# Independent review of the 59-coordinate ordinary-abc quartic ## Conclusion and scope The normalized construction in `build59.py` has an exact quadratic compilation in 59 rational coordinates, including the input. Its 51 quadratic residuals have a sum of squares of degree four, integer content one, and 873 coordinate monomials. The coefficient height for every positive integer $B$ is ```math \max\{22838305425411, B^2+4092\}. ``` The transfer from the retained 62-coordinate construction is valid under ordinary rational $abc$ and the elliptic, formal comparison, trace-index, simultaneous prime-pattern, and simultaneous-twist premises explicitly retained in the subject paper. The proof gives every sufficiently large odd $B$, with an ineffective sufficient threshold. The coefficient height formula does not provide that threshold. The translated-secant chart, shared detector and Laurent packing originate in notes of **Gabe Conant with Astra 6.2**, audited in v78. The single derivative quotient, derivative interpolation and ratio normalizer are the subsequent v79 transfer. This review checks the new transfer and its edge cases. It does not replace the retained arithmetic premises by numerical evidence. The independent checker certifies the exact algebra described below; the universal valuation and existence arguments are mathematical proofs. ## Exact algebra and source equivalence Run, from the repository root: ```text python -B computation/variable_reductions_v79/abc/check59.py ``` The default invocation is read-only. It imports neither a builder nor an earlier checker. Its dependencies are the pinned parent `computation/coauthor_audit_v78/work_abc/small_gate62/primitive/quadrics.json`, the four files in `quartic59`, and SymPy. It compares the result with `independent59.json`. The optional `--write` regenerates that certificate only after an intentional review. The check has the following distinct parts. 1. It reconstructs the 31 nonunit retained physical equations directly in $\mathbb Q(\sqrt2)$, using the supplied physical labels and stated Laurent weights, and checks each up to its nonzero constant row multiplier. It checks the literal $cc^{-1}=1$ equation separately. 2. It verifies all 32 retained rows against the pinned parent after the declared coordinate relabeling, and independently constructs all 19 new rows from the displayed mathematical formulas. 3. It derives $G,P,Q$ by exact polynomial differentiation, verifies the derivative converse and all new physical contact scalings, and checks the native pole input normalization. 4. It expands the complete sum of squares in SymPy, independently of the producer's hand-written sparse convolution, and compares every integer coefficient of every parameter-separated term. 5. It verifies the support, all 59 used coordinates, the coefficient-one $X^4$ term, the mixed coefficient profiles, the slope leading terms, the local normalizer identity, and the exact rational height constants. The retained chart equations are unchanged. Thus all their tangent, zero-ordinate, finite two-torsion difference, and reference-chart branches have exactly the same algebraic meaning as in the parent. The arithmetic interpretation of that complete chart remains a retained premise. The first native pole has the physical-to-stored map ```math D=c z_{42},\quad b=c z_{34},\quad e=c z_{35},\quad u_0=c z_{36},\quad v_0=c z_{38},\quad p=z_{37},\quad q=z_{39}. ``` Its square helpers satisfy $p^2=c z_{40}$, $q^2=c z_{41}$. Multiplying its three main rows by $c^2$ gives precisely ```math e^2+bp^2=1,\qquad u_0v_0=e(1-b),\qquad u_0^2-v_0^2=bq^2. ``` The new pole-input row is $z_{34}=z_{33}^2$, where ```math z_{33}=\tau,\qquad b=c\tau^2=\frac{u g^2}{(1+u)^2},\qquad u=c^{-1}. ``` There is no extra factor $c^{-2}$ in this physical input. The contact quotient $u$ and the native pole coordinate $u_0$ are different quantities. ## Contact interpolation, guards and count For the shifts ```math \{-8,-7,-2,-1,3,4,5,6\}, ``` the polynomial is ```math G(t)=t^8-102t^6-144t^5+3009t^4+7200t^3 -17308t^2-32976t+40320. ``` The shifts are distinct, nonzero, have sum zero, and no two have zero sum, including a shift paired with itself. With $a=M^4$, $M\ge B\ge3$, every pair sum $2a+r+s$ is at least $146$. Thus the two eight-contact lists are disjoint before prime selection. Set ```math P=tG'-8G,\qquad Q=(4t^2-102)G'-32tG. ``` Both have degree at most six. Every power occurring in $G,P,Q$ is a product of at most two among $1,t,t^2,t^4$. After imposing the two power equations and the exact value equation, the two derivative equations give ```math t(H-G')=0,\qquad (4t^2-102)(H-G')=0. ``` Their coefficients have the constant combination ```math (4t^2-102)-4t\cdot t=-102. ``` Consequently $H=G'(t)$ at every rational point, including $t=0$. There is no division by $t$, $x$, $y$, or $xy$, and no deleted zero-coordinate fiber. This saves the two cubic-power coordinates. Each of the ten stored contact coordinates has physical value $c$ times its stored value. The derivative sum has physical value $c^2$ times its stored value. The quotient row is exactly ```math F_a+16cF=0, \qquad F=G(x)G(y),\quad F_a=G'(x)G(y)+G(x)G'(y). ``` The unit equation gives $c\ne0$. If $F=0$, exactly one of $G(x)$, $G(y)$ vanishes: both vanishing would force $x+y=-(r+s)$, incompatible with $x+y=2M^4>16$. Since the roots of $G$ are simple, exactly one vanishing factor makes $F_a\ne0$, a contradiction. If $F_a=0$, the quotient equation first gives $F=0$, already excluded. Thus the system recovers ```math u=-16F/F_a\ne0, ``` with no additional inverse registers. The 44 retained coordinates and 15 new coordinates give 59 in total. The new coordinates are $M,M^2$, ten contact coordinates, the derivative sum, and the two ratio-normalizer helpers. The 32 retained plus 19 new residuals give 51 equations. No assertion of algebraic independence is needed for this census or for the sum-of-squares equivalence over $\mathbb Q$. ## The ratio normalizer and its denominators The physical normalizer is ```math g=\frac{1+uD^2}{1+u^3D^2},\qquad \tau=\frac{g}{1+c},\qquad b=c\tau^2. ``` In stored coordinates put $J_s=D_s^2=D^2/c^2$ and $\lambda=g/c$. The cleared ratio equation is ```math \lambda(c+J_s)=1+cJ_s. ``` If $c+J_s=0$, this equation forces $c^2=1$. The case $c=1$ would require the rational square $J_s=-1$, which is impossible. The case $c=-1$ is $u=-1$. The derivative radical lower bound below excludes $u=-1$ for every sufficiently large $B$, before either the pole soundness argument or the conclusion that $X$ is integral is used. Thus both $c+J_s$ and $1+c$ are automatically nonzero in the defining range. Equivalently the physical denominator $1+u^3D^2$ is nonzero. No inverse of $g$ is used in soundness. Its numerator is nonzero at each consumed odd-valuation prime by parity; in the integer construction it is nonzero because $c>0$ and $g\to1$ at the real preparation. The fiber $D=0$ is allowed throughout and gives $g=1$. The shared trace denominator also has an automatic guard. If $\operatorname{Tr}W=0$, the equation $D\operatorname{Tr}W=\operatorname{Nm}(W-1)$ would imply $W=1$ in the field, contradicting $\operatorname{Tr}1=2$. This also excludes $W=0$. ## Universal odd-valuation proof Let $p$ be a prime, $\mu=v_p(u)$ odd, and $d=v_p(D)$, with $d=+\infty$ permitted. When $d$ is finite, both $\mu+2d$ and $3\mu+2d$ are odd and cannot equal zero. Hence ```math v_p(g)=\min(0,\mu+2d)-\min(0,3\mu+2d). ``` For $\mu<0$, the first argument of the minimum is larger than the second. Therefore $v_p(g)\ge0$ for every $d$, regardless of poles in the actual auxiliary point coordinates. Since $v_p(u/(1+u)^2)=-\mu$, ```math v_p(b)=-\mu+2v_p(g)>0 ``` is odd. This uniform conclusion is why the denominator $1+u^3D^2$ is needed; the unnormalized expression $g=1+uD^2$ does not supply it. For $\mu>0$, there are three intervals for $2d$. | Interval | $v_p(g)$ | $v_p(b)$ | |---|---:|---:| | $2d>-\mu$ | $0$ | $\mu$ | | $-3\mu<2d<-\mu$ | $\mu+2d$ | $3\mu+4d$ | | $2d<-3\mu$ | $-2\mu$ | $-3\mu$ | The endpoints cannot occur because $2d$ is even and $\mu$ is odd. For $D=0$, the first case applies with infinite $d$. The native pole equations exclude a negative odd valuation of $b$, including all zero-coordinate cases. Indeed the first equation forces $v_p(e)=0$: a negative valuation of $e^2$ is even, while a negative valuation of $bp^2$ is odd, so they cannot cancel; the latter term must have positive valuation if it is nonzero. Then the middle equation gives $v_p(u_0)+v_p(v_0)=v_p(b)$, an odd integer. Their valuations cannot be equal. The third equation then has a left side of even valuation and a right side of odd valuation, a contradiction. If the right side is zero, it would instead force equal valuations, again a contradiction. The last interval is therefore impossible. In the middle interval the odd integer $3\mu+4d$ must be positive. All surviving cases give ```math d>-3\mu/4. ``` This proof is universal. The finite valuation grid in the checker is only an additional regression check of the formula and inequality. ## Extraction from the shared detector At a consumed good prime, $v_p(c)=-\mu$, with $\mu>0$ odd. If both slot valuations $w_1,w_2$ of $W$ are negative, then ```math v_p\operatorname{Nm}(W-1)=w_1+w_2,\qquad v_p\operatorname{Tr}W\ge\min(w_1,w_2), ``` so $v_p(D)\le\max(w_1,w_2)$. This includes an inert prime, where the two valuations coincide. If either slot is already integral there is nothing to prove. In either case the preceding inequality for $D$ gives a slot with $v(W)>-3\mu/4$. At that slot the four nonconstant terms in ```math W=1+cC^4+c^2J^2-c^3\overline I^{\,4}-c^4\overline{K'}^{\,2} ``` have respective valuation classes $-\mu,-2\mu,-3\mu,0$ modulo four. The even valuations of $J,K'$ are the retained reciprocal-abscissa property. A negative minimum is consequently unique and cannot cancel. Every one of these terms has valuation greater than $-3\mu/4$, giving ```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. ``` All four raw tests are positive. A zero raw test has infinite valuation and satisfies the same conclusion. The retained inverse-parameter extraction theorem now gives local integrality of $X$ and $X(m-X)\equiv0\pmod p$. No extra nontorsion, nontangent or nonzero ordinate hypothesis has been introduced. ## Height argument and uniformity in the shift The intermediate estimate proved in v63 is ```math \log r^+_{\mathrm{odd},S}(u) \ge (k-6-3\epsilon)\max\{h(x),h(y)\} -(2+\epsilon)h(M)-O(1). ``` Its proof precedes the optional replacement of the larger slot height by half the product height, and never divides by $xy$. Thus it applies here when $x=0$ or $y=0$. Its hypotheses are satisfied by the fixed eight shifts and the nonzero contact values just checked. Its constant and exceptional set depend on the fixed shifts, not on $B$. Since $x+y=2M^4$, $\max(h(x),h(y))\ge2h(M)-O(1)$. With $k=8$ and $\epsilon=1/100$, ```math \log r^+_{\mathrm{odd},S}(u)\ge\frac{193}{100}h(M)-O(1). ``` In particular $u=-1$, whose radical on the left is one, forces $h(M)$ bounded by a fixed constant. Since $M\ge B$, this proves the eventual guard independently of all pole conditions. For a nonintegral input, $X(m-X)\ne0$ because $m\in\mathbb Z$. The extraction conclusion implies that the consumed radical divides its reduced numerator, so ```math \log r^+_{\mathrm{odd},S}(u) \le2h(X)+\log(1+|m|). ``` The retained signed mixed-power estimate with $\kappa=1001/1000$ gives ```math h(X)\le\frac{2\kappa-1}{8-6\kappa} h(X^8+T_{\mathrm{sc}}^2)+O(1) \le\frac{501}{997}h(M)+O(1). ``` The second inequality is uniform in positive integer $B$: adding $B$ to a nonnegative rational leaves its reduced denominator unchanged and does not decrease the height. Also $2h(T_{\mathrm{sc}})\le h(X^8+T_{\mathrm{sc}}^2)+8h(X)+O(1)$. Together with the retained trace-index bound, this gives $m^2=O(h(M)+1)$. The exact strict margin is ```math \frac{193}{100}-2\frac{501}{997} =\frac{92221}{99700}>0. ``` Consequently a nonintegral solution would bound $h(M)$ by a fixed constant, whereas $h(M)\ge\log B$. This proves exclusion of all nonintegral inputs for sufficiently large $B$. ## Integer completeness and the order of choices Fix a nonzero integer $X$. The retained construction first fixes the root difference $A=[X]R_0$, its scaled trace, $M$, and $a=M^4$. It fixes bounded real and local neighborhoods for the remaining point functions. Thus both eight-contact lists and all their collision values are fixed before contact prime selection. Clear the rational contact denominators and enlarge the **preparation** set by their coefficient, denominator and collision primes. This is a finite set depending on the chosen integer witness. It is separate from the fixed exceptional set in the soundness height estimate. Use the retained simultaneous prime-pattern theorem on the sixteen signed linear forms, with finite avoidance and the same formal-index refinements. For a sufficiently large real slope, the first eight forms can be negated to have positive leading sign; this does not change their prime support. More explicitly, choose $D_0>0$ with $D_0^2(a+r)\in\mathbb Z$ for all eight shifts and use the permitted prepared form ```math z=\frac{N}{D_0^2L^2},\qquad p_{r,\pm}=N\pm D_0^2(a+r)L^2. ``` The supplied sixteen primes give the exact factorization ```math F=\frac{\prod_r p_{r,-}p_{r,+}}{D_0^{32}L^{32}}. ``` The eight minus signs from the quadratic factors cancel. This identity accounts for every numerator prime of the contact product; it does not assume that the subsequently chosen auxiliary points have no new denominator primes. At each selected simple contact exactly one linear factor has valuation one and all other contact factors are units. The derivative identity therefore gives $v_p(u)=1$. At the successful slot the retained preparation gives $v(C),v(\overline I)\ge3$ and $v(J),v(\overline{K'})\ge4$, so $W\equiv1$. At the other slot the reciprocal products are units and the fourth term has the unique minimum $-4$. Therefore $v_p(D)>0$, $g$ is a unit and $v_p(b)=1$. Outside the finite preparation set, positive odd valuations of $u$ can only occur at these selected primes. For an integral slope at a prime, $F,F_a$ are integral and a positive valuation of their quotient requires a numerator factor of $F$. If the slope has valuation $-h<0$, its contact product has valuation $-16h$, while its derivative has valuation at least $-14h$, so $v_p(u)\le-2h<0$; cancellation in $F_a$ can only decrease this quotient valuation further. The leading factor 16 is a unit outside the preparation set. Thus a new denominator prime cannot become an unprepared positive odd contact prime. New **negative** odd valuations of $u$, including those introduced by $F_a$, are harmless for every actual point denominator by the universal ratio-normalizer proof. Their valuations of $b$ are positive odd. If $v_p(u)$ is even, then $v_p(u/(1+u)^2)$ and hence $v_p(b)$ are even. This last statement includes $v_p(u)=0$ and an arbitrarily deep cancellation in $1+u$. It remains to handle the finite preparation set and real infinity. The exact leading terms are ```math F=z^{16}+O(z^{14}),\qquad F_a=-16az^{14}+O(z^{12}), \qquad u\sim z^2/a,\quad c\sim a/z^2. ``` They hold at real infinity and at every fixed nonarchimedean place as the slope denominator is deepened. On the fixed point neighborhoods the raw tests remain bounded, so ```math W-1=O(c),\qquad \operatorname{Tr}W\to2,\qquad D=O(c^2). ``` The trace limit is nonzero also over $\mathbb Q_2$; its valuation is eventually one, a fixed loss absorbed by increasing the depth. Writing $D=c^2E$, with $E$ bounded locally, gives the exact identity ```math g=\frac{1+c^3E^2}{1+cE^2}\longrightarrow1. ``` This bound is uniform on those fixed neighborhoods, so no control of auxiliary denominators at subsequently discovered primes is assumed. At the real place $c>0$ for sufficiently large slope, and $b\sim c>0$ tends to zero. Thus $00$ even. All these are finitely many local and real conditions specified before the selected prime choices, exactly as allowed by the retained simultaneous prime-pattern premise. Every negative valuation of the actual rational $b$ is now even, so its reduced denominator is a square. The small-input pole lemma from v78 applies to this actual $b$: it requires only $0