\documentclass[11pt,a4paper]{article} \usepackage[T1]{fontenc} \usepackage{lmodern,amsmath,amssymb,amsthm,mathtools,booktabs,microtype} \usepackage[margin=25mm]{geometry} \usepackage[colorlinks=true,linkcolor=blue,urlcolor=blue,citecolor=blue]{hyperref} \newtheorem{theorem}{Theorem}[section] \newtheorem{lemma}[theorem]{Lemma} \newtheorem{proposition}[theorem]{Proposition} \newtheorem{corollary}[theorem]{Corollary} \newcommand{\Q}{\mathbb Q}\newcommand{\Z}{\mathbb Z} \DeclareMathOperator{\Disc}{Disc}\DeclareMathOperator{\supp}{supp} \setlength{\emergencystretch}{2em} \hypersetup{pdftitle={Separating Quadratic Roots in Finite Algebras and Smaller Degrees for Diophantine Presentations}} \title{Separating Quadratic Roots in Finite Algebras\\and Smaller Degrees for Diophantine Presentations} \author{}\date{13 November 2026} \begin{document}\maketitle \begin{abstract} We give an explicit polynomial encoding a rational point of a finite algebra subject to a nonzero condition. A triangular linear combination separates quadratic roots without a lower bound on their nonzero radicands. Separate bounds for the base roots and the radicands keep the polynomial weights small. Characteristic polynomials are truncated to the dimension of the relevant nonzero part. The construction includes vanishing radicands, nonreal roots, repeated characteristic values, and nonreduced excluded components. Applied to the existing arithmetic sources, it gives degree bounds below $2^{78}$ in 25 variables, below $2^{80}$ in the fully specified 27-variable construction, and below $2^{38}$ in both conditional 13-variable constructions. The fully specified coefficient-height bound improves to $2^{2^{92}}$. The variable counts, arithmetic hypotheses and sufficient coefficient parameters are unchanged. \end{abstract} \section{Applications and their precise scope} Every variable count includes the distinguished input. The degree is total degree, and coefficient height is the largest absolute integer coefficient. The following are bounds for new polynomials with exactly the same rational input projections as the source constructions in the repository. \begin{center}\small \begin{tabular}{lrrr} \toprule Source & Variables & Degree at most & Support less than\\\midrule Inherited arithmetic premises &25&185517288563368898320200&$2^{1815}$\\ Fully specified, inherited premises &27&679086760181946108087584&$2^{2009}$\\ Weak $abc$ at $121/100$ &13&207969001841&$2^{437}$\\ Weak $abc$ at $16/15$ &13&199368913265&$2^{436}$\\\bottomrule \end{tabular}\end{center} The last two rows also retain the elliptic group and height, pole-realization, and prime-pattern inputs of their sources. The first two retain their additional modular and contact-height inputs. This paper changes none of those arithmetic statements. It proves a new algebraic conversion of the same sources. It is not an independent verification of their arithmetic foundations. For 25 variables the sufficient odd coefficient parameter remains ineffective. For 27 variables the existing certified value $L=20483$ remains sufficient under the same premises. In both 13-variable families the sufficient odd shift $B$ remains ineffective; the numerical $B=3$ examples are not certified defining members. Multiplication by the input includes zero in the two 13-variable constructions, exactly as in their sources. The source polynomials have not been expanded. Thus the displayed degrees and support sizes are upper bounds, not exact degrees or counts. The coefficient specification in Section~\ref{sec:formula} consists entirely of finite integer-polynomial operations. \section{Finite algebras and the nonzero condition} Let $R=\Z[x_1,\ldots,x_a]$. Let \[ B=R[z_1,\ldots,z_p]/(f_1(z_1),\ldots,f_p(z_p)),\qquad A=B[q_1,\ldots,q_s]/(q_j^2-c_j)_{j=1}^s, \] where $p,s\ge1$, the $f_i\in R[Z]$ are monic of degrees $d_i\ge2$, and $c_j\in B$. Suppose each $f_i$ is separable at every rational specialization of the base coordinates. Put \[ b=\prod_i d_i,\qquad n=2^s b. \] The usual monomials of exponent below the defining degrees are bases of $B$ and $A$. Write $M_v^B$ or $M_v^A$ for multiplication by $v$ in these bases. Fix $g\in B$ and set \[ G=gq_1\cdots q_s,\qquad T=g^2c_1\cdots c_s\in B. \] At a rational base point our target condition is \begin{equation}\label{eq:target} \text{there exists a $\Q$-algebra map $A\longrightarrow\Q$ with $G\ne0$.} \end{equation} There are no additional residual equations. The existing source certificates verify this form of the nonzero condition and the independence of each $c_j$ from the $q$-generators. At a rational base point $B$ is an etale algebra. On its factors where $T\ne0$, all the $c_j$ and $g$ are nonzero, and the corresponding part of $A$ is also etale. The other factors of $A$ may have nilpotents. They will not be discarded by an unjustified division. \section{A triangular combination of quadratic roots} \begin{lemma}\label{lem:triangular} Let $S\ge1$ be real and let $c_1,\ldots,c_s\in\mathbb C^\times$ satisfy $|c_j|\le S$. For each choice $q_j^2=c_j$, put \[ \phi(q_1,\ldots,q_s) =\frac1{2S}\sum_{j=1}^s(4S)^{1-j} \left(\prod_{k0$. Put $\epsilon=D/K$ and $L=C/D$. Then $0<\epsilon\le1$ and $L=1+4Z_0/\epsilon$. \begin{lemma}\label{lem:pointsep} Any two distinct roots of any $f_i$ differ in absolute value by at least $\epsilon$. \end{lemma} \begin{proof} If $\Delta$ is the chosen root difference, the discriminant product gives \[ |\delta_i|\le |\Delta|^2(2Z_0)^{d_i(d_i-1)-2}. \] For $t>0$, $t^2/(1+t^2)\le\sqrt t$. Therefore \[ \epsilon\le \frac{\delta_i^2}{1+\delta_i^2}(2Z_0)^{-\kappa} \le\frac{\sqrt{|\delta_i|}} {(2Z_0)^{(d_i(d_i-1)-2)/2}}\le|\Delta|. \] The exponent is the maximum needed for a single polynomial, not the sum of all discriminant exponents. \end{proof} Define in $A$ \begin{gather*} J=4S,\qquad Q=2S J^{s-1},\qquad \Xi=\sum_{j=1}^sJ^{s-j}\left(\prod_{k0$, then $a_r\ne0$ and $h_r(X)$ is the product of $X-T(\chi)$ over the $r$ nonzero character values, with repetitions retained. Hence $E_r=a_r$ on this part and $E_r=0$ on $T=0$. If $r>r_0$, then $a_r=0$ and $h_r(X)=X^{r-r_0}h_{r_0}(X)$, with $h_0=1$ when $r_0=0$. This vanishes at every value of $T$. Thus $E_r=0$ in $B$, and also in $A$; $H_r=1$. This explains why no additional factor $a_r^2$ is required in $\omega_r$. If $rr}a_j^2$ strictly positive at every rational $U$. Consequently any rational zero of~\eqref{eq:polynomial} belongs to the unique correct case $r=r_0>0$. In that case $E_r$ kills the entire excluded part, including any nilpotents. The remaining part is etale of dimension $d_r$. On it $\omega_r=a_r\Omega$ has distinct nonzero character values by Proposition~\ref{prop:sep}. Thus the full determinant has degree $d_r$, so the truncation is exact. Its rational zero gives a nonzero rational character value of $\omega_r$ of multiplicity one. The corresponding one-dimensional rational kernel of $I-UM_{\omega_r}$ is preserved by every native multiplication matrix. Their actions on this line are rational scalars satisfying the algebra relations. They define a rational character on which $G\ne0$. Conversely, a rational character with $G\ne0$ has a nonzero rational value of $\omega_{r_0}$. Its reciprocal is a root of $H_{r_0}$, and all higher $a_j$ vanish. This gives a zero of $P$. If $r_0=0$, every $H_r=1$ and $P$ has no zero. All assertions concern the complete rational zero set, not a generic open subset. \end{proof} \section{Degree and support bounds} Give the base coordinates weight one, the $z_i$ weights $w_i$, and the $q_j$ weights $v_j$, so that monic reduction does not increase weight. Suppose $\operatorname{wt}(G)\le w_G$. Put \[ t=2w_G,\quad u=\sum_i d_i(d_i-1)w_i,\quad b_w=\sum_i(d_i-1)w_i. \] The scalar bounds in~\eqref{eq:bounds} have degree at most \[ z_d=2\max_i d_iw_i,\qquad s_d=2\max_j(2v_j+b_w). \] Indeed an entry of multiplication by $c_j$ has degree at most $2v_j+$ column weight $-$ row weight. Define \[ d_d=2u,\qquad c_d=(\kappa+1)z_d+2u, \] and \begin{align*} A_0&=ss_d+\max\left\{(p+1)c_d, \max_{1\le i\le p}(id_d+(p+1-i)c_d+w_i)\right\},\\ A_1&=(p+1)d_d+\max_{1\le j\le s} \left((s-j)s_d+2\sum_{k