# Independent review of the 58-variable ordinary-abc quartic ## Result and scope The new construction has **58 rational variables, including its input, and 50 integral quadratic residuals**. Their sum of squares has degree four, content one, 867 coordinate monomials, and 870 terms when powers of the coefficient parameter are kept separate. For every positive integer $B$, its coefficient height is ```math \max\{22838305425411,\ B^2+4091\}. ``` The defining theorem holds for every sufficiently large odd $B$, with an ineffective sufficient threshold, under ordinary rational $abc$ and the retained elliptic, formal-comparison, all-point trace, and prime-pattern inputs. Native-pole completeness uses the precise square classes in [Burungale, arXiv:2611.16130v1, Theorem B](https://arxiv.org/html/2611.16130v1). It does not assume the class-one extension used explicitly as an additional premise by the historical 59-variable construction. The shared chart and detector originate in notes of **Gabe Conant with Astra 6.2**. The derivative transfer is the subsequent v79 work, and the one-helper normalizer is the v80 improvement. The review has two different parts: exact computation establishes the specified finite identities and coefficients; the argument below checks the logical transfer and its quantified arithmetic hypotheses. Neither part independently proves $abc$ or the cited arithmetic theorems. ## Source maps, complete coefficients, and derivative boundaries Run from the repository root: ```text python -B computation/variable_reductions_v80/abc/check58.py ``` This invocation is read-only. The checker imports no producer or earlier checker. Its physical reconstruction of the unchanged v78 rows is adapted explicitly from the audited v79 checker; that inheritance is recorded, not represented as a new independent discovery. It independently derives the changed equations, their physical interpretation, and the two-curve pole identities. SymPy expands all 50 squares, independently of the producer's sparse convolution, and compares every coefficient with the published table. The checker verifies 47 rows unchanged from v79, and the three new rows ```math \alpha=c^{-1}+D_s^2,\qquad \tau(1+\alpha)=\alpha,\qquad b_s=7\tau^2. ``` The physical values are $D=cD_s$, $u=c^{-1}$, and $b=cb_s$. Thus ```math \alpha=u+u^2D^2,\qquad b=7c\tau^2=\frac7u\left(\frac{\alpha}{1+\alpha}\right)^2. ``` The five native-pole rows retain $e=ce_s$, $u_0=cu_s$, $v_0=cv_s$, $p=p_s$, $q=q_s$, $p^2=cP_s$, and $q^2=cQ_s$. They give exactly ```math e^2+bp^2=1,\qquad u_0v_0=e(1-b),\qquad u_0^2-v_0^2=bq^2. ``` In particular the factor seven belongs to the physical input; there is no additional unstated power of $c$. The polynomial ```math G(t)=t^8-102t^6-144t^5+3009t^4+7200t^3 -17308t^2-32976t+40320 ``` has the eight stated distinct shifts, with zero sum and no opposite pair. The exact identities $P=tG'-8G$ and $Q=(4t^2-102)G'-32tG$ have degree at most six. After recovering the stored powers and value of $G$, the two derivative rows imply ```math t(H-G')=0,\qquad(4t^2-102)(H-G')=0. ``` Their coefficients generate the nonzero constant $-102$, so $H=G'$ also at $t=0$. Neither contact coordinate is divided out. For $B\ge3$, the positive midpoint $x+y=2M^4$ prevents both contact polynomials from vanishing. A single zero makes $F_a$ nonzero because the corresponding root is simple; the quotient equation then excludes that zero. Thus the compiled source recovers $F F_a\ne0$ without a new inverse coordinate. Every monomial and all 58 coordinates occur. The $X^4$ coefficient is one. The only mixed coefficient profiles are $B^2+4091$, $8B+1$, and $2B+1$. The identities ```math B^2+4091-(8B+1)=(B-4)^2+4074, \qquad B^2+4091-(2B+1)=(B-1)^2+4089 ``` prove the asserted maximum for every positive $B$, together with the checked maximum of the fixed coefficients. This algebraic statement does not identify a sufficient defining value of $B$. ## Universal valuation argument and zero cases First suppose $v_p(b)$ is negative and odd. In the first native equation, the valuations of $e^2$ and $bp^2$ have opposite parity. They cannot cancel, so both are integral; the second term is either zero or has positive valuation, and $e$ is a unit. The middle equation forces $u_0,v_0\ne0$ and $v_p(u_0)+v_p(v_0)=v_p(b)$, an odd integer. Their valuations differ. Consequently $u_0^2-v_0^2$ has finite even valuation, whereas $bq^2$ has odd valuation or is zero. This contradiction covers every prime, including two, and every zero-coordinate boundary. Now let $p\ne7$ and let $\mu=v_p(u)$ be odd. Put $d=v_p(D)$, with $v_p(0)=+\infty$. The two summands of $\alpha$ have respective valuations $\mu$ and $2\mu+2d$, of opposite parity. Therefore ```math A:=v_p(\alpha)=\min\{\mu,2\mu+2d\}. ``` If $\mu<0$, then $A<0$, so $v_p(\tau)=0$ and $v_p(b)=-\mu>0$. This is uniform in all actual point denominators, including denominator primes discovered only after prime selection. If $\mu>0$, both $A<0$ and $A=0$ would give a negative odd valuation of $b$. In the latter case an arbitrarily deep cancellation in $1+\alpha$ only makes $v_p(\tau)$ smaller. Thus $A>0$ and the native obstruction forces $-\mu+2A>0$. It follows that ```math A>\mu/2,\qquad d>-3\mu/4. ``` These are universal case arguments. The finite valuation grid in the checker is supplementary regression evidence only. It explicitly includes the $A=0$ denominator-cancellation cases. The gate itself excludes $\alpha=-1$, since its residual would equal one. The branch $D=0$ is retained and gives $\tau=u/(1+u)$ and $b=7c/(1+c)^2$. The abstract branch $\alpha=0$ is not removed by a fictitious guard: it would imply $u=-(uD)^2$. For sufficiently large $B$, the derivative-height estimate, before using any pole conclusion, supplies a positive odd valuation of $u$ and excludes this branch. No denominator of the new normalizer depends on that eventual argument. ## Detector extraction and uniform heights The cleared equation $D\operatorname{Tr}W=\operatorname{Nm}(W-1)$ forces $\operatorname{Tr}W\ne0$, since otherwise $W=1$ in the field $\mathbb Q(\sqrt2)$ while its trace is zero. If both slot valuations of $W$ are negative, the norm and trace inequalities give $v_p(D)\le\max(v_1(W),v_2(W))$. Otherwise a slot is already integral. There is therefore a slot with $v(W)>-3\mu/4$. The argument also covers an inert prime, where the two slots have the same valuation. The four nonconstant detector terms have distinct classes modulo four. This uses the even valuations of the reciprocal-abscissa products $J,K'$. A negative minimum is unique, so every term has valuation greater than $-3\mu/4$. Hence ```math v(C)>\mu/16,\quad v(J)>5\mu/8,\quad v(\overline I)>9\mu/16,\quad v(\overline{K'})>13\mu/8. ``` Zero tests satisfy these inequalities with infinite valuation. The exact parent extraction theorem applies, including tangent fibers, zero ordinates, and finite two-torsion differences, and gives local integrality of $X$ and $X(m-X)\equiv0\pmod p$. The v63 derivative proof gives the intermediate estimate in $\max(h(x),h(y))$ before replacing that height by a bound involving $xy$. It remains valid when one slot is zero: one may choose the larger height slot first, and $G(0)\ne0$. Its multiple-contact estimate uses the squarefree homogeneous form $AB\prod_j(2q_jA^4+p_jB^4)$ and loses $(2+\epsilon)h(M)$, not a factor four more. The polynomial and its constants depend on the fixed shifts, not on $B$. Together with $x+y=2M^4$, $\epsilon=1/100$ gives $\log r^+(u)\ge(193/100)h(M)-O(1)$. The unchanged signed mixed-power bound gives $h(X)\le(501/997)h(M)+O(1)$. The strict margin is $92221/99700>0$. The all-point trace-index bound in v65, including its separate treatment of the difference $(0,0)$, then bounds $m$ whenever $X\ne0,m$. This contradicts $h(M)\ge\log B$ for sufficiently large $B$. Excluding seven changes only the fixed exceptional set and constants. ## Exact prime-pattern and CRT quantifiers The required prime statement is explicitly available for every fixed finite list in [Sun, arXiv:2610.20691v3, Theorem 4.1](https://arxiv.org/html/2610.20691v3). It permits distinct integral coefficients, prescribed divisibility of the square-denominator parameter, and arbitrary fixed lower bounds for the primes and slope. Its Gaussian ray conditions and cross-order coprimality are the ones used by the retained parent. No bound uniform in the number of coefficients or their sizes is required here. The logical order of the completeness choices is as follows. 1. Fix the integer input $n\ne0$, its root difference $[n]R_0$, and the resulting scaled trace, $M$, and $a=M^4$. 2. Choose bounded real and finite-place neighborhoods of the required point functions, avoiding their poles. Fix the associated congruence progression for the shared index, as in v65. Enlarge the temporary preparation set by every coefficient, denominator, and collision prime of the now-fixed contact positions, and by the numerator and denominator primes of $a$. 3. Choose $D_0$ clearing the sixteen coefficients $\pm D_0^2(a+r)$, and prescribe enough denominator depth and a real slope bound for all these fixed neighborhoods. The coefficients are distinct because $a$ exceeds the fixed contact diameter. 4. Apply the prime theorem once to this fixed list and preparation data. Refine the shared elliptic index on the compatible CRT progression, retaining the chosen finite-place neighborhoods and real component. The arithmetic-dependencies paper v8, Section 2, states the relevant deductions explicitly. Dividing the two reduction orders by four gives coprime cross-sign annihilation moduli. Their odd parts avoid the fixed progression modulus; the only remaining compatibility is the existing power-of-two condition on the chosen index. At seven the prescribed Gaussian residue gives both adjacent norms nonzero modulo seven, even though seven is omitted from the theorem's odd avoidance modulus. The parent halving lemma puts $R_0$ in the doubled reduction subgroups, so these divided orders are valid annihilators. Finally the compact-group progression lemma retains real returns after the compatible finite congruences are imposed. This is the parent deduction, not an extra approximation assumption for the new detector. With $z=N/(D_0^2L^2)$ the sixteen distinct prime values give ```math F=\frac{\prod_r (N-D_0^2(a+r)L^2)(N+D_0^2(a+r)L^2)}{D_0^{32}L^{32}}. ``` At a selected prime exactly one factor vanishes simply, so $v_p(u)=1$. The successful slot has the unchanged error depths at least three and reciprocal-product depths at least four. At the other slot the last detector term has the unique minimum $-4$. Thus $v_p(D)>0$ and $v_p(\alpha)=v_p(\tau)=v_p(b)=1$. At an unselected prime outside the temporary set, an integral slope makes $F$ a unit and $F_a$ integral, hence $v_p(u)\le0$. A nonintegral slope gives $v_p(u)=2v_p(z)$ from the leading terms $F=z^{16}+O(z^{14})$ and $F_a=-16az^{14}+O(z^{12})$. At the temporary fixed places, sufficiently deep denominators give $v_p(u)=2v_p(z)-v_p(a)$, which is even since $a=M^4$. This accounts for every positive odd valuation of $u$; no unspecified auxiliary denominator is assumed to have been fixed in advance. The one-helper lemma handles every negative odd valuation outside seven, and even valuations of $u$ give even valuations of $b$ there. At fixed places, bounded point neighborhoods give $D=O(c^2)$. Writing $D=c^2E$ with bounded $E$ gives the exact formula ```math \tau=\frac{1+c^3E^2}{1+c+c^3E^2}\longrightarrow1. ``` At two, $c/(a/z^2)$ may be made to lie in $1+8\mathbb Z_2$ and $v_2(c)$ arbitrarily large positive even; hence $b\in7\mathbb Q_2^{\times2}$ and $v_2(b)\ge4$. At seven, $v_7(b)=1+v_7(c)>0$. All negative valuations of $b$ are therefore even. Real positivity and smallness follow already from $u>0$, $\alpha\ge u$, and $0<\tau<1$, giving $0