{"id":"2be8703f-c0e3-45db-96fb-92bdea047ca0","arxiv_id":"2501.06779","paper_version":2,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The slopes of exceptional bundles on P2 are precisely the Markov fractions.","lead":"This paper proves that Springborn's Markov fractions, a set of rationals tied to Markov numbers, are exactly the slopes of exceptional vector bundles on the projective plane. It connects two previously separate classification results and offers a shorter proof that the ranks of such bundles are Markov numbers.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the matching of the Drézet–Le Potier recursion to Springborn's mediant rule is algebraically correct, and the remaining reliance on the DLP classification is external but standard.","rationale":"The paper's main theorem is a bridge between two known classifications. The proof consists of a local algebraic identity (15) and a tree-induction argument. The identity is verified independently and is correct: the DLP recursion and the Springborn mediant agree when applied to adjacent Markov fractions, and the induction over dyadic rationals produces the full Markov fraction tree because both index sets are the same infinite binary tree. The reader's weakest assumption identifies the external completeness of the Drézet–Le Potier classification as the main supporting import; this is legitimate and standard, and I found no internal misstatement of (14). The rank corollary is stated briefly, and the paper does not spell out how ranks follow from the slope identification alone, but ranks being Markov is already part of the cited DLP/Rudakov results, and the paper clearly presents the rank statement as a consequence of that framework rather than as an independent derivation. Overall, the central claim is well supported and no load-bearing concern lands.","tokens_in":9566,"tokens_out":26558,"duration_ms":248095,"concrete_test":"Recompute the recursion (14) from the seeds ε(0)=0/1 and ε(1/2)=1/2 (equivalently ε(1)=1/1 by the translation property) for dyadic levels up to n=6, and compare the resulting fractions with Springborn's Markov fraction tree (Fig. 2); any mismatch would signal a transcription error in (14) or in the seed normalization.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that the set E of slopes of exceptional bundles on P2 equals the set of Markov fractions. The proof imports from Drézet–Le Potier that E is the image of the function ε on dyadic rationals defined by the recursion (14), then proves that (14) is exactly the Springborn mediant rule (1) for neighbouring Markov fractions. I rechecked the key identity (15) and the induction: the algebra reduces correctly to the Markov equation, and the base pair 0/1, 1/2, as well as subsequent pairs such as 2/5, 1/2 and 5/13, 2/5, all give the expected Markov fractions. The dyadic levels of the binary tree are cofinal in the same infinite rooted binary tree used by Springborn, so the image of ε on dyadic rationals in [0,1/2] is exactly MFR once (14) is matched with (1). The only genuine reliance is on the completeness of the Drézet–Le Potier classification and on the exact transcription of (14); these are standard external facts rather than an internal gap. No load-bearing flaw was identified.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper identifies the set of slopes of exceptional vector bundles on the projective plane with the Markov fractions recently defined by Boris Springborn. The author recalls the Markov fraction tree and the Springborn mediant rule, then proves the algebraic identity (15) showing that the Drézet–Le Potier recursion (14) for the exceptional-slope function on dyadic rationals is exactly the Springborn mediant rule. Theorem 3.1 then follows from the Drézet–Le Potier classification of exceptional slopes, and Rudakov's theorem that the ranks of exceptional bundles are Markov numbers is obtained as a corollary. The final sections discuss the relation to Minkowski's question-mark function, the Frobenius unicity conjecture, and possible generalizations to del Pezzo surfaces.","tokens_in":9781,"tokens_out":19333,"duration_ms":156466,"significance":"If correct, the main theorem provides a new and conceptually simple bridge between algebraic geometry and Diophantine approximation: the exceptional slopes on P^2 are the worst approximable rationals introduced by Springborn. The central identity (15) is derived cleanly, with the Markov equation entering exactly in the last step, and it gives a genuine simplification of Rudakov's route to the Markov-number ranks. The paper is explicit that the completeness of the Drézet–Le Potier classification is an external input, so the contribution is the identification of two existing descriptions rather than a new classification. The reformulation of the Frobenius unicity conjecture in terms of exceptional bundles is attractive and potentially useful. The presentation is terse in places and contains several typos, but no load-bearing mathematical flaw was identified.","major_comments":[],"minor_comments":[{"comment":"The final step 'Now the theorem follows' is very terse: it should state explicitly that the properties ε(-x)=-ε(x) and ε(x+n)=ε(x)+n give ε(0)=0 and ε(1)=1, and that, by induction, each dyadic interval [m/2^n,(m+1)/2^n] is mapped by ε to an adjacent edge of the Markov fraction tree, so the set of all values on dyadic rationals is exactly the node set of that tree. This is an expositional clarification rather than a correction, since the argument is clear from the preceding identity.","section":"§3, proof of Theorem 3.1"},{"comment":"In the sentence 'which we claim to be the set MF of Markov fractions from [0,1/2]', the notation should be MFR, or explicitly MF∩[0,1/2], since MF was defined in (5) as the full affine orbit.","section":"§3"},{"comment":"In the list of the bottom left branch, the fraction 89/233 after 408/985 is a typo: the Pell branch continues with 2378/5741, while 89/233 belongs to the Fibonacci branch. Please correct the entry.","section":"§2, Prop. 2.4"},{"comment":"In the proof of Lemma 2.2, the displayed formula for q'_1 is garbled by a duplicated 'q′1 = q′1 ='; it should read q'_1 = (q_2^2+q_3^2)/(p_2q_3-p_3q_2).","section":"§2, Lemma 2.2"},{"comment":"The sentence 'The relation (9) now follows from the construction of the tree' is vague; a direct reference to Markov's theorem on the Conway topograph (Fig. 3) would clarify why the triples (q_1,q_2,q_3) satisfy the Markov equation and hence the Vieta relation (7).","section":"§2, Prop. 2.1"},{"comment":"The proof of Proposition 4.1 is only a reference to the McShane identity; please spell out the normalization that makes the saltus formula (21) match µ(0)=0 and µ(1)=1.","section":"§4, Prop. 4.1"}],"recommendation":"minor_revision","confidential_remarks":"The main claim appears correct and the paper is suitable for a number theory or algebraic geometry audience. The required changes are local clarifications and typo fixes; no load-bearing issue was found."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper proves that Springborn's Markov fractions are exactly the slopes of exceptional bundles on P2. The new thing is the identification itself, plus a simpler route to Rudakov's result that exceptional ranks are Markov numbers. The proof does what it says: it takes the Drézet–Le Potier recursion (14), shows it is equivalent to the Springborn mediant rule (1) via identity (15), and then relies on the DLP classification for completeness. I rechecked the algebra around (15); it reduces correctly to the Markov equation using Lemma 2.2, and the dyadic levels are cofinal in the same tree, so the image argument is sound.\n\nThe paper is honest about its debts. It states outright that the proof can essentially be extracted from Rudakov, and the main lift is from [8]. That is not a flaw; it is accurate framing. The direct derivation of (15) is a genuine contribution and makes the correspondence transparent. The note that the DLP function and Springborn's function are two descriptions of the same set is well placed.\n\nSoft spots are minor. The completeness of the DLP classification is external; if that classification had a gap, the main theorem would inherit it, but there is no reason to doubt such a standard result. Proposition 2.1 is terse, especially the step proving integrality of p'_1, though the argument is recoverable. There are a few typos (the duplicated q'_1 in the displayed formula in Lemma 2.2 is one). Section 4, on extensions to real numbers and the question mark function, is interesting but somewhat tangential; the McShane identity argument for Proposition 4.1 is only sketched. The final conjectures reformulate Frobenius unicity in bundle language but do not resolve it, which the paper does not claim.\n\nThe citation pattern looks fine. Self-citation appears only in Section 4 and is appropriate. The paper is a solid, checkable piece of mathematics: the central claim is well supported, the algebra is correct, and the limitations are stated rather than hidden.\n\nThis deserves a serious referee. It is not a breakthrough, but it is a useful unification that will interest both algebraic geometers and number theorists. I would accept it after minor revision, asking for a slightly expanded proof of Proposition 2.1 and cleanup of the typos.","headline":"A clean identification of Springborn's Markov fractions with the exceptional slopes on P2; the core algebra checks out and the main external input is the DLP classification.","tokens_in":10265,"tokens_out":1008,"would_cite":true,"duration_ms":11295,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["14J60","11J06"],"pacs":[],"model":"deepseek-v4-flash","headline":"The slopes of the exceptional vector bundles on the projective plane are exactly the Markov fractions.","keywords":["Markov fractions","exceptional vector bundles","slopes","Markov numbers","Springborn mediant","Drézet–Le Potier classification","Conway topograph","Diophantine approximation"],"falsifier":"Generate the Markov fraction tree to a fixed depth and compute the Drézet–Le Potier function $\\epsilon$ on all dyadic rationals with denominator up to a matching bound; the theorem predicts the two lists coincide, so the first rational appearing in only one list would refute it. Alternatively, exhibiting an exceptional bundle on $\\mathbb{P}^2$ whose slope is not a Markov fraction, or a Markov fraction for which no exceptional bundle exists, would settle the matter directly.","tokens_in":9372,"feed_emoji":"🔢","tokens_out":11800,"duration_ms":100896,"temperature":0.7,"pith_summary":"Springborn's Markov fractions and the slopes of the exceptional vector bundles on the complex projective plane are the same set of rational numbers. The paper proves this by showing that the Drézet–Le Potier recursion, which describes all exceptional slopes as values of a function on dyadic rationals, is exactly the Springborn mediant rule that builds the Markov fraction tree. If correct, the result unifies two independent-looking classifications: one from algebraic geometry (rigid stable bundles on $\\mathbb{P}^2$) and one from Diophantine approximation (rationals with denominators given by Markov numbers). It also gives a simpler proof of Rudakov's theorem that the ranks of these bundles are Markov numbers, and it ties the Frobenius unicity conjecture to a statement about exceptional bundles being determined by their rank.","feed_headline":"Markov fractions equal the slopes of exceptional bundles on P^2","feed_subtitle":"The worst-approximable rationals are the same ones that classify rigid bundles on the projective plane.","key_machinery":"The load-bearing object is the Springborn mediant, which replaces the Farey mediant $(p_1/q_1)\\oplus(p_2/q_2)=(p_1+p_2)/(q_1+q_2)$ with $(p_1q_1+p_2q_2)/(q_1^2+q_2^2)$; iterated from $0/1$ and $1/2$ on the Farey tree, it generates the Markov fractions. The other half of the machinery is the Drézet–Le Potier recursion (14), which defines $\\epsilon$ on dyadic rationals. The proof's key identity (15) shows that, for neighbouring Markov fractions, the Drézet–Le Potier rule and the Springborn rule produce the same rational; the algebra uses the Markov equation $q_1^2+q_2^2+q_3^2=3q_1q_2q_3$ and the tree relation (9) from Lemma 2.2. Once the two recursions are identified, the Drézet–Le Potier classification transfers the equality to the full slope set.","core_discovery":"The central claim is Theorem 3.1: the set $\\mathcal{E}$ of slopes of exceptional vector bundles on $\\mathbb{P}^2$ coincides with the set of all Markov fractions. Exceptional bundles are the stable bundles with $\\operatorname{Ext}^1(E,E)=0$; their slope is the ratio of first Chern class to rank, $\\mu(E)=c_1(E)/r(E)$, and it is known from [8] that the slope determines the bundle. Drézet and Le Potier described $\\mathcal{E}$ as the image of a function $\\epsilon$ defined on dyadic rationals by the recursion (14). The paper's proof shows that this recursion is equivalent to the Springborn mediant rule, so the Markov fraction tree enumerates exactly the exceptional slopes. In particular, the denominators of Markov fractions are the Markov numbers, recovering Rudakov's result on ranks, and every numerator satisfies $p^2+1\\equiv 0\\pmod q$.","pith_inferences":["An implicit consequence is that Diophantine statistics of Markov numbers—prime factors, growth rates, and the density results for prime Markov numbers—become statements about which Chern classes can occur for rigid bundles on $\\mathbb{P}^2$.","The equality also suggests a practical enumeration: the Springborn tree is a much simpler data structure than the dyadic recursion (14), so it could serve as an indexing scheme for exceptional slopes in computational geometry.","The saltus-function picture hints that the limiting Markov irrationalities with Lagrange number $<3$, left open by Springborn, might be realized as limits of slopes along infinite chains of exceptional bundles.","The del Pezzo analogues discussed at the end of the paper suggest a template for other surfaces: match a mutation tree for ranks to a recursion for slopes, and the exceptional-slope set may again be governed by a Markov-type Diophantine equation."],"forward_implications":["The ranks of exceptional bundles on $\\mathbb{P}^2$ are Markov numbers, with a proof simpler than Rudakov's.","The Markov fraction tree is an explicit recursive enumeration of all exceptional slopes, so the Diophantine properties of Markov fractions—being the worst approximable rationals with approximation constants $\\ge 1/3$—automatically describe the geometry of exceptional bundles.","The slope-counting function $\\mu$ is a saltus function: it is exactly the sum of its jumps, and its derivative vanishes almost everywhere, so the set of exceptional slopes has measure-zero structure.","Given the theorem, the Frobenius unicity conjecture for Markov numbers is equivalent to the statement that every exceptional bundle on $\\mathbb{P}^2$ is determined up to the natural $\\mathrm{Aff}_1(\\mathbb{Z})$ action by its rank.","For an exceptional bundle of rank $q$ and first Chern class $p$, one has $p^2+1\\equiv 0\\pmod q$; for prime Markov ranks the congruence has a unique solution up to sign, confirming the unicity conjecture in that case."],"supporting_citations":[{"why":"Provides the Drézet–Le Potier classification of exceptional slopes as values of the function ε on dyadic rationals, which the paper matches to Markov fractions.","marker":"[8]"},{"why":"Introduces Markov fractions, the Springborn mediant rule, and the bijection between rationals in [0,1] and Markov fractions, giving the target set of the theorem.","marker":"[30]"},{"why":"Supplies Rudakov's theorem that ranks of exceptional bundles on P2 are Markov numbers, reproved here as Corollary 3.2.","marker":"[26]"},{"why":"Gives Markov's recursive description of Markov triples via Vieta involutions, used to show denominators of Markov fractions are Markov numbers.","marker":"[20]"},{"why":"Provides the Conway-topograph recursion for Markov numbers used in Lemma 2.2 and in drawing the Markov fraction tree.","marker":"[4, 11]"},{"why":"Supplies standard facts about Markov numbers, including the Fibonacci and Pell identifications of the special branches of the tree.","marker":"[1]"}],"fun_headline_variants":["Markov fractions equal exceptional slopes on the projective plane","Exceptional bundle slopes are precisely Markov fractions","On P^2, exceptional slopes match Markov fractions","Markov fractions are the slopes of rigid bundles on P^2","Simplifying Rudakov: exceptional ranks are Markov numbers"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that the Drézet–Le Potier classification is complete, namely that the recursion (14) really lists every slope of an exceptional bundle on $\\mathbb{P}^2$; the paper shows this list coincides with the Markov fractions but does not re-prove the completeness of the classification.","fun_headline_variants_meta":{"raw":{"variants":["Markov fractions equal exceptional slopes on the projective plane","Exceptional bundle slopes are precisely Markov fractions","On P^2, exceptional slopes match Markov fractions","Markov fractions are the slopes of rigid bundles on P^2","Simplifying Rudakov: exceptional ranks are Markov numbers"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000563,"raw_usage":{"total_tokens":2595,"prompt_tokens":795,"completion_tokens":1800,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":411,"completion_tokens_details":{"reasoning_tokens":1724}},"tokens_in":411,"tokens_out":1800,"duration_ms":14252,"temperature":1.0,"reasoning_tokens":1724,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T20:53:42.289369+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Generate the Markov fraction tree to a fixed depth and compute the Drézet–Le Potier function $\\epsilon$ on all dyadic rationals with denominator up to a matching bound; the theorem predicts the two lists coincide, so the first rational appearing in only one list would refute it. Alternatively, exhibiting an exceptional bundle on $\\mathbb{P}^2$ whose slope is not a Markov fraction, or a Markov fraction for which no exceptional bundle exists, would settle the matter directly.","supporting_citations":[{"cited_title":"Dr` ezet and J","cited_arxiv_id":null,"evidence_quote":"Provides the Drézet–Le Potier classification of exceptional slopes as values of the function ε on dyadic rationals, which the paper matches to Markov fractions."},{"cited_title":"Springborn The worst approximable rational numbers","cited_arxiv_id":null,"evidence_quote":"Introduces Markov fractions, the Springborn mediant rule, and the bijection between rationals in [0,1] and Markov fractions, giving the target set of the theorem."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies Rudakov's theorem that ranks of exceptional bundles on P2 are Markov numbers, reproved here as Corollary 3.2."},{"cited_title":"Markov Sur les formes quadratiques binaires ind´ efinies","cited_arxiv_id":null,"evidence_quote":"Gives Markov's recursive description of Markov triples via Vieta involutions, used to show denominators of Markov fractions are Markov numbers."},{"cited_title":"Aigner Markov’s Theorem and 100 Years of the Uniqueness Conjecture: A Math- ematical Journey from Irrational Numbers to Perfect Matchings","cited_arxiv_id":null,"evidence_quote":"Supplies standard facts about Markov numbers, including the Fibonacci and Pell identifications of the special branches of the tree."}],"review_version":1}