{"id":"898a91e1-d9ec-4f87-bddf-9d7fae1d766b","arxiv_id":"2510.08093","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A general cubic rational self-map of P^2 is surjective when its indeterminacy locus has at most six points, and two new explicit surjective cubic maps are proved.","lead":"The paper claims that a generic cubic rational self-map of the complex projective plane is surjective exactly when its indeterminacy locus has 1–6 points (in the body; the abstract says \"at least 3\"), and it constructs two new explicit surjective cubic maps by combining finite-field enumeration with neural-network heuristics. The geometric proof via projections of del Pezzo surfaces is attractive, but one essential genericity step is asserted rather than proved.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2.6 relies on an unproven genericity assertion: the residual curve R must meet E_i in two distinct points with ψ|_R unramified there. The text merely says this is 'immediate' by generality, so the generic classification is conditional on a missing Bertini-type argument.","rationale":"The reader's weakest_assumption identifies exactly the same point: the proof of Theorem 2.6 declares by 'generality of f (and Σ)' that the required unramified/non-tangency condition holds, but gives no explicit open condition. My analysis of the proof confirms that this is the step on which the surjectivity criterion for δ ≥ 3 hinges. The explicit examples in Propositions 3.2 and 3.6 are proved directly and appear correct, so they do not rescue the generic classification. The ML portion is not reproducible as shipped because output.txt is missing, and the paper itself hedges on the reliability of the neural-network predictions; however, this is not part of the core geometric claim. The abstract's 'at least 3' statement is inconsistent with Corollary 2.3, but that is a presentational error that can be corrected without changing the mathematics. The key unresolved issue is whether the missing genericity assertion can be made rigorous by a standard parameter count; my proposed computational test on the degree-4 case would settle whether the bad locus is genuinely negligible or whether the theorem's conclusion can fail. Since the reader already marked the paper CONDITIONAL with moderate confidence and this concern supports that assessment, the verdict should remain unchanged.","tokens_in":12256,"tokens_out":13709,"duration_ms":113804,"concrete_test":"Take the degree-4 del Pezzo surface X ⊂ P^4 from Section 3.1, fix one exceptional line E_i, and parameterize the projection center Σ as a line in G(1,4) disjoint from X. For each Σ, set H = span(Σ, E_i) and R = X ∩ H − E_i. Compute, in Plücker coordinates, the closed condition that R is tangent to E_i at some point, or that at an intersection point o ∈ R ∩ E_i the tangent line T_o R meets Σ. Determine whether this bad locus B ⊂ G(1,4) is a proper closed subset by saturating its ideal and evaluating at one explicit rational Σ outside it. If B is proper, the genericity step is valid and the proof gap is fillable; if B = G(1,4), the degree-4 case would contradict Theorem 2.6 and the classification would need revision.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is inside the proof of Theorem 2.6. For a fixed exceptional line E_i, the proof considers the hyperplane H with X ∩ H = E_i + R and needs R ∩ E_i = {o_1, o_2} with distinct o_j, and needs ψ|_R unramified at both points. Lemma 2.7 only gives smoothness of R. The equality R · E_i = 2 does not imply two distinct intersection points: tangency would give one point of multiplicity 2. The proof then states 'this is immediate — again by generality of f (and Σ)' without specifying any open condition. This is not cosmetic: the conclusion ψ^{-1}(o) ⊄ E_i for o ∈ ψ(E_i) depends on having a reduced preimage point in R outside E_i. If R were tangent to E_i at an intersection point, or if the tangent line to R met the projection center Σ so that ψ|_R ramified, the local-isomorphism argument would fail and the criterion from [15, Prop. 3] would not be verified. Thus, as written, the central classification 'generic f is surjective iff δ ≥ 3' is only conditional on an unproven genericity claim. There is also a presentation inconsistency: the abstract's 'cardinality at least 3' contradicts Corollary 2.3, since #I_f = 7 satisfies 'at least 3' yet is non-surjective.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies surjective rational endomorphisms f: P^2 ⇢ P^2 given by cubics with nonempty indeterminacy locus I_f. For a general set I_f of 9−δ points in general position, the main theorem (Theorem 2.6) claims that the induced map is surjective exactly when δ ≥ 3, i.e. #I_f ≤ 6; Corollary 2.3 handles δ = 2 (#I_f = 7) by constructing an unruly pencil. The proof uses the blow-up of the indeterminacy points to realize f as a projection of a del Pezzo surface Xδ ⊂ P^δ. The paper also presents two explicit cubic surjective maps (Propositions 3.2 and 3.6) found by finite-field experiments, together with a neural-network-based heuristic 'measure of surjectivity'. The abstract states the criterion as '#I_f has cardinality at least 3', which contradicts the body.","tokens_in":12686,"tokens_out":8213,"duration_ms":66523,"significance":"If the genericity step in Theorem 2.6 can be rigorously justified, the paper would provide a clean classification for generic cubic plane rational endomorphisms with prescribed general indeterminacy locus, extending the quadratic case of Kulikov–Zhdanovskiy. The two explicit surjective cubic maps are nontrivial and correctly proven modulo small gaps; they are genuine contributions and illustrate that experimental finite-field search can identify interesting examples. The ML/NN part is anecdotal: it gives no reproducible data, error bars, or a precise predictive claim, so it should be viewed as motivational rather than as a load-bearing component of the paper.","major_comments":[{"comment":"The proof asserts that R·E_i = 2 implies R∩E_i = {o_1,o_2} with distinct points, and then that ψ|_R is unramified at o_1,o_2, saying 'this is immediate — again by generality of f (and Σ)'. This is not an explicit open condition. Intersection number 2 does not exclude tangency, and Lemma 2.7 only proves smoothness of R. If R is tangent to E_i or if ψ|_R ramifies at the intersection points, the criterion ψ^{-1}(o) ⊄ E_i from [15, Prop. 3] is not verified. The proof needs a concrete genericity statement (e.g. an open condition on the plane Π in the Grassmannian, or on (f,Σ)) guaranteeing that R∩E_i is reduced and that ψ|_R is unramified there. As written, Theorem 2.6 — the central classification — is conditional on this missing argument.","section":"§2.4, proof of Theorem 2.6"},{"comment":"The abstract states that a general non-regular cubic endomorphism is surjective iff I_f has cardinality at least 3. This is inconsistent with the body: Corollary 2.3 says the generic map is not surjective when #I_f = 7, while Theorem 2.6 says it is surjective when #I_f ≤ 6. Since #I_f = 7 satisfies 'at least 3', the abstract condition is false. The correct statement is '#I_f ≤ 6' (equivalently δ ≥ 3), or the abstract must be rephrased to match Corollary 2.3 and Theorem 2.6.","section":"Abstract; Corollary 2.3"},{"comment":"In the line L = (y=0) part, the proof claims that for [1:0:a] ∈ L\\{P,Q}, f([0:1:a−2]) = [1:0:a]. For a = 2, the proposed preimage is [0:1:0], which lies in I_f, so the map is not defined there. Thus the given argument does not cover the point [1:0:2]. This can be repaired separately (for instance [1:2:0] maps to [1:0:2]), so the proposition may still be true, but the proof as written has a gap. Please fix this and check whether any other value of a is affected.","section":"§3.2, proof of Proposition 3.2"}],"minor_comments":[{"comment":"The experimental claims lack quantitative support: no dataset size, no number of test samples, no variance/error bars, and the value 0.0737 is reported as a single average with no standard deviation. The file output.txt is not included, and the provided code has formatting and truncation issues. Since these experiments are not used to prove the main theorems, this does not block acceptance, but the claims should be marked as anecdotal or supplemented with reproducible details.","section":"§3.4, ML experiment"},{"comment":"The statement that 'the openness property ... does not hold over F2' is too strong for the evidence given; an average prediction value near 0 does not establish that no Zariski-open surjectivity set exists over F2. Please soften the claim or provide a precise finite-field statement.","section":"§1.5, §3.4"},{"comment":"After the transformation (a,b) ↦ (a−b,b), the displayed map is written as a map in (x,y), but the right-hand side depends on x and z (e.g. x^2 + z − xz over x − z). This is presumably a typo for (x,z). Please clarify the variable names to avoid confusion.","section":"§3.6, proof of Proposition 3.6"},{"comment":"The relation to the author's previous paper [12] on surjective rational maps and del Pezzo surfaces should be clarified: the current geometric setup overlaps with [12], and the reader is not told which statements are new and which are recalled from there.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The central theorem is interesting but its proof has a genuine gap at the genericity step, and the abstract contradicts the body. I believe the theorem is likely true and the gap can be fixed by a standard Bertini/open-orbit argument, so I recommend major revision rather than rejection. The ML part is not load-bearing and should be clearly separated from the rigorous results. The author should also clarify the overlap with [12] and the provenance of the finite-field table, since the examples are exhibited as 'found experimentally' but the raw data are not provided."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the two explicit cubic surjective maps in Propositions 3.2 and 3.6 are genuinely checked and are new; the ML part is a heuristic that doesn't do much; the main theorem (generic surjectivity iff delta >= 3) is plausible but not proved as written because a genericity step is asserted rather than argued.\n\nWhat the paper does well: it frames the cubic case as the natural next step after the quadratic classification of Kulikov–Zhdanovskiy, and the del Pezzo interpretation (blow up the base points, f becomes a linear projection) is the right framework. Propositions 3.2 and 3.6 are elementary but careful: the authors reduce surjectivity to roots of a quartic and check the edge cases. Those arguments appear correct, and the examples give usable test cases.\n\nThe soft spots are real. In Theorem 2.6, after defining R via X ∩ H = E_i + R, the proof needs R ∩ E_i to consist of two distinct points and needs ψ|_R to be unramified at them. The text says this is immediate by generality of f and Σ. But R·E_i = 2 only gives a zero-cycle of length 2; tangency would give one point of multiplicity 2, and ramification would break the local isomorphism argument. No open condition is specified, so as written the theorem is conditional on an unproven Bertini-type statement. This is load-bearing, not cosmetic: the criterion from [15, Prop. 3] requires ψ^{-1}(o) not subset E_i. A referee should ask for a precise genericity statement.\n\nThere is also a plain inconsistency between the abstract ('cardinality at least 3') and Corollary 2.3, since #I_f = 7 satisfies 'at least 3' but the body says those maps are not surjective. The correct statement is #I_f ≤ 6. That needs fixing.\n\nThe experimental part is the weakest. The code and the output.txt file are not shipped reproducibly, there are no error bars, and the claim that openness fails over F2 rests on a small orthonormal sample and a neural-network prediction. The authors acknowledge this is heuristic, so it doesn't damage the geometry, but it's not something a referee can verify. The overlap with the author's previous del Pezzo paper [12] is also not clarified, so the novelty, while real, is hard to calibrate.\n\nWho should read it: people working on rational maps, del Pezzo surfaces, or algebraic statistics. The explicit examples are worth having even if the theorem remains conditional. My recommendation: send it to peer review, but the referee should require a rigorous genericity argument for Theorem 2.6 and a corrected abstract.","headline":"The two explicit cubic surjective maps are proven and look correct, but the generic-surjectivity theorem has a load-bearing genericity gap and the abstract contradicts the body.","tokens_in":13040,"tokens_out":2404,"would_cite":true,"duration_ms":20430,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14E05","14D05","68-04"],"pacs":[],"model":"deepseek-v4-flash","headline":"For generic cubic rational maps P^2 ⇢ P^2, surjectivity is exactly the condition that the indeterminacy locus has at most six points; seven points generically give non-surjective maps.","keywords":["surjective rational map","rational endomorphism","cubic map","indeterminacy locus","del Pezzo surface","unruly pencil","machine learning","projective plane"],"falsifier":"Run the finite-field enumeration from Section 3 over a prime p ≥ 5 for a fixed generic set of six base points; any plane Π ⊂ Λ containing a pencil ℓ with base locus Bs(ℓ) = I_f gives a cubic endomorphism with #I_f = 6 that is not surjective, contradicting Theorem 2.6. Equivalently, in the del Pezzo model, look for a hyperplane H with X∩H = E_i + R where R is singular or meets E_i non-transversely; showing such H occur for a non-empty family of generic centers would break the proof's central step.","tokens_in":12191,"feed_emoji":"🎯","tokens_out":11171,"duration_ms":93156,"temperature":0.7,"pith_summary":"This paper asks when a cubic rational endomorphism of the complex projective plane — a map defined by three cubic polynomials but undefined at its indeterminacy locus I_f — nevertheless hits every point of the target plane. The main claim, proved in the body, is that for a generic cubic map with a fixed generic set of base points, surjectivity is determined by one number: the map is surjective exactly when #I_f ≤ 6, equivalently when the blow-up of I_f is a del Pezzo surface of degree at least 3, while the generic map with seven indeterminate points is not surjective. The argument blows up I_f, views f as a linear projection of a del Pezzo surface, and shows that every fiber of that projection escapes the exceptional divisors. The paper also develops a finite-field enumeration coupled with a neural-network predictor, producing two new explicit surjective cubic maps. A reader should care because the result reduces a qualitative question about rational maps to a sharp cardinality threshold, and shows that machine-generated candidates can be certified by direct geometry.","feed_headline":"Cubic maps hit everything when their blind spot has at most six points","feed_subtitle":"Generic cubic maps of P^2 are surjective exactly when the indeterminacy locus has at most six points.","key_machinery":"The central device is the del Pezzo surface X = Bl_{I_f} P^2 of degree δ = 9 − #I_f. For δ ≥ 3 the anticanonical linear system |−K_X| embeds X as a degree-δ surface in P^δ = Λ^*, and the cubic endomorphism becomes the linear projection ψ from a codimension-3 subspace Σ onto the target plane Π^*; surjectivity is governed by the condition that no fiber ψ^{-1}(o) lies inside an exceptional divisor E_i. The key local identity is the hyperplane section X∩H = E_i + R, where R is a smooth residual curve with R·E_i = 2; checking that ψ|_R is unramified at the two intersection points R∩E_i rules out the obstruction. Equivalently, in the dual plane, f fails to be surjective exactly when Π contains an","core_discovery":"For a generic cubic rational endomorphism f: P^2 ⇢ P^2 with indeterminacy locus I_f = {P_1,...,P_{9−δ}} in general position, the paper proves that surjectivity is equivalent to δ ≥ 3, i.e. #I_f ≤ 6. For δ = 2 (#I_f = 7), Proposition 2.2 constructs an 'unruly pencil' ℓ ⊂ Π whose base locus equals I_f, so the generic f is not surjective (Corollary 2.3). For δ ≥ 3, the blow-up X of I_f is a del Pezzo surface of degree δ, and f becomes a regular projection ψ: X → Π^*; Theorem 2.6 shows ψ^{-1}(o) ⊄ E_i for every point o and exceptional divisor E_i, hence f is onto. Its proof uses the hyperplane section X∩H = E_i + R with R smooth and R·E_i = 2, and the fact that ψ|_R is unramified at R∩E_i. Propo","pith_inferences":["The abstract's formulation ('cardinality at least 3') is inconsistent with the body's proved condition δ ≥ 3, which is #I_f ≤ 6; if the body is correct, the intended threshold is 'at most six points,' and the theorem also covers maps with one or two base points, whereas the abstract's wording would exclude them.","The neural network's output is a real-valued 'measure of surjectivity' rather than a certificate; a natural strengthening would be to train on a graded invariant — for example, the minimal degree of a pencil whose base locus equals I_f — so that predictions come with error estimates instead of binary labels.","The connection drawn to rational elliptic surfaces and Painlevé families suggests a testable extension: when the projection ψ is viewed as an elliptic fibration, surjectivity may coincide with the existence of a section, and one could look for such a section in the two explicit examples."],"forward_implications":["For any fixed generic set of at most six points in P^2, every sufficiently general cubic map with exactly those indeterminacy points covers the whole target plane.","For seven points in general position, generic cubic maps through them are not surjective; the cutoff is sharp for δ = 2.","Surjectivity of a cubic rational map has an explicit geometric certificate: for every exceptional divisor E_i and every point o, the fiber ψ^{-1}(o) must not be contained in E_i; equivalently, Π contains no unruly pencil.","The finite-field search yields two new explicit surjective cubic endomorphisms; one of them is found even though every plane tested over F_2 was labelled non-surjective, showing that the finite-field heuristic can be wrong while the complex map is still surjective."],"fun_headline_variants":["Surjective cubic maps need ≤6 base points","Cubic endomorphisms: surjectivity iff at most 6 indeterminate points","Cubic maps hit everything when indeterminacy has ≤6 points","P^2 cubic surjectivity: #I_f ≤ 6","Generic cubic maps are onto when base points ≤6"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The proof of Theorem 2.6 depends on an unverified generality assertion: for a generic map and center Σ, the residual curve R in X∩H = E_i + R is smooth and the projection ψ is unramified at the two points of R∩E_i — if a non-empty family of generic projections had R tangent to E_i at an intersection point, the argument would not settle surjectivity.","fun_headline_variants_meta":{"raw":{"variants":["Surjective cubic maps need ≤6 base points","Cubic endomorphisms: surjectivity iff at most 6 indeterminate points","Cubic maps hit everything when indeterminacy has ≤6 points","P^2 cubic surjectivity: #I_f ≤ 6","Generic cubic maps are onto when base points ≤6"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001,"raw_usage":{"total_tokens":4063,"prompt_tokens":729,"completion_tokens":3334,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":473,"completion_tokens_details":{"reasoning_tokens":3260}},"tokens_in":473,"tokens_out":3334,"duration_ms":22859,"temperature":1.0,"reasoning_tokens":3260,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T10:49:13.915246+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the finite-field enumeration from Section 3 over a prime p ≥ 5 for a fixed generic set of six base points; any plane Π ⊂ Λ containing a pencil ℓ with base locus Bs(ℓ) = I_f gives a cubic endomorphism with #I_f = 6 that is not surjective, contradicting Theorem 2.6. Equivalently, in the del Pezzo model, look for a hyperplane H with X∩H = E_i + R where R is singular or meets E_i non-transversely; showing such H occur for a non-empty family of generic centers would break the proof's central step.","supporting_citations":[],"review_version":1}