{"id":"49559369-1be2-4b4d-8b47-7e78ec59cae5","arxiv_id":"2505.03993","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Over a regular base with a normal crossings divisor, the p-divisible group (and n-torsion) of a degenerating abelian scheme extends uniquely to a log p-divisible group (log finite group scheme).","lead":"This paper proves that torsion points of an abelian scheme that degenerates to a semi-abelian scheme along a normal crossings divisor extend canonically to 'log' finite group schemes and log p-divisible groups. It generalizes a result of Kato and Zhao from discrete valuation rings to higher-dimensional regular bases, which matters for studying degenerations of abelian varieties in arithmetic geometry.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.7 uses a finite-level uniqueness step for p^n-torsion that the cited results do not supply.","rationale":"The central claim, existence and uniqueness of the log p-divisible group, is very likely correct: the local complete case is handled by log 1-motives, and the descent/purity strategy is sound. However, the written proof of Theorem 4.7 relies on a finite-level uniqueness statement for A[p^n]^log that is not covered by Lemma 4.1(2). Theorem 4.6's finite-level uniqueness used Lemma 4.1(1), which requires n invertible at the generic points of D; for p^n over a p-adic component this condition fails. The reader's verdict of CONDITIONAL is therefore appropriate but for a slightly more specific reason than the reader's stated weakest assumption: the missing piece is not just the uncited surjectivity in Lemma 4.1(1), but the absence of a truncated p-power analogue of that lemma. My proposed test would settle whether [BWZ24, Lemma 4.8] supplies this missing piece or whether an additional argument is needed; until then, the proof should be regarded as conditional. I do not see grounds to reject the theorem itself, and no machine-checked proof or independent verification is offered, so ACCEPT or UNVERDICTED are not warranted. The existing CONDITIONAL verdict should stand unchanged.","tokens_in":17837,"tokens_out":38438,"duration_ms":423047,"concrete_test":"Check whether [BWZ24, Lemma 4.8] actually proves full faithfulness of the restriction functor on the category of log finite group schemes killed by p^n over a DVR with standard log, or only on log p-divisible groups. Then test the finite-level analogue directly: take a Tate curve over Z_p with period π, and compare the log finite group schemes arising from log 1-motives with monodromy pairings n and n+p. If their p-torsion extensions over Z_p are non-isomorphic while their restrictions to Q_p are isomorphic, the finite-level uniqueness asserted in Theorem 4.7 fails; if they are isomorphic, verify that the same holds at every p^n level and that the argument 'same as Theorem 4.6' can be made rigorous using only the cited lemmas.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 4.7 says: 'For every n≥1, by the same argument as Theorem 4.6, there exists a unique log finite group A[p^n]^log over (X,M_X)'. But the uniqueness argument in Theorem 4.6's first paragraph is specifically for an integer n that is invertible at each generic point of D; it invokes Lemma 4.1(1), which uses Lemma 2.8 and the surjectivity of Gal(Kbar/K)→π_1^{két}(Spec R,M_R). For p^n with p non-invertible on a component of D, that hypothesis fails. Lemma 4.1(2), the only replacement offered for the p-power case, is stated for log p-divisible groups, not for truncated log finite group schemes killed by p^n. Full faithfulness for p-divisible groups does not formally imply full faithfulness for each p^n-torsion level: a generic-fiber map of G[p^n]→H[p^n] need not come from a compatible system of maps of p-divisible groups. Thus the proof of finite-level uniqueness over DVRs, which is needed to glue extensions from the generic points of D, is left unsupported. The surrounding argument also contains a smaller slip: Theorem 4.7 ends by citing Proposition 3.14(3), whereas the statement that Qlog[p^∞] is a log p-divisible group is Proposition 3.14(4). This is consistent with the proof treating the finite-level construction as a black box rather than presenting the necessary p-power rigidity.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves two extension statements for torsion of abelian schemes degenerating to semi-abelian schemes over a regular base with a normal crossings divisor. Theorem 4.6 (Theorem B) states that for an integer n that is invertible at the generic points of the divisor, the finite flat group scheme A_U[n] extends uniquely to a log finite group scheme over the log scheme (X, M_X). Theorem 4.7 (Theorem A) states that the p-divisible group A_U[p^∞] extends uniquely to a log p-divisible group, without any invertibility hypothesis on p. The strategy is to reinterpret Mumford's degeneration theory over a complete regular local base in terms of log 1-motives (Theorem 3.20), then to pass to discrete valuation rings using Beauville-Laszlo gluing, and finally to globalize by purity and strict fpqc descent. The paper is clearly written and the choice of tools is coherent, but the proof of Theorem 4.7 contains a finite-level uniqueness step that is not supported by the stated results.","tokens_in":18057,"tokens_out":8092,"duration_ms":84197,"significance":"If the main theorems are correct, they give a canonical log p-divisible group attached to any semi-abelian degeneration of an abelian scheme over a regular base with normal crossings divisor, extending the DVR results of Kato, Zhao, and Würthen-Zhao to higher-dimensional bases. The paper also contributes a new categorical equivalence, Theorem 3.20, between Mumford's degeneration data and log 1-motives, which is a useful reinterpretation and may be of independent interest. The approach avoids the full theory of log abelian varieties and is therefore likely to be more accessible. The paper does not rely on circular reasoning: the construction is validated by the known rigidity of Mumford's degeneration theory and by the log 1-motive interpretation. However, the p-power finite-level uniqueness argument in Theorem 4.7 is a genuine gap, and the citation for a key full-faithfulness input exceeds the stated hypotheses of the cited theorem.","major_comments":[{"comment":"The proof of Lemma 4.1(1) rests on the assertion that the natural map Gal(Kbar/K) → π_1^{két}(Spec R, M_R) is surjective, stated without proof or citation. This is not a purely formal fact: for the standard log structure on a discrete valuation ring, the Kummer étale fundamental group is related to the tame quotient, and the identification requires an argument. This surjectivity is load-bearing because it is the uniqueness mechanism for n-torsion extensions in Proposition 4.2(1) and in the first paragraph of Theorem 4.6. Please provide a proof or a precise reference; if only a tameness-restricted statement is true, the uniqueness statements need to be reformulated accordingly.","section":"§4, Lemma 4.1(1)"},{"comment":"The sentence \"For every n≥1, by the same argument as Theorem 4.6\" is not justified. In Theorem 4.6 the hypotheses that n is finite and invertible at the generic points of D are used through Lemma 4.1(1) to obtain unique finite-level extensions. For n = p^r, that invertibility hypothesis is not assumed and may fail; the argument of Theorem 4.6 therefore does not apply to the finite levels A[p^r]^log. Lemma 4.1(2), the only p-power replacement offered, gives full faithfulness for entire log p-divisible groups, not for truncated log finite group schemes killed by p^r. Full faithfulness for p-divisible groups does not formally imply full faithfulness for each p^r-torsion level: a map on the generic fiber of A[p^r]^log need not come from a compatible system of maps of p-divisible groups. This finite-level uniqueness and gluing step is load-bearing, since it is exactly what allows the direct system A[p^r]^log to be formed and then identified with Qlog[p^r]. The proof needs either a p-power analogue of Lemma 4.1(1) at finite levels or a different strategy that constructs the p-divisible extension before passing to p^r-kernels.","section":"§4, Theorem 4.7"},{"comment":"The invocation of [BWZ24, Theorem 5.19] goes beyond the hypotheses stated there: that theorem assumes mixed characteristic (0,p) and a perfect residue field. The parenthetical remark that [BWZ24, Lemma 4.8] works without these assumptions is a claim, not a proof. Since Proposition 4.2(2) uses this full-faithfulness statement to obtain uniqueness of the DVR-level p-divisible extension, and Theorem 4.7 ultimately depends on it, this point needs to be substantiated by a proof or by a reference whose hypotheses cover the same class of discrete valuation rings used in the paper.","section":"§4, Lemma 4.1(2)"}],"minor_comments":[{"comment":"The proof cites Proposition 3.14(3), but the statement that Qlog[p^∞] is a log p-divisible group is Proposition 3.14(4). Please correct the citation or explain the intended reference.","section":"§4, Theorem 4.7, last paragraph"},{"comment":"The heading contains the typo \"fnite\"; it should read \"finite\".","section":"Corollary 4.5"},{"comment":"The sentence \"It is an important to understand degeneration of abelian varieties\" is grammatically incomplete; please rephrase.","section":"Introduction, first paragraph"},{"comment":"The condition that D⊗Z Z[1/n] be dense in D is used without defining the tensor notation. Please spell out its meaning at first use.","section":"Theorems 4.6 and 4.7"},{"comment":"The proof states that the same argument as in [Kat21] gives strict fpqc descent although loc. cit. proves strict fppf descent. Since this is used for the Beauville-Laszlo gluing, please provide the few-line argument or a precise reference for strict fpqc descent of finite Kummer log flat schemes.","section":"Proposition 2.7"}],"recommendation":"major_revision","confidential_remarks":"This is a promising paper with a natural main result and a useful new reinterpretation in terms of log 1-motives. The central issue is the p-power finite-level uniqueness step in Theorem 4.7; the proof as written does not supply the necessary truncated full-faithfulness statement. I would support publication after the authors either prove such a statement or restructure the argument to construct the log p-divisible group first and then take its finite-level kernels. The reliance on [BWZ24] beyond its stated hypotheses should also be checked carefully."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear Colleague,\n\nThe short version: this paper generalizes Kato-Zhao from DVRs to regular bases with a normal crossings divisor, and the broad strategy is sound. But Theorem 4.7 has a genuine gap: the proof of uniqueness for finite-level p^n-torsion extensions over DVRs is not supplied, and the invocation of 'the same argument as Theorem 4.6' does not cover the p-power case.\n\nWhat's new: Theorems A and B are genuinely new. Theorem C, an equivalence between Mumford's degeneration data and log 1-motives over complete regular local rings, is a clean and useful reformulation. The method is appealing: local log 1-motive construction, Beauville-Laszlo gluing, purity for homomorphisms, and strict fpqc descent. The introduction is honest about the relation to Kajiwara-Kato-Nakayama's log abelian varieties and Question 1.1.\n\nThe main soft spot: In the proof of Theorem 4.7, the author says for every n≥1, 'by the same argument as Theorem 4.6,' there exists a unique log finite group A[p^n]^log. But Theorem 4.6's uniqueness argument uses Lemma 4.1(1), which requires n invertible at the generic points of D. For p^n with p equal to the residue characteristic of a component of D, that hypothesis fails. Lemma 4.1(2) covers full p-divisible groups, not finite levels; full faithfulness for p-divisible groups does not imply uniqueness at each truncated level. Without finite-level uniqueness over DVRs, the gluing step lacks support. This is a real gap, though probably repairable—one would need either a finite-level analogue of Lemma 4.1(2) or a different argument tying finite-level extensions to the p-divisible group.\n\nThere are also smaller issues: Lemma 4.1(1) asserts surjectivity of Gal(Kbar/K)→π_1^{két}(Spec R, M_R) without proof or citation; that's likely true here, but needs a reference. Lemma 4.1(2) extends [BWZ24, Theorem 5.19] beyond its stated hypotheses; the author claims the argument works, but that deserves checking. And the final citation of Proposition 3.14(3) should be (4), a minor slip.\n\nWho it's for: anyone working on log p-divisible groups, degeneration of abelian varieties, or log abelian varieties. If Theorem A holds, it's a useful structural result that bypasses the heavier KKN machinery. The gap is in the proof, not the concept.\n\nI would send this to a serious referee, and ask the author to repair the finite-level uniqueness argument before publication.","headline":"A useful and likely correct generalization of Kato-Zhao, but Theorem 4.7's proof has a real gap in the finite-level uniqueness for p-power torsion.","tokens_in":18681,"tokens_out":10796,"would_cite":false,"duration_ms":92031,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14K10","14L15","14F35","14G20"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that the p-divisible group of an abelian scheme degenerating along a normal crossings divisor extends uniquely to a log p-divisible group over the regular base.","keywords":["log p-divisible groups","semi-abelian degeneration","log 1-motives","log finite group schemes","Kummer log flat topology","normal crossings divisor","degeneration theory of abelian varieties"],"falsifier":"Find two log finite group schemes over a discrete valuation ring, killed by an integer $n$ invertible on the ring, that are isomorphic over the fraction field but not over the log scheme; such a pair would contradict Lemma 4.1(1) and break the uniqueness part of both main theorems. Equivalently, produce a Kummer log étale cover of $\\operatorname{Spec} R$ not dominated by any Galois cover of the fraction field.","tokens_in":17546,"feed_emoji":"📐","tokens_out":15170,"duration_ms":129907,"temperature":0.7,"pith_summary":"This paper establishes a higher-dimensional generalization of a known one-dimensional result: when an abelian scheme degenerates to a semi-abelian scheme along a normal crossings divisor (locally a union of coordinate hyperplanes) in a regular base, the $p$-divisible group of the abelian scheme extends uniquely to a log $p$-divisible group over the whole base. The analogous statement holds for the $n$-torsion when $n$ is invertible at the generic points of the divisor. This matters because the naive torsion of the semi-abelian scheme is only quasi-finite and loses the monodromy information of the degeneration, whereas the log object is finite flat after suitable covers and records that information. The reader should take away that degenerating abelian varieties still carry a well-behaved torsion theory in the logarithmic world.","feed_headline":"Semi-abelian degenerations carry canonical log p-divisible groups","feed_subtitle":"Abelian scheme degenerating along a normal crossings divisor: its p-power torsion extends uniquely as a log object.","key_machinery":"The central object is a log 1-motive, a homomorphism $Y \\to G_{\\log}$ from a locally constant lattice $Y$ to the log-semi-abelian sheaf attached to a split semi-abelian scheme $G$; its torsion $Q_{\\log}[n]$ is a log finite group scheme. The workhorse is the equivalence $\\mathrm{DEG}(X,U)\\simeq \\mathrm{DD}(X,U)\\simeq \\mathrm{DD}^{\\log}(X,U)$ over complete regular local rings, which converts degeneration data into log 1-motives; the $n$-torsion (or the $p$-power torsion) of the resulting log 1-motive is the required canonical extension. Uniqueness is carried by fully faithful restriction functors (Lemma 4.1 over discrete valuation rings, Corollary 4.5 in codimension at least 2), so any two candidate extensions agree as soon as they agree on the generic fiber.","core_discovery":"Theorem 4.7 states that, for an fs log scheme $(X,M_X)$ (a fine and saturated log scheme) defined by a locally noetherian regular scheme $X$ and a normal crossings divisor $D$, if $A$ is a semi-abelian scheme over $X$ whose restriction to $U=X\\setminus D$ is an abelian scheme, then the $p$-divisible group $A_U[p^\\infty]$ over $U$ uniquely extends to a log $p$-divisible group over $(X,M_X)$. Theorem 4.6 gives the same uniqueness for the $n$-torsion $A_U[n]$ when $n$ is invertible at the generic points of $D$. The extension is not the naive system $\\{A[p^n]\\}$ of quasi-finite flat group schemes attached to the semi-abelian scheme; it is a log finite group scheme in the Kummer log flat topology, the topology generated by adjoining roots of monomials, and it retains the monodromy information that the naive system loses.","pith_inferences":["The uniqueness suggests the log $p$-divisible group is intrinsic to the generic abelian scheme together with the chosen normal crossings compactification, and should be unchanged under modifications of the divisor that do not change $U$.","A testable extension would be to non-regular bases: after passing to a log regular resolution, the same argument should produce a log $p$-divisible group, and the question would be whether the result is independent of the resolution.","If Theorem 4.7 is combined with log Dieudonné theory, each semi-abelian degeneration should give a semi-stable or log prismatic Galois representation whose monodromy is the pairing encoded in the log 1-motive.","The proof strategy also suggests that the natural question of whether every such degeneration comes from a log abelian scheme (Question 1.1 in the paper) should have an affirmative answer in general, since the torsion-level structure already exists canonically."],"forward_implications":["The log $p$-divisible group $A[p^\\infty]^{\\log}$ is independent of auxiliary choices: any two semi-abelian models of the same abelian scheme over $U$ yield isomorphic log $p$-divisible groups, because the extension is unique.","For every $n$ invertible at the generic points of $D$, the finite flat group scheme $A_U[n]$ extends canonically to a log finite group scheme over $(X,M_X)$.","The objects $A[p^n]^{\\log}$ assemble into a log $p$-divisible group with exact sequences and Weil pairings inherited from the log 1-motive, so structural results for $p$-divisible groups transfer to these degenerations.","Over complete regular local rings, degeneration data for semi-abelian schemes are equivalent to log 1-motives ($\\mathrm{DEG}\\simeq\\mathrm{DD}\\simeq\\mathrm{DD}^{\\log}$), giving a log-geometric description of the monodromy pairing.","The canonical log extension is available even where the semi-abelian torsion system has non-constant rank, which makes it the right input for log versions of Dieudonné theory and for arithmetic compactifications."],"supporting_citations":[{"why":"Proves the one-dimensional case over a complete discrete valuation ring that this paper generalizes to higher-dimensional bases.","marker":"[Zha21, Theorem 5.2]"},{"why":"Supplies the notion of log finite group schemes and log p-divisible groups, and the one-dimensional complete DVR uniqueness result.","marker":"[Kat23, §4.3]"},{"why":"Provides the degeneration theory of abelian varieties and the equivalence between semi-abelian degenerations and degeneration data that Theorem 3.20 reinterprets via log 1-motives.","marker":"[FC90]"},{"why":"States the form of the degeneration theory used in Theorem 3.19 to obtain the equivalence between semi-abelian degenerations and degeneration data.","marker":"[Mad19, (1.2.2)]"},{"why":"Gives the construction of $Q_{\\log}[n]$ as a log finite group scheme from a log 1-motive, the object that provides the desired extension.","marker":"[WZ24, Proposition 3.5]"},{"why":"Provides strict fpqc descent for log finite group schemes, used to glue local extensions in the proof of Theorem 4.6.","marker":"[Kat21, Theorem 7.1 and Theorem 8.1]"},{"why":"Provides the gluing lemma used in Proposition 2.7 to reduce the discrete valuation ring case to the complete case.","marker":"[BL95]"},{"why":"Supplies the full-faithfulness theorem for log p-divisible groups over discrete valuation rings used in Lemma 4.1(2).","marker":"[BWZ24, Theorem 5.19]"}],"fun_headline_variants":["Unique log p-divisible group extends abelian torsion in degeneration","Log p-divisible groups keep monodromy in semi-abelian degeneration","Torsion extends uniquely as log object in semi-abelian degeneration","Semi-abelian degeneration yields canonical log p-divisible group"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the claim that over a discrete valuation ring the absolute Galois group of the fraction field maps onto the Kummer log étale fundamental group of the log ring, together with the extension of a known full-faithfulness theorem for log $p$-divisible groups beyond the hypotheses under which it was stated.","fun_headline_variants_meta":{"raw":{"variants":["Unique log p-divisible group extends abelian torsion in degeneration","Log p-divisible groups keep monodromy in semi-abelian degeneration","Torsion extends uniquely as log object in semi-abelian degeneration","Semi-abelian degeneration yields canonical log p-divisible group"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001808,"raw_usage":{"total_tokens":7050,"prompt_tokens":809,"completion_tokens":6241,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":425,"completion_tokens_details":{"reasoning_tokens":6163}},"tokens_in":425,"tokens_out":6241,"duration_ms":40877,"temperature":1.0,"reasoning_tokens":6163,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T23:42:11.390027+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find two log finite group schemes over a discrete valuation ring, killed by an integer $n$ invertible on the ring, that are isomorphic over the fraction field but not over the log scheme; such a pair would contradict Lemma 4.1(1) and break the uniqueness part of both main theorems. Equivalently, produce a Kummer log étale cover of $\\operatorname{Spec} R$ not dominated by any Galois cover of the fraction field.","supporting_citations":[],"review_version":1}