{"id":"11528a5b-f45f-4227-a21a-6ccecf9e9ba0","arxiv_id":"2506.23812","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For n≥15, the moduli space of n-pointed genus-3 curves is of general type, completing the Kodaira classification of pointed genus-3 moduli spaces.","lead":"This paper proves that the space of genus-3 curves with 15 or more marked points is of general type, the 'maximally complicated' birational class. It completes the Kodaira dimension classification for genus 3, resolving the last open low-genus case.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The singularity analysis computes ages on H0(K^2(D)) but applies the Reid-Tai criterion to the dual deformation space; the two age functions differ, so the junior-element classification in Prop. 3.3 is not automatically the one needed.","rationale":"The reader's conditional verdict is well supported. The bigness computation in Section 5 is consistent: the chosen constants cancel λ and ψ, and the remaining boundary coefficients are positive for n ≥ 15. The main unresolved risk is indeed the singularity analysis in Section 3, but my concern is more specific than the reader's general warning about the case analysis. Section 2.2 and Prop. 3.4 explicitly take the local model to be V*/G with V = H0(Ω⊗ω(Σ p_i)), while Lemma 3.1 and Props. 3.2–3.3 compute ages on V itself. Since dualizing a representation changes fractional parts from r to 1-r on nontrivial eigenspaces, the set of elements with age < 1 can change. Table 1 row (6) gives a concrete example: age_V = 3/4 but age_{V*} = 5/4. The paper does not supply an inequality showing that age_{V*} < 1 implies age_V < 1 in the cases that occur, nor does it dualize its eigenvalue computations. This is a localized, checkable gap rather than a demonstrated false theorem. It affects the proof of Prop. 3.3 and therefore the identification of the non-canonical locus in Cor. 3.5 and the lifting statement Prop. 1.4. Because the issue can be settled by a dualized recomputation and a finite automorphism check, it does not by itself justify rejection of the theorem; it strengthens the conditionality of the reader's accept. Hence I keep the verdict unchanged as CONDITIONAL, with the added condition that the dual-age classification be supplied or verified.","tokens_in":18413,"tokens_out":25020,"duration_ms":316500,"concrete_test":"Recalculate the singularity analysis with ages evaluated on the actual deformation space V*: replace each nontrivial eigenvalue e^{2πi a/N} on H0(K^2(D)) by its inverse e^{-2πi a/N}, and then apply Props. 2.2 and 2.4 to the resulting eigenvalues. In particular, recheck Table 1 row (6), where the H0-age is 3/4 but the dual age is 5/4, and recheck the elliptic-tail order-6 lifting argument in Example 2.6. Then run a small explicit enumeration of automorphism groups of stable genus-3 pointed curves with n ≤ 3, searching for an automorphism outside Δ1,∅ with age_{V*} < 1 and age_V ≥ 1. If such an automorphism exists, Prop. 3.3 is false; if none exists, the fix is to state explicitly that all ages are taken on the tangent representation and to adjust the eigenvalue tables and inequalities accordingly.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 3.2 and Lemma 3.1 compute ages and eigenvalue lists on V = H0(C, Ω_C ⊗ ω_C(Σ p_i)), and Proposition 3.3 then calls an automorphism 'junior' when this H0-age is < 1. But the quotient singularity used for the Reid-Shepherd-Barron-Tai criterion (Prop. 2.2) and the CCM lifting criterion (Prop. 2.4) is V*/G, where V is the deformation space. The relevant age is therefore the age on the dual representation V*, not the age on V. These two ages need not agree: for an order-4 automorphism with eigenvalues {-1, i} on V, age_V = 1/2 + 1/4 = 3/4, while age_{V*} = 1/2 + 3/4 = 5/4. Table 1 row (6) is of this type, so Table 1 is not literally a table of junior elements for the quotient V*/G. Overcounting junior elements is harmless, but the proof of Prop. 3.3 needs the opposite direction: it must show that no automorphism outside Δ1,∅ has age_{V*} < 1 while age_V ≥ 1. Prop. 3.2 only bounds age_V, so it does not by itself justify Prop. 3.3, Cor. 3.5, or the conclusion that non-canonical singularities are contained in J0. Since Prop. 3.4 and hence Proposition 1.4 depend on this exclusion, the no-adjunction theorem has a load-bearing gap unless all age computations in §3 are understood to be dualized; in that case the eigenvalue lists and the inequalities in Props. 3.2 and 3.4 must be recomputed.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper aims to complete the Kodaira classification of the moduli space of pointed genus-3 curves by proving that \\overline{M}_{3,n} is of general type for n \\ge 15. The strategy is standard: first show that the canonical class K_{\\overline{M}_{3,n}} is big, and then show that the singularities of \\overline{M}_{3,n} impose no adjunction conditions. The bigness argument is an explicit divisor-class computation using the pullback of the hyperelliptic locus in \\overline{M}_3 and a Farkas-type divisor on \\overline{M}_{3,14}; the no-adjunction argument is a Reid–Shepherd–Barron–Tai analysis of the local actions of automorphism groups of pointed stable curves, together with a test-curve argument identifying a rigid component of the canonical divisor. The paper concludes that Theorem 1.1 follows from Theorem 1.2 (bigness) and Theorem 1.3 (no adjunction).","tokens_in":18641,"tokens_out":34284,"duration_ms":351582,"significance":"If Theorem 1.1 is correct, it is a clean and valuable completion of the Kodaira classification in genus 3, closing the last open genus in which only finitely many cases were known. The paper also proves a structural statement, Theorem 1.3, about adjunction conditions for all n \\ge 1, which is of independent interest. The bigness computation in Section 5 is explicit, arithmetically checkable, and, modulo the cited divisor classes, correct; the choice of the constants t and s is not fitted to the conclusion but is forced by matching the \\lambda- and \\psi-coefficients. The singularity analysis is extensive and is the part of the paper that carries the main technical risk; as explained in the major comments, that part is not yet established as written.","major_comments":[{"comment":"The age used for the Reid–Tai criterion is computed on the wrong representation. Proposition 3.2 and Proposition 3.3 compute ages on V = H^0(C, \\Omega_C \\otimes \\omega_C(p_1+\\cdots+p_n)). Proposition 3.4, however, recalls that a neighborhood of a point of \\overline{M}_{3,n} is isomorphic to a neighborhood of the origin in V^*/G, where G = Aut(C,p_1,\\ldots,p_n), and its Case 1 invokes Proposition 3.3 to conclude that G has no junior elements for the quotient singularity V^*/G. The age on the dual representation is not equal to the age on V: replacing an eigenvalue e^{2\\pi i r} by e^{-2\\pi i r} replaces the fractional part r by 1-r when r>0. For example, the eigenvalue pair (\\zeta_4,-1) in Table 1 row (6) has age 3/4 on V but age 5/4 on V^*. Conversely, an automorphism with age_V(\\varphi) \\ge 1 can have age_{V^*}(\\varphi) < 1, so the bound in Proposition 3.2 does not, by itself, rule out junior elements of V^*/G outside \\Delta_{1,\\emptyset}. Since Proposition 3.4, Corollary 3.5, and hence Proposition 1.4 and Theorem 1.3 depend on this exclusion, the no-adjunction theorem is not proved by the present argument. The eigenvalue lists and inequalities in Section 3 need to be recomputed for the dual representation, or a proof must be given that the two ages agree in all cases that occur.","section":"Section 3, Propositions 3.2–3.4"},{"comment":"The use of Example 2.6 to conclude that the general point of J_0 is a non-canonical singularity is also affected by the same dual-representation issue. The text says that a generator for the elliptic-tail case acts on H^0(\\Omega_C \\otimes \\omega_C(\\sum p_i)) with eigenvalues (\\zeta_6, \\zeta_6^2, 1, \\ldots, 1). On the deformation space V^*, the corresponding eigenvalues are (\\zeta_6^{-1}, \\zeta_6^{-2}, 1, \\ldots, 1), with fractional parts (5/6, 4/6, 0, \\ldots, 0). After quotienting by the quasi-reflection g^3, the induced generator has two eigenvalues with fractional part 4/6, so its age is 4/3, not 2/3 as in Example 2.6. Thus Example 2.6, as stated, does not establish non-canonicity of the singularities at the general point of J_0. If the eigenvalue lists in Case 2 of Proposition 3.4 are intended to be eigenvalues on V^* rather than on V, this must be stated explicitly and the lists must be derived from the exact sequence (3.2) for the dual representation; as written, the derivation is for H^0(\\Omega_C \\otimes \\omega_C(\\sum p_i)).","section":"Section 3, Corollary 3.5 and Example 2.6"}],"minor_comments":[{"comment":"The displayed normalization n = (2r+1)(g+1) in the definition of D^r_{g,n} is inconsistent with the subsequent use of D^3_{3,14}: for g=3 and r=3, the displayed formula gives n=28, whereas the proof of Theorem 1.2 uses n=14. The computation is consistent with the normalization n = (2r+1)(g-1), which gives n=14; please correct the text and state the intended normalization explicitly.","section":"Section 5, definition of D^r_{g,n}"},{"comment":"In the node-orbit contribution computation, the phrase 'the residue of l modulo m/N' should read 'modulo N/m'; as written the fraction is inverted.","section":"Section 3, Proposition 3.2"},{"comment":"The proof of Lemma 3.1 and the final paragraphs of Proposition 3.2 contain several terse 'one checks' and 'similarly' passages, for example in Cases B2–B5 of Lemma 3.1 and in the exclusion of cases (1), (3), (4) and (5) in Proposition 3.2. Since the completeness of these case splits is load-bearing for the singularity analysis, the checks should be expanded or relegated to a clearly verifiable appendix.","section":"Lemma 3.1 and Proposition 3.2"},{"comment":"References [AB] and [BMS] appear to list the same Agostini–Barros paper under two different keys; the bibliography should be corrected to avoid a duplicated entry.","section":"References"},{"comment":"The phrase 'has a non-canonical singularities' should be corrected to 'has a non-canonical singularity'.","section":"Corollary 3.5"}],"recommendation":"major_revision","confidential_remarks":"This is a promising paper with a clear and mostly sound bigness argument, but the Reid–Tai analysis in Section 3 has a representation-theoretic gap: ages are computed on H^0(\\Omega \\otimes \\omega(\\sum p_i)) while the quotient singularity is V^*/G. The author needs to redo the junior-element classification for the dual representation and adjust Examples 2.5–2.6 and Corollary 3.5 accordingly. This is a substantial revision, but the overall approach appears salvageable, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe main result is genuinely new and, on the face of it, right: M_{3,n} is of general type for n ≥ 15, completing a case that had been open. The bigness argument in Section 5 is clean and I checked the arithmetic; the chosen constants t and s do cancel λ and ψ, and the boundary coefficients become positive for n ≥ 15. The rigid component argument for 4m·Δ_{1,∅} is also standard and the test-curve computation in Lemma 4.2 checks out.\n\nThe soft spot is in the singularity analysis, and it is not merely a matter of unexpanded “one checks” steps. The paper computes ages on V = H^0(C, Ω⊗ω(p_1+...+p_n)), but the local quotient model is V^*/G where V is that space of differentials. The age on the dual is generally different: for eigenvalues {−1, i} on V the age is 3/4, but on V^* it is 5/4. Proposition 3.2 only bounds the age on V, yet Proposition 3.3 and Corollary 3.5 use this to conclude that outside Δ_{1,∅} there are no junior elements for the quotient V^*/G. That inference does not follow as written. The proof needs to rule out age_{V^*}(φ) < 1 for automorphisms outside Δ_{1,∅}, and the paper does not compute that.\n\nThe likely fix is straightforward: age_{V^*}(φ) < 1 is the same as age_V(φ^{-1}) < 1, so the same geometric case analysis should go through after inverting the automorphism, with the eigenvalue lists in Table 1 dualized. But the manuscript does not contain that step, so the no-adjunction theorem has a genuine gap in its current form.\n\nThere are also two minor textual issues: the reference [BMS] appears to duplicate [AB], and the general definition of the Farkas divisor says n = (2r+1)(g+1) while the proof uses D^3_{3,14}, consistent with (2r+1)(g−1). These are easy corrections.\n\nBottom line: the paper deserves a serious referee. The bigness part is solid and the final theorem is important, but the singularity analysis needs to be reworked or clarified on the dual-age issue before the proof is complete.","headline":"Genuinely new result with a solid bigness computation, but the singularity analysis applies the Reid-Tai criterion to the wrong dual representation.","tokens_in":19362,"tokens_out":7304,"would_cite":true,"duration_ms":73141,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14H10","14E08"],"pacs":[],"model":"deepseek-v4-flash","headline":"For every n≥15, the moduli space $M_{3,n}$ of stable pointed genus-3 curves is of general type, completing the Kodaira classification in genus 3.","keywords":["moduli of curves","Kodaira dimension","general type","genus 3","pointed curves","Reid-Tai criterion","canonical divisor","singularities of moduli spaces"],"falsifier":"Take a stable pointed genus-3 curve $[C,p_1,\\ldots,p_n]$ with $n \\geq 1$ and $[C,p_1,\\ldots,p_n] \\notin \\Delta_{1,\\emptyset}$. Compute the eigenvalues of a nontrivial automorphism $\\varphi \\in \\operatorname{Aut}(C,p_1,\\ldots,p_n)$ acting on $V = H^0(C, \\Omega_C \\otimes \\omega_C(p_1+\\cdots+p_n))$ and their age. If any such computation yields $0 < \\operatorname{age}(\\varphi) < 1$, Proposition 3.3 is contradicted; the entries of Table 1 are directly checkable by diagonalizing the listed elliptic, hyperelliptic, and rational cases.","tokens_in":18046,"feed_emoji":"📐","tokens_out":12639,"duration_ms":124461,"temperature":0.7,"pith_summary":"This paper completes the Kodaira classification of moduli spaces of stable pointed curves in genus 3. Its theorem is that $M_{3,n}$ is of general type whenever $n \\geq 15$, and since $M_{3,n}$ was already known to be rational for $n \\leq 14$, the Kodaira dimension of every genus-3 moduli space is now determined. The proof separates into two statements: the canonical class of $M_{3,n}$ is big for $n \\geq 15$, and the singularities of $M_{3,n}$ impose no adjunction conditions for $n \\geq 1$, so the big canonical class actually forces general type. The singularity analysis is the delicate part: it identifies the non-canonical locus as exactly the locus of curves with an unmarked elliptic tail of j-invariant 0.","feed_headline":"Fifteen marked points push genus-3 moduli into general type","feed_subtitle":"With n≤14 already known rational, the Kodaira classification of genus-3 moduli spaces is now complete.","key_machinery":"The argument is carried by two mechanisms. First, the Reid--Shepherd-Barron--Tai criterion and its refinements: a finite quotient $V/G$ has canonical singularities when $G$ contains no junior elements, where an element is junior when the sum of the fractional parts of its eigenvalues on $V$ lies strictly between 0 and 1. A variant from [CCM] gives an explicit inequality involving the order of vanishing of a pluri-canonical form along coordinate hyperplanes, allowing such forms to lift even at non-canonical quotient singularities. Second, a classification of all possible junior automorphisms of pointed genus-3 curves: Lemma 3.1 and Proposition 3.2 produce a finite list of cases, showing that outside the locus of unmarked elliptic tails of j-invariant 0, no junior automorphisms occur. For bigness, the machinery consists of the canonical divisor formula $K_{M_{3,n}} = 13\\lambda + \\psi_1 + \\cdots + \\psi_n - 2\\delta - \\delta_{1,\\emptyset}$, the test-curve class obtained by gluing a fixed genus-2 pointed curve to a variable elliptic tail, the effective divisors coming from the hyperelliptic locus and from $D^3_{3,14}$, and the bigness of $a\\lambda + b\\psi$ for positive $a,b$.","core_discovery":"The central claim is Theorem 1.1: for every $n \\geq 15$, the moduli space $M_{3,n}$ is of general type. This is deduced from Theorem 1.2, which shows that the canonical class $K_{M_{3,n}} = 13\\lambda + \\psi_1 + \\cdots + \\psi_n - 2\\delta - \\delta_{1,\\emptyset}$ is big by writing it as a positive combination of big classes, effective divisor classes pulled back from the hyperelliptic locus and from the divisor $D^3_{3,14}$, and boundary classes; and from Theorem 1.3, which shows that all global $m$-canonical forms lift from the coarse moduli space to a resolution, so the singularities impose no adjunction conditions. The key singularity result is Corollary 3.5: for $n \\geq 1$, the locus of non-canonical singularities is exactly $J_0$, the locus of pointed curves admitting an unmarked elliptic tail of j-invariant 0. Proposition 1.5 adds the needed rigidity: the pluri-canonical divisor $mK_{M_{3,n}}$ contains the boundary divisor $4m\\Delta_{1,\\emptyset}$ as a rigid component, so every global $m$-canonical form vanishes along $\\Delta_{1,\\emptyset}$ to order at least $4m$, which feeds into the lifting criterion at the non-canonical locus.","pith_inferences":["Beyond the paper, the refined lifting criterion should transfer to higher genera: failure of the Reid--Tai hypothesis at a non-canonical locus need not be fatal if one can exhibit a rigid boundary component of high multiplicity, so the known thresholds for higher genera could be attacked by the same combination.","Extending the test-curve computation of Proposition 1.5 to $M_{g,n}$ for $g \\geq 4$ would show whether the factor $4m$ along $\\Delta_{1,\\emptyset}$ is a general low-genus phenomenon; if it is, the adjunction step in higher genera becomes purely a statement about bigness of the canonical class.","The sharp jump from rational at $n=14$ to general type at $n=15$ in genus 3 suggests that for other fixed $g$, the Kodaira dimension may also jump directly from $-\\infty$ to maximal as $n$ grows, with no intermediate values; checking this pattern at the current thresholds in higher genera would test the numerical mechanism behind the bigness proof."],"forward_implications":["For every $n \\geq 15$, $M_{3,n}$ has maximal Kodaira dimension equal to its dimension $6+n$, so it is not uniruled and not rational.","Combined with the known rationality of $M_{3,n}$ for $n \\leq 14$, the theorem gives the complete Kodaira classification in genus 3: Kodaira dimension $-\\infty$ for $n \\leq 14$ and maximal for $n \\geq 15$.","For $n \\geq 1$, the non-canonical singular locus of $M_{3,n}$ is exactly $J_0$, the locus of pointed curves with an unmarked elliptic tail of j-invariant 0; all other quotient singularities are canonical.","Every global $m$-canonical form on $M_{3,n}$ vanishes along $\\Delta_{1,\\emptyset}$ to order at least $4m$, a rigid-boundary phenomenon that makes the adjunction condition automatic.","The theorem confirms the expectation that in each fixed genus only finitely many pairs $(g,n)$ fail to be of general type, with genus 3 now settled at the threshold $n=15$."],"supporting_citations":[{"why":"Supplies the canonical divisor formula on $M_g$, the age-based framework for adjunction conditions, and the hyperelliptic divisor class used in the bigness proof.","marker":"[HM]"},{"why":"Establishes the general-type thresholds for $g \\geq 4$ and the bigness criterion for $a\\lambda + b\\psi$ used at the end of Theorem 1.2.","marker":"[Lo]"},{"why":"Computes the class of the divisor $D^r_{g,n}$ of curves admitting a section of $\\omega^{\\otimes(r+1)}(-\\sum p_i)$; this supplies the effective input $D^3_{3,14}$ in the bigness argument.","marker":"[Fa]"},{"why":"Provides the refined lifting criterion (Proposition 2.4) that lets pluri-canonical forms with controlled vanishing lift even at non-canonical quotient singularities.","marker":"[CCM]"},{"why":"Gives the classical Reid--Tai criterion: finite quotients without junior elements are canonical, the primary test used to locate non-canonical singularities.","marker":"[Ta]"},{"why":"Provides the companion statement of the Reid--Tai criterion on canonical singularities of finite quotient varieties.","marker":"[Re]"},{"why":"Supplies the Picard basis, the canonical class formula on $M_{g,n}$, the Mumford relation $\\kappa_1 = 12\\lambda + \\psi - \\delta$, and the intersection numbers of the test curve.","marker":"[ACG]"},{"why":"Establishes the rational Picard group basis of $M_{g,n}$ on which all divisor class computations rely.","marker":"[AC]"},{"why":"Proves rationality of $M_{3,n}$ for $n \\leq 14$, the complementary half of the Kodaira classification in genus 3.","marker":"[CF]"},{"why":"Supplies the lemma used to identify a rigid component of an effective divisor from intersection numbers with a curve class, the mechanism behind Proposition 1.5.","marker":"[BMS]"}],"fun_headline_variants":["Fifteen points push genus-3 moduli to general type","Genus-3 moduli: general type for n≥15","Kodaira classification of genus-3 moduli now complete","Canonical class big: genus-3 moduli general type at n=15","No adjunction conditions: genus-3 moduli general type"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the classification of junior automorphisms collected in Lemma 3.1 and Proposition 3.2 is exhaustive: if even one nontrivial automorphism of a pointed genus-3 curve outside the exceptional loci has age below 1 and is missing from the tables, the identification of the non-canonical locus with $J_0$ and the lifting argument both break down.","fun_headline_variants_meta":{"raw":{"variants":["Fifteen points push genus-3 moduli to general type","Genus-3 moduli: general type for n≥15","Kodaira classification of genus-3 moduli now complete","Canonical class big: genus-3 moduli general type at n=15","No adjunction conditions: genus-3 moduli general type"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000159,"raw_usage":{"total_tokens":1223,"prompt_tokens":933,"completion_tokens":290,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":549,"completion_tokens_details":{"reasoning_tokens":199}},"tokens_in":549,"tokens_out":290,"duration_ms":3531,"temperature":1.0,"reasoning_tokens":199,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T21:33:11.750791+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a stable pointed genus-3 curve $[C,p_1,\\ldots,p_n]$ with $n \\geq 1$ and $[C,p_1,\\ldots,p_n] \\notin \\Delta_{1,\\emptyset}$. Compute the eigenvalues of a nontrivial automorphism $\\varphi \\in \\operatorname{Aut}(C,p_1,\\ldots,p_n)$ acting on $V = H^0(C, \\Omega_C \\otimes \\omega_C(p_1+\\cdots+p_n))$ and their age. If any such computation yields $0 < \\operatorname{age}(\\varphi) < 1$, Proposition 3.3 is contradicted; the entries of Table 1 are directly checkable by diagonalizing the listed elliptic, hyperelliptic, and rational cases.","supporting_citations":[],"review_version":1}