By @josebroxSourceGitHub
Fork
Clone
git clone https://github.com/JoseBrox/diophantine-z-q_improvements.git
Cite
Version link
https://openversions.com/josebrox/diophantine-z-q-improvements/commit/570cac434e7daf15e46853e80a5a7c609892b8da
Version identifier (commit hash)
570cac434e7daf15e46853e80a5a7c609892b8da

Fork to create your own copy with credit to the source. Clone to get a Git copy on your computer.

Earlier version

Checks complete

Canonical expanded TXT artifacts and complete manifest, with BB-oracle index procedure Use Elliot's explicit-record formatter (x1 input, sorted exponent-index monomials, decimal integer coefficients) for exact 163-N1, 179-fixed and 56-B3 quartics, with sizes 60,907 / 66,905 / 11,564 bytes and symbolic BB(t) profile with 60,931 bytes. Keep the source tables and mathematical hypotheses unchanged. Give the rigorous BB-oracle stopping procedure that specifies one successful 163-variable index, while explicitly declining to certify a numerical fixed t without a bound on the weighted trace constant. Add an independent reference-format checker, the full 2,503-file distribution manifest and the README expanded-file-size column. Restore historical citation bytes for all prior verification gates.

Committed

Claims

Findings

There is a general few-variable family of integer polynomials in 25 total variables of total degree ≤87,778,394,112\le 87,778,394,112 whose rational input projection is the integers, with no additional height hypothesis beyond the retained arithmetic premises, using a four-contact scalar and projective two-root encoder with an ineffective sufficient odd exponent.

There is a general quartic family of integer polynomials in 163 total variables of exact degree 4, using 137 quadratic squares and 2,789 expanded monomials, with exact coefficient height H(FN)=max⁡{126616122944180358034867798936,  1494N+2}H(F_N)=\max\{126616122944180358034867798936,\;149^{4N+2}\} for N≥1N\ge 1, whose rational input projection is the integers for a sufficiently large ineffective NN under the retained arithmetic premises.

There is a fully specified general integer polynomial in 27 total variables of total degree ≤335,647,080,448\le 335,647,080,448 with explicit L=143L=143 and coefficient height <2246< 2^{2^{46}} whose rational input projection is the integers under the retained arithmetic premises.

There is a fully specified quartic integer polynomial in 179 total variables of exact degree 4, using 151 quadratic squares and 3,036 monomials, with exact 748-bit coefficient height C02C_0^2 where C0=A0B∗4C_0 = A_0 B_*^4, A0=645904574641049503754777843805872291231896003979A_0 = 645904574641049503754777843805872291231896003979, and B∗=14904184670611231B_* = 14904184670611231, whose rational input projection is the integers under the retained arithmetic premises.

Under the ordinary rational abc conjecture, there is a few-variable family of integer polynomials in 11 total variables of total degree ≤6,361,985\le 6,361,985 using twelve contacts at the first height scale with an ineffective sufficient fixed odd shift.

Under the ordinary rational abc conjecture, there is a quartic family of integer polynomials in 56 total variables of exact degree 4, using 46 quadratic squares and 809 monomials, with exact coefficient height B2+377888002299B^2 + 377888002299 for every positive integer BB, where every sufficiently large fixed odd BB is defining.

There is a ten-variable definition of the rational subfield inside a number field KK under abc over KK, an elliptic curve over Q\mathbb{Q} whose positive rank stays unchanged over KK, and the stated published pole theorem; composing with the eleven-variable rational construction gives 110 variables for the integers inside KK, and 220 for the ring of integers when KK is quadratic.

Methods

Nonlinear contact rechart and direct use of the published pole theorem to reduce the ordinary-abc quartic variable count from 62 to 56.

Source excerpts

“The [v81 continuation](docs/results_20261119_v81.md) further reduces the ordinary-abc quartic through **62→59→58→56 variables**, using a nonlinear contact rechart and the published pole theorem directly.” · Show in paper

Complete ordered witness constructions to reduce general quartics to 163 and 179 variables.

Source excerpts

“The [v82 continuation](docs/results_20261120_v82.md) reduces the general quartics to **163 and 179 variables**, with complete ordered witness constructions.” · Show in paper

Amplifying odd valuations using the cubic rational map (T3−2)/(T3−3)(T^3-2)/(T^3-3) and encoding six algebraic roots with one guarded determinant coordinate to define the rational subfield inside a number field KK.

Source excerpts

“The proof amplifies odd valuations, uses the cubic rational map `(T³−2)/(T³−3)`, and encodes six algebraic roots with one guarded determinant coordinate.” · Show in paper

Computational reconstruction of exact identities, guards, coefficient tables, and finite certificates using Python 3.12, SymPy 1.14.0, and PassageMath PARI 10.8.14 (including exact algebraic rank bounds and PARI descent).

Source excerpts

“Use **Python 3.12**, **SymPy 1.14.0**, and **PassageMath PARI 10.8.14**. The new genus-one certificate uses exact algebraic rank bounds from PARI/GP.” · Show in paper

“The verifiers reconstruct exact identities, guards, coefficient tables and finite certificates in temporary directories. Their receipts identify what was checked.” · Show in paper

Research files

Commit by @josebrox

Code files

Other files

Polynomial presentations of the integers over the rationals

Complete proofs and exact computational specifications for integer polynomials

F(X,Y1,…,Ym)F(X,Y_1,\ldots,Y_m)

whose rational input projection is the integers, under the arithmetic premises stated in each construction.

Every variable count includes the input. Degree means total degree. Coefficient height is the maximum absolute integer coefficient after the fixed parameters have been substituted. We optimize variable count, then degree, then coefficient height. No global minimum is claimed.

Current constructions — 20 November 2026

The reductions 193→165, 229→180, 68→62 and 13→11 come from notes of Gabe Conant with Astra 6.2. They have been audited and integrated here, with further degree, coefficient and hypothesis improvements. The source audit distinguishes those contributions from the subsequent work. The v81 continuation further reduces the ordinary-abc quartic through 62→59→58→56 variables, using a nonlinear contact rechart and the published pole theorem directly. The v82 continuation reduces the general quartics to 163 and 179 variables, with complete ordered witness constructions.

“None” in the height-hypothesis column means no additional conjecture beyond the arithmetic premises explicitly retained in the proof. A complete independent audit of all inherited elliptic, prime-pattern, pole-realization and height foundations has not been established.

ConstructionAdditional height hypothesisTotal variablesDegreeCoefficients and effectivenessExpanded TXT size (kB)
General few-variable familyNone25≤ 87,778,394,112Four-contact scalar and projective two-root encoder; sufficient odd exponent ineffective.Not expanded; <2^848 (L=143 fixture bound)
General quartic familyNone1634, exact137 quadratic squares; 2,789 monomials; defining threshold ineffective.60.907 (N=1 fixture; BB formula)
Fully specified general polynomialNone27≤ 335,647,080,448Explicit L=143; coefficient height < 2^(2^46).Not expanded; <2^965 (L=143 upper bound)
Fully specified quarticNone1794, exact151 quadratic squares; 3,036 monomials; 748-bit exact height.66.905
Few-variable family under abcOrdinary rational abc11≤ 6,361,985Twelve contacts at the first height scale; sufficient fixed odd shift ineffective.Not expanded; <2^227 (B=3 fixture bound)
Quartic family under abcOrdinary rational abc564, exact46 quadratic squares; 809 monomials; sufficient fixed odd shift ineffective.11.564 (B=3 fixture)

The last column gives decimal kB (1 kB = 1,000 bytes) for single-line ASCII, fully expanded TXT files with x1 as the input, x2, ... as other variables, and no final newline. Numeric quartic sizes are measured exactly. The 163-variable N=1 and 56-variable B=3 members are algebraic fixtures, not certified defining members. The other entries are extremely loose upper bounds for the indicated stored circuit member, not measurements or predictions of the first defining member. See fully expanded TXT files and exact size formulas.

The unrestricted-degree bounds describe complete finite determinant specifications; those enormous polynomials have not been fully expanded. All displayed quartics are exactly expanded. Every construction has integer content one, with primitive normalization where needed. Small coefficient fixtures and coefficient-height cutoffs are not sufficient defining parameters.

The results JSON records the exact specifications, source assumptions and parameter qualifications. Weaker fixed-exponent versions of abc and alternative coefficient presentations are retained in the detailed proof guide.

Coefficients and explicit files

The general quartic family has exact coefficient height

H(FN)=max⁡{126616122944180358034867798936,  1494N+2},N≥1.H(F_N)=\max\{126616122944180358034867798936,\;149^{4N+2}\},\qquad N\ge1.

A sufficient defining N exists under the retained premises, but its value is ineffective. The Busy Beaver coefficient specialization gives an exact symbolic specification of all 2,789 expanded monomials: only the coefficients of x78^2*x90, x78*x90 and x78^2 vary, respectively as -6615898*149^(2*BB(t)), -298*149^(2*BB(t)), and 22201*149^(4*BB(t)). For every sufficiently large fixed Busy Beaver index t, the polynomial defines the integers under the inherited premises. The current ineffective threshold does not certify any particular index, including 2^1000. A Busy Beaver–oracle index procedure specifies one working index rigorously, but does not evaluate it to a decimal numeral. The symbolic expanded TXT is 60.931 kB, but its three coefficients are not in decimal. The exact size of the true decimal-expanded file is given in the linked supplement.

The complete 163-variable coefficient table and quadratic equations are supplied.

For the fully specified 179-variable quartic, put C0=A0*B_*^4, where A0=645904574641049503754777843805872291231896003979 and B_*=14904184670611231. Its exact coefficient height is C0², of 748 bits. The expanded coefficient table, quadratic equations, and independent reconstruction are supplied. This reduces the variable count while increasing the height from the preceding 108-bit, 180-variable member; that member and the original presentation from Gabe Conant with Astra 6.2 remain in their dated packages.

The 56-variable conditional quartic has exact height B²+377888002299 for every positive integer B; every sufficiently large fixed odd B is defining under its stated premises. The threshold is ineffective. Its expanded coefficient table, quadratic equations, and independent reconstruction are supplied. The previous coefficient presentations and all their checks remain in their dated packages.

Contributors and citation

This repository belongs to the ℤℚ Diophantine Project, which uses the collective agentic notes from Stefan Baranczuk, Jose Brox, Gabe Conant, Sohail Farhangi, James Freitag, Elliot Glazer, Jennifer Park, and Caroline Terry. The main repository for the project is Elliot Glazer's.

This repository includes updated results and more related work: polynomial witnesses of unrestricted degree but lower number of variables, results conditional on the abc conjecture, etc.

Cite this work as "ℤℚ Diophantine Project contributors. ℤ is Diophantine in ℚ. 2026. https://github.com/elliotglazer/diophantine-z-q". If you want to include any of the related results specific to this repository, please add "https://github.com/JoseBrox/diophantine-z-q_improvements" to the end of the reference.

Source attribution

The new shared-product quartics, positive-factor encodings, weak-abc single-pole construction and related realization notes were supplied in Gabe Conant's notes with Astra 6.2, in H10-quartic-constructions.zip and H10-eleven-variable-constructions.zip. The attribution record identifies the original material and subsequent audit work. The supplied source bytes and checksum lists are preserved.

The earlier 198-variable archive retains its own Gabriel Conant and ChatGPT attribution. The arithmetic inputs include the separately identified Sun, Jaswal and Burungale results and the supplied simultaneous-twist strengthening. Their exact statements, versions, use and review limitations are recorded in arithmetic inputs, the individual proofs and THIRD_PARTY.md.

Original project contributions are under CC BY 4.0. Third-party material retains its rights and attribution. See CONTRIBUTING.md and CITATION.cff.

Number-field results

The Pasten–Vidaux continuation gives a separate ten-variable definition of the rational subfield inside K, under abc over K, an elliptic curve over Q whose positive rank stays unchanged over K, and the stated published pole theorem. The proof amplifies odd valuations, uses the cubic rational map (T³−2)/(T³−3), and encodes six algebraic roots with one guarded determinant coordinate. Zywina's theorem supplies the stable-rank curve for quadratic K. Composing with the eleven-variable rational construction gives 110 variables for the integers inside K, and 220 for the ring of integers when K is quadratic, retaining that construction's arithmetic premises. These are existence bounds over K, with an ineffective correction; they do not change the rational-variable table above.

Open reductions and proved obstructions

The unrestricted-degree counts 25 and 11, and the ordinary-abc quartic count 56, remain unchanged. The exact-square checkpoint isolates the remaining actual-trace requirement and proves further prepared-fibre obstructions. The general quartic improvements retain the shared square coordinate for R²; eliminating it is still open. A proved parity obstruction rules out one proposed full-pair compression in the retained integral Laurent chart; it does not establish a lower bound for arbitrary quartics.

The new elliptic divisibility detector uses five coordinates for a prepared subrelation in which every nonzero integral quotient occurs, under ordinary abc and the published pole realization. It does not identify that quotient with an external rational input. A one-parameter genus-one criterion would give ten coordinates if its actual detector fibers have rational points. Every prepared scalar fiber is locally soluble, but an exact prepared counterexample rules out inferring rational points from those local conditions alone. Neither result is a new definition of the integers.

The complete general and conditional 21/9-variable targets remain unproved. The main unresolved step is global rational realization at the actual elliptic detector values, with every physical lift and denominator condition satisfied. Local solubility, a rational point on an auxiliary cover, and successful finite searches do not establish that step.

The research directions, actual-trace criteria, physical-lift checkpoint, and reference-normalization theorem give precise formulations. Proved obstructions exclude specified constructions and specializations; they do not establish a global lower bound on the number of variables.

For number fields, the definability reductions relate the ring of integers, the rational integers and the rational subfield. No counterexample to Diophantine definability for a number field, or theorem covering every number field, is established here.

Historical results

The historical results JSON and dated proof guides preserve superseded constructions and their full checks. Historical numerical bounds apply to their named sources, rather than the current table.

  • v65, v66 and v67: earlier quartic constructions and point-coordinate audits.
  • v68 and its canonical polynomial package: the preceding fully expanded explicit quartic.
  • v69 and v70: source audits, trace arithmetic and geometric obstructions.
  • v71, v72, v73, v74 and v75: enlarged point charts, reconstructed quartics and even-valuation improvements.

Reproduction

Use Python 3.12, SymPy 1.14.0, and PassageMath PARI 10.8.14. The new genus-one certificate uses exact algebraic rank bounds from PARI/GP. From the repository root:

python -m pip install sympy==1.14.0 passagemath-pari==10.8.14
python -B scripts/verify_v84.py --report /tmp/v84-verification.json
python -B scripts/verify_v83.py --report /tmp/v83-verification.json
python -B scripts/verify_v82.py --report /tmp/v82-verification.json
python -B scripts/verify_v81.py --report /tmp/v81-verification.json
python -B scripts/verify_v80.py --report /tmp/v80-verification.json
python -B scripts/verify_v79.py --report /tmp/v79-verification.json
python -B scripts/verify_v78.py --report /tmp/v78-verification.json
python -B scripts/verify.py --full --receipt /tmp/legacy-verification.json

The full current sequence, including every intervening regression gate, is specified in the CI workflow. The reproducibility guide describes source restoration, coefficient formats and optional arithmetic checks.

The verifiers reconstruct exact identities, guards, coefficient tables and finite certificates in temporary directories. Their receipts identify what was checked. Computational reconstruction does not prove the external arithmetic premises or constitute independent non-author mathematical review. The v82 gate includes the exact PARI descent for its new counterexample; the older optional PARI checks remain opt-in.

The complete TeX manuscripts are the mathematical deliverables. Affected papers are compiled and visually inspected before publication; scripts/update_manifest.py records the distributed file hashes.