{"id":"cf60406e-cc42-429e-a75c-f6bd95961cc9","arxiv_id":"2607.26323","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Two-dimensional dominoes in de Rham–Witt cohomology are classified by orbits of Frobenius-skew polynomials, all dominoes admit a triangular normal form, and the resulting theory bounds the p-primary Brauer exponent of supersingular abelian varieties in terms of the a-number.","lead":"This paper classifies 'dominoes' — non-finitely-generated pieces of de Rham–Witt cohomology that carry the nonzero differentials of the slope spectral sequence — in dimension two, and puts dominoes of any dimension into a normal form. It uses this structure to compute Brauer-type invariants of supersingular abelian varieties and to bound the p-primary Brauer exponent in terms of the a-number, for every prime p, answering a question of Grammatica–Skorobogatov–Yang.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.22 Step 3 equality φ(N)=V_K^{-1}M is the load-bearing step; it depends on [BO78, Thm 8.26(1)] to identify reduction of the Nygaard filtration with the Hodge filtration. If this identification fails, Theorems D–F and the a-number bounds collapse.","rationale":"The reader's weakest assumption identified the Mazur–Ogus/F-crystal reconstruction chain; my stress-test narrows this to a single unverified off- cited equality in Theorem 4.22 Step 3. This is the point where the abstract's geometric payoff—the a-number bounds on the Brauer p-exponent and σ_Art—connects to the algebraic classification. The structural results in §3 appear internally coherent, and the lower-bound arguments in §5 are carefully derived once this reconstruction is granted. However, the proof of φ(N)=V_K^{-1}M is not self-contained: it depends on a specific interpretation of [BO78, Thm 8.26(1)] that is not demonstrated in the text. Because this equality is necessary for K^{-1}=ker(dV) and hence for Theorems D–F, I recommend conditional acceptance pending an independent check on the simplest nontrivial cases. This is not a claim of error, only a precise identification of the least secure load-bearing link.","tokens_in":56094,"tokens_out":34498,"duration_ms":287196,"concrete_test":"Compute both sides for a concrete supersingular abelian surface A (e.g., the cyclic supergeneral surface of Example 5.12) without invoking [BO78]: from M=H^2_crys(A/W) and its Φ, compute L=∩Φ^{-r}(p^rM), V_K^{-1}M, and φ(N) using the explicit Nygaard complex WΩ•(-1); verify φ(N)=V_K^{-1}M and that the induced pair equals (ker d, ker dV) for U^{0,2}_A=U_2. Also test a superspecial surface where U=U_1. If either test fails, Theorem 4.22 Step 3 is false.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central geometric results (Theorems D–F) reconstruct ker(dV) from the F-crystal via K^{-1}=V_K^{-1}M/(L∩V_K^{-1}M). The proof of the key equality φ(N)=V_K^{-1}M in Theorem 4.22 Step 3 is compressed: it reduces to showing that u∈Fil^1_N M=Φ^{-1}(pM) maps to zero in H^2(X,O_X), and invokes [BO78, Thm 8.26(1)] for the assertion that the reduction of Fil^1_N M is the Hodge filtration. This is asserted, not derived. If the relevant form of Mazur's theorem requires, say, a lift of X to W_2 or a slope-filtration statement different from Φ^{-1}(pM), the identification K^{-1}=ker(dV) fails, and p-exp(U^{0,2}_A)=e(N) and σ_Art(A)=lgth(N/L) no longer follow. The rest of the paper's §5 bounds rest on this single deduction. The structural §3 arguments appear internally consistent; this is the least secure link.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a structure theory for dominoes, the non-finitely-generated pieces of de Rham–Witt cohomology lying over the nonzero slope-spectral-sequence differentials. It classifies two-dimensional indecomposable dominoes by orbits of Frobenius-skew polynomials (Theorem A), gives a fixed-type presentation theorem for arbitrary-dimensional dominoes (Theorem 3.18), and identifies a distinguished maximal-exponent family (Theorem 3.20). It then introduces Ekedahl modules and Nygaard pairs, and claims that for a Mazur–Ogus variety the F-crystal H^2_crys(X/W) functorially reconstructs the diagonal slice H^2(X,WΩ^•)[0,0] and hence the domino U^{0,2}_X (Theorem 4.22). For supersingular abelian varieties this is turned into a combinatorial algorithm for cyclic F-crystals, closed computations for two families (Theorems 5.13, 5.14), and bounds on the p-exponent and the degree-two Artin invariant in terms of the a-number (Theorem F). The paper also attaches formal and perfect unipotent groups to each domino and proves equality of their isogeny partitions (Theorem C).","tokens_in":1491,"tokens_out":1428,"duration_ms":144052,"significance":"If correct, this is a substantial advance. The two-dimensional classification and the fixed-type normal form are parameter-free and internally coherent, and the explicit cyclic algorithm gives computable answers for natural families of supersingular abelian varieties. The sharpness remarks in §5.21 and §5.25 provide independent support for the exponent bounds. The geometric half of the paper, however, depends on a chain of reconstructions whose most delicate link is Theorem 4.22; the a-number bounds and the Artin-invariant bounds collapse if that link fails. The structural §3 arguments are detailed and appear sound, and I found no fitted-parameter circularity.","major_comments":[{"comment":"The equality φ(N)=V_K^{-1}M is the load-bearing step for Theorems D–F and for all of §5. The proof is compressed: it asserts that the reduction map sends Fil^1_N M to the Hodge filtration Fil^1_H via [BO78, Thm 8.26(1)], and that this implies u=V(n). The cited theorem is not stated with its exact hypotheses, and the compatibility of the Nygaard modification WΩ•_X(-1) with the Mazur–Ogus condition is not verified. If this identification fails, p-exp(U^{0,2}_A)=e(N) and σ_Art(A)=lgth_W(N/L) no longer follow. Please expand Step 3 into a separate lemma giving the precise Berthelot–Ogus statement used and proving the edge identifications ε_0 and ε_{-1}.","section":"§4.2, Theorem 4.22, Step 3"},{"comment":"The reconstruction and all of §5 assume that every supersingular abelian variety is Mazur–Ogus. The text cites [GSY25, Def. 1.4] for ‘straight’ but does not show that ‘straight’ is equivalent to Definition 4.8, nor does it give a proof that abelian varieties have torsion-free crystalline cohomology and Hodge–de Rham degeneration in the exact sense used by Ekedahl. Since the a-number bounds apply to every supersingular abelian variety, this premise is load-bearing. Please either prove the needed statement directly or give a precise standard reference, and clarify the relation between ‘straight’ and the intrinsic Hodge–Witt filtration formula of Proposition 4.14.","section":"§4.1, Example 4.11 and Definition 4.8"},{"comment":"The isogeny partition theorem is advertised as a central structural result, but its proof is only a sketch. The identification im[p]^r_{G^{perf}(U)} ≃ G^{perf}(p^rU) is asserted via exactness of the F-fixed realization and Proposition A.1, and the Zink/Serre decompositions are cited without checking that the dimensions d_r(U) are exactly the Witt-part contributions. Since Theorem C is one of the main claims, please expand this proof or state the theorem as an assembly of the cited standard results with the exactness check carried out.","section":"Appendix A.2, Theorem A.2"}],"minor_comments":[{"comment":"The notation W_{2,σ}[[V]]⊕kF is confusing: kF looks like a field extension rather than a one-dimensional k-vector space spanned by F. Please clarify the notation.","section":"Example 3.10"},{"comment":"The theorem statement says ‘Assume g≥3’ only in the proof paragraph; please state the hypothesis explicitly in the theorem block. Also, the a=2 improvement to g−2 is stated in Theorem F but appears only at the end of the proof; this is easy to miss.","section":"Theorem 5.24"},{"comment":"This remark concludes that a rank-6 cyclic F-crystal is not principally quasi-polarizable. It would be cleaner to state explicitly that the blockwise algorithm of Theorem 5.6 is being applied to an abstract cyclic slope-one F-crystal before drawing a conclusion outside the abelian-variety setting.","section":"Remark 5.31"},{"comment":"Several references are to preprints or very recent papers ([GSY25], [LY26], [Yan26]). Please date them and, where possible, indicate the specific theorem or section being used, especially [GSY25, Def. 1.4] and [BO78, Thm 8.26(1)].","section":"General references"}],"recommendation":"major_revision","confidential_remarks":"I found the structural §3 section convincing and the paper’s computations impressive. The main uncertainty is Theorem 4.22, Step 3, and the external Mazur–Ogus/straightness assumption; these are load-bearing for Theorems D–F and §5. I recommend major revision rather than rejection because the issues are local and fixable by expanding the proof or stating the precise external theorems. No circularity or fitted-parameter issue was found."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe short version: the structural half of this paper is genuinely new and looks solid; the geometric half is a real payoff but rests on a cited Mazur–Ogus/Berthelot–Ogus input that is asserted rather than derived. The reader's take is about right: accept is defensible, with the verification risk centered in §4.2.\n\nWhat's actually new: Theorem 3.9 classifies 2-dimensional indecomposable dominoes as frame-change orbits of skew polynomials; Theorem 3.18 gives a fixed-type presentation by strictly upper triangular matrices; Theorem 3.20 identifies the distinguished domino with maximal p-exponent; Theorem A.2 (isogeny partition) is a clean and useful result, as are the a-number bounds in §5. The paper is carefully organized, discloses its limitations honestly (no full indecomposable list for j≥4, the non-standard 'perfect Brauer group' terminology, the principal-polarization assumption for the upper σ_Art bound), and the proofs in §3 are detailed and internally consistent. Credit where due: this is the first classification beyond dimension one and the normal form gives real computational access.\n\nThe soft spot is exactly where the stress-test note points. Theorem 4.22 Step 3 identifies ker(dV) with V_K^{-1}M/(L ∩ V_K^{-1}M). The proof reduces to showing that the reduction of Fil^1_N M is the Hodge filtration, invoking [BO78, Thm 8.26(1)]. That invocation is compressed — it's asserted, not derived — and if the relevant Mazur's theorem requires conditions not met by every supersingular abelian variety, Theorems D–F and the §5 bounds collapse. I don't think this is a fatal flaw: the Mazur–Ogus property of abelian varieties is standard in the field (Ekedahl, GSY25), and [BO78, Thm 8.26(1)] is a concrete published theorem. But a referee should verify that the application conditions are met, especially the slope-filtration identification with Φ^{-1}(pM). The structural results in §3 and Appendix A are independent of this step and should survive regardless.\n\nWho this is for: anyone working on de Rham–Witt cohomology, Brauer groups of supersingular abelian varieties, or unipotent group schemes in characteristic p. The paper deserves a serious referee; I'd send it out. If I were the referee, I'd focus on Theorem 4.22 Step 3 and ask the authors to expand the reduction argument. Also worth checking: the principal-polarization assumption on the upper bound and the 'cyclic supergeneral' Dieudonné module claim in Theorem 5.13 — but those look standard.\n\nRecommendation: engage with it. It is a strong preprint with one compressed load-bearing citation.","headline":"Strong structure theory for dominoes; the geometric payoffs are real but hang on a compressed Berthelot–Ogus step that a referee should check carefully.","tokens_in":57053,"tokens_out":2971,"would_cite":true,"duration_ms":27854,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14F30","14K15"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that every domino in de Rham–Witt cohomology admits a normal form, classifies the two-dimensional ones, and uses this to reconstruct Brauer-type invariants of supersingular abelian varieties from crystalline cohomology, bou","keywords":["de Rham–Witt cohomology","dominoes","slope spectral sequence","Raynaud ring","crystalline cohomology","supersingular abelian varieties","Brauer group","a-number"],"falsifier":"Exhibit a principal polarization on the abelian variety whose H^1 Dieudonné module is the rank-6 cyclic F-crystal with valuation word (0,0,1,0,1,1); the paper's upper bound σ_Art(A) ≤ 7 makes such a polarization impossible, so constructing one would refute Theorem 5.30.","tokens_in":55767,"feed_emoji":"🎲","tokens_out":10467,"duration_ms":89356,"temperature":0.7,"pith_summary":"The paper studies a characteristic-p phenomenon in the cohomology of smooth proper varieties: the de Rham–Witt cohomology contains a canonical piece, the domino, that is not finitely generated and that carries exactly the nonzero differentials of the slope spectral sequence. Previously only one-dimensional dominoes were classified; the paper proves that every domino has a normal form given by a strictly upper-triangular matrix over a skew polynomial ring, and that the two-dimensional indecomposable ones are classified by a single Frobenius-skew polynomial up to rescaling. On this structural basis, it reconstructs the degree-two diagonal slice of de Rham–Witt cohomology from the crystalline F-crystal for Mazur–Ogus varieties, and for supersingular abelian varieties it computes the domino explicitly in two families and proves that the p-exponent of the p-primary Brauer group is bounded by ceil((g-1)/a) ≤ e ≤ g-a+1, with e = g-1 iff a = 1, for every prime p. A sympathetic reader would care because this is the first classification beyond dimension one, and it turns a structural gap into explicit numerical bounds on Brauer groups.","feed_headline":"Dominoes beyond dimension one classified and normalized","feed_subtitle":"This normal form yields explicit Brauer groups and a-number exponent bounds for supersingular abelian varieties.","key_machinery":"The central object is the domino U, a two-term module U^0 → U^1 over the Raynaud ring—the ring that packages Frobenius F, Verschiebung V, and the de Rham differential d—killed by a power of p but not finitely generated over the Witt vectors; it is the part of de Rham–Witt cohomology that carries the nonzero slope-spectral-sequence differentials. The structural load is borne by three tools: the type sequence J(U), the canonical filtration whose graded pieces are elementary dominoes U_j; the presentation theorem over the skew polynomial ring k_σ[V], which writes any domino of fixed type as a strictly upper-triangular matrix of Ext^1-classes and describes isomorphisms as upper-triangular change","core_discovery":"On the paper's own terms, the discovery is that higher-dimensional dominoes are not an unmanageable wilderness: every domino of fixed type is presented, uniquely up to explicit upper-triangular changes of generators, by a strictly upper-triangular matrix of extension classes over k_σ[V] (Theorem 3.18), and in dimension two the classification is the orbit space of a nonzero polynomial under the Frobenius-skew frame-change f ↦ σ(b_0) f c_0^{-1} (Theorem 3.9). The paper then shows that each domino carries two unipotent realizations—a formal group and a perfect group—with identical isogeny partitions, so the two Brauer-type invariants are two functors of one object. The geometric payoff is Theor","pith_inferences":["The normal-form theorem suggests that the 'type sequence plus extension matrix' is the right invariant of a domino, so any geometric invariant that depends only on isogeny (like the formal Brauer group's isogeny class) can forget the matrix, while finer invariants (like the actual domino) remember it; one testable consequence is that the extension matrix should be recoverable from the differential","The reconstruction theorem implies that, at least in the Mazur–Ogus setting, the domino contributes no new information beyond the F-crystal of degree-two crystalline cohomology; a natural extension is to test this for non-Mazur–Ogus varieties, where the diagonal slice need not be nft and the Nygaard-pair description should fail in a measurable way.","Because the upper bound on σ_Art is proved via a self-dual chain and a discriminant-length estimate, the same strategy may yield bounds on other length invariants attached to Hodge–Witt filtrations, such as the lengths of torsion in H^{2d-2} of varieties with nondegenerate duality.","The principal-polarization obstruction in §5.3 turns the size of a domino into a numerical witness against polarizability; one could systematically search cyclic F-crystals for degree/p-exponent combinations that violate the upper bound and thereby produce new non-polarizable Dieudonné modules."],"forward_implications":["Two-dimensional classification: an indecomposable 2-dimensional domino is an extension U_{j_2} → U → U_{j_1} with gap at least 2, and its isomorphism class is a Frobenius-skew polynomial f(V) up to the rescaling f ↦ σ(b_0) f c_0^{-1}; a full list is possible in this dimension.","Normal form in all dimensions: for any type sequence, the isomorphism classes of dominoes are orbits of upper-triangular presentation matrices under explicit upper-triangular changes of generators; this turns classification into a (generally wild) matrix problem but gives a canonical language for all further computations.","Unipotent realizations: each domino yields a formal unipotent group and a perfect unipotent group with the same isogeny partition; in degree two these are the formal Brauer group and the perfect Brauer group, so the two Brauer invariants are governed by a single object.","Reconstruction: for Mazur–Ogus varieties, H^2(X,WΩ^•_X)[0,0] and in particular the domino U^{0,2}_X are determined by the F-crystal H^2_crys(X/W); the domino is therefore an invariant of crystalline cohomology, not an additional piece of structure.","Brauer bounds: for every supersingular abelian g-fold and every prime p, the p-exponent of the p-primary Brauer group lies between ⌈(g-1)/a⌉ and g-a+1, and the degree-two Artin invariant σ_Art(A) is bounded by g(g-1)-binom(a,2) from below and by floor(e g(2g-1)/2) from above under a principal polarization; a=1 characterizes the maximal exponent g-1."],"fun_headline_variants":["2D dominoes classified; normal forms for all dimensions","Dominoes obtain normal forms in every dimension","Brauer group exponent bound from domino classification","Dominoes tied to unipotent groups and Brauer bounds"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that every supersingular abelian variety is Mazur–Ogus—crystalline cohomology torsion-free and the Hodge–de Rham spectral sequence degenerating—so that the Hodge–Witt filtration is read off from the F-crystal by the formula Fil^i = ∩ φ^{-r}(p^{ir}M); if any supersingular abelian variety violates this, the identification of the Brauer exponent with the lattice exponent (and of σ_Art with half the discriminant length) collapses.","fun_headline_variants_meta":{"raw":{"variants":["2D dominoes classified; normal forms for all dimensions","Dominoes obtain normal forms in every dimension","Brauer group exponent bound from domino classification","Dominoes tied to unipotent groups and Brauer bounds"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001429,"raw_usage":{"total_tokens":5577,"prompt_tokens":697,"completion_tokens":4880,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":441,"completion_tokens_details":{"reasoning_tokens":4826}},"tokens_in":441,"tokens_out":4880,"duration_ms":34905,"temperature":1.0,"reasoning_tokens":4826,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T00:10:07.439214+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhibit a principal polarization on the abelian variety whose H^1 Dieudonné module is the rank-6 cyclic F-crystal with valuation word (0,0,1,0,1,1); the paper's upper bound σ_Art(A) ≤ 7 makes such a polarization impossible, so constructing one would refute Theorem 5.30.","supporting_citations":[],"review_version":1}