# Sun v1 and the 214-variable proposal: mathematical audit Date: 13 November 2026. Author review and new derivations, not a non-author acceptance receipt. ## Decision and exact scope No essential counterexample or invalid implication was found in the portions reconstructed below. This is **not complete acceptance of either integer-definition theorem**. Sun's main reduction has been examined section by section, but its imported arithmetic foundations have not all been recursively audited. For PR30, the selected construction, changed local argument, uniform-height argument and product-height averaging application have been examined; the entire 703-page assembly and every imported source have not been checked. Unread dependencies are review limitations, not assertions that they are false. Only JoseBrox/diophantine-z-q_improvements is modified by this work. The branch is based on fa09c9c6c97982c13265e7088d95cdb61b3e90cf. No issue, comment, review, branch or pull request was created in Elliot's repository. Existing main-branch numerical records and verification certificates are unchanged. ## Sources and version boundaries Sun: Zhi-Wei Sun, *Z is Diophantine over Q with 28 unknowns*, arXiv:2611.14605v1, submitted 10 November 2026. The versioned HTML, main mathematical sections and the indicated PDF pages were inspected. Source: https://arxiv.org/html/2611.14605v1 . This is a preprint; no journal acceptance is inferred. PR30: elliotglazer/diophantine-z-q, head c84bbfa4dbe8f32fd3bf29df029f7b7849d36202, revision 3 of the receiving manuscript. The reviewed source arguments are: | File | Git blob SHA | Scope | |---|---|---| | research/quartic214/PROOF.md | 64d2de534529bfe735dbb94a446c968dbb6cc9a9 | Complete assembly note read; its dependency claims are not themselves proof certificates. | | research/quartic214/proofs/uniform-existence.md | 05f8f589be24ae5fbbcd38a97e49a2a470db12c8 | P01-P20 read; the principal arithmetic transitions are assessed below. | | research/quartic214/proofs/height-averaging.md | 2279774ec08274150486f5868c7566725e64ccc3 | Complete application read; multiplicity, base field, normalization and isogeny direction checked at the stated-premise level. | The PR metadata says 214 total coordinates, 188 quadratic rows, 2919 collected monomials and an ineffective defining threshold. Those complete coefficient tables were **not independently reconstructed in this audit**. Its condition N>=53 concerns coefficient height, not a numerically certified defining member. Revision 3 supersedes revision 2 as the receiving edition; older review decisions are not silently rebound to it. The product construction used in the averaging argument was directly checked in the authors' September 2022 version of Gaudron--Remond, *Nouveaux theoremes d'isogenie*, Lemma 16.2(1), printed p.104, text and page image. It constructs a product over the fixed field, allows repeated factors, and does not increase stable height. The 2023 publication is Memoires SMF 176, DOI 10.24033/msmf.484. This audit does **not** claim a byte-for-byte or page-by-page comparison with the final published edition. ## Sun: section-by-section assessment | Passage | Checks and disposition | |---|---| | Section 1, theorem and counting convention | The 28 variables are existential unknowns, excluding the input: 29 total coordinates in this project's convention. It is an existence construction with a fixed multiplier, not a displayed numerical coefficient table. This does not improve our 25-total-coordinate record. | | Section 2.1, normalized curve and transported inputs | The change of coordinates scales x,y by the second and third powers of a fixed unit. Rank, group structure and the exponent-24 contact-height theorem are retained inputs. Their full original proofs are not newly verified here. | | Lemma 2.1, odd-index certificate | Reconstructed the squareclass argument, independence of the three classes by norms and real signs, and the half-point equations. The norm/sign argument requires the stated group decomposition. Nonzero denominators follow before duplication. The choices are finite algebraic choices, not additional continuous parameters. | | Lemma 2.2 and Corollary 2.4 | Supplied a new self-contained proof of a strictly stronger result: product-height equality is equivalent to equality of the indices' 2-adic valuations, including positive integer weights. The three-scale consequence extends uniformly to a common valuation. No rank theorem is used for this auxiliary result. | | Section 2.3, pole predicate | Checked the distinction between two free structural parameters and finite square-root choices, the nonzero guards, and the direction in which the negative-odd-valuation obstruction and prepared converse are consumed. The original restricted-converse/descent proof is an outstanding recursive source check. | | Section 2.4, Gaussian prime patterns | Checked that contact coefficients, denominator clearing, avoidance and divisibility data are fixed before invoking the theorem. Repeated contact positions must be merged with all occurrences retained. The analytic prime-pattern proof and all its quantitative suppliers are not re-proved here. | | Section 2.5, relation combining | The nonzero square inputs and separate zero branches are essential. The final construction also uses a separating finite algebra, not just a characteristic polynomial with arbitrary rational roots. | | Section 3, sixteen tests | Reconstructed the two-anchor congruence. Integrality of one exact ratio comes first; positive-order comparison gives integrality of the other, and only then may relative logarithmic precision be used as absolute precision. Identity numerator points are allowed. The anchor difference is a unit at consumed primes. | | Lemma 4.1, contact change | The fractional linear map and normalized primitive linear-form valuations give the same odd radical away from a fixed set. The pole used for normalization is not a contact. Enlarging the exceptional set changes a constant, not the power. | | Lemma 4.2, ordinary norm and trace | Valuation parity prevents cancellation. A maximizing place in the trace bound supplies all tests at one place. Divisibilities are separate for each detector; multiplying radicals across detectors is not justified or needed. The square scale factor is essential with an odd number of contacts. | | Lemma 5.1, scalar certificate | Reconstructed every step in the quadratic norm tree, including zero leaves. The coefficients of the seven strict leaf bounds are 1/16,9/16,5/16,13/16,3/8,7/16,15/16. The formal ratio implies divisibility of s-1 without assuming the comparison index is a unit. | | Lemma 6.1, real preparation | Checked the required order of choices: fixed real bounds precede congruence moduli. The nonempty regular part of each relevant real orbit component is a substantive imported compactness/chart assertion; full recursive verification of that supplier remains open here. | | Lemma 6.2, finite preparation | Exact valuations of the reference index and deeper scalar index are compatible with later choices. Coefficientwise divisibility at the ramified prime 2 is required; divisibility merely by a local maximal ideal would be insufficient. | | Lemmas 6.3-6.4, CRT and root labels | Checked the division of selected primes into conjugate reduction orders and the compatibility with prescribed exact local valuations. Prime data are fixed after the root labels and scales, not before. Completeness uses odd nonzero labels and the Lucas construction, not arbitrary rational labels. The imported prime theorem remains a premise. | | Proposition 6.5, joint completeness | Reviewed selected-prime trace residues, other-prime parity, preparation at the exceptional set, and the uniform real smallness bound. Both denominators and nonzero normalizers have explicit reasons. No existence proof from local solvability alone was substituted for the pole converse. | | Lemmas 7.1-7.2 and Propositions 7.3-7.4 | Reconstructed the strict height comparisons and the denominator argument for beta=2X^2+3. The multiplier must be fixed after obtaining constants uniform in the multiplier. New proof gives orders 49,13,25 instead of 64,16,48. | | Lemmas 8.1-8.2, point and group-law charts | Checked the F1!=0 quartic and all three F1=0 branches, the distinction between finite roots and free coordinates, and the need for identity/torsion branches in the finite atlas. Separability is imposed on the actual specialization, not only generically. | | Lemma 8.3, finite algebra | Reconstructed separation before gcd filtering, the finite list of separating linear combinations, and componentwise invertibility of discriminants. A rational separating root determines a rational tuple by Galois invariance. Dependent roots are adjoined in a tower. The bounded-degree branching procedure must retain all degree drops and zero branches. | | Equations (8.2)-(8.3), guards | Direct substitution gives Psi=(D^2 G W)^d Pi((D^2 G W)^(-1)). At DG=0 or W=0, Psi=1, so no spurious guarded solution occurs. Translating by 0 or 1 supplies a nonzero root. Sums of rational squares encode conjunction and products encode finite disjunction with one shared final variable. | | Lemma 9.1 and Remark 9.2 | Division-polynomial recurrences and the ordinate formula were checked against exact repeated addition for m=1,...,12. This is finite implementation evidence. The universal identities are the classical division-polynomial identities; the domain is nontorsion, so denominators do not vanish. An unspecified sufficiently large M is not a numerical coefficient file. | | Proposition 10.1 | The all-specializations qualifier is essential and is supplied by the preceding finite stratification, not by generic separability. The conclusion remains relative to the arithmetic inputs used in soundness and completeness. | ### Findings and repairs **S1 — No essential flaw established in the reconstructed reduction.** This is a scoped outcome, not a complete verification of the underlying arithmetic theorem. A missing recursive source check is recorded as such, not classified as a mathematical disproof. **S2 — Bibliographic version ambiguity.** Reference [10] gives an unversioned arXiv source whose successive versions have different titles, counts and numbering. The theorem applications should be bound to a precise version and exact statements. This is a repairable citation/application issue unless comparing the intended source reveals a genuinely weaker theorem. We have not inferred such a failure merely from the version ambiguity. **S3 — Potential misuse of the product-height statement is ruled out.** It is false for unrestricted nonzero indices. The pair P,2P has exact defect log(41). Sun imposes oddness, so this is not a counterexample to his lemma or theorem. The new exact criterion identifies the full valid index range. **S4 — Precision is unnecessarily high for the final inequalities.** Under the same inputs, orders 49,13,25 suffice. The full proof, including the local and real completeness checks for the changed normalizers, is in the precision paper. The degree-64 logarithm is retained. No numerical degree or coefficient improvement is claimed without compiling the changed source. ## PR30: the changed mathematical arguments ### Algebra and local valuations (P03-P07) The source uses beta=51x^2+2504, c=|a|, the reference curve over Q(sqrt(2)), and the quarter-trace of a **sum** of four abscissas. These are not Sun's beta, scalar index or product scale. The q-inverse equations remove q=0 and do not claim a bijection on that discarded stratum. The argument t=1, where q=-1, remains allowed; contact points are excluded algebraically before division. For every positive odd contact depth nu, the generic factor terms have disjoint valuation classes: odd, 2 modulo 4, and 4 modulo 8. In the special factor the two odd classes are -nu and nu modulo 4. Zero tests are harmless. The exact checker verifies all residue classes for odd nu modulo 8, which proves the congruence classification; arbitrary valuations with those residues are covered by the written minimum argument. The trace/pole inequality gives ell>-nu/2 at a maximizing place. It supplies order at least three for the deep comparison and Lucas tests, positive order for the input test, and positive even reciprocal-abscissa valuations. Integrality precedes the formal-logarithm congruences. Half-integral reference indices cause no division by 2 problem at consumed odd primes. The resulting M_j^3|Delta and M_j|J are separate divisibilities, without an extra scalar factor or assumed coprimality between scales. ### Aggregate height and uniform soundness (P08-P12) The sum argument is not invalidated by the mixed-valuation product counterexample. The proposed geometric proof excludes constant translates using positive same-block pole coefficients and independence of the two conjugate generating points. The passage from this condition to the published family-height theorem, its exact embedding hypothesis and all source hypotheses still requires a complete primary-source review before whole-theorem acceptance; reading the application note is not a substitute. At the stated-premise level, the exponent calculation and quantifiers are correct. The full-family supplier gives alpha=2103/1024. Two disjoint blocks give alpha/2. The published note weakens alpha to 7/3, obtaining Lucas exponent 14/9. Keeping alpha gives 701/512, and the input-height coefficient improves from 7/5 to 2103/1993. The sufficient-threshold power becomes 512/323 instead of 9/4. These are proved algebraic consequences of the same supplier, not new verification of the supplier. They neither change the 214 coordinates nor make the threshold effective. ### Completeness and the base field (P13-P20) The scalar real estimate is uniform in the integer input because r4>=1 and R_s>=A_N/(1+149^2). Later point labels, moduli, primes and witnesses may depend on N and x; this does not alter the quantifier order exists N0, for every N>=N0, for every x. The changed factors have the stated positive depths at selected primes. Elsewhere the even number of contacts makes each scale's contribution a square, including nonsquare scales. The real q>0 witnesses satisfy the inverse-weight equations. The pole realization consumes the stated non-CM family with j=35152/9, not the CM curve used for integer labels. Full Selmer corank is needed before any finite-Shafarevich conclusion. This audit does not certify the recursive restricted-converse, Kato or descent foundations. Those are the main remaining source obligations, together with the Gaussian prime-selection and modular-height chains. ### Height averaging The application selects r separate occurrences of the A-isogeny class over the actual field L', not merely the set of distinct isogeny classes and not an auxiliary evaluation field. Height is additive on the actual product. The small isogeny is then applied to a dimension-two factor in the direction factor-to-A, retaining its height cost. Negative heights do not invalidate the averaging. A useful sharpening is h_F(C)<=(h_F(J)+3g/2)/r-3. It follows by retaining the dimension g-2r of the complementary product. This is an elementary improvement of the average, not a substitute for the existence of the fixed-field product or the proof of multiplicity. The companion precision paper states the general lemma with all hypotheses. ### Remaining PR30 checks Not completed here: a source-first inventory of all 703 pages; recursive proofs of rank, restricted converse, Kato, Gaussian transference/selection, geometric cover, normalized modularity, conductor and integral metric/height suppliers; final-publisher comparison for every theorem; reconstruction of all 188 quadratic rows and 2919 coefficients; review of every historical objection and its current incorporation; current CI and the protected independent-review decision. No favorable older review has been represented as our independent acceptance. ## New files, validation and next mathematical step The product-height paper proves a universal theorem independent of the inherited definability premises. The precision paper proves explicit conditional implications and the changed-order application. The Python checker uses only exact standard-library arithmetic; its certificate records 25 completed stages and clearly labels bounded tests. This is not the earlier 126-stage repository suite, which was not rerun in this environment. The papers were compiled twice and their rendered pages inspected. New files have their own manifest; the main release manifest and numerical table are untouched. The next necessary audit is to bind the exact prime-pattern and pole-realization suppliers to every consuming hypothesis, including merged contact incidences and the full Selmer-corank condition. Only after those and the remaining height-source checks are closed should either whole definition be promoted as independently audited. For optimization, compile the reduced-order rational source and measure its actual elimination degree before announcing a numerical gain.