa varieties and torsion-freeness of the relevant localized $\\mathbb{Z}_\\ell$-cohomology force the mod-$\\ell$ complex to be concentrated in one degree, which pins down the shift left indeterminate by the local computation.","core_discovery":"The central claim is Theorem 7.6. Fix an $L$-parameter $\\varphi$ satisfying (A1), so that $S_\\varphi \\simeq \\mathbb{G}_m^r$ and the component $[C_\\varphi]$ is isomorphic to $[(\\mathbb{G}_m^r \\times \\mu_\\Lambda)/\\mathbb{G}_m^r]$, where $\\mu_\\Lambda$ records the $\\ell$-Sylow factors of the residue-field tori attached to $\\varphi$. Let $A_\\varphi$ be the ring of global functions on $[C_\\varphi]$. Then, with $\\Lambda \\in \\{\\mathbb{F}_\\ell,\\mathbb{Q}_\\ell\\}$, the paper establishes an orthogonal decomposition\n$$D_{\\mathrm{lis}}^{[C_\\varphi]}(\\mathrm{Bun}_G,\\Lambda)^\\omega \\simeq \\bigoplus_{\\chi\\in \\mathrm{Irr}(S_\\varphi)} D\\big($i^{{\\mathrm{ren}}$}_{b_\\chi!}\\,\\mathrm{Rep}_\\Lambda(s_{\\varphi,\\chi})\\big)^\\omega,$$\nwhere $b_\\chi$ is the basic element indexed by $\\chi$ and $s_{\\varphi,\\chi}$ is the inertial class of the corresponding supercuspidal pair. Equivalently, the spectral action gives an equivalence\n$$\\mathrm{IndPerf}^{\\mathrm{bqc}}_{\\mathrm{Coh}}([C_\\varphi]) \\simeq D_{\\mathrm{lis}}^{[C_\\varphi]}(\\mathrm{Bun}_G,\\Lambda)^\\omega,$$\nand for every coherent complex $L$ on $[C_\\varphi]$ and every character $\\chi$,\n$$L(\\chi)\\star F_W(\\mathrm{Id}) \\simeq i_{b_\\chi!}\\big(\\delta_{b_\\chi}^{-1/2}\\otimes \\pi_{L(\\chi)}\\big)[-d_\\chi].$$\nIn particular the sheaf $\\mathcal{G}_\\varphi = \\bigoplus_\\chi F_\\chi$ is a nontrivial Hecke eigensheaf for the parameter $\\varphi$. This is the strongly generic part of the categorical local Langlands conjecture for $\\mathrm{GL}_n$ with torsion coefficients.","pith_inferences":["Because $S_\\varphi$ is a torus, the block equivalences are all to module categories over a single ring $A_\\varphi$ with nilpotent directions $Y_i$; one could test whether the nilpotent directions control the size and support of torsion classes by comparing blocks with different $\\ell$-Sylow groups.","The global-lifting step uses one carefully chosen Shimura variety to fix shifts; if concentration and torsion-freeness can be established for a family of levels, the same method might remove the (A1) disjointness condition or cover non-banal primes.","The explicit Harris-Viehmann formula gives a concrete recursion for the mod-$\\ell$ cohomology of basic Rapoport-Zink spaces; for small $n$ this could be turned into a numerical check of the vanishing ranges, which would be a direct test of the categorical equivalence at the level of generators."],"forward_implications":["For every parameter satisfying (A1), the spectral action on the Whittaker sheaf is computed explicitly, so the strongly generic categorical local Langlands equivalence holds with $\\mathbb{F}_\\ell$ and $\\mathbb{Q}_\\ell$ coefficients.","The sheaf $\\mathcal{G}_\\varphi = \\bigoplus_\\chi F_\\chi$ is a non-zero Hecke eigensheaf for $\\varphi$, giving torsion-coefficient eigensheaves for these parameters.","The Harris-Viehmann conjecture is verified for local Shimura varieties of $\\mathrm{GL}_n$ attached to (A1) parameters: $R\\Gamma_c(G,b,\\mu)[\\delta_b^{1/2}\\otimes\\pi_b] \\simeq \\pi_1\\boxtimes \\mathrm{Hom}_{S_\\varphi}(\\chi_b^{-1}, r_{-\\mu}\\circ\\varphi|_{\\hat G})[-h]$, and the parabolic-induction version follows from the semisimple decomposition of $r_{-\\mu}$.","For type-A Shimura varieties, the $\\varphi$-block of $R\\Gamma_c(\\mathrm{Sh}_{K^p},\\mathbb{F}_\\ell)$ is concentrated in degrees $[0,d]$, in degree $d$ when the variety is compact or $\\varphi$ is irreducible, and in the non-generic case below $d+r-1$; compactness then forces the range $[d-r+1, d+r-1]$.","The supercuspidal part of the mod-$\\ell$ cohomology of basic Rapoport-Zink spaces for unitary groups is computed explicitly, giving an analogue of the Kottwitz conjecture with $\\mathbb{F}_\\ell$ coefficients."],"supporting_citations":[{"why":"Supplies the entire framework: the stack $\\mathrm{Bun}_G$, the lisse-étale sheaf categories $D_{\\mathrm{lis}}$, the spectral action, and the categorical local Langlands conjecture being proved.","marker":"[FS21]"},{"why":"Gives the $\\mathbb{Q}_\\ell$-coefficient version of the result and the proof strategy (induction on rank and on $|\\chi|$, plus modification arguments on the Fargues-Fontaine curve) that this paper adapts to $\\mathbb{F}_\\ell$.","marker":"[Ngu23a]"},{"why":"Establishes the equivalence of the relevant Bernstein block with the category of $A_\\varphi$-modules, constructs the compact pro-generator, and proves uniqueness of the supercuspidal $\\mathbb{F}_\\ell$-representation with a given parameter.","marker":"[Dat12]"},{"why":"Classifies irreducible smooth $\\mathbb{F}_\\ell$-representations of $\\mathrm{GL}_n$ and its inner forms by supercuspidal multisegments; this is used to construct $\\pi_\\chi$ and to prove Lemma 6.1 and the irreducibility of the relevant parabolic inductions.","marker":"[MS14a]"},{"why":"Provides the block decomposition of the category of smooth $\\Lambda$-representations of inner forms of $\\mathrm{GL}_n$ on which the local categorical decompositions rest.","marker":"[SS16]"},{"why":"Supplies the banal-case modular local Langlands results, the lifting of supercuspidal representations to integral characteristic-0 representations, and the Igusa-stack filtrations used in the global step.","marker":"[DHKM24]"},{"why":"Provides the product formula for cohomology of Shimura varieties, affineness of Igusa varieties, and the torsion-vanishing framework used to localize and bound $R\\Gamma_c(\\mathrm{Sh}_{K^p},\\mathbb{F}_\\ell)$.","marker":"[HL23]"},{"why":"Supplies the Igusa-stack form of the product formula (Theorem 8.5.7) and the Bernstein decomposition in families used to isolate the supercuspidal block.","marker":"[DHKZ24]"},{"why":"Computes characteristic-0 cohomology of EL-type Rapoport-Zink spaces, serving as the known base case that fixes the shift after the mod-$\\ell$ lift.","marker":"[Shi12]"}],"fun_headline_variants":["Spectral action proves torsion Langlands for GL_n","Torsion Langlands for GL_n from spectral action","Spectral action computes torsion Shimura cohomology","Torsion vanishing for type-A Shimura varieties","Categorical Langlands torsion verified for GL_n"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of the modular supercuspidal step assumes that for every relevant supercuspidal parameter one can choose a compact simple type-A Shimura variety satisfying the product formula and such that the localized $\\mathbb{Z}_\\ell$-cohomology is concentrated in one degree and torsion-free, allowing mod-$\\ell$ cohomology to be lifted to characteristic $0$.","fun_headline_variants_meta":{"raw":{"variants":["Spectral action proves torsion Langlands for GL_n","Torsion Langlands for GL_n from spectral action","Spectral action computes torsion Shimura cohomology","Torsion vanishing for type-A Shimura varieties","Categorical Langlands torsion verified for GL_n"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000909,"raw_usage":{"total_tokens":3966,"prompt_tokens":1066,"completion_tokens":2900,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":682,"completion_tokens_details":{"reasoning_tokens":2824}},"tokens_in":682,"tokens_out":2900,"duration_ms":18389,"temperature":1.0,"reasoning_tokens":2824,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T21:05:19.716362+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small explicit type-A Shimura variety at a non-banal prime $\\ell$ (for instance a $\\mathrm{GU}(1,n-1)$ case), localize its mod-$\\ell$ cohomology at an $L$-parameter $\\varphi$ satisfying (A1), and check whether classes appear outside the asserted degrees: outside degree $d$ when the variety is compact or $\\varphi$ is irreducible, or outside the range $[0,d]$ in general. A single such class would falsify Theorem 8.1. Alternatively, test the identity $C_{\\chi_i}\\star F_\\chi \\simeq F_{\\chi\\otimes\\chi_i}$ from Theorem 6.5 directly by computing the restriction of $C_{\\chi_i}\\star F_\\chi$ to the stratum of $b_{\\chi\\otimes\\chi_i}$ and checking that all other stratum restrictions vanish.","supporting_citations":[],"review_version":1}