{"id":"057304f3-cab2-405a-9859-4d7d5513de73","arxiv_id":"2507.15175","paper_version":1,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A de Rham-Higgs comparison for mixed-Hodge-module-inspired subcomplexes is established in special positive-characteristic cases, yielding decomposition and vanishing theorems when the logarithmic cotangent bundle splits into small-rank subbundles.","lead":"This paper proposes and partially proves a positive-characteristic analogue of the Deligne-Illusie decomposition theorem for subcomplexes inspired by mixed Hodge modules, relating Frobenius-pushed de Rham complexes to Higgs complexes. The main results give new decomposition, Kunneth, and vanishing theorems under a logarithmic cotangent splitting condition, including new cases for abelian varieties.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.4's proof requires every coordinate change fixing f to factor into determinant-one elementary matrices, but on X=G_m×G_m with f=xy and m2=x^2/y the matrix A has det -3, so the claimed chain (53) of f-compatible splittings does not exist as written.","rationale":"I read the paper in good faith. The central positive results are conditional: Theorem 4.2 and Corollary 1.3 assume the strong splitting hypothesis (21), and the reader correctly flags this as limiting their scope. However, the single most load-bearing step for the stronger claim Theorem 4.4, which is supposed to avoid the splitting hypothesis, is the reduction to elementary-matrix coordinate changes. The determinant obstruction is not a matter of disagreement with consensus; it is an internal algebraic inconsistency in the proof as written. It does not prove the theorem false, only that the current argument is incomplete and needs a substantial fix. I therefore keep the reader's CONDITIONAL verdict, so verdict_should_be is UNCHANGED. My agreement with the reader is partial: the splitting limitation is real, but my identified concern is the concrete false step in the proof of Theorem 4.4.","tokens_in":48837,"tokens_out":14967,"duration_ms":181658,"concrete_test":"Verify the factorization claim directly: for X = G_m × G_m, p ≠ 3, f = xy, take old basis (dlog f, dlog y) and new basis (dlog f, dlog(x^2/y)). The transition matrix is [[1,2],[0,-3]], determinant -3, while every elementary matrix allowed in the proof of Theorem 4.4 has determinant 1. Thus A is not a product of the allowed matrices, so the chain (53) used to prove Ho_Ω,IV,a = Ho_Ω',IV,a does not exist for this pair of coordinate systems. If this determinant obstruction is real, the proof of Theorem 4.4 must be revised; if the authors can exhibit a different chain or normalize determinants, the concern is answered.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The unprotected step is in the proof of Theorem 4.4 (§4.4). After choosing two systems of coordinates with f = t1...tn = m1...mn, the proof encodes the coordinate change by a matrix A satisfying (dlog f, dlog m2, ..., dlog mn) = (dlog f, dlog t2, ..., dlog tn) A and asserts, by 'basic linear algebra and shrinking X', that A is a product of the listed elementary matrices, each of determinant one. Since every listed factor has determinant 1, such a factorization forces det A = 1. But det A need not be 1. Over k with p ≠ 3, take X = G_m × G_m, f = xy, t2 = y, m1 = y^2/x, m2 = x^2/y. Then m1 m2 = f, (m1, m2) is a coordinate system, and dlog m2 = 2 dlog f - 3 dlog y, so A = [[1,2],[0,-3]] with det A = -3. Shrinking X to any Zariski open cannot change this stalk value. Hence the claimed chain (53) of f-compatible splittings, with successive splittings related by (40), is not justified; Proposition 4.1 cannot be applied along the chain. The theorem may still be true, but this concrete false reduction is a load-bearing gap in the current proof.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a positive-characteristic generalization of the Deligne–Illusie decomposition theorem in which the complexes under comparison are subcomplexes inspired by mixed Hodge modules (weight filtrations, intersection complexes, Kontsevich subcomplexes), twisted by a global function f. The main construction combines the inverse Cartier transform with a deformed Higgs field Θ, yielding a formal-local comparison near the zeros of certain coefficients of f (Theorem 3.1). A global comparison is then proved under the assumption that the logarithmic cotangent bundle splits as a direct sum of subbundles of rank < p−ℓ (Theorem 4.2), with results on independence of the splitting (Proposition 4.1), compatibility with Hodge pairs of types I–IV (Theorem 4.3), and two-term truncations for arbitrary a (Theorem 4.4). The final sections derive Künneth, E1-degeneration, and vanishing consequences in positive characteristic and characteristic zero.","tokens_in":49085,"tokens_out":7582,"duration_ms":87561,"significance":"If the main results are correct, they give a substantial extension of the Ogus–Vologodsky and Sheng–Zhang comparisons and a new route to Deligne–Illusie-type decompositions for twisted complexes, with applications to abelian varieties and semistable families. The formal-local theorem of Section 3 is a genuine contribution, with an explicit deformation via Artin–Hasse exponentials. However, the global theorems are heavily conditional on the splitting assumption (21), which is strong and fails for general smooth varieties; consequently Corollary 1.3 and the applications in Section 5 are limited to a special class of varieties. More importantly, several load-bearing steps in the proofs of Sections 4.2 and 4.4 are either omitted or based on incorrect assertions, so the global comparison theorems are not established as written.","major_comments":[{"comment":"The proof asserts that, after shrinking X, the change-of-basis matrix A defined by (dlog f, dlog m2, ..., dlog mn) = (dlog f, dlog t2, ..., dlog tn)A can be written as a product of elementary matrices of the three listed types. Since every listed transformation has determinant ±1, such a factorization would force det A = ±1. This is false in general. For example, on X = G_m × G_m over a field k of characteristic p ≠ 3, take f = xy, t2 = y, m1 = y^2/x, and m2 = x^2/y. Then m1m2 = xy = f, and dlog m2 = 2 dlog f − 3 dlog y, so A = [[1,2],[0,−3]] has determinant −3. Passing to a Zariski open subset cannot change this determinant because it is a nonzero constant in k. Hence the required chain (53) of f-compatible splittings does not exist by this argument, and Proposition 4.1 cannot be applied along such a chain. This is a load-bearing gap in the proof of Theorem 4.4.","section":"§4.4, proof of Theorem 4.4, chain (53)"},{"comment":"Lemma 4.2 states that Construction 4.3 is independent of the choice of basis, but its proof is given as 'similar to that of Lemma 4.1, and we omit the details.' This lemma is the key input to Proposition 4.1, which in turn underlies the compatibility theorem and the two-term truncation theorem. The verification is not a routine detail: Construction 4.3 involves the coefficient C_o and the index set H_{Ω,o}, both of which are substantially more complicated than the corresponding data in Lemma 4.1. Without a complete proof, the homotopy between Ho_Ω and Ho_Ω′ is not established.","section":"§4.2, Lemma 4.2"},{"comment":"Even if Lemma 4.2 were available, the proof of Proposition 4.1 relies on a very long coefficient comparison. The identities c^L_{S,i,j} = c^R_{S,i,j}, c^L_{l,q,i,S,i,j} = 0, and c^L_{u,q,i,S,i,j} = 0 are verified only in 'typical situations', and the remaining cases are dismissed by 'direct computation'. Since this proposition is the only mechanism for proving that the de Rham–Higgs comparison is independent of the chosen splitting, the proof as written is not complete. A full verification, or a more conceptual argument, is needed before Theorem 4.3 and Theorem 4.4 can be accepted.","section":"§4.2, proof of Proposition 4.1"},{"comment":"The proof of Theorem 4.4 explicitly treats only Hodge pairs of type (IV,0) and says that the first three types 'can be checked in a similar manner.' Given the intricate compatibility and independence arguments in Sections 4.2–4.3, this is not sufficient. In particular, type (II,ℓ) involves the intersection subcomplexes whose compatibility is proved separately via [SZ2], and type (III,ℓ) involves the Kontsevich subcomplexes; the analogous chain-of-splittings argument is not supplied for these cases.","section":"§4.4, Theorem 4.4, proof for all types"}],"minor_comments":[{"comment":"The phrase 'de Rhm-Higgs comparison' should read 'de Rham-Higgs comparison'.","section":"Introduction, Question 1.1"},{"comment":"The word 'semistble' is a typo for 'semistable'.","section":"§2.6, type (III)"},{"comment":"The sentence 'Since ˆO_{X,x} is fully faithful flat over O_{X,x}' should say 'faithfully flat'; the argument only requires faithful flatness to detect non-vanishing of coherent sheaves.","section":"Corollary 3.2, proof"},{"comment":"In the sentence 'V (t^α ω_I) = V (γ_i) or V (−γ_i), i ≤ i ≤ n', the index range should be '1 ≤ i ≤ n'.","section":"§3.2, Definition 3.2 and following"},{"comment":"In the proof of Theorem 5.1, the text says 'For i = 1, 2, 3, we set H^*_i ...' but only i = 1, 2 are used; this is confusing and should be corrected.","section":"§5.1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of the journal and contains promising ideas, especially the formal-local comparison of Section 3. However, the proof of the global comparison theorems depends on omitted or questionable computational steps, and the proof of Theorem 4.4 contains a concrete false linear-algebra claim. I would recommend asking the authors to supply a complete proof of Proposition 4.1 (including Lemma 4.2) and a corrected argument for the independence of the splitting in Theorem 4.4, and to clarify exactly which applications survive without the full untruncated comparison."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Zebao's paper has real content. The abelian-variety decomposition for varieties that are not quasi-F-split (Corollary 1.3), the Künneth formula for the Deligne–Illusie decomposition (Theorem 5.1), and the inverse Cartier transform for p-connections (Section 5.2) are genuinely new, and Section 3's formal-neighborhood comparison with exponential twisting is a clean extension of Ogus–Vologodsky. The higher-homotopy method from Sheng–Zhang is pushed quite far.\n\nThree things to flag before you invest time. First, the global theorems all depend on the splitting hypothesis (21): the logarithmic cotangent bundle must split into subbundles of rank < p−ℓ. That is strong and essentially automatic only for abelian varieties, so the title overstates the scope. No real mixed Hodge modules appear; the K* complexes are subcomplexes inspired by them. Second, Lemma 4.2 is stated without proof, and the independence Proposition 4.1 is verified only in representative combinatorial cases. That pattern is a concern, but it becomes concrete in Theorem 4.4. There the proof asserts that any coordinate change fixing f factors into determinant-one elementary matrices via a chain like (53). On X=G_m×G_m with f=xy and m2=x^2/y, the relevant matrix A has determinant −3, so no such factorization exists even after shrinking X. The chain of f-compatible splittings is therefore not justified, and the two-term truncation theorem currently lacks a proof.\n\nThe paper deserves a serious referee: the machinery is substantial, the byproducts are valuable, and the formal-neighborhood and support results are likely correct. But I would not accept this version. I'd ask for a corrected proof of Theorem 4.4, a full proof of Lemma 4.2, and a softened title. After that, Corollary 1.3 and the Künneth formula would be worth citing.","headline":"Valuable new results, but Theorem 4.4's proof has a false determinant-one factorization claim, so the paper needs a serious revision before it can be trusted.","tokens_in":49681,"tokens_out":3488,"would_cite":false,"duration_ms":36427,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14F30","14F40","14G17","14D07"],"pacs":[],"model":"deepseek-v4-flash","headline":"Frobenius pushforwards of twisted de Rham subcomplexes match Higgs complexes at every closed point and globally when the cotangent splits.","keywords":["positive characteristic","de Rham–Higgs comparison","mixed Hodge modules","inverse Cartier transform","Deligne–Illusie decomposition","Higgs bundles","Frobenius pushforward","E1-degeneration"],"falsifier":"Check a supersingular abelian variety of dimension $> p$ (for instance a supersingular abelian fourfold over $\\mathbb{F}_2$: dimension $4 > p = 2$ and $p$-rank $0 < g-1 = 3$, so it is not quasi-$F$-split). Corollary 1.3 predicts $F_\\ast\\Omega^\\bullet_{A/k}$ is decomposable in $D(A)$; computing its Frobenius pushforward through Hodge--Witt cohomology or the de Rham--Dieudonné theory cited in the paper and finding no direct-sum splitting would refute the central claim, while confirming a split where the quasi-$F$-split theorem does not apply would corroborate it.","tokens_in":48535,"feed_emoji":"🧩","tokens_out":25155,"duration_ms":222006,"temperature":0.7,"pith_summary":"This paper tries to show that the Frobenius pushforward of a twisted de Rham complex can be identified, truncation by truncation or even in full, with the corresponding Higgs complex, for the subcomplexes a mixed Hodge module theorist would single out: weight filtrations, intersection complexes, and Kontsevich subcomplexes. The comparison is posed with an arbitrary twisting function $f$, matching the operator $\\nabla + df\\wedge$ against $\\theta - df'\\wedge$. The paper proves the formal statement at every closed point by deforming the Higgs field through a variant of the $\\alpha$-transform, then obtains the global untruncated comparison under a splitting hypothesis: the logarithmic cotangent bundle $\\Omega^1_{X/k}(\\log D)$ must decompose into subbundles of rank $< p - \\ell$. A byproduct is that the de Rham complex of every abelian variety, even a supersingular one of dimension greater than $p$, is decomposable after Frobenius. Two-term truncations, with no splitting hypothesis, are shown to match for nilpotent level at most $p-3$, and applications include a Künneth formula, $E_1$-degeneration, and vanishing theorems.","feed_headline":"Full Frobenius de Rham–Higgs match under cotangent splitting","feed_subtitle":"Generalizes Deligne–Illusie, works for all abelian varieties, and needs no cotangent splitting for two-term truncations.","key_machinery":"Two mechanisms carry the argument. The first is a deformed Higgs field, a variant of the $\\alpha$-transform: on the completion along the zero scheme of the coefficients $f_{i_1\\cdots i_n}$, the Higgs field $\\hat{\\theta} - df\\wedge$ is replaced by $\\Theta := \\hat{\\theta} - \\sum_{i_1,\\ldots,i_n}\\sum_{j\\ge 0} f_{i_1\\cdots i_n}^{p^j-1}\\, df_{i_1\\cdots i_n}\\wedge$, whose Higgs complex is isomorphic to the original twisted one while its inverse Cartier transform is isomorphic to $(\\hat{H}, \\hat{\\nabla} + df\\wedge)$; the isomorphism of Cartier transforms is multiplication by the Artin--Hasse exponential $\\prod \\mathrm{AH}(f_{i_1\\cdots i_n})$. The second is the splitting hypothesis $\\Omega^1_{X/k}(\\log D) = \\bigoplus_{i=1}^{\\beta}\\Omega_i$ with $\\mathrm{rank}\\,\\Omega_i < p - \\ell$, which feeds an explicit $L$-indexed $\\infty$-homotopy $Ho_\\Omega$ built from higher-homotopy formulas $\\varphi^{(r,s)}$ -- combinatorial sums over Frobenius liftings with coefficients $C(i,S,j)$ -- realizing the quasi-isomorphism $\\varphi_\\Omega: K^\\bullet_{\\mathrm{Hig}} \\to \\check{C}(U', F_\\ast K^\\bullet_{\\mathrm{dR}})$. The same formulas decompose $F_\\ast\\Omega^\\bullet_{X/k}(\\log D)$ into Koszul complexes $K^\\ast_v$ that are acyclic except at $v = 0$; compatibility of the homotopy with the divisor or the semistable map (Definition 4.2) makes the isomorphism restrict to the subcomplexes, and independence of the splitting up to homotopy yields the two-term truncation theorem.","core_discovery":"The central claim is that the paper's Question 1.1 has an affirmative answer in three regimes. Formally: for every closed point $x$ and every global function $f$, $F_\\ast(K^\\bullet_{\\mathrm{dR}}, \\nabla + df\\wedge) \\otimes \\hat{\\mathcal{O}}_{X',x'} \\cong (K^\\bullet_{\\mathrm{Hig}}, \\theta - df'\\wedge) \\otimes \\hat{\\mathcal{O}}_{X',x'}$ in the derived category, for all three kinds of subcomplexes; a corollary is that the cohomology supports of the two twisted complexes agree, and if that support is a finite set of closed points the local isomorphisms glue to a global isomorphism. Globally and without truncation: if $\\Omega^1_{X/k}(\\log D)$ splits as $\\bigoplus_i \\Omega_i$ with $\\mathrm{rank}\\,\\Omega_i < p - \\ell$, then $F_\\ast\\Omega^\\ast(H,\\nabla) \\cong \\Omega^\\ast(E,\\theta)$ in $D(X)$, and more generally $\\tau_{<q}K^\\bullet_{\\mathrm{Hig}} \\cong \\tau_{<q}F_\\ast K^\\bullet_{\\mathrm{dR}}$ for the mixed-Hodge-type subcomplexes, with $q = \\infty$ under the full splitting and $q = p - \\ell$ otherwise; the isomorphism is independent of the chosen splitting up to homotopy. Unconditionally: for Hodge pairs of types I--IV with nilpotent level at most $p-3$, the two-sided truncations agree, $\\tau_{[a,a+1]}F_\\ast K^\\bullet_{\\mathrm{dR}} \\cong \\tau_{[a,a+1]}K^\\bullet_{\\mathrm{Hig}}$ in $D(X')$ for every $a \\ge 0$, giving isomorphisms $F_\\ast H^a(K^\\bullet_{\\mathrm{dR}}) \\cong H^a(K^\\bullet_{\\mathrm{Hig}})$ of cohomology sheaves.","pith_inferences":["Because the formal-local comparison needs no hypothesis on $f$ or on the splitting, the global obstruction is purely a gluing problem for the deformed Higgs fields $\\Theta$; any geometric condition that makes the local isomorphisms compatible on overlaps -- not only a cotangent splitting -- should yield the same untruncated comparison, a testable route to weaken the splitting hypothesis.","The abelian-variety corollary suggests the splitting hypothesis is a condition on the cotangent sheaf (a direct sum of small pieces), not on $F$-splitting; one could test whether smooth varieties in characteristic $p$ whose tangent bundle splits into line bundles -- the positive-characteristic analogue of split-tangent varieties -- all satisfy the untruncated decomposition.","The two-term truncation theorem is unconditional and is the part most likely to extend: if the argument is stable under variation of the twisting function $f$, it could yield a mod-$p$ proof of degeneration of the full irregular Hodge filtration in characteristic 0, going beyond the dimension equalities and vanishing statements derived in Section 5."],"forward_implications":["Under the splitting hypothesis the truncation disappears: $F_\\ast\\Omega^\\ast(H,\\nabla) \\cong \\Omega^\\ast(E,\\theta)$ in $D(X)$, a full untruncated de Rham--Higgs comparison that extends the Deligne--Illusie decomposition beyond the range where truncation is usually necessary.","If $\\Omega^1_{X/k}(\\log D)$ splits into subbundles of rank $< p$, then $F_\\ast\\Omega^\\bullet_{X/k}(\\log D)$ is decomposable; in particular every abelian variety, including supersingular ones of dimension $> p$ with $p$-rank $< g-1$ (hence not quasi-$F$-split), has decomposable Frobenius-pushed de Rham complex.","For Hodge pairs of types I--IV with nilpotent level at most $p-3$, one gets $\\tau_{[a,a+1]}F_\\ast K^\\bullet_{\\mathrm{dR}} \\cong \\tau_{[a,a+1]}K^\\bullet_{\\mathrm{Hig}}$ in $D(X')$ for every $a \\ge 0$, and hence $F_\\ast H^a(K^\\bullet_{\\mathrm{dR}}) \\cong H^a(K^\\bullet_{\\mathrm{Hig}})$.","Applications include a Künneth formula for the decomposition, $E_1$-degeneration of Hodge-to-de Rham spectral sequences for Fontaine--Faltings modules (recovering Oda's degeneration and Mumford's vanishing for supersingular abelian varieties), and mod-$p$ proofs of characteristic-0 results: dimension equalities for twisted de Rham cohomology and Kodaira--Saito vanishing."],"supporting_citations":[{"why":"Supplies the decomposition theorem for truncated Frobenius pushforwards of de Rham complexes that this paper generalizes, along with the W2(k)-liftability framework.","marker":"[DI]"},{"why":"Establishes the inverse Cartier transform and the truncated de Rham--Higgs comparison that Theorem 4.2 removes the truncation from.","marker":"[OV]"},{"why":"Extends the nonabelian Hodge correspondence and the truncated comparison to logarithmic settings with the divisor D.","marker":"[Sche]"},{"why":"Provides the intersection de Rham subcomplexes and the infinity-homotopy Cech method that Section 4 refines into the splitting-dependent quasi-isomorphism.","marker":"[SZ2]"},{"why":"Constructs the single-splitting morphism that Proposition 1.2 compares the general homotopy against.","marker":"[SZ1]"},{"why":"Contributes the Cech construction and abstract Koszul complex technique behind the higher homotopies and the two-term truncation refinement.","marker":"[AS]"},{"why":"Introduces Fontaine--Faltings modules, p-connections, and the inverse Cartier transform variant used in the mod p^n results of Section 5.","marker":"[Fa]"},{"why":"Supplies the alpha-transform analogue and the prismatic viewpoint on Drinfeld's two-sided truncation decomposition.","marker":"[O24]"},{"why":"Proves full decomposability for quasi-F-split varieties, the result that Corollary 1.3 extends to abelian varieties that are not quasi-F-split.","marker":"[P2]"}],"fun_headline_variants":["De Rham–Higgs for mixed Hodge modules in char p","Deligne–Illusie generalized to mixed Hodge modules","No cotangent splitting needed for truncations","De Rham–Higgs match for all abelian varieties","Two-term truncations match without full splitting"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the logarithmic cotangent bundle $\\Omega^1_{X/k}(\\log D)$ splits into a direct sum of subbundles, each of rank strictly below $p - \\ell$ and, for the subcomplex results, with the splitting compatible with the divisor or the semistable map; the global untruncated theorems have no stated proof when this splitting fails.","fun_headline_variants_meta":{"raw":{"variants":["De Rham–Higgs for mixed Hodge modules in char p","Deligne–Illusie generalized to mixed Hodge modules","No cotangent splitting needed for truncations","De Rham–Higgs match for all abelian varieties","Two-term truncations match without full splitting"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000804,"raw_usage":{"total_tokens":3575,"prompt_tokens":1029,"completion_tokens":2546,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":645,"completion_tokens_details":{"reasoning_tokens":2465}},"tokens_in":645,"tokens_out":2546,"duration_ms":20906,"temperature":1.0,"reasoning_tokens":2465,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:39:22.938111+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Check a supersingular abelian variety of dimension $> p$ (for instance a supersingular abelian fourfold over $\\mathbb{F}_2$: dimension $4 > p = 2$ and $p$-rank $0 < g-1 = 3$, so it is not quasi-$F$-split). Corollary 1.3 predicts $F_\\ast\\Omega^\\bullet_{A/k}$ is decomposable in $D(A)$; computing its Frobenius pushforward through Hodge--Witt cohomology or the de Rham--Dieudonné theory cited in the paper and finding no direct-sum splitting would refute the central claim, while confirming a split where the quasi-$F$-split theorem does not apply would corroborate it.","supporting_citations":[],"review_version":1}