{"id":"9606ff19-abf2-40af-a1cb-8612584cff9a","arxiv_id":"1909.01898","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Generic mod ℓ cohomology of non-compact unitary Shimura varieties is concentrated above the middle degree, and fibers of the compactified Hodge-Tate period map are compactified Igusa varieties.","lead":"This paper proves that the generic part of the mod ℓ cohomology of non-compact unitary Shimura varieties is concentrated above the middle degree. The result is a key input for potential automorphy theorems without self-duality, including the Sato-Tate conjecture for elliptic curves over CM fields.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1.1 is proved only at levels divisible by N0; descent to arbitrary K is missing and non-formal in mod ℓ.","rationale":"The central claim is a sharp concentration theorem for the generic part of the cohomology, and the geometric engine—compactified Hodge–Tate period maps with fibers equal to compactified Igusa varieties—is compelling. The reader correctly identified hypothesis (ii), the length-at-most-2 condition on ρ_m, as the most fragile input; the paper itself flags it as critical. My stress-test found a different, more structural concern: the proof of Theorem 1.1 depends on trace-formula results that are stated only for levels divisible by a sufficiently large N0, while the theorem is claimed for arbitrary neat K. The missing descent from deep level to arbitrary level is non-formal in mod ℓ cohomology precisely because the covering index may be divisible by ℓ. This is not an objection to the geometric results, which appear well supported by Sections 3–4, nor to the trace-formula computations in Section 5, but it is a genuine gap in the proof of the theorem as stated. The paper contains many real strengths: the affineness results for compactified Igusa varieties, the explicit toroidal boundary charts, the semiperversity theorem, and the careful attribution of prior and concurrent work all point to a correct core. My recommended adjustment to CONDITIONAL reflects the need to either provide the missing level-descent argument or restrict the theorem’s statement to the levels for which the proof actually works.","tokens_in":68760,"tokens_out":19134,"duration_ms":220949,"concrete_test":"In the full arXiv version, search for a reduction of Theorem 1.1 to the N0-divisible level used in Theorems 2.8.6 and 5.1.2. If no such reduction exists, test the descent concretely for U(1,1) over an imaginary quadratic field with ℓ dividing [K(N):K(NN0)]: check whether a nontrivial class in H^1(X_{K(N)}, Fℓ)_m with ρ_m of length at most 2 can pull back to zero in H^1(X_{K(NN0)}, Fℓ)_m. If such a class exists, the theorem as stated is unproved; otherwise, document why the descent is valid modulo ℓ.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proof of Theorem 1.1 in §2.8 invokes Theorem 2.8.6 and Theorem 5.1.2, whose statements require the level N to be divisible by a sufficiently large N0 (see Remark 5.4.5). Theorem 1.1 itself is stated for an arbitrary neat compact open K, with no level condition. No reduction from arbitrary K to an N0-divisible principal level is supplied. This is not a formal step: for the finite cover X_{K(NN0)} → X_{K(N)}, the standard Hochschild–Serre descent for mod ℓ cohomology fails when the covering degree is divisible by ℓ, because Fℓ is not a direct summand of the permutation module Fℓ[G] over Fℓ and the trace map can be zero. Thus the full statement of Theorem 1.1 is not established by the provided argument unless the scope is silently restricted to levels divisible by N0.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves a vanishing theorem for the m-localized mod ℓ cohomology of non-compact unitary Shimura varieties associated to quasi-split unitary groups of even dimension. Under the hypotheses (i) F^+ ≠ Q, (ii) the attached semisimple Galois representation ρ_m has length at most 2, and (iii) there is a totally split prime p ≠ ℓ with generic Frobenius eigenvalues, the theorem asserts that H^i(X_K, F_ℓ)_m = 0 for i < d and H^i_c(X_K, F_ℓ)_m = 0 for i > d. The proof proceeds through the geometry of the Hodge–Tate period map on minimal and toroidal compactifications, whose fibers are shown to be compactified Igusa varieties, combined with semiperversity of nearby cycles, a trace-formula computation of Igusa cohomology, and a boundary-cohomology computation. The paper also derives consequences for Z_ℓ coefficients, torsion-freeness in the middle degree, and applications to Eisenstein classes.","tokens_in":68722,"tokens_out":9348,"duration_ms":96177,"significance":"If correct, the main theorem is an important extension of the authors' previous compact case [CS17] to non-compact Shimura varieties, and it provides exactly the control over torsion cohomology needed in [ACC+18] for potential automorphy theorems without self-duality. The geometric core, Theorem 1.10, identifying fibers of the compactified Hodge–Tate period map with compactified Igusa varieties, is a substantial and reusable contribution. The paper is carefully structured: the main theorem is reduced in Section 2.8 to four named inputs, each developed in later sections, and the trace-formula and boundary computations are made quite explicit. The main weakness is a level-descent gap: the proof of Theorem 1.1 establishes the statement only for principal levels divisible by an auxiliary integer N0, while the theorem is stated for arbitrary neat compact open K.","major_comments":[{"comment":"The proof establishes the concentration statement only for principal levels K(N) with N divisible by an auxiliary integer N0. Theorem 2.8.6 (Corollary 5.1.3) and Theorem 2.8.7 both carry this hypothesis, as noted in Remark 5.4.5, and the trace-formula computation of Section 5 is stated only under it. Theorem 1.1, however, is asserted for an arbitrary neat compact open K. No reduction from arbitrary K to such a principal level is given. This is not a formal Hochschild–Serre descent: when the finite cover X_{K(N)} → X_K has degree divisible by ℓ, the F_ℓ[G]-permutation module is not semisimple and the trace map can be zero, so the m-localized F_ℓ-cohomology at the deeper level need not determine the cohomology at K. The same gap affects the compact-support statement, which is obtained by Poincaré duality at the end of the proof. The theorem should be restricted to levels divisible by N0 (or to levels whose index in K(N0) is prime to ℓ), or a genuine level-descent argument must be supplied.","section":"§2.8, proof of Theorem 1.1"},{"comment":"The statements of these two theorems in Section 2.8 omit the level condition that is explicitly present in their proofs: Theorem 5.1.2 and Corollary 5.1.3 assume N is divisible by N0, and Section 6.4 inherits this. As stated, the theorems appear to hold for all neat levels, which is exactly what makes the gap in Theorem 1.1 easy to overlook. The statements should either include the N0 hypothesis or explain why the arguments extend to all levels.","section":"§2.8, Theorems 2.8.6 and 2.8.7"}],"minor_comments":[{"comment":"The remark asserts that the theorem and its consequences extend to nontrivial Z_ℓ-coefficient systems via the Hochschild–Serre spectral sequence; this extension is not proved and would need the same level conditions as the main theorem.","section":"Remark 1.5"},{"comment":"The symbol q is used both for a rational prime and for the cardinality of a residue field in the displayed characteristic polynomial; this is mildly confusing and should be disambiguated.","section":"Corollary 5.1.3"},{"comment":"The phrase 'choose a rank 1 point of F_ℓ of dimension d−d_b' is not immediately clear, since rank 1 points are usually described by their residue fields and 'dimension' of a point is not standard; please clarify whether this means a rank 1 point whose closure has that dimension.","section":"Lemma 2.8.4"},{"comment":"The arXiv source contains several OCR-style spacing artifacts in the title and abstract (e.g., 'P ART', 'V ARIETIES'); these should be corrected in the final version.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The level-descent gap is the only serious obstruction I see to the main theorem as stated; if the authors either restrict the statement to levels divisible by N0 or add a descent argument, the paper would be acceptable. I would also ask them to state explicitly which level version is used in [ACC+18], since that determines whether the missing descent affects the applications."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new content is the geometry: compactified Igusa varieties, the identification of fibers of the compactified Hodge–Tate period map with them (Theorem 1.10), and the semiperversity result. The trace-formula computation of Igusa cohomology and the boundary computation in Section 6 are substantial and look careful. The paper is also commendably honest about its restrictions—Remark 1.3 says the length-at-most-2 hypothesis is critical, Remark 1.2 says F+ ≠ Q could be removed with extra work, and Remark 2.6.3 retracts an earlier incorrect claim. That is the right way to write mathematics. The problem is the level. Theorem 1.1 is stated for an arbitrary neat compact open K, but the proof in §2.8 only treats principal level K(N), and the inputs (Corollary 5.1.3, hence Theorem 2.8.6) require N divisible by a sufficiently large N0, as noted in Remark 5.4.5. No descent from K(N) to arbitrary K is supplied. This is not a formality: when the covering degree is divisible by ℓ, Hochschild–Serre does not descend mod ℓ cohomology, because Fℓ is not a direct summand of Fℓ[G]. So the theorem as stated is not established. The gap is fixable—restrict the statement to levels divisible by N0, or prove a transfer argument—but it is load-bearing for the statement as written. The length-at-most-2 condition is a real restriction, but the authors flag it as critical and it seems necessary for the method; that is not a defect in the proof. The heavy reliance on the authors' prior work (CS17) and on the concurrent ACC+18 is appropriate: the dependencies are explicit and the cited results are published or in wide circulation. Who gets value from this? People working on torsion in Shimura variety cohomology and the Calegari–Geraghty method. The applications in ACC+18 rely on the geometric results and on the concentration statement at the levels actually used there; the full generality of Theorem 1.1 may not be needed for those applications. The paper deserves a serious referee. I would send it out, with the level question as the main point to resolve before acceptance.","headline":"Genuinely new geometry and a serious but fixable level problem: Theorem 1.1 is only proved for levels divisible by N0, not arbitrary neat K.","tokens_in":741,"tokens_out":2120,"would_cite":true,"duration_ms":47429,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11G18","11F80","14G35"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that the generic part of mod $\\ell$ cohomology of non-compact unitary Shimura varieties is concentrated above the middle degree, and compactly supported cohomology below it.","keywords":["Shimura varieties","mod-ℓ cohomology","Hodge-Tate period map","Igusa varieties","unitary groups","torsion cohomology","Galois representations","Eisenstein cohomology"],"falsifier":"A direct way to test the claim is to compute, for one maximal ideal $\\mathfrak m$ satisfying the three hypotheses, the localized groups $H^i(X_K,\\mathbb{F}_\\ell)_{\\mathfrak m}$ for $i<d$ and $H^i_c(X_K,\\mathbb{F}_\\ell)_{\\mathfrak m}$ for $i>d$ using the trace formula and the Section 6 boundary formula; any nonzero group in a forbidden degree would refute Theorem 1.1. In particular, the Section 6 boundary formula should give zero for every intermediate-rank cusp label after localization at such an $\\mathfrak m$, and a nonzero contribution there would pinpoint where the proof's single-stratum mechanism fails.","tokens_in":68336,"feed_emoji":"📐","tokens_out":12519,"duration_ms":125381,"temperature":0.7,"pith_summary":"This paper establishes that, for even-dimensional quasi-split unitary Shimura varieties attached to a CM field $F$ with $F^+\\neq\\mathbb{Q}$, the mod $\\ell$ cohomology localized at a ``generic'' Hecke eigensystem is concentrated above the middle degree: $H^i(X_K,\\mathbb{F}_\\ell)_{\\mathfrak m}$ can be nonzero only for $i\\ge d$, while $H^i_c(X_K,\\mathbb{F}_\\ell)_{\\mathfrak m}$ can be nonzero only for $i\\le d$, where $d=[F^+:\\mathbb{Q}]n^2$. The genericity condition is that the associated semisimple Galois representation $\\rho_{\\mathfrak m}$ has length at most two and has Frobenius eigenvalues at a totally split prime $p$ satisfying $\\alpha_{i,v}\\neq p\\alpha_{j,v}$ for $i\\neq j$. The same conclusion is deduced with $\\mathbb{Z}_\\ell$ coefficients, including torsion-freeness of $H^d(X_K,\\mathbb{Z}_\\ell)_{\\mathfrak m}$, together with a clean splitting of the excision sequence for boundary cohomology. The reason to care is that the boundary of these non-compact spaces contains the locally symmetric spaces for $GL_n/F$, so the result controls even Eisenstein cohomology classes coming from $GL_n$; this control was the new input used for local-global compatibility and potential automorphy without self-duality. The decisive geometric tool is a description of the fibres of the Hodge--Tate period map on compactified Shimura varieties as compactified Igusa varieties.","feed_headline":"Generic mod-ℓ cohomology sits above the middle degree","feed_subtitle":"New theorem covers non-compact unitary Shimura varieties and tames Eisenstein classes from GL_n, with torsion-free middle degree.","key_machinery":"The load-bearing object is the Hodge--Tate period map from the infinite-level perfectoid Shimura variety to the flag variety of totally isotropic $F$-stable subspaces, together with the Igusa varieties attached to Newton strata. An Igusa variety is the moduli space of isomorphisms between the $p$-divisible group of the universal abelian variety and a fixed $p$-divisible group $X$, up to $p$-power isogeny. The paper's Theorem 1.10 identifies the fibres of the Hodge--Tate period map on the minimal and toroidal compactifications with partial minimal and toroidal compactifications of these Igusa varieties, as open immersions with the same rank-1 points; this transfers cohomology computations from the Shimura variety to the compactified Igusa varieties. On those varieties, the argument combines affineness of the partial minimal compactifications, giving an Artin-vanishing upper bound, with a semiperversity result for nearby cycles giving a lower bound, and with a trace-formula computation of the $\\mathbb{Q}_\\ell$-cohomology of Igusa varieties that is used to attach Galois representations and to force non-ordinary Newton strata to disappear. The length-at-most-two hypothesis is what guarantees that only the full-rank boundary stratum contributes, making the excision sequence split cleanly around the middle degree.","core_discovery":"On the paper's own terms, the central discovery is Theorem 1.1. For a quasi-split unitary group of even dimension with $F^+\\neq\\mathbb{Q}$, fix a maximal ideal $\\mathfrak m$ in the abstract Hecke algebra such that the attached semisimple Galois representation $\\rho_{\\mathfrak m}$ has length at most two and is generic at a totally split prime $p\\neq\\ell$ in the sense that no two Frobenius eigenvalues have ratio $p$. Then the localization of the cohomology of the Shimura variety at $\\mathfrak m$ vanishes below the middle degree, and the compactly supported cohomology vanishes above the middle degree, with $d=[F^+:\\mathbb{Q}]n^2$. The companion geometric result, Theorem 1.10, is that the fibres of the Hodge--Tate period map on both the minimal and toroidal compactifications are canonically the partial compactifications of Igusa varieties, so comparing cohomology of the fibres with cohomology of Igusa varieties remains valid at the boundary. The paper argues that the boundary cohomology is then governed by a single full-rank $GL_n$ stratum, and that the boundary comparison theorem forces boundary contributions to be confined to the two sides of degree $d$ in exactly the way described by the excision sequence around the middle degree.","pith_inferences":["Beyond the paper: if the length-at-most-two hypothesis really is the obstruction it appears to be, the natural next step is a version of the boundary comparison theorem that permits a second boundary stratum while still controlling its possible degrees; the Section 6 boundary formula gives a computational tool for testing when such a stratum can survive localization.","Beyond the paper: the clean boundary-splitting picture suggests that the folklore vanishing range for $GL_n/F$ with torsion coefficients may be accessible by reduction to unitary Shimura varieties, provided the remaining cases of local-global compatibility at $\\ell$ are supplied; the present theorem contributes exactly the boundary control that such a reduction would need.","Beyond the paper: because the fibre comparison on compactifications is proved at the level of perfectoid spaces with the same rank-1 points, the same open immersions should transport other cohomology theories on the compactified Shimura variety, such as intersection cohomology or nearby cycles on integral models, to compactified Igusa varieties."],"forward_implications":["The generic part of the mod $\\ell$ cohomology of a non-compact even-dimensional unitary Shimura variety is nonzero only at or above the middle degree, while compactly supported cohomology is nonzero only at or below it.","The same concentration statements hold with $\\mathbb{Z}_\\ell$ coefficients, and the middle-degree group $H^d(X_K,\\mathbb{Z}_\\ell)_{\\mathfrak m}$ is torsion-free.","The excision long exact sequence splits boundary cohomology cleanly: below the middle degree everything maps into compactly supported cohomology, above the middle degree everything comes from ordinary cohomology, and an explicit exact sequence of length four holds around degree $d$.","For a non-Eisenstein eigensystem, ordinary and compactly supported cohomology coincide and are concentrated in degree $d$, matching the folklore vanishing prediction for the locally symmetric space of $GL_n/F$ in this setting.","Because the theorem covers Eisenstein classes coming from the $GL_n$ boundary, it gives control over torsion classes in the cohomology of that boundary, and this control feeds into local-global compatibility and potential automorphy results without any self-duality hypothesis."],"supporting_citations":[{"why":"The compact-case predecessor whose method is extended; supplies the perfectoid Hodge--Tate period map and the Igusa-variety comparison on the open Shimura variety.","marker":"[CS17]"},{"why":"Supplies a theorem used to attach the semisimple Galois representation to the Hecke eigensystem, and is the announced application of the main result.","marker":"[ACC+18]"},{"why":"Provides the perfectoid Shimura varieties at infinite level and the construction of Galois representations for torsion classes in cohomology of locally symmetric spaces.","marker":"[Sch15]"},{"why":"Classifies $p$-divisible groups over the ring of integers of a complete algebraically closed field, identifying points of the flag variety with the $p$-divisible groups whose fibres are computed by Igusa varieties.","marker":"[SW13]"},{"why":"Establishes that Oort central leaves are well-positioned, giving the partial toroidal and minimal compactifications of leaves used throughout Section 3.","marker":"[LS18]"},{"why":"Provides the stable trace formula for Igusa varieties that is the starting point for computing their $\\mathbb{Q}_\\ell$-cohomology as a virtual representation.","marker":"[Shi10]"},{"why":"Supplies the base-change and Galois-representation results for unitary groups used to turn the trace-formula description into attached Galois representations.","marker":"[Shi11]"},{"why":"The boundary-cohomology formula for Shimura varieties adapted in Section 6 to compute the cohomology of the boundary of Igusa varieties.","marker":"[Pin92]"},{"why":"Gives the affineness of partial minimal compactifications of Ekedahl--Oort strata, which underlies the affineness of Igusa minimal compactifications and the Artin-vanishing upper bound.","marker":"[Box15]"}],"fun_headline_variants":["Unitary Shimura generic cohomology: above middle degree, non-compact","Non-compact unitary Shimura: generic cohomology above middle","Generic cohomology above middle degree for unitary Shimura varieties","Cohomology of unitary Shimura: generic part floats above middle","Eisenstein classes controlled: generic cohomology above middle"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Galois representation attached to the Hecke eigensystem has length at most two; if it had length three or more, multiple boundary strata could contribute and the concentration conclusion is not established.","fun_headline_variants_meta":{"raw":{"variants":["Unitary Shimura generic cohomology: above middle degree, non-compact","Non-compact unitary Shimura: generic cohomology above middle","Generic cohomology above middle degree for unitary Shimura varieties","Cohomology of unitary Shimura: generic part floats above middle","Eisenstein classes controlled: generic cohomology above middle"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00132,"raw_usage":{"total_tokens":5376,"prompt_tokens":949,"completion_tokens":4427,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":565,"completion_tokens_details":{"reasoning_tokens":4334}},"tokens_in":565,"tokens_out":4427,"duration_ms":29302,"temperature":1.0,"reasoning_tokens":4334,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:06:13.906253+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A direct way to test the claim is to compute, for one maximal ideal $\\mathfrak m$ satisfying the three hypotheses, the localized groups $H^i(X_K,\\mathbb{F}_\\ell)_{\\mathfrak m}$ for $i<d$ and $H^i_c(X_K,\\mathbb{F}_\\ell)_{\\mathfrak m}$ for $i>d$ using the trace formula and the Section 6 boundary formula; any nonzero group in a forbidden degree would refute Theorem 1.1. In particular, the Section 6 boundary formula should give zero for every intermediate-rank cusp label after localization at such an $\\mathfrak m$, and a nonzero contribution there would pinpoint where the proof's single-stratum mechanism fails.","supporting_citations":[],"review_version":1}