{"id":"86529347-29fd-4f73-9202-30949f078840","arxiv_id":"1908.07073","paper_version":5,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For KHT-free Hecke maximal ideals, the nilpotency partition of local monodromy at v agrees modulo l with the characteristic-zero partition, so level lowering at v is impossible.","lead":"This paper proves a higher-dimensional version of Ribet's level-lowering obstruction for Galois representations over CM fields: under certain freeness hypotheses, the local monodromy at a prime v is unchanged after reduction modulo l. This 'level fixing' phenomenon blocks level lowering at v and is linked to level raising and Ihara's lemma.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1.2 rests on proposition 3.1, whose proof of torsion-freeness of the E1 terms delegates a key freeness assertion to the unpublished preprint [9] and a two-line lemma 3.2; if that freeness fails, the mod-l monodromy identification breaks.","rationale":"The reader's weakest_assumption identifies proposition 3.1 as the load-bearing step, and my reading agrees: the proof of theorem 1.2 in §5 explicitly begins with 'By KHT-freeness of m, we are in the situation of proposition 3.1', and every subsequent step (the saturated filtration, the mod-l identification of N^{coho}_{v,m} with N_{v,m}, and the application of proposition 4.1) depends on it. The proof of proposition 3.1 has two weak points: lemma 3.2, whose proof is a two-line sketch asserting that the p- and p+-intermediate extensions coincide for characters, and the freeness of the sheaf cohomology germs of Fil^1_{!,χ_v}(Ψ), which is asserted via the Lubin-Tate/Drinfeld comparison and [28] but is also the content of the submitted preprint [9]. I do not see an internal contradiction in the surrounding argument. One apparent overstatement in the proof of proposition 3.1, 'as e_v(l)>d', is not load-bearing: the theorem's chain condition only gives t<e_v(l) for each Jordan block, and that weaker bound is what the proof of proposition 4.1 needs. Because the concern is a verification gap rather than a demonstrated error, the appropriate verdict remains CONDITIONAL, matching the reader's assessment.","tokens_in":21802,"tokens_out":10108,"duration_ms":106784,"concrete_test":"Make [9] available and verify its proposition 3.1.3 together with the asserted freeness of the cohomology of the Lubin-Tate tower; in parallel, recompute the simplest nontrivial instance of proposition 3.1 (d=2, Iwahori level at v, χ_v a character with t<e_v(l)) using only published results [3], [6], [20] and [28], checking that every E1 term is free over Z_l and vanishes outside degree d-1. If the terms are free, the concern reduces to a citation gap; if they are not, theorem 1.2 is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The theorem's engine is proposition 3.1: after localization at m, the E1 terms of the nearby-cycles spectral sequence (2.11) are torsion-free and vanish outside middle degree. The proof of proposition 3.1 is where the argument is least secure. Its first half invokes lemma 3.2 to identify the p- and p+-intermediate extensions of Harris-Taylor local systems; the proof of that lemma is a terse smoothness argument that does not spell out the required perverse cohomology vanishing on the boundary strata. Its second half, in proposition 3.12, must show that the germs of Fil^1_{!,χ_v}(Ψ) are free Z_l-modules. The text says these germs are described by Lubin-Tate tower cohomology and then, via the Faltings-Fargues comparison [20], by Drinfeld tower cohomology, citing [28] for freeness; this is exactly the assertion that the author's submitted preprint [9] is said to prove in full generality, and no published proof is supplied in this paper. If any of these freeness assertions fails, the E1 terms can carry torsion, the saturated filtration of H^{d-1} used in §5 is not established, and the equality of maximal nilpotency between N^{coho}_{v,m} mod l and N_{v,m} is no longer forced. This is a verification gap, not evidence of falsehood: the theorem may be true, but the central claim currently depends on an unpublished input and a sketched lemma.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies level lowering/fixing for Galois representations attached to cohomological automorphic forms on a similitude group of signature (1,d-1), in the framework of Kottwitz-Harris-Taylor Shimura varieties. The central result, Theorem 1.2, states that under hypotheses on the maximal ideal m (irreducibility of the residual Galois representation, character-type local subquotients with a no-chain condition, and KHT-freeness), the maximal nilpotency of the local monodromy operator at a place v is the same modulo l as in characteristic zero, so that the level at v cannot be lowered. The proof proceeds by analyzing the nearby-cycles spectral sequence and its localized E1 terms, showing they are torsion-free and concentrated in middle degree (Proposition 3.1), then comparing the monodromy filtration with the cohomological monodromy via a Carayol-type tensor product decomposition (Theorem 2.12). The final sections draw consequences for automorphic congruences and for a relation between Ihara's lemma and level raising.","tokens_in":22132,"tokens_out":2765,"duration_ms":30906,"significance":"If the proof is completed, the paper would establish a genuinely higher-dimensional analogue of Mazur's principle for level lowering, a phenomenon that has been well understood for GL2 but largely open in higher dimension. The main theorem gives a clean, falsifiable criterion under which the mod-l monodromy order is forced to match its characteristic-zero counterpart, and the applications to Ihara's lemma and level raising in Sections 6-7 are natural and interesting. The paper also makes its hypotheses explicit and acknowledges the dependence of the proof on external inputs. However, the central derivation currently rests on an unpublished preprint [9] and on a very compressed lemma (Lemma 3.2), and these are load-bearing rather than cosmetic gaps.","major_comments":[{"comment":"The proof of Proposition 3.1 depends essentially on unpublished material. The splitting of the nearby-cycles sheaf and the torsion-freeness of the cohomology of the summands are attributed to [9, Proposition 3.1.3], and the proof of Proposition 3.12 delegates the key freeness assertion for the germs of the sheaf cohomology to [9], via the Lubin-Tate and Drinfeld towers. Since [9] is a submitted preprint and no proof of these freeness statements is included here, the derivation of the saturated filtration used in §5 is not self-contained. If the freeness assertions in [9] fail, the E1 terms can carry torsion and the equality of maximal nilpotency between N^coho_{v,m} mod l and N_{v,m} is not forced. The manuscript itself implicitly concedes this issue in the remark after Theorem 1.2.","section":"§3, Proposition 3.1"},{"comment":"Lemma 3.2 is proved in two sentences but is used to identify the p- and p+-intermediate extensions of Harris-Taylor local systems for characters, which is a key step in showing that the graded pieces of the filtration are free and that the resolution (3.4) is strict. The proof states that P is perverse for the two t-structures and that the displayed perverse cohomology vanishings hold, but it does not justify these vanishings on the boundary strata of the Newton stratification. Smoothness of the open Newton strata, as cited from [24], does not by itself establish the required adjunction properties for the intermediate extension, and no computation of the perverse cohomology sheaves on the complement is supplied. A full argument or a precise reference with proof is needed.","section":"§3, Lemma 3.2"},{"comment":"The proof of Proposition 4.1 is too compressed at the point where the strictness of the monodromy map is reduced to the assertion that the reduction modulo l of a certain isomorphism is non-zero and then to an explicit check in Iwahori level using [11, §3.1]. The passage from non-vanishing of the reduction to an isomorphism of the relevant graded pieces is not justified, and the current level hypotheses on m are not visibly compatible with the settings of [11] and [23]. Since Proposition 4.1 is what transfers the local monodromy statement to the cohomological monodromy operator in the proof of Theorem 1.2, this gap is load-bearing.","section":"§4, Proposition 4.1"}],"minor_comments":[{"comment":"There are numerous typographical errors and infelicities, e.g. 'on can also define', 'infer informations', 'o perator', 'strati cat ion', 'galoisian'; the paper would benefit from a careful proofreading pass.","section":"Abstract and Introduction"},{"comment":"The phrase 'the cohomology groups of the Kottwitz-Harris-Taylor Shimura variety ... localized at m, are free' is ambiguous: it should specify that the groups are free as Z_l-modules and should indicate which cohomological degrees and which level structures are meant.","section":"Definition 1.1"},{"comment":"In the remark comparing with GL2, the statement 'our second hypothesis is q_v not congruent to -1 mod l' appears without a precise derivation or reference to a specific equation; a short explanation would improve readability.","section":"§1, after Definition 1.3"},{"comment":"The sentence 'We then just need to verify that every map is strict' is not followed by the promised verification; the rest of the proof invokes the comparison theorem and [28] without detailing how the freeness of the Lubin-Tate and Drinfeld tower cohomology implies strictness of the maps in the resolution (3.13).","section":"§3, proof of Proposition 3.12"}],"recommendation":"major_revision","confidential_remarks":"The main theorem is plausible and the overall strategy is sound, but the verification currently depends on an unpublished submission by the same author ([9]) and on a lemma whose proof is only sketched. I would recommend requesting that the author either include a complete proof of the needed freeness assertions and of Lemma 3.2, or replace [9] by a published or otherwise publicly available and accepted reference. The paper also relies heavily on a cluster of the author's own earlier papers; while this is not itself a defect, the referee should verify that [9] has actually appeared or is available in a form suitable for citation. If the gaps cannot be filled, the paper should not be accepted in its current form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: this is a genuine higher-dimensional Mazur-principle theorem, but the proof is not self-contained; the main engine, proposition 3.1, leans on the author's submitted preprint [9] for a key torsion-freeness statement, so as written the theorem is conditional.\n\nThe genuinely new content is theorem 1.2: for KHT Shimura varieties in dimension d>2, under irreducibility plus a condition on the set of characters in the restriction to the decomposition group, the nilpotency order of the monodromy at v is already maximal modulo l, so the level cannot be lowered. That goes beyond Ribet's GL2 result and Sorensen's level raising. Section 7's equivalences between this level-raising property and Ihara's lemma are also new and worth having. The strategy is coherent—localize at a KHT-free maximal ideal, use the nearby-cycles spectral sequence, and prove the E1 terms are torsion-free so the mod-l monodromy is forced by the local perverse sheaf—and the author is transparent about where [9] is used. The introduction explicitly says that the general version of lemma 3.2 is one of the main results of [9].\n\nThe soft spot is real and it is exactly the one the stress-test flags. Proposition 3.1 is load-bearing; its proof invokes lemma 3.2, whose two-line proof does not spell out the required perverse cohomology vanishing on the boundary strata, and then passes to Lubin-Tate and Drinfeld tower cohomology, asserting freeness of the localized E1 terms by citing [9] and [28]. Proposition 3.12's proof is similarly compressed and ends by saying the general case is proved in [9]. If those freeness assertions fail, the saturated filtration used in section 5 is not established and theorem 1.2 does not follow. That is a verification gap, not evidence of falsehood: the argument is structurally sound and the author has proved related freeness results before. But a referee cannot check the proof from the text alone, and the paper's reliance on a dense corpus of the author's own earlier work, including a submitted preprint, makes the verification burden high.\n\nThis paper is for specialists in Shimura varieties, nearby cycles, and the Langlands program. It deserves a serious referee; a specialist should be asked to determine whether proposition 3.1 can be established using the existing [9], and whether lemma 3.2 is as straightforward as claimed. My recommendation: send it to peer review, and require the author to either include the needed results from [9] or state explicitly which parts of theorem 1.2 are conditional on [9].","headline":"Genuine higher-dimensional level-fixing theorem, but the proof's central torsion-freeness input rests on an unpublished same-author preprint.","tokens_in":22648,"tokens_out":3886,"would_cite":false,"duration_ms":38679,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11F70","11F80","11F85","11G18","20C08"],"pacs":[],"model":"deepseek-v4-flash","headline":"Under generic conditions, mod l monodromy at v already determines the level, so level lowering is impossible.","keywords":["Shimura varieties","level lowering","Galois representations","monodromy","nearby cycles","Ihara's lemma","Hecke algebra","Kottwitz-Harris-Taylor"],"falsifier":"Compute the spectral sequence of proposition 3.1 for one explicit KHT-free maximal ideal $\\mathfrak m$ satisfying the hypotheses and check whether every $E_1^{p,q}$ localized at $\\mathfrak m$ is free and zero outside $p+q=d-1$; a single non-zero torsion class outside the middle degree would invalidate proposition 3.1 and with it the equality of maximal nilpotency. More directly, exhibit $\\mathfrak m$ with the stated local characters for which two lifts $\\widetilde{\\mathfrak m}_1,\\widetilde{\\mathfrak m}_2$ have monodromy partitions at $v$ with different largest parts; theorem 1.2 predicts this cannot happen.","tokens_in":21590,"feed_emoji":"🧮","tokens_out":11770,"duration_ms":108309,"temperature":0.7,"pith_summary":"This paper proves a higher-dimensional analogue of the classical level-lowering principle for Galois representations supplied by automorphic forms on unitary similitude groups. It shows that, for a maximal ideal of the relevant Hecke algebra whose mod $\\ell$ Galois representation is irreducible and whose local constituents at a bad-reduction place $v$ are characters avoiding one short chain, the nilpotency order of the local monodromy operator modulo $\\ell$ equals the same order in characteristic zero. Therefore no automorphic lift of the mod $\\ell$ representation can have a lower level at $v$: the level is fixed. This matters because level-lowering congruences are the engine of the modular method for Diophantine equations, and here the obstruction is read directly from local monodromy and from the freeness of the relevant cohomology.","feed_headline":"Mod l monodromy already fixes the level at v","feed_subtitle":"Generic local conditions force mod l monodromy to match characteristic zero, so no lift can lower the level at v.","key_machinery":"The engine of the proof is the nearby-cycles spectral sequence on the special fiber at $v$ of a Kottwitz-Harris-Taylor Shimura variety. The nearby-cycles perverse sheaf $\\Psi_v$ carries a filtration by Newton strata, and the associated spectral sequence computes the localized cohomology; proposition 3.1 asserts that under the theorem's hypotheses all $E_1$ terms are torsion free and vanish outside the middle degree, so the monodromy operator on $H^{d-1}$ is purely local. A tensor-product theorem identifies this localized cohomology, as a module for the Hecke algebra and the Galois group, with $\\sigma_{\\mathfrak m}\\otimes\\rho_{\\mathfrak m}$, so nilpotency statements about the cohomological monodromy pass to $N_{\\mathfrak m,v}$. KHT-freeness of $\\mathfrak m$ means the localized cohomology groups are free $\\mathbb Z_\\ell$-modules; the character-chain hypothesis guarantees the Harris-Taylor local systems involved admit a unique stable lattice, which is what makes the comparisons of nilpotency unambiguously defined.","core_discovery":"The central claim is theorem 1.2. Let $\\mathfrak m$ be a KHT-free maximal ideal of the anemic Hecke algebra, let $\\rho_{\\mathfrak m}$ be the associated mod $\\ell$ Galois representation, and assume $\\rho_{\\mathfrak m}$ is irreducible, that the irreducible subquotients of its restriction to the decomposition group at $v$ are characters, and that the set $S_v(\\mathfrak m)$ contains no chain $\\chi_v,\\chi_v(1),\\dots,\\chi_v(e_v(\\ell)-1)$. Then the partition $d_{\\mathfrak m,v}$ associated to the mod $\\ell$ unipotent monodromy at $v$ and the partition $d_{\\widetilde{\\mathfrak m},v}$ associated to any characteristic-zero lift $\\widetilde{\\mathfrak m}\\subset\\mathfrak m$ have the same maximal Jordan-block size. The content of this equality is that the nilpotency order of $N_v$ at $v$ is already determined modulo $\\ell$; hence the level at $v$ cannot be lowered while keeping the same mod $\\ell$ Satake parameters.","pith_inferences":["The character-chain condition functions as a genericity condition: it is precisely what forces uniqueness of the stable lattice in the Harris-Taylor local system. The paper's own remark indicates that, once the freeness of Lubin-Tate cohomology is available, the same theorem should hold with 'character' replaced by 'supercuspidal representation whose mod $\\ell$ reduction remains supercuspidal'; te","The equivalence between Ihara's lemma and level raising in the $q_v\\equiv 1$ case suggests that a failure of Ihara's lemma should be visible as torsion in the $E_1$ page of the nearby-cycles spectral sequence. Constructing such torsion classes explicitly would give a direct geometric obstruction to level raising.","The method is not obviously limited to KHT Shimura varieties: any setting with a Newton stratification, a local monodromy operator, and free localized cohomology should admit a similar level-fixing statement, so the paper's dichotomy (level fixed vs. level raised) could become a general principle for Shimura varieties of Hodge type."],"forward_implications":["Under the hypotheses of theorem 1.2, any characteristic-zero lift $\\widetilde{\\mathfrak m}$ of $\\rho_{\\mathfrak m}$ has, at $v$, a monodromy operator with the same maximal nilpotency as the mod $\\ell$ operator; in particular no lift can have strictly smaller Jordan blocks at $v$, so the level at $v$ is fixed.","This gives a higher-dimensional counterpart of the classical principle that level lowering is controlled by the mod $\\ell$ Galois representation: the obstruction is read off from local constituents of $\\rho_{\\mathfrak m}$ and from KHT-freeness.","When $q_v\\equiv 1\\pmod{\\ell}$ and $\\ell>d$, the level-raising property at $v$ is equivalent to the appropriate Ihara lemma, and Ihara's lemma for compact unitary groups implies it for KHT unitary groups.","Corollary 6.3 yields explicit automorphic congruences: from one lift with local component $\\mathrm{St}_h(\\chi_v)\\widehat{\\times}\\Psi_v$ at $v$, one obtains lifts with $\\mathrm{St}_h(\\chi'_v)\\widehat{\\times}\\Psi'_v$ for every character $\\chi'_v$ congruent to $\\chi_v$ modulo $\\ell$, at the same level."],"supporting_citations":[{"why":"Supplies the freeness of Lubin-Tate tower cohomology and the splitting of the nearby-cycles sheaf on which proposition 3.1 depends.","marker":"[9]"},{"why":"Provides the tensor-product description of localized cohomology used in theorem 2.12 to transfer monodromy from cohomology to $\\rho_{\\mathfrak m}$.","marker":"[29]"},{"why":"Gives the GL2 prototype of the tensor-product theorem and the original framework for comparing characteristic-zero and mod $\\ell$ monodromy.","marker":"[15]"},{"why":"Sets up the partition comparison $d_{\\mathfrak m,v}$ versus $d_{\\widetilde{\\mathfrak m},v}$ and the use of the nearby-cycles spectral sequence in this context.","marker":"[11]"},{"why":"Defines KHT-freeness and lists hypotheses that guarantee it, which theorem 1.2 assumes.","marker":"[8]"},{"why":"Describes the $\\mathbb Q_\\ell$-cohomology and resolutions of Harris-Taylor sheaves over $\\mathbb Q_\\ell$ used in proposition 3.3.","marker":"[3]"},{"why":"Constructs the $\\mathbb Z_\\ell$-filtration of the nearby-cycles perverse sheaf whose graded pieces carry the monodromy action.","marker":"[6]"},{"why":"Establishes torsion-freeness of cohomology of p-adic symmetric spaces, used in proving strictness in proposition 3.12.","marker":"[28]"}],"fun_headline_variants":["Mod l monodromy pins the level, no lowering allowed","Maximal Jordan block fixed mod l, so level holds","Higher-dimensional Mazur principle: mod l decides level","Mod l monodromy forces maximal block, no level lowering","Mod l monodromy already rules out level lowering"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument stands on the claim that the localized nearby-cycles cohomology groups are torsion free and lie only in the middle degree; this is delegated to a submitted preprint about Lubin-Tate spaces and a lemma whose proof is only a two-line sketch, so if that freeness fails the level-fixing conclusion can fail too.","fun_headline_variants_meta":{"raw":{"variants":["Mod l monodromy pins the level, no lowering allowed","Maximal Jordan block fixed mod l, so level holds","Higher-dimensional Mazur principle: mod l decides level","Mod l monodromy forces maximal block, no level lowering","Mod l monodromy already rules out level lowering"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000767,"raw_usage":{"total_tokens":3399,"prompt_tokens":944,"completion_tokens":2455,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":560,"completion_tokens_details":{"reasoning_tokens":2371}},"tokens_in":560,"tokens_out":2455,"duration_ms":17821,"temperature":1.0,"reasoning_tokens":2371,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:27:20.825447+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the spectral sequence of proposition 3.1 for one explicit KHT-free maximal ideal $\\mathfrak m$ satisfying the hypotheses and check whether every $E_1^{p,q}$ localized at $\\mathfrak m$ is free and zero outside $p+q=d-1$; a single non-zero torsion class outside the middle degree would invalidate proposition 3.1 and with it the equality of maximal nilpotency. More directly, exhibit $\\mathfrak m$ with the stated local characters for which two lifts $\\widetilde{\\mathfrak m}_1,\\widetilde{\\mathfrak m}_2$ have monodromy partitions at $v$ with different largest parts; theorem 1.2 predicts this cannot happen.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the freeness of Lubin-Tate tower cohomology and the splitting of the nearby-cycles sheaf on which proposition 3.1 depends."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the tensor-product description of localized cohomology used in theorem 2.12 to transfer monodromy from cohomology to $\\rho_{\\mathfrak m}$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the GL2 prototype of the tensor-product theorem and the original framework for comparing characteristic-zero and mod $\\ell$ monodromy."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Sets up the partition comparison $d_{\\mathfrak m,v}$ versus $d_{\\widetilde{\\mathfrak m},v}$ and the use of the nearby-cycles spectral sequence in this context."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines KHT-freeness and lists hypotheses that guarantee it, which theorem 1.2 assumes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Describes the $\\mathbb Q_\\ell$-cohomology and resolutions of Harris-Taylor sheaves over $\\mathbb Q_\\ell$ used in proposition 3.3."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Constructs the $\\mathbb Z_\\ell$-filtration of the nearby-cycles perverse sheaf whose graded pieces carry the monodromy action."},{"cited_title":"Schneider and U","cited_arxiv_id":null,"evidence_quote":"Establishes torsion-freeness of cohomology of p-adic symmetric spaces, used in proving strictness in proposition 3.12."}],"review_version":1}