{"id":"f38ddba8-a5fc-4aaa-9eac-04fcbdd92474","arxiv_id":"1908.01933","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The Severi line equality, already known in characteristic zero, is now characterized in all characteristics: equality holds exactly for canonical models that are flat double covers of Abelian surfaces.","lead":"This paper proves that, over any algebraically closed field, a minimal surface of general type with maximal Albanese dimension achieves the Severi equality only when its canonical model is a flat double cover of an Abelian surface. It also proves a refined Severi inequality with an explicit constant that detects when the Albanese morphism is a double cover.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 5.13(2) rests on an unverified citation: the genus-change formula from [9, Prop. 2.2] may not apply to the fibrations ϕ_n, whose generic geometric fibre is admitted to be possibly singular; if that formula fails, the c0-based case of the refined inequality is unproved.","rationale":"Good-faith reading: the paper's goal is to extend the Severi-line characterization to all characteristics. The overall architecture—Pardini covering trick, slope inequality, descent of double covers, flat-double-cover analysis—is coherent, and Theorem 3.1 and Section 6 are plausible. The most exposed point is Proposition 5.13. In case (1), where φ_n descends to a fixed π_i, the computation is clear. In case (2), the proof has two external inputs: the genus-change formula from [9] and Lemma 5.8's torsion bound. The second seems locally checkable and the proof via Lemma 5.9 is reasonable, though it must be verified carefully. The first is exactly the reader's weakest assumption. Because the paper explicitly admits the generic geometric fibre of ϕ_n may be singular, any hypothesis in [9] requiring smoothness would immediately invalidate the application. The formula as quoted is also being applied to a minimal resolution of a rational map, not to a straightforward quotient, so there is room for hidden assumptions. This does not make the theorem false; it makes the proof not yet verifiable at that step. The numerical error in Example 7.3 is real: the divisor defined by x1^{2n}+x2^{2n+1}=0 is principal, whereas the paper claims it belongs to the ample class 2nΓ1+(2n+1)Γ2. This shows the examples need correction but does not bear on Theorem 1.5 itself. My recommendation: keep the reader's CONDITIONAL verdict; the central claim remains plausible but should not be accepted until the hypotheses of [9, Prop. 2.2] are shown to hold for these fibrations, or an independent proof of the estimate in Proposition 5.13(2) is supplied.","tokens_in":22631,"tokens_out":22174,"duration_ms":224269,"concrete_test":"Extract from [9, §2.1] the exact statement and hypotheses of Prop. 2.2, and check them one by one for the fibration ϕ_n: X̃_n→P^1 constructed in §5.2 with Ξ_n chosen as in Lemma 5.8: (i) smoothness of the generic geometric fibre; (ii) the inseparable degree-2 map φ_n being the normalised relative Frobenius or satisfying the flatness condition used in [9]; (iii) whether the formula applies to the minimal resolution Z'_{n,Ξ} or only to the quotient before resolution. Then run the local model F: y^2 = x^3 + x over K = k(t) in characteristic 2, with quotient P^1_K: compute both the quotient genus and dim_K(Ω_{F/K,tor})_η and test whether g' = g − (1/4)dim holds; if it does not, identify the missing term and correct Proposition 5.13(2), or supply a counterexample to that step.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The theorem's 'on the Severi line ⇒ double cover' direction passes through Theorem 5.14, whose proof in one characteristic-2 case is Proposition 5.13(2). There the paper applies a genus-change formula, g'_n = g_n − (1/4) dim_{k(t)}(Ω_{X̃_n/P^1,tor})_η, cited to [9, §2.1, Prop. 2.2], with no statement of its hypotheses. The surrounding construction flags a relevant failure: before (13) the authors note that 'ϕ_n can inevitably have singular generic geometric fibre since the Bertini theorem is not as strong as in characteristic 0'. If [9, Prop. 2.2] assumes smooth generic fibres, a flat/normalised inseparable double cover, or computes the genus of the quotient rather than of a minimal resolution of the image, then the application to ϕ_n and to Z'_{n,Ξ} is unjustified. Also, Lemma 5.8's upper bound on the torsion dimension and its use of Lemma 5.9 must be re-checked, since that bound is the second input to the same estimate. If either input fails, the limit lim g'_n/g_n ≥ c0(X,L) collapses, and with it Corollary 5.15 in exactly the separable-a_X/inseparable-φ_n case that motivated the c0 correction. Separately, Example 7.3 is numerically wrong: x1^{2n}+x2^{2n+1}=0 is a principal divisor, not a member of |2nΓ1+(2n+1)Γ2|; this does not touch the main theorem but signals the example section needs repair.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies minimal surfaces of general type with maximal Albanese dimension over algebraically closed fields of arbitrary characteristic. It introduces a constant c(X,L) defined from the finite set of rational double covers of X relative to the Albanese map, together with a characteristic-2 correction term c0(X,L) when the Albanese map is separable, and proves a refined Severi inequality K_X^2 ≥ (4 + min{c(X,L), 1/3})χ(O_X) (Theorem 5.14). As a consequence, equality K_X^2 = 4χ(O_X) implies that q = 2 and the Albanese map is a double cover (Corollary 5.15). In Section 6 the authors analyze flat double covers of abelian surfaces and prove the converse, so that a minimal surface is on the Severi line if and only if its canonical model is a flat double cover of an abelian surface (Corollary 6.3 and Theorem 1.5). Section 7 gives two characteristic-2 examples, one with an inseparable and one with a separable Albanese morphism.","tokens_in":23032,"tokens_out":16753,"duration_ms":159732,"significance":"If the proof can be made fully rigorous, the result completes the characterization of the equality case of the Severi inequality in all characteristics, settling a conjecture of Lopes–Pardini and Manetti in the positive-characteristic case and generalizing the characteristic-zero results of Barja–Pardini–Stoppino and Lu–Zuo. The refined inequality with the explicit constant c(X,L) is a new quantitative statement, and the treatment of inseparable double covers via 1-foliations is a useful contribution. The paper is not circular: the constant c(X,L) is defined independently and the refined inequality is proved rather than imported. However, the proof rests on several delicate technical steps whose hypotheses are not fully stated or verified in the current manuscript.","major_comments":[{"comment":"The proof of Theorem 3.1(2) uses an index ℓ in the definition of ℓ'' = min{ℓ' ≤ i < ℓ | r_i ≥ b_i + 2} and in the summation ranges of inequalities (10) and (11), but ℓ is never defined in the manuscript. I conjecture that the intended definition is ℓ = min{i | γ_i = 1}, i.e., the first index for which the relative canonical map is birational. If so, this must be stated explicitly and its existence justified for the fibrations under consideration. As written, the chain of inequalities proving K_f^2 ≥ (4+c)χ_f is incomplete, and this is a load-bearing step for Theorem 5.14.","section":"Section 3, proof of Theorem 3.1(2)"},{"comment":"The proof of Proposition 5.13(2) invokes the genus change formula g'_n = g_n − 1/4 dim_{k(t)}(Ω_{\\tilde X_n/P^1,tor})_η, citing [9, §2.1, Prop. 2.2], without stating the hypotheses under which this formula is valid. The authors themselves note before (13) that the fibrations ϕ_n can have singular generic geometric fibre because Bertini is not as strong as in characteristic 0. If the cited formula requires a smooth generic fibre, or computes the arithmetic genus of the quotient rather than that of the minimal resolution Z'_{n,Ξ}, then its application to the fibrations constructed in Section 5.2 is not justified. Since case (2) is exactly the situation needed for Corollary 5.15 when a_X is separable but φ_n is inseparable, the authors must verify the hypotheses of [9, Prop. 2.2] for the constructed ϕ_n or give a self-contained proof of the formula in this setting.","section":"Section 5.2, Proposition 5.13(2)"},{"comment":"The inequality dim_{k(t)}(Ω_{\\tilde X_n/P^1,tor})_η ≤ c_1(Ω_{X_n/Alb_X})·L_{X_n} in Lemma 5.8 is the second input to the estimate in Proposition 5.13(2), but the proof is too terse at the key comparison. In particular, the step 'It then follows from Lemma 5.9, that the length is not larger than the length of the torsion sheaf Ω_{X_n/k,ξ_j}/(df_1,...,df_l)' requires an explicit argument: Lemma 5.9(2) gives the valuation of a generic linear combination, and one then needs to compare the torsion length of the quotient by that single element with the torsion length of the quotient by all the df_i. Please expand this step so that the direction of the inequality is transparent. Also, the assertion that the horizontal torsion components are among the strict transforms of the P_j should be justified in more detail, since the exceptional divisors of ~X_n → X_n include sections of ϕ_n and the exclusion of these horizontal components from the torsion support is not immediate.","section":"Section 5.2, Lemma 5.8"}],"minor_comments":[{"comment":"In the three-bullet selection rule, the second and third bullets both say 'a_n : X → Alb_X is inseparable'; the third bullet should presumably say 'separable', as it introduces Lemma 5.8. Please correct this typo.","section":"Section 5.2, choice of Ξ_n after Lemma 5.7"},{"comment":"The divisor defined by x_1^{2n}+x_2^{2n+1}=0 is the divisor of a rational function on the abelian surface A, hence it is a principal divisor. It cannot be a member of the non-trivial linear system |2nΓ_1+(2n+1)Γ_2|. The example needs to be recast using an actual section of the relevant line bundle. This error is peripheral to the main theorem but invalidates the example as written.","section":"Section 7.3, Example 7.3"},{"comment":"In the displayed formula for λ_{ϕ_n} in equation (13), the notation 'H' is used in the numerator while 'L_X' is used in the denominator for the same pullback class; please make the notation consistent. Also, the heading 'non-hyperellitic' in Section 3 should read 'non-hyperelliptic', and the reference [21] has a typo 'slop inequalities'.","section":"Equation (13) and Section 3 heading"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a significant open problem and the overall strategy is promising. However, the current write-up has load-bearing gaps: the undefined index ℓ in the proof of Theorem 3.1(2), and the unstated hypotheses of the cited genus-change formula in Proposition 5.13(2). The numerical error in Example 7.3, while peripheral, suggests that the example section needs careful checking. I recommend major revision and would be willing to re-review once these points are addressed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The new thing here is real: Theorem 1.5 finishes the equality case of the Severi inequality in positive characteristic, and the p=2 case genuinely needs new machinery—inseparable double covers, 1-foliations, and a descent argument for double covers over multiplication-by-n maps. The refined inequality K_X^2 ≥ (4 + min{c(X,L), 1/3})χ(O_X) with the explicitly defined constant is also new. The paper does the honest work of proving the slope inequality rather than importing it, and the main self-citations are used as tools, not as a way to smuggle in the conclusion.\n\nWhat I like: the overall strategy is clear and the proof is organized around well-defined objects. The definition of c(X,L) is natural, and Proposition 5.2 correctly identifies when it vanishes. Section 6, on flat double covers of abelian surfaces, is a useful and fairly self-contained analysis, including the inseparable case. The examples in Section 7 are a good-faith attempt to show the theorem is not empty in characteristic 2.\n\nWhere I would push back: the load-bearing step is Proposition 5.13(2). The paper applies the genus-change formula g'_n = g_n − (1/4) dim(Ω_{X̃_n/P^1,tor})_η, cited to [9, Prop. 2.2], without stating its hypotheses. The surrounding text admits that the fibrations ϕ_n can have singular generic geometric fibre because Bertini is weaker in positive characteristic. If [9, Prop. 2.2] assumes smooth generic fibres, or a different normalization, then the application is not justified. This matters exactly for the c0(X,L) branch—the case where a_X is separable but φ_n is inseparable. I do not see a hidden circularity or a fitted constant, but I also could not verify this step from the text. The referee should ask the authors to state the formula's hypotheses and either prove it in the needed generality or adjust the argument. Lemma 5.8, which controls the torsion dimension, is a second input to the same estimate and should be checked alongside.\n\nThe example in Section 7.3 has a clear numerical problem: the divisor defined by x1^{2n} + x2^{2n+1} = 0 is principal, not a section of |2nΓ1 + (2n+1)Γ2|. That does not touch the main theorem, but it signals the example section was not carefully proofread.\n\nNet: the paper is a serious contribution and deserves a serious referee. It is not a desk reject. I would send it out with a request to verify the genus-change hypotheses, fix Example 7.3, and clarify Lemma 5.8. If the authors can close that gap, the theorem is very likely correct and is a substantial result.","headline":"A serious, mostly sound completion of the Severi-line classification in all characteristics, but the referee should check the unstated hypotheses on the genus-change formula in Proposition 5.13(2) and fix the numerical error in Example 7.3.","tokens_in":23515,"tokens_out":2173,"would_cite":true,"duration_ms":27291,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14J29","14J17","14K05"],"pacs":[],"model":"deepseek-v4-flash","headline":"The equality case of the Severi inequality occurs exactly when the canonical model is a flat double cover of an abelian surface, over every algebraically closed field.","keywords":["Severi inequality","Severi line","surfaces of general type","maximal Albanese dimension","flat double cover","positive characteristic","characteristic 2","slope inequality"],"falsifier":"Construct a characteristic-2 surface with a separable Albanese morphism, $K_X^2=4\\chi(\\mathcal O_X)$, and no degree-2 map to an abelian surface; then compute the genus change in the fibration $\\phi_n$ produced by the covering trick and check whether the inequality $g'_n\\ge n^{2q-2}(2K_X-R_X)\\cdot L_X/4+1$ from Proposition 5.13 holds, where $R_X=c_1(\\det(\\Omega_{X/\\mathrm{Alb}_X}))$. A single failure of that inequality would refute the refined Severi inequality and hence Theorem 1.5.","tokens_in":22466,"feed_emoji":"📐","tokens_out":10895,"duration_ms":100109,"temperature":0.7,"pith_summary":"This paper closes the equality case of the Severi inequality in all characteristics. For a minimal surface of general type whose Albanese morphism is generically finite onto its image, the inequality $K_X^2 \\ge 4\\chi(\\mathcal O_X)$ is sharp exactly when the canonical model is a flat double cover of an abelian surface. The proof establishes a refined inequality $K_X^2 \\ge (4+\\min\\{c(X,L),\\frac13\\})\\chi(\\mathcal O_X)$, where $c(X,L)$ is a nonnegative constant that vanishes precisely when the Albanese morphism is a double cover of an abelian surface, so equality in the original inequality forces $c(X,L)=0$. A separate analysis of flat double covers, including inseparable and wildly ramified cases in characteristic 2, proves the converse and supplies explicit examples. The result matters because it completes the classification of surfaces attaining the Severi bound, removing the last characteristic-$p$ obstructions.","feed_headline":"Severi equality means a flat double cover of an abelian surface","feed_subtitle":"Sharp case of the Severi inequality is now classified over every algebraically closed field","key_machinery":"The load-bearing mechanism is a refined slope inequality for non-hyperelliptic fibrations: if a fibration of fibre genus $g$ has the property that every degree-2 component in its Harder-Narasimhan filtration has fibre genus at least $c g$, then its slope satisfies $K_f^2\\ge (4+c)\\frac{g-1}{g+2}\\chi_f$. Into this inequality the paper feeds a family of fibrations $\\phi_n\\colon \\widetilde X_n\\to \\mathbb P^1$ produced by the covering trick—base change by multiplication by $n$ on the Albanese variety and resolution of a pencil of divisors—whose slopes converge to $K_X^2/\\chi(\\mathcal O_X)$. The decisive step, Proposition 5.13, shows the limiting ratio of fibre genera is bounded below by $c(X,L)$; the proof either descends the double covers to $X$ or, when the fibration is inseparable, uses the genus-change formula $g'_n=g_n-\\frac14\\dim_{k(t)}(\\Omega_{\\widetilde X_n/\\mathbb P^1,\\mathrm{tor}})_\\eta$ together with a torsion estimate in terms of $c_1(\\Omega_{X_n/\\mathrm{Alb}_X})$. The converse rests on the canonical resolution of flat double covers, with rank-one foliations accounting for inseparable covers in characteristic 2.","core_discovery":"Over an algebraically closed field of any characteristic, let $X$ be a minimal surface of general type with maximal Albanese dimension, meaning the Albanese morphism $a_X\\colon X\\to \\mathrm{Alb}_X$ is generically finite onto its image. The central discovery is the complete equality case: $K_X^2=4\\chi(\\mathcal O_X)$ holds if and only if the canonical model $X_{\\mathrm{can}}$ is a flat double cover of an abelian surface. The forward direction is obtained from a refined Severi inequality whose extra term is a constant $c(X,L)\\ge 0$ built from the ratios $K_{Y_i}\\cdot h_i^*L\\,/\\,K_X\\cdot a_X^*L$ over the finitely many rational double covers of $X$ relative to its Albanese map, together with a characteristic-2 correction $c_0$ involving $\\det(\\Omega_{X/\\mathrm{Alb}_X})$. Since $c(X,L)=0$ exactly when $\\dim\\mathrm{Alb}_X=2$ and $a_X$ is a double cover, equality in the Severi inequality forces $X$ to be a double cover of an abelian surface; the converse and the identification of the canonical model come from a study of flat double covers in all characteristics, where inseparable covers are handled through foliations.","pith_inferences":["Beyond the paper: if the refined inequality is stable under specialization, it would give a uniform gap statement for families of irregular surfaces approaching the Severi line in any characteristic: the limit either is a flat double cover or has normalized self-intersection at least $4\\chi + \\frac13\\chi$ in the appropriate sense.","Beyond the paper: the $c_0(X,L)$ term suggests that in characteristic 2 the failure of the classical double-cover reduction is measured by torsion of relative Kähler differentials; one could test this by computing $c_0$ for explicit families of inseparable fibrations and checking whether equality forces the asymptotic $c_0$ to vanish.","Beyond the paper: the wild branch-divisor examples indicate that Severi-line surfaces in characteristic 2 form larger moduli strata than their characteristic-0 counterparts, since the branch locus may acquire arbitrarily high multiplicity while the canonical model stays smooth; this has consequences for any attempted moduli description.","Beyond the paper: the same slope-inequality mechanism might extend to higher-dimensional varieties of maximal Albanese dimension or to other sharp inequalities, with the role of the double cover replaced by a higher-degree fibration; that extension is not attempted in the paper."],"forward_implications":["In every characteristic, a surface of general type with maximal Albanese dimension attains $K_X^2=4\\chi(\\mathcal O_X)$ only by being a flat double cover of an abelian surface; no exotic positive-characteristic counterexamples remain.","The refined inequality gives an explicit quantitative gap: whenever the Albanese morphism is not a double cover of an abelian surface, $K_X^2 \\ge (4+\\min\\{c(X,L),\\frac13\\})\\chi(\\mathcal O_X)$, bounding how close such surfaces can come to the Severi line.","Classification of Severi-line surfaces reduces to classifying flat double covers of abelian surfaces, including inseparable and wildly ramified covers, and the two characteristic-2 examples show such covers exist with both separable and inseparable Albanese morphisms.","The canonical models of Severi-line surfaces have at worst A-D-E (rational double point) singularities in characteristic 2, and their branch divisors can be wildly singular, unlike in characteristic zero."],"supporting_citations":[{"why":"Supplies the original covering-trick and slope-inequality route to the Severi inequality, which the present proof adapts to positive characteristic.","marker":"[18]"},{"why":"Establishes the Severi inequality itself over fields of all characteristics, the inequality whose equality case is being classified.","marker":"[24]"},{"why":"Gives the characteristic-zero equality classification and the double-cover reduction that the present paper replaces in characteristic 2.","marker":"[14]"},{"why":"Supplies the genus-change formula for inseparable fibrations used in Proposition 5.13.","marker":"[9]"},{"why":"Establishes the correspondence between rank-one foliations and inseparable double covers, used for the inseparable cases and for the characteristic-2 correction c0.","marker":"[7]"},{"why":"Provides the flat-double-cover formalism and the A-D-E singularity criterion used in the Section 6 classification.","marker":"[6]"},{"why":"Guarantees the regular model needed to stop the canonical resolution for separable double covers.","marker":"[13]"},{"why":"Shows the slope inequalities used here remain valid in positive characteristic, underpinning Theorem 3.1.","marker":"[21]"}],"fun_headline_variants":["Severi equality means flat double covers of abelian surfaces","Equality in Severi: abelian surface double covers","Sharp Severi iff flat abelian double cover","All characteristics: Severi equality = abelian double cover","Flat double covers: the only way to hit Severi equality"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of the refined inequality in the inseparable-fibration case relies on a cited genus-change formula for inseparable fibrations, which computes the drop in fibre genus from the torsion of relative Kähler differentials; if that formula has hidden hypotheses not satisfied by the fibrations built from the covering trick, the equality-implies-double-cover direction would remain unproved in that case.","fun_headline_variants_meta":{"raw":{"variants":["Severi equality means flat double covers of abelian surfaces","Equality in Severi: abelian surface double covers","Sharp Severi iff flat abelian double cover","All characteristics: Severi equality = abelian double cover","Flat double covers: the only way to hit Severi equality"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000877,"raw_usage":{"total_tokens":3824,"prompt_tokens":1008,"completion_tokens":2816,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":624,"completion_tokens_details":{"reasoning_tokens":2750}},"tokens_in":624,"tokens_out":2816,"duration_ms":19425,"temperature":1.0,"reasoning_tokens":2750,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:59:54.744349+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct a characteristic-2 surface with a separable Albanese morphism, $K_X^2=4\\chi(\\mathcal O_X)$, and no degree-2 map to an abelian surface; then compute the genus change in the fibration $\\phi_n$ produced by the covering trick and check whether the inequality $g'_n\\ge n^{2q-2}(2K_X-R_X)\\cdot L_X/4+1$ from Proposition 5.13 holds, where $R_X=c_1(\\det(\\Omega_{X/\\mathrm{Alb}_X}))$. A single failure of that inequality would refute the refined Severi inequality and hence Theorem 1.5.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the original covering-trick and slope-inequality route to the Severi inequality, which the present proof adapts to positive characteristic."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the Severi inequality itself over fields of all characteristics, the inequality whose equality case is being classified."},{"cited_title":"and Zuo, K.: On Severi type inequalities for irregular surfaces , Inter- national Mathematics Research Notices, 2019 (2017): 231–248","cited_arxiv_id":null,"evidence_quote":"Gives the characteristic-zero equality classification and the double-cover reduction that the present paper replaces in characteristic 2."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the genus-change formula for inseparable fibrations used in Proposition 5.13."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the correspondence between rank-one foliations and inseparable double covers, used for the inseparable cases and for the characteristic-2 correction c0."},{"cited_title":"and Dolgachev, I.: Enriques surfaces, Progress in Mathematics, 76","cited_arxiv_id":null,"evidence_quote":"Provides the flat-double-cover formalism and the A-D-E singularity criterion used in the Section 6 classification."},{"cited_title":"118 (1999): 61-102","cited_arxiv_id":null,"evidence_quote":"Guarantees the regular model needed to stop the canonical resolution for separable double covers."},{"cited_title":"and Zhou, M.: Remarks on Xiao’s approach of slop inequali- ties, Asian J","cited_arxiv_id":null,"evidence_quote":"Shows the slope inequalities used here remain valid in positive characteristic, underpinning Theorem 3.1."}],"review_version":1}