By @josebrox

613b16a / docs/results_20261120_v82.md

Raw
# Scalar-point sharing, elliptic divisibility, and actual rational realization

This continuation proves **163-variable general quartics** and a **fully specified 179-variable quartic**, under the arithmetic premises identified in their source papers. Every count includes the input. The unrestricted-degree bounds **25 and 11** and the ordinary-abc quartic bound **56** remain unchanged.

The inherited constructions are based on **notes of Gabe Conant with Astra 6.2**. The new partial-trace scalar, scaled scalar-point chart, weighted aggregate, elliptic divisibility detector, and genus-one analysis are subsequent project work. Original source bytes, attribution, historical records, and review hashes are preserved.

## Complete quartic constructions

The [manuscript](../papers/quartic/scalar_point163_179_20261120_v82.tex) gives the local argument, inverse maps, ordered integer witnesses, and the two global height arguments. The [coefficient package](../computation/variable_reductions_v82/general/README.md) contains every physical row, every coordinate map, every expanded coefficient, and an independent checker.

| Construction | Total coordinates | Quadratic equations | Monomials | Degree | Content |
|---|---:|---:|---:|---:|---:|
| General family | 163 | 137 | 2789 | 4 | 1 |
| Fully specified member | 179 | 151 | 3036 | 4 | 1 |

### Partial trace and a smaller point chart

In the retained field $K=\mathbb Q(\sqrt2)$, write $f=1/x(H)$ and use the positive scalar $R$. The new equations are

$$
J=\frac{Z^2}{R^5}+\frac{f^2}{R^4},\qquad
\tau\operatorname{Tr}(Z^2/R^5)=N_{K/\mathbb Q}(J),\qquad
b_s=R(R+9792)\tau^2.
$$

At a consumed odd scalar valuation $\mu=v_p(R)>0$, the native pole implies, in one common place above $p$,

$$
v(Z)>\frac94\mu,\qquad v(f)>\frac74\mu.
$$

Positivity excludes both $Z=0$ and a zero partial trace. With $\widehat J=R^4J$, $L=\widehat J-f^2$, and $\widehat\tau=R^4\tau$, the scalar equations become the quadrics

$$
RL=Z^2,\qquad \widehat\tau\operatorname{Tr}(L)=N(\widehat J).
$$

Scaling the scalar point chart by $Z$ gives the linear identity $M=-2S-W$. Both rational components of $M$ can therefore be deleted. The inverse point map is proved on every rational solution; the primary denominator is forced nonzero, whereas the alternate zero-ordinate branch remains allowed.

In the general family, the aggregate uses $R x(Q_2)$ instead of $x(Q_2)$. The retained coordinate $P^*$ already equals this product, so no reciprocal of $R$ is needed. Four actual input rows force

$$
\rho=r(X)>0,\qquad R(1+149^2\rho)=149C^2\rho.
$$

This establishes $R>0$ before using any Laurent inverse and excludes an extra component at $R=0$. The fixed member retains its unweighted aggregate and one inverse coordinate. The net changes are consequently $165\to163$ and $180\to179$.

**The shared $R^2$ coordinate remains.** The native relation $\widehat\tau\sigma=R^4$ is represented by $\widehat\tau\sigma=R_2^2$, $R_2=R^2$. This construction does not claim the requested deletion of that register.

### Comparison precision and soundness

The scaled comparison is

$$
E_2^*=A\overline{(V+M)}+2\overline{T_H}\,\overline Z
      =-\overline{Zt_2}\,C_2.
$$

Its fourth-power term receives exponent 11; the companion term retains exponent 5. The scalar label $a$ compensates for the potentially nonintegral multiplier. The proof establishes

$$
v_p\bigl(a(A-a)\bigr)\ge3,\qquad M_j^3\mid a\Delta,
$$

before transferring integrality to the other comparisons. Since $|a\Delta|\le14B_0^5$, the retained radical-height argument gives exponent $35/18<2$.

For 163 variables, the [positive weighted trace theorem](../papers/trace_heights/weighted_trace_20261114_v27.tex), applied to fixed weights $(1,r,1,1)$, gives

$$
B_0^2\le A\bigl(h(R)+h(T)\bigr)+J.
$$

The constants are independent of the fixed family parameter $N$. The underlying published result is [von Buhren, Theorem 1.1, p. 308](https://www.numdam.org/article/JTNB_2017__29_1_307_0.pdf); the repository proof checks its projective embedding and exceptional-translate hypotheses. Its additive constant is ineffective.

Set $L_N=\lfloor9(2N-1)/4\rfloor+1$. The scalar forces $B_0\ge|a|\ge149^{L_N}$ and $h(\beta)\le K_0 149^{-L_N}|a|$. Together with

$$
h(R)\le4h(\beta)+(2N+1)\log149+\log1801500,
$$

this makes the extra scalar-height term uniformly $o(B_0^2)$. Fix the height constants first, absorb that term, use $35/18<2$ to bound $B_0$ independently of $N$, and only then enlarge $N$. The input argument and Lucas separation finish the proof. No effective threshold is asserted.

### Ordered global witnesses

The manuscript fixes the input and scalar, then real neighborhoods and the finite support of odd valuations of $R(R+9792)$. The reference is chosen sufficiently deep at those places before the scalar label and Lucas roots are fixed. Exact formal depths and the real bound prepare the scalar pole. Prime selection occurs only after all root coordinates and avoidance primes are known. The synchronized reference refinement retains those earlier conditions and supplies the changed comparison precision already at depth one.

The denominator check includes newly introduced poles of $\tau$: outside the prepared support, $v_p(R(R+9792))+2v_p(\tau)$ is even. The global witness argument therefore does not stop at finitely many local approximations. It finishes with the retained native-pole realization theorem and explicitly recovers every physical helper.

### Coefficients and the fixed endpoint

For $C=149^N$, every $N\ge1$ has exact coefficient height

$$
\max\{126616122944180358034867798936,149^{4N+2}\}.
$$

The fully specified 179-variable member uses

$$
A_0=645904574641049503754777843805872291231896003979,
\quad B_*=14904184670611231,\quad C_0=A_0B_*^4.
$$

The fixed scalar gives $A_0^3B_*^9\mid a$, hence $|a|\ge2^{960}$. Exact inequalities close the effective endpoint using the retained unweighted height theorem. Its coefficient height is exactly $C_0^2$, of 748 bits. The lower variable count takes priority over the larger coefficient height. The preceding 180-variable, 108-bit member is preserved as a historical construction.

## A new elliptic divisibility detector

The [complete proof](../papers/elliptic_divisibility/prepared_eds_divisibility_20261120_v82.tex) and [exact checks](../computation/variable_reductions_v82/eds/README.md) use

$$
E:y^2=x^3-2x,\qquad
b(H,U)=\frac{y(H)+x(U)^2(1+x(U)^2)}{(1+x(U)^2)^2}.
$$

For $U=[m]H$, negative valuations of this expression are even outside a fixed finite set. At a primitive denominator prime where $H$ has odd formal depth and $U$ is nonformal, its valuation is negative and odd. Ordinary rational abc supplies an eventual primitive exponent-one divisor; the manuscript also gives a direct proof with positive mass margin $77/303>1/5$, following the denominator argument in [Cornelissen–Zahidi, Proposition 3.10(iii)](https://arxiv.org/pdf/math/0412473).

For every nonzero integral quotient, an ordered real and finite-place choice of $H$ prepares $d=(112/841)b$ for the published intrinsic pole theorem. Two one-coordinate point charts, two pole parameters, and a rank-eight finite-algebra encoder give **five coordinates for this prepared divisibility subrelation**. The [LMFDB record 256.b1](https://www.lmfdb.org/EllipticCurve/Q/256/b/1) supplies the recorded generator assertion; exact PARI checks independently recover rank and torsion.

This is not a five-variable definition of the integers. It need not accept every preassigned divisible pair, and it does not identify an external input $X$ with $j/h$. That identification is the remaining constructive task.

## A sharper ten-coordinate criterion and a global obstruction

The [genus-one manuscript](../papers/pole_parameters/genus_one_sound_criterion_20261120_v82.tex) considers the actual source functions

$$
d=\frac{7u}{(1+u)^2},\qquad H=V^2+d,
\qquad Y^2=d(s^4-H^2).
$$

At a positive odd contact $\mu=v_p(u)$, this equation forces $2v_p(V)<\mu$, exactly the scalar inequality needed by the existing soundness proof. A non-torsion rational point at an actual prepared detector tuple for each integer would replace two free pole parameters by one and give ten total coordinates. **That actual-input rational-point assertion remains unproved.**

A sharper sufficient target is now proved: for one fixed admissible source constant, it is enough to realize these actual fibers at all but finitely many inputs $cm^2+b$, with fixed positive squarefree $c$ and $b\in\mathbb Z$. Substitute $T=cX^2+b$ in the sound ten-coordinate polynomial. If $X=a/q$ is reduced and $T$ is integral, then $q^2\mid c$, so $q=1$. Finitely many missing integer inputs are restored by multiplying their linear factors. This preserves the ten-coordinate count. No such progression is yet known to have the required actual rational witnesses.

Every prepared scalar fiber is nevertheless locally soluble at every place. The proof covers negative and positive valuations, the prime two, and the real place. This stronger local conclusion is still insufficient for global realization.

An exact counterexample uses

$$
u=\frac{6137}{256},\quad
d=\frac{10997504}{40870449},\quad
V=\frac{2698}{2131},\quad
e=\frac{4d}{V^2}=\frac{30464}{45369}.
$$

These values have all stated real, denominator, and two-adic preparation properties, with nonsquare $u$. Scaling identifies the associated elliptic curve with $E_{6305215}$. Certified algebraic descent gives rank exactly one. Its full rational Kummer image forces equal valuation parity at 17 and 10597 for every nonzero shifted abscissa $x,x-N,x+N$. The required square class 119 has unequal parity. The obstruction survives every constant multiplier supported on $\{2,3,5,7\}$ and every power of $N$.

The [read-only checker](../computation/variable_reductions_v82/abc/check_genus_one.py) reconstructs the three curve maps, both isogenous algebraic-rank bounds, all eight Kummer classes, and the exact prepared tuple. It uses PARI's algebraic bounds, not an analytic-rank or BSD inference. The second descent uses the same PARI implementation and is not presented as an independent implementation. The counterexample is **not asserted to occur at the actual contact detector or actual trace**.

The [separate independent review](../computation/variable_reductions_v82/abc/independent_review/proof_review.md) reconstructs the Kummer image using actual group additions and checks the entire 64-class target family. It also verifies the quadratic-progression corollary. The [common-level scope note](../computation/variable_reductions_v82/abc/common_level_scope.md) proves that one rank-one congruent-number curve of level $n\equiv5\pmod8$ cannot provide two distinct positive nonsquare abscissa classes both prepared in one fixed 2-adic class $1$ or $7$. This excludes that naive way of sharing the general construction’s two pole levels; sharing a different parameter remains open.

## Verification and remaining work

The v82 gate independently reconstructs both quartics, rebuilds the producer outputs in a temporary directory, checks the elliptic fixtures and exact genus-one descent, and binds the proof-review and TeX-review receipts. The corrected genus-one reference and new corollary have a separate followup receipt; exact reverse edits reconstruct the initially reviewed source without changing its historical hashes. Historical arithmetic, coefficient checks, and review hashes remain active in the full CI sequence.

The next substantial targets are actual rational realization for the ten-coordinate criterion or another pole compression; identification of the external input in the new elliptic divisibility route; removal of the shared $R^2$ register with a nonlinear reference chart; and a further contact rechart below 56. No result here proves a new 25/11 bound, a universal lower bound, or a complete independent audit of all inherited arithmetic foundations.