{"id":"9fccef2f-b934-4a5c-a817-e6a89217ac2f","arxiv_id":"2507.08568","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A mostly expository article that re-proves known p-adic comparison theorems with elementary tools and adds a new corollary about multiplication by n on fppf cohomology of abelian varieties.","lead":"This paper gives an elementary account of Bhatt-Lurie's quasisyntomic descent and Nygaard filtration, and uses it to re-prove Illusie's comparison between fppf and crystalline cohomology in positive characteristic. It also derives new structure results for fppf cohomology groups, including the action of multiplication-by-n on abelian varieties.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.5 rests on an unproved deformation-lifting input in Theorem 3.2.3; if [BeM07] does not cover Frobenius-smooth algebras, the descent step for the triangle fails.","rationale":"The reader identified the same weakest assumption: the descent theorem for crystalline cohomology along R → R_perf depends on a deformation-theoretic input quoted from [BeM07]. I agree that this is the most load-bearing unproved step: every later reduction in the proof of Theorem 4.5 — from smooth X to eqrsp algebras — passes through Theorem 3.2.3, and without the lifting statement the Čech arguments in Construction 4.3 and Corollary 3.2.6 cannot be made. The paper is otherwise self-aware about its reliance on Bhatt–Lurie and states that Section 4 is a rewriting of [BhL22, Section 7], so the mathematical result is not in doubt. The concern is therefore about the completeness of the elementary proof rather than the validity of the theorem. Since the cited result is standard and the central theorem is already established in the literature, this does not change the acceptance verdict; it does suggest that the author should either prove the lifting lemma or state it explicitly as an imported theorem with precise hypotheses.","tokens_in":34788,"tokens_out":37271,"duration_ms":432796,"concrete_test":"Verify the cited deformation input against the precise hypotheses needed for Theorem 3.2.3: for a Frobenius-smooth F_p-algebra R and a PD thickening (A, I) with p^n = 0, does R → A/I lift to R_n → A? Work the explicit example R = B[[x]] with B a perfect F_p-algebra, constructing the lift via a p-basis and the universal property of the divided power envelope. If the lift exists in this example and [BeM07, Cor 1.2.7, Prop 1.2.6] indeed covers it, the concern is resolved; if the lift fails, Theorem 3.2.3 needs a restricted hypothesis or a new proof.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The exact triangle for general smooth X is assembled by descent along R → R_perf (Theorem 3.2.3). In the proof of Theorem 3.2.3(1), the existence of compatible lifts (R_n, F_n) to W_n(k) and of a lift R_n → A for every PD thickening (A, I) is quoted from [BeM07, Cor 1.2.7, Prop 1.2.6] without proof. This is the sole mechanism that upgrades étale descent to coperfection descent for all Frobenius-smooth F_p-algebras, and it is the step that allows the reduction of Theorem 4.5 to eqrsp algebras. The paper's Frobenius-smooth class includes non-finite-type algebras such as B[[x_1,...,x_n]] with B perfect, whereas the cited Berthelot–Messing results are formulated in a Dieudonné-theoretic setting for smooth algebras over perfect fields. If the lifting statement does not literally apply to this larger class, then Theorem 3.2.3 collapses and the proof of Theorem 4.5 for arbitrary smooth X cannot be assembled. This is a dependence on a quoted external input rather than an observed internal inconsistency; however, it is the least secure point in an otherwise elementary presentation, and it is exactly the load-bearing assumption identified by the reader.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops an elementary, ∞-category-free treatment of quasisyntomic descent and the Nygaard filtration in positive characteristic. Its central theorem (Theorem 4.5) states an exact triangle RΓ_fppf(X, Z_p(1)) → F^1_N RΓ_cris(X/W(k)) → RΓ_cris(X/W(k)) for smooth X over a perfect field k, from which Illusie's comparison H^i_fppf(X, Q_p(1)) ≅ H^i_cris(X/W(k))[1/p]^{F=p} is deduced. The proof reduces by cohomological descent to elementary quasiregular semiperfect algebras, where all three terms are computed explicitly. The paper also gives a descent proof of Ogus' comparison between infinitesimal cohomology and the unit-root part of crystalline cohomology, derives structural results for fppf cohomology (a free part plus a p-group of finite p-exponent), computes the multiplication-by-n action on fppf cohomology of abelian varieties, and works out two explicit examples.","tokens_in":35051,"tokens_out":12682,"duration_ms":145938,"significance":"If correct, the paper provides a valuable, more accessible route to Bhatt–Lurie's comparison theorem and its consequences, avoiding ∞-categories. The explicit Acris computations for eqrsp algebras are clear and useful, as are the worked examples (ordinary abelian varieties and E×E) and the new proof of the [n]-action on fppf cohomology. The proof of Ogus' theorem via descent is elegant. The genuinely new results are modest, however, and much of the paper is expository or a re-proof of known theorems; its main value is pedagogical and organizational rather than groundbreaking.","major_comments":[{"comment":"The proof of coperfection descent delegates the two crucial deformation-theoretic inputs to [BeM07, Cor. 1.2.7 and Prop. 1.2.6] without stating their precise hypotheses. This is load-bearing: the lifting lemma is what upgrades étale descent to descent along R → R_perf for every Frobenius-smooth F_p-algebra, and Theorem 4.5 uses it for arbitrary smooth X. Since the class of Frobenius-smooth algebras includes non-finite-type rings such as B[[x_1,…,x_n]] with B perfect, the reader cannot verify from the text that the cited Dieudonné-theoretic statements apply verbatim. Please state the cited results, justify their applicability to the full class of Frobenius-smooth algebras, or replace the quotation with a direct proof.","section":"§3.2, Theorem 3.2.3(1)"},{"comment":"The exactness of (4.3), which is the explicit algebraic heart of the comparison theorem, is written out only for the one-variable algebra C = B[x^{p^{-∞}}]/(x). The passage to the general eqrsp case is dismissed with the sentence 'the proof adapts as is, it is only more tedious to keep track of all indices'. Since Theorem 4.5 reduces all smooth schemes to exactly these multivariable algebras, the omitted verification is load-bearing. Please provide the multivariable computation in full or give a formal reduction (e.g. an explicit isomorphism or induction) that turns the asserted adaptation into a complete proof.","section":"§4, Theorem 4.4(2)"}],"minor_comments":[{"comment":"The proof headings appear to be interchanged: the paragraph labeled 'Proof of Theorem 3.3.1' in fact proves the unit-root description that constitutes Theorem 3.3.2, while Theorem 3.3.1 is Grothendieck's characteristic-zero statement quoted earlier. Please relabel the proofs.","section":"§3.3"},{"comment":"In the statement, 'H^i_fppf(X, Z_p(1))/tors' should presumably be 'H^i_fppf(A, Z_p(1))/tors'. The proof would also benefit from a one-line diagram chase noting that the map H^i_fppf(A) → F^1_N H^i_cris(A) is injective only after quotient by torsion, so that the action on the free quotient is indeed determined by the action on F^1_N H^i_cris(A).","section":"§5.2, Corollary 5.2.4"},{"comment":"In the proof, the algebra 'k[y^{p^{-∞}}]/(y)' should be written 'F_p[y^{p^{-∞}}]/(y)' (or with a new symbol) for consistency, since k is the base perfect field and the algebra under consideration is an F_p-algebra.","section":"§2.4, Proposition 2.4.9"},{"comment":"The rank formula g·binom(g, i-1) for H^i_fppf(A, Z_p(1)) needs a stated range for i, because for i = 0 the binomial coefficient is undefined.","section":"§5.2, Proposition 5.2.5(3)"}],"recommendation":"major_revision","confidential_remarks":"The paper is honest about its expository nature; the genuinely new results are modest. The two major comments concern load-bearing points: the applicability of the cited Berthelot–Messing deformation results to Frobenius-smooth algebras, and the unproved multivariable case of the key exactness computation. Both are fixable within the scope of the manuscript. If the journal seeks primarily large original advances, the fit is borderline, but the clarity and usefulness of the exposition justify considering a revised version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this paper is a well-written expository account of Bhatt–Lurie's approach to Illusie's comparison theorem, and it is honest about what is new. The genuinely new items are Corollary 5.2.4 (multiplication-by-n acts as n^i on fppf cohomology of abelian varieties, answering Skorobogatov's question) and the explicit computation H^3_fppf(E×E) = k for a supersingular elliptic curve E, plus a derived-category formulation of Ogus' theorem. The paper's claim to originality is modest and correct: Sections 1–4 contain only the new proof of Ogus' theorem, Section 4 is a rewriting of part of Bhatt–Lurie, and most of Section 5 is new proofs of known results.\n\nWhat the paper does well: the descent framework is kept light, the explicit algebra for elementary quasiregular semiperfect rings is spelled out in real detail (at least in one variable), and the examples at the end show the machinery actually computes something. The writing is clear and the authors are appropriately careful about credits.\n\nSoft spots: the multi-variable case of Theorem 4.4 is dismissed with 'adapts as is', which is probably fine but should be checked. More importantly, Theorem 3.2.3 — the coperfection descent theorem that the whole proof of Theorem 4.5 depends on — quotes [BeM07, Cor 1.2.7, Prop 1.2.6] for the existence of compatible lifts (R_n, F_n) and lifts to PD thickenings for Frobenius-smooth algebras. The class includes rings like B[[x_1,...,x_n]] with B perfect, which is outside the smooth finite-type setting I associate with Dieudonné theory. The stress-test note is right: if that reference does not literally cover Frobenius-smooth algebras, the descent statement collapses and the argument for arbitrary smooth X cannot be assembled. I cannot check the original right now, but the authors should be asked to state the hypotheses of the cited results and explain why they apply. It is a quotable external input rather than an internal inconsistency, so it does not sink the paper, but it is the weakest link.\n\nBottom line: for a reader who wants an elementary path to Illusie's comparison and the Nygaard exact triangle without infinity-categories, this is useful. For a specialist, the new corollary and the supersingular example are worth having. I would cite it and I would send it to a referee who can verify the Berthelot–Messing citation. Recommendation: accept, with that verification.","headline":"Honest, useful re-derivation of known comparisons plus one new corollary and a nice explicit example; the load-bearing lifting input in Theorem 3.2.3 deserves a check, but the paper is worth refereeing.","tokens_in":35593,"tokens_out":2804,"would_cite":true,"duration_ms":31920,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14F30","14F20","14K15"],"pacs":[],"model":"deepseek-v4-flash","headline":"The Nygaard filtration gives an elementary route to Illusie's p-adic comparison theorem.","keywords":["Nygaard filtration","quasisyntomic descent","crystalline cohomology","fppf cohomology","Illusie comparison","Frobenius-smooth algebras","elementary quasiregular semiperfect algebras","Brauer group"],"falsifier":"Exhibit a Frobenius-smooth $\\mathbb{F}_p$-algebra $R$ and a divided-power thickening $(A,I)$ over $W_n(k)$ for which the map $R \\to A/I$ does not lift to a map $R_n \\to A$ compatibly with Frobenius lifts; that would break Theorem 3.2.3 and the descent argument underlying Theorem 4.5. Alternatively, find an elementary quasiregular semiperfect algebra $C$ for which the mod-$p$ sequence $(1+J)^\\times/p \\to F^1_N A_{\\mathrm{cris}}(C)/p \\to A_{\\mathrm{cris}}(C)/p$ has a nonzero kernel, contradicting Theorem 4.4.","tokens_in":2441,"feed_emoji":"🔺","tokens_out":7939,"duration_ms":115664,"temperature":0.7,"pith_summary":"This paper establishes a new proof of Illusie's comparison theorem: for a smooth proper variety $X$ over an algebraically closed field $k$ of characteristic $p$, the fppf cohomology with $\\mathbb{Q}_p(1)$ coefficients is isomorphic to the slope-1 Frobenius-isotypical part of crystalline cohomology, $H^i_{\\mathrm{cris}}(X/W(k))[1/p]^{F=p}$. The proof goes through an exact triangle relating fppf cohomology with $\\mathbb{Z}_p(1)$ to the first Nygaard filtration piece of crystalline cohomology via the map $F/p-1$, using quasisyntomic descent to reduce to elementary quasiregular semiperfect algebras. Along the way the paper gives a new proof of Ogus' comparison between infinitesimal and crystalline cohomology, derives structural results on $H^i_{\\mathrm{fppf}}(X,\\mathbb{Z}_p(1))$ (a direct sum of a free module and a $p$-torsion group of finite exponent), and determines the action of multiplication-by-$n$ on the fppf cohomology of abelian varieties, answering a question of Skorobogatov. This approach avoids the formalism of $\\infty$-categories, replacing Bhatt-Lurie's machinery with canonical complexes built from perfections.","feed_headline":"Nygaard filtration yields new route to Illusie comparison theorem","feed_subtitle":"An elementary descent proof, plus explicit structure of fppf cohomology of varieties and abelian varieties.","key_machinery":"The named machinery is the first piece of the Nygaard filtration, $F^1_N R\\Gamma_{\\mathrm{cris}}(X/W(k))$, defined as the cohomology of the kernel $I_{\\mathrm{cris}}$ of the surjection of crystalline structure sheaves $O_{\\mathrm{cris}} \\to \\mathbb{G}_a$; it sits in an exact triangle $F^1_N R\\Gamma_{\\mathrm{cris}} \\to R\\Gamma_{\\mathrm{cris}} \\to R\\Gamma(X,O_X)$. A completed first Chern class $\\hat{c}_1 : R\\Gamma_{\\mathrm{fppf}}(X,\\mathbb{Z}_p(1)) \\to F^1_N R\\Gamma_{\\mathrm{cris}}$ refines the usual Chern class, and the map $F/p-1$ (Frobenius divided by $p$ minus identity, defined on elementary quasiregular semiperfect algebras via explicit descriptions of $A_{\\mathrm{cris}}$) completes the triangle. Descent along $R \\to R_{\\mathrm{perf}}$ for crystalline cohomology, proved using lifts of Frobenius-smooth algebras to $W_n(k)$ quoted from Berthelot-Messing, reduces everything to algebras $C = B[x_1^{p^{-\\infty}}, \\dots, x_n^{p^{-\\infty}}]/(x_1,\\dots,x_n)$, where $A_{\\mathrm{cris}}(C)$ is an explicit divided-power power series ring and exactness of the triangle is checked modulo $p$.","core_discovery":"Theorem 4.5 asserts that for $X$ smooth over a perfect field $k$, the sequence $R\\Gamma_{\\mathrm{fppf}}(X,\\mathbb{Z}_p(1)) \\to F^1_N R\\Gamma_{\\mathrm{cris}}(X/W(k)) \\xrightarrow{F/p-1} R\\Gamma_{\\mathrm{cris}}(X/W(k))$ is an exact triangle in $D(\\mathbb{Z}_p)$. From this triangle the paper recovers Illusie's comparison: when $k$ is algebraically closed and $X$ is smooth and proper, $H^i_{\\mathrm{fppf}}(X,\\mathbb{Q}_p(1)) \\cong H^i_{\\mathrm{cris}}(X/W(k))[1/p]^{F=p}$. The triangle is proved by descent: first to affine $X$, then along $X_{\\mathrm{perf}} \\to X$ to elementary quasiregular semiperfect algebras, where crystalline cohomology is $A_{\\mathrm{cris}}(C)$, the Nygaard filtration is explicit, and exactness becomes a direct computation with power series.","pith_inferences":["The paper's replacement of $\\infty$-categories by canonical complexes built from perfections suggests the same descent package could be adapted to other $p$-adic cohomology functors that also have explicit descriptions on semiperfect algebras.","For straight varieties the Nygaard filtration is determined by the $F$-crystal only under the Mazur-Ogus Newton-above-Hodge hypothesis; the paper's examples suggest that the exact triangle itself may still determine fppf cohomology from the $F$-crystal when that hypothesis fails.","The computation $H^3_{\\mathrm{fppf}}(E\\times E,\\mathbb{Z}_p(1)) \\cong k$ for a supersingular elliptic curve $E$ shows that torsion fppf cohomology can carry positive-dimensional information (the unipotent group $\\mathbb{G}_a$), which may have consequences for Brauer-group computations on supersingular surfaces.","Corollary 5.2.4 is stated for abelian varieties; a testable extension is whether the same $n^i$ action holds on the fppf cohomology of any straight variety carrying a multiplication-by-$n$ endomorphism."],"forward_implications":["Illusie's comparison theorem $H^i_{\\mathrm{fppf}}(X,\\mathbb{Q}_p(1)) \\cong H^i_{\\mathrm{cris}}(X/W(k))[1/p]^{F=p}$ follows as a direct corollary of the exact triangle, with properness used only to make the maps $F/p-1$ surjective after inverting $p$.","For every $i$, $H^i_{\\mathrm{fppf}}(X,\\mathbb{Z}_p(1))$ is the direct sum of a free $\\mathbb{Z}_p$-module of rank $\\operatorname{rank} H^i_{\\mathrm{cris}}(X)^{F=p}$ and a $p$-torsion group of finite $p$-exponent; for $i=1,2$ the groups are finite-type $\\mathbb{Z}_p$-modules.","For straight varieties (torsion-free crystalline cohomology and a degenerating Hodge-de Rham spectral sequence, e.g. abelian varieties, K3 surfaces, complete intersections), fppf cohomology is completely determined by the $F$-crystal $H^i_{\\mathrm{cris}}$, and $F^1_N H^i = F^{-1}(pH^i)$.","On an abelian variety $A$, multiplication-by-$n$ acts as $n^i$ on both $H^{i+1}_{\\mathrm{fppf}}(A,\\mathbb{Z}_p(1))_{\\mathrm{tors}}$ and $H^i_{\\mathrm{fppf}}(A,\\mathbb{Z}_p(1))/\\mathrm{tors}$, answering Skorobogatov's question.","The same descent formalism gives a new proof of Ogus' theorem $R\\Gamma_{\\mathrm{inf}}(X/W(k)) \\simeq R\\lim_F R\\Gamma_{\\mathrm{cris}}(X/W(k))$."],"supporting_citations":[{"why":"Theorem 7.3.5 is the source of the exact triangle being proved; the paper rewrites its proof without infinity-categories.","marker":"[BhL22]"},{"why":"Théorème II.5.5 is the comparison theorem being reproved, and the paper also draws from it the isocrystal surjectivity lemma for F/p-1 after inverting p.","marker":"[Ill79]"},{"why":"Corollary 1.2.7 and Proposition 1.2.6 supply the deformation-theoretic lifts of Frobenius-smooth algebras on which descent along R to R_perf rests.","marker":"[BeM07]"},{"why":"Gives the explicit description of A_cris for elementary quasiregular semiperfect algebras used in Theorem 4.4.","marker":"[Dri20]"},{"why":"Theorem 8.26 (Mazur-Ogus) is the input for identifying the Nygaard filtration with F^{-1}(pH^i) on straight varieties.","marker":"[BeO78]"},{"why":"Supplies the infinitesimal-cohomology comparison theorem that the paper reproves in Section 3.3.","marker":"[Og75]"},{"why":"Acknowledged as the origin of the F/p-1 triangle via the syntomic topology, to which Theorem 1.5 goes back.","marker":"[FoM87]"}],"fun_headline_variants":["Elementary descent proof of Illusie's comparison theorem","Nygaard filtration simplifies p-adic cohomology comparisons","Quasisymptotic descent yields new fppf cohomology proofs","Answering Skorobogatov on abelian variety fppf cohomology","Exact triangle links fppf and crystalline cohomology"],"cache_read_input_tokens":37760,"weakest_assumption_plain":"The descent theorem for crystalline cohomology along $R \\to R_{\\mathrm{perf}}$ assumes that every Frobenius-smooth $\\mathbb{F}_p$-algebra $R$ has compatible lifts $(R_n, F_n)$ to $W_n(k)$ and that every divided-power thickening lifts as well, a deformation-theoretic input quoted from Berthelot-Messing; if that input fails, the descent step that assembles the exact triangle for general smooth $X$ collapses.","fun_headline_variants_meta":{"raw":{"variants":["Elementary descent proof of Illusie's comparison theorem","Nygaard filtration simplifies p-adic cohomology comparisons","Quasisymptotic descent yields new fppf cohomology proofs","Answering Skorobogatov on abelian variety fppf cohomology","Exact triangle links fppf and crystalline cohomology"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000214,"raw_usage":{"total_tokens":1432,"prompt_tokens":956,"completion_tokens":476,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":572,"completion_tokens_details":{"reasoning_tokens":384}},"tokens_in":572,"tokens_out":476,"duration_ms":5098,"temperature":1.0,"reasoning_tokens":384,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T18:15:40.490846+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhibit a Frobenius-smooth $\\mathbb{F}_p$-algebra $R$ and a divided-power thickening $(A,I)$ over $W_n(k)$ for which the map $R \\to A/I$ does not lift to a map $R_n \\to A$ compatibly with Frobenius lifts; that would break Theorem 3.2.3 and the descent argument underlying Theorem 4.5. Alternatively, find an elementary quasiregular semiperfect algebra $C$ for which the mod-$p$ sequence $(1+J)^\\times/p \\to F^1_N A_{\\mathrm{cris}}(C)/p \\to A_{\\mathrm{cris}}(C)/p$ has a nonzero kernel, contradicting Theorem 4.4.","supporting_citations":[],"review_version":1}