{"id":"b2d8f00f-dfe5-4a14-b579-96f8f84447b5","arxiv_id":"1908.06411","paper_version":4,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"For every N and every prime l, the l-primary subgroup of the rational cuspidal divisor class group of X_0(N) is a direct sum of explicit cyclic subgroups generated by the new divisors Z_l(d).","lead":"This paper finds explicit generators and orders for every prime-by-prime piece of the rational cuspidal divisor class group of the modular curve X_0(N), for every N. It completes a long-running program aimed at the conjectural description of the rational torsion of the Jacobian J_0(N).","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 6.7's splitting for levels with v2(N) >= 5 relies on an unproved injectivity of the degeneracy pullback π1*; if that citation fails, Theorem 1.5 is not established for N divisible by 32.","rationale":"After reading the full argument, the main line is coherent: the linear-independence criteria, vector constructions, and order formulas are consistent with the claimed structure. The reader's conditional verdict is appropriate. The most dangerous step is not any algebraic identity inside the main induction, but the external injectivity assertion in Section 6.7, because (a) it is exactly the bridge needed for the hard N divisible by 32 case, (b) it is not proved in this paper, and (c) the corresponding map may have degree divisible by 2, so injectivity on 2-primary cuspidal classes is not a consequence of general principles. Other delegated computations, such as the beta-case proofs in Lemmas 2.21 and 2.22 and the genus-zero check of Theorem 3.13, are checkable but numerous; they increase the cost of verification without changing the nature of the concern. No evidence of internal inconsistency was found; the concern is about an external citation and could be settled by a targeted check.","tokens_in":92265,"tokens_out":12911,"duration_ms":133909,"concrete_test":"Locate [32, Rem. 2.7] and check exactly what map it proves injective. Independently, for N=96 compute the linear-equivalence class of π*_1(B2(5,5)) by applying the paper's V(C) algorithm: form C as the pullback of B2(5,5) (the generator of C(32) of order 4) under π1(96,32), compute V(C)=Υ(96)·Φ96(C), then apply Theorem 3.13 to get its order. If the order is 4, injectivity on this generator holds and the §6.7 step is sound for N=96; if the order is 1 or 2, the split (6.7) is false and the main theorem fails for N divisible by 32.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The weakest point is in the proof of Theorem 6.1, Section 6.7, where the non-squarefree splitting for N divisible by 32 is reduced to the isomorphism π*_1(C(2^r)) ≃ ⊕_{f=3}^r ⟨π*_1(B2(r,f))⟩. The justification is a bare citation: \"Since π*_1 is injective (cf. [32, Rem. 2.7])\". For a finite morphism of curves, pullback on Pic^0 is not automatically injective; composing with pushforward gives f_* f^* = deg(f), so any torsion of order dividing deg(f) can lie in the kernel. Here deg(π1(96,32)) = 4 and C(32) ≅ Z/4, so injectivity on this 2-group is a nontrivial fact rather than a formality. If the cited remark is inapplicable or false, then the direct sum in (6.7) can collapse and the main theorem is not established for every N divisible by 32. The paper also leaves some routine beta-case computations to the reader (Lemmas 2.21 and 2.22) and the genus-zero check in Theorem 3.13; these are secondary but add to the verification burden.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops an explicit vector method to determine the rational cuspidal divisor class group C(N) of X0(N) for every positive integer N. The main results are Theorem 1.5 (splitting of C(N) into a squarefree part plus cyclic non-squarefree summands), Theorem 1.6 (ℓ-primary structure of the squarefree part via divisors Y2(d)), and Theorem 1.7 (ℓ-primary structure of all of C(N) via divisors Zℓ(d)). The technical engine is a computation of V-vectors attached to rational cuspidal divisors, an order formula (Theorem 3.13), tensor and degeneracy-map constructions of divisor families (A_p, B_p, B_2, Z, Y_i, D), and linear-independence criteria (Theorems 5.1–5.3). The proof is remarkably complete in its overall architecture, with explicit generators, explicit order formulas n(N,d) and N(N,d), and an explicit treatment of the difficult 2-primary cases. Several computational checks are, however, delegated to the reader, and one splitting step in Section 6.7 relies on a cited injectivity statement that is not proved or quoted in the text.","tokens_in":1923,"tokens_out":2784,"duration_ms":111737,"significance":"If correct, the paper completely determines the structure of the rational cuspidal divisor class group of X0(N) for all N, with explicit generators and orders. This is the strongest unconditional evidence to date for the generalized Ogg conjecture. The paper has substantial strengths: it reproves the necessary Ligozat criterion, gives a self-contained treatment of cusps and degeneracy maps, provides explicit lower-triangular and lower-ℓ-unipotent matrices that make the direct-sum decompositions concrete, and honestly flags the hard cases (N divisible by 32, the Y2(A(1)) case). The main risk is that the theorem rests on a large body of hand-verified vector identities, several of which are delegated to the reader, and on one injectivity claim whose cited justification is not reproduced. These issues are fixable but need to be addressed before the central claim can be considered fully established.","major_comments":[{"comment":"The proof of (6.7) for N divisible by 32 uses the isomorphism π1^*(C(2^r)) ≃ ⊕_{f=3}^r ⟨π1^*(B2(r,f))⟩, justified only by 'Since π1^* is injective (cf. [32, Rem. 2.7])'. For a finite morphism of curves, pullback on Pic^0 is not automatically injective: composing with pushforward gives f_* f^* = deg(f), so elements of torsion order dividing deg(π1)=2^r can lie in the kernel. Since C(2^r) is a 2-group, this is a genuinely nontrivial assertion. The cited remark should be stated and proved, or a direct argument supplied. This step is load-bearing: if injectivity fails, the direct sum in (6.7) can collapse and Theorem 1.5 is not established for N divisible by 32. The author's own Remark 6.28 acknowledges that the proof for this case is indirect, which makes the need for a complete justification more pressing.","section":"§6.7, Eq. (6.7)"},{"comment":"The order formula in Theorem 3.13 is foundational for every later order computation, but its proof says that for the fourteen genus-zero levels N ∈ {1,2,3,4,5,6,7,8,9,10,12,13,16,18,25} 'we can easily verify the formula, which we leave to the readers.' This is a finite check, but it is an omitted proof of a result used throughout the paper. Please include the verification, for instance as a short table of genus-zero levels with the relevant orders, or a uniform argument that J0(N)=0 in those cases suffices.","section":"§3.3, Theorem 3.13"},{"comment":"The proofs of Lemmas 2.21 and 2.22 state that the β-map computations are 'similar' and leave the details to the reader. These formulas are used later in Proposition 5.4 and hence feed directly into the construction of the vectors A_p(r,f) and B_p(r,f) and the generation claims. Since the manuscript's policy is generally to be self-contained, the β-case computations should be written out or at least summarized in an appendix.","section":"§2.6, Lemmas 2.21 and 2.22"},{"comment":"The central structural argument consists of a long sequence of asserted identities for V-vectors: that certain entries are zero, that certain diagonal entries are ±1 or ℓ-adic units, and that certain Pw_p sums have the claimed parity. A single wrong entry can convert a direct sum into a proper extension, so these computations are load-bearing. The text often says 'by direct computation' or refers to a displayed table, but does not provide machine-checkable code or a systematic verification certificate. I did not independently verify all of these identities within the review budget. I recommend that the author supply a verification appendix, e.g., computer algebra scripts checking the finite collection of vector identities, or a more detailed case-by-case derivation for the least transparent steps (notably Lemma 6.37 and Lemma 6.40).","section":"§6.5–§6.6, Propositions 6.14, 6.24, 6.25, 6.35, Lemmas 6.37 and 6.40"}],"minor_comments":[{"comment":"The statement of Theorem 1.7 does not give the order of Zℓ(d); the order is only defined in a footnote through the prime-to-ℓ parts of n(N,d) and N(N,d′). It would improve readability to state the order of Zℓ(d) directly in the theorem.","section":"§1.1, Theorem 1.7"},{"comment":"The orderings ≺ and ⊳ are intricate, and the proof that ι is a bijection in Lemma 6.8 is compressed. A small example illustrating ι for a level with t=3 and u=1 would be very helpful to the reader.","section":"§6.1, Definitions 6.2–6.6"},{"comment":"In Definition 1.17, the expression H(N,p_I) := 2 if I∈(F^1_u∪G^1_u∪{A(1)}) \\ (F_s∪G_s), and 1 otherwise, is easy to misread because F_s and G_s themselves depend on s. Please spell out at least one example for the 2-primary case.","section":"§1.4, Definition 1.17"},{"comment":"The sentence 'Since Tu = {I_f : 3≤f≤r}' is correct only when r≥5 and u is the index of the prime 2; for r≤4 the earlier degenerate case is handled separately. Making this conditional explicit at that point would prevent confusion.","section":"§6.7, end of proof of Theorem 6.1"}],"recommendation":"major_revision","confidential_remarks":"The paper is technically ambitious and, if the vector computations are correct, it solves a natural and previously open problem. My main concern is structural rather than stylistic: the proof for N divisible by 32 depends on an injectivity statement about π1^* that is only cited, not proved. I could not verify the large body of hand computations within the review budget, and the delegation of several checks to the reader increases the risk. If the author can provide a proof of the injectivity claim and make the computational core independently verifiable, I would support publication, possibly after a shorter second round."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is the paper that finishes the Ogg-Mazur program for C(N). For every N and every prime l, Yoo constructs divisors Z_l(d) and proves C(N)[l^∞] is the direct sum of their cyclic classes. If the main theorems hold, this is the strongest evidence so far for the generalized Ogg conjecture. The advance over the cited literature is real: prime powers, two primes, powers of 2, and squarefree N were all known cases; this handles arbitrary N by splitting off the squarefree part and then doing the l-primary decomposition.\n\nThe paper is unusually careful. The order algorithm (Theorem 3.13) is a genuine simplification of Ligozat's method, and the linear independence criteria in Section 5 are clean and general. The author names the hard cases: Remark 6.28 admits the direct strategy fails for levels divisible by 32 and an indirect argument is used; Remark 6.41 begins to state another limitation but is truncated mid-sentence, which is frustrating and should be fixed.\n\nNow the soft spots, in proportion. First, a large verification burden. The whole structure theorem runs on hand-checked V-vectors and GCD computations. Several subcomputations are explicitly left to the reader: the beta cases in Lemmas 2.21 and 2.22 and the genus-zero check in Theorem 3.13. I did not independently verify any of this, and a referee needs to. This is not a flaw in the argument as written, but it is a high bar for certification.\n\nSecond, and more load-bearing, the proof of the v2(N) >= 5 case in Section 6.7 reduces to an injectivity statement for the degeneracy pullback pi_1^* on C(2^r): \"Since pi_1^* is injective (cf. [32, Rem. 2.7])\". The stress-test note is right that this is not a formality. For finite morphisms of curves, f_* f^* = deg(f), so anything of order dividing deg(f) can in principle be in the kernel. Here deg(pi_1(96,32)) = 4 and C(32) is Z/4, so injectivity on this 2-group is a nontrivial fact. If the cited remark is inapplicable, Theorem 1.5 is not established for N divisible by 32. The author should either reprove injectivity in this setting or state the exact lemma.\n\nOverall, the architecture is coherent and I found no contradiction. The claim is plausible and the paper deserves a serious referee. My recommendation: send it to review, but require the referee to verify the delegated computations (Lemmas 2.21/2.22, genus-zero checks) and confirm the [32, Rem. 2.7] citation. Also fix Remark 6.41. For readers, if you work on torsion of Jacobians, this will be the standard reference; cite it.","headline":"Yoo's paper completes the program for the rational cuspidal divisor class group of X0(N) with explicit generators; worth serious referee time, but the 2-adic and delegated hand computations need checking.","tokens_in":93192,"tokens_out":2782,"would_cite":true,"duration_ms":26252,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11G16","11G18","14G05"],"pacs":[],"model":"deepseek-v4-flash","headline":"For every positive integer N, the rational cuspidal divisor class group of X0(N) is now completely determined.","keywords":["rational cuspidal divisor class group","modular curve X0(N)","cuspidal subgroup","generalized Ogg conjecture","eta quotients","l-primary subgroup","direct sum decomposition","Jacobian torsion"],"falsifier":"Check the construction at a moderately composite level, e.g. $N=2^5\\cdot 3$ and $\\ell=2$: compute the normalized V-vectors of the divisors $Y_2(d)$ and $Z_2(d)$ exactly as defined in Sections 4.5 and 6.6, verify that the matrix is lower 2-unipotent with the claimed diagonal units, and compare the orders $n(N,d)$ and $N(N,d)$ with an independent modular-symbol computation of $C(N)[2^\\infty]$. A mismatch in any entry of the matrix, or in any order, would falsify the central decomposition; equivalently, verifying the generation propositions for one level with several prime factors would decide the claim.","tokens_in":91836,"feed_emoji":"🔢","tokens_out":10236,"duration_ms":91235,"temperature":0.7,"pith_summary":"Every prime-part of the rational cuspidal divisor class group C(N) of the modular curve X0(N) is now computed. For each nontrivial divisor d of N and each prime ℓ, the paper constructs an explicit degree-zero rational cuspidal divisor Z_ℓ(d), gives a closed formula for the order of its class, and proves that C(N)[ℓ^∞] is the direct sum of the cyclic subgroups generated by these classes. The consequence is that the finite abelian group C(N) is completely known for all N, not merely up to order, settling the structure question on the cuspidal side of the conjectured equality with the rational torsion of J0(N). The computation is carried by a vector method: divisors are converted into vectors through an integral eta-quotient matrix, and direct-sum decompositions are certified by lower-triangular matrices of normalized vectors.","feed_headline":"Every rational cuspidal divisor class group of X0(N) is now known","feed_subtitle":"Explicit divisors give cyclic generators and orders for every prime part of the group.","key_machinery":"The operating object is the V-vector. For each degree-zero rational cuspidal divisor C, the paper forms the coefficient vector in $S_2(N)$ and multiplies by an explicit integral matrix $\\Upsilon(N)$—the tensor product of tridiagonal matrices derived from the eta-quotient order matrix—to obtain $V(C)\\in S_1(N)$; the normalized vector $\\mathcal{V}(C)=V(C)/\\gcd$ behaves like a coordinate vector. The order of C is read off from $\\gcd(V(C))$ and the parity of sums $Pw_p(\\mathcal{V}(C))$ (Theorem 3.13). The linear-independence criteria (Theorems 5.1–5.3) then say: if the normalized vectors of a list of divisors form a lower-triangular matrix with diagonal ±1 or ℓ-adic units and the parity conditions hold, the subgroups add in direct sum. The rest of the paper is the construction of divisors—$A_p(r,f)$, $B_p(r,f)$, $B_2(r,f)$, $D(p_i^{r_i},p_j^{r_j})$ and their tensor products—whose V-vectors have exactly this shape.","core_discovery":"At the paper's own level of claim, Theorem 6.1 states the complete structure: C(N) is isomorphic to the direct sum of the squarefree part and the cyclic groups generated by Z(d) for non-squarefree d, and for each prime ℓ the ℓ-primary subgroup of the squarefree part is isomorphic to the direct sum over squarefree d of the cyclic groups generated by the divisors Y_2(d). Combining the two, for every prime ℓ the ℓ-primary subgroup of all of C(N) is a direct sum of cyclic groups generated by explicitly defined divisors Z_ℓ(d), one per nontrivial divisor of N, with orders given by the formulas n(N,d) and N(N,d), which are numerators of $G\\cdot H/24$ with explicit $G$ and $H$. This gives the exact isomorphism type of C(N), not a bound or a conjecture.","pith_inferences":["If the result is correct, the explicit generators give a ready-made set of candidate torsion points on J0(N); comparing these cyclic summands with Hecke-module or Eisenstein-ideal computations would test the conjectured equality with the rational torsion subgroup on a much larger range than before.","The V-vector method is not obviously tied to the particular family X0(N): the same ingredients—orbit-sum cuspidal divisors, an integral order matrix from eta quotients, and degeneracy maps—exist for X1(N) and Atkin–Lehner quotients, so an analogous decomposition may hold there.","Because the proof leaves some vector computations to the reader, a machine-checked verification of the relevant tables for all N up to a chosen bound would be a natural companion; it would confirm the lower-unipotent patterns on which the whole argument rests.","The splitting into squarefree and non-squarefree parts suggests that the obstruction to a single global basis for C(N), rather than a prime-by-prime ℓ-adic basis, lives entirely in the squarefree component; understanding the transition from ℓ-adic bases to an integral basis is the natural next question."],"forward_implications":["For every positive integer N, the finite abelian group C(N) is now known: explicit generators and orders for every prime component, not merely the order of the group.","The rational cuspidal subgroup of J0(N) is determined whenever it equals C(N), in particular for N=4M and N=8M with M odd squarefree, giving explicit instances of the conjectured equality with rational torsion.","The splitting theorem isolates the non-squarefree part of C(N) as a direct sum of cyclic groups generated by Z(d); the squarefree part is reduced, prime by prime, to cyclic summands generated by Y_2(d).","The new linear-independence criteria give a general, order-theoretic method for detecting relations among rational cuspidal divisors, replacing previous case-by-case searches for relations.","All constructed generators are supported only at cusps and are defined by explicit formulas, so they can be written down and used in further computations without solving auxiliary equations."],"supporting_citations":[{"why":"Supplies the eta-quotient criterion and the matrix relating divisor coefficients to exponents; the paper's order formula and $\\Upsilon(N)$ matrix invert it.","marker":"[7, Sec. 3.2]"},{"why":"Gives the finiteness of cuspidal divisor classes, so the quotient by divisors of modular units is a finite torsion group.","marker":"[10, Cor. 3.6]"},{"why":"Gives the complementary finiteness result used with [10] to define the cuspidal subgroup.","marker":"[4]"},{"why":"Describes the Galois action on cusps, yielding the rational cuspidal divisors $(P_d)$ and the basis $C_d$ built from them.","marker":"[24, Th. 1.3.1]"},{"why":"Supplies the canonical $\\mathbb{Q}$-model of $X_0(N)$ used throughout the definitions.","marker":"[23, Ch. 6]"},{"why":"Computes the prime-power case; the paper's $B_p(r,f)$ construction extends and repairs that method.","marker":"[8]"},{"why":"Computes the power-of-2 case and supplies the baseline for the defective 2-primary vectors that motivate $B_2(r,f)$.","marker":"[22, Th. 10]"},{"why":"Computes the order of C(N) for squarefree N and is the starting point for the squarefree-part decomposition by $Y_2(d)$.","marker":"[26, Th. 6.1]"}],"fun_headline_variants":["Cuspidal divisor class group on X0(N): complete structure from explicit divisors","Cyclic generators give full classification on X0(N) for cuspidal divisors","Explicit divisors determine cuspidal class group on X0(N) exactly","Full structure of cuspidal divisor class group on X0(N) now explicit","Cuspidal divisor class groups on X0(N) fully pinned down by generators"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the paper's extensive hand-verified vector arithmetic is correct: the constructed divisors integrally generate the relevant divisor groups, and every normalized V-vector has exactly the claimed zero, ±1, or ℓ-adic-unit entries in the prescribed lower-triangular pattern; a single wrong entry could turn a direct sum into a proper extension, and the text leaves some auxiliary computations, including a beta-case and a genus-zero check, to the reader.","fun_headline_variants_meta":{"raw":{"variants":["Cuspidal divisor class group on X0(N): complete structure from explicit divisors","Cyclic generators give full classification on X0(N) for cuspidal divisors","Explicit divisors determine cuspidal class group on X0(N) exactly","Full structure of cuspidal divisor class group on X0(N) now explicit","Cuspidal divisor class groups on X0(N) fully pinned down by generators"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001241,"raw_usage":{"total_tokens":5046,"prompt_tokens":854,"completion_tokens":4192,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":470,"completion_tokens_details":{"reasoning_tokens":4082}},"tokens_in":470,"tokens_out":4192,"duration_ms":26567,"temperature":1.0,"reasoning_tokens":4082,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:48:10.178375+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Check the construction at a moderately composite level, e.g. $N=2^5\\cdot 3$ and $\\ell=2$: compute the normalized V-vectors of the divisors $Y_2(d)$ and $Z_2(d)$ exactly as defined in Sections 4.5 and 6.6, verify that the matrix is lower 2-unipotent with the claimed diagonal units, and compare the orders $n(N,d)$ and $N(N,d)$ with an independent modular-symbol computation of $C(N)[2^\\infty]$. A mismatch in any entry of the matrix, or in any order, would falsify the central decomposition; equivalently, verifying the generation propositions for one level with several prime factors would decide the claim.","supporting_citations":[{"cited_title":"Appl., 7 (1973), 155–156","cited_arxiv_id":null,"evidence_quote":"Gives the complementary finiteness result used with [10] to define the cuspidal subgroup."},{"cited_title":"Kubert and Serge Lang, Modular units, Grundlehren der mathematischen Wissenschaften, Vol","cited_arxiv_id":null,"evidence_quote":"Computes the prime-power case; the paper's $B_p(r,f)$ construction extends and repairs that method."}],"review_version":1}