\documentclass[11pt]{article} \usepackage[T1]{fontenc} \usepackage[utf8]{inputenc} \usepackage[a4paper,margin=1in]{geometry} \usepackage{amsmath,amssymb,amsthm} \usepackage{enumitem} \usepackage{microtype} \usepackage{xcolor} \usepackage{hyperref} \hypersetup{colorlinks=true,linkcolor=blue,urlcolor=blue} \urlstyle{same} \setlength{\emergencystretch}{2em} \newtheorem{theorem}{Theorem}[section] \newtheorem{lemma}[theorem]{Lemma} \newtheorem{proposition}[theorem]{Proposition} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{definition} \newtheorem{definition}[theorem]{Definition} \newtheorem{construction}[theorem]{Construction} \newtheorem{example}[theorem]{Example} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \newcommand{\Q}{\mathbb Q} \newcommand{\N}{\operatorname N} \newcommand{\sig}{\operatorname s} \newcommand{\LM}{\operatorname{LM}} \newcommand{\LT}{\operatorname{LT}} \newcommand{\LC}{\operatorname{LC}} \newcommand{\Nil}{\operatorname{Nil}} \newcommand{\Tr}{\operatorname{Tr}} \title{An Algebraic Proof of Hilbert Nullstellensatz over Rationals} \author{% Liang Zhang\thanks{Corresponding author: \href{mailto:liang.zhang@unidt.com}{\texttt{liang.zhang@unidt.com}}} \and Xiaohua(Michael) Xuan\thanks{Email: \href{mailto:michael.xuan@unidt.com}{\texttt{michael.xuan@unidt.com}}}% } \date{} \begin{document} \maketitle \begin{abstract} % We with the aid of AI, prove a finite approximate form of the weak % Nullstellensatz over the Gaussian rationals. This extends the % one-variable result of the first paper in this series to several % variables within Wildberger's finitist framework. % Let $f_1,\ldots,f_m$ be polynomials in $n$ variables with Gaussian % rational coefficients. Suppose there are no polynomials % $g_1,\ldots,g_m$ in the same variables with Gaussian rational % coefficients such that $1=g_1f_1+\cdots+g_mf_m$. % (A finite Gr\"obner basis computation decides whether such polynomials exist.) % Then for every positive rational number $\varepsilon$ there is a point % $z$ with $n$ Gaussian rational coordinates such that % \[ % \sum_{j=1}^{m}\N(f_j(z))<\varepsilon^2, % \qquad \N(a+ib)=a^2+b^2. % \] % The construction uses polynomial arithmetic, Gaussian elimination, % Gr\"obner bases, finite grids, and natural-number induction. % It requires no completion, algebraic closure, maximal ideal, compactness, % limiting process, or infinite set. We, with the aid of AI, prove a finite approximate form of the weak Nullstellensatz over the Gaussian rationals: Let $f_1,\ldots,f_m$ be polynomials in $n$ variables with rational complex coefficients. If there are no polynomials $g_1,\ldots,g_m$ in the same variables with rational complex coefficients such that $1=g_1f_1+\cdots+g_mf_m$. (A finite Gr\"obner basis computation decides whether such polynomials exist.), then, for any positive rational number $\varepsilon$, there exists a pont $z$ with $n$ rational complex coordinates such that \[ \sum_{j=1}^{m}\N(f_j(z))<\varepsilon^2, \] where $\N(a+ib)=a^2+b^2$. The proof is constructive and relis only on finite rational arithmetics without using real numbers or infinite sets concepts, and extends authors' earlier one-vaiable result on Fundemental Theorem of Algerba to several-varaibles case. \end{abstract} \setcounter{tocdepth}{1} \tableofcontents \clearpage \section{Introduction} The existence of polynomial zeros is central to algebra. In one variable, the Fundamental Theorem of Algebra has a history that includes Gauss's dissertation of 1799 \cite{BasuVelleman2017}. In 1893 Hilbert proved the Nullstellensatz in his work on invariants~\cite{Hilbert1893}, linking polynomial ideals to their common zeros. Its early development through elimination and ideal theory is surveyed in \cite{Stevens2023}. Gr\"obner bases make ideal membership and elimination computational \cite{BeckerWeispfenning}. This raises a foundational question: how much of a polynomial existence theorem follows from finite algebraic data and explicitly justified finite operations? Motivated by Wildberger's emphasis on finite objects and rational arithmetic \cite{Wildberger2005,WildbergerFTA}, the first paper in this series \cite{FiniteFTA} gave a finite rational proof of an approximate version of the Fundamental Theorem of Algebra. Here we extend this approach to several variables: a proper polynomial ideal admits rational complex points where its given generators are simultaneously as small as prescribed. Algebraic types are interpreted through finite data, without completed infinite collections or limiting processes. Section~\ref{sec:finite-setting} introduces the finite setting and states the main theorem. Sections~\ref{sec:finite-bounds}--\ref{sec:groebner} develop the elementary estimates and polynomial reduction tools. Sections~\ref{sec:finite-quotient}--\ref{sec:primitive} reduce the problem to a finite-dimensional algebra and then to one polynomial in one variable. Sections~\ref{sec:univariate}--\ref{sec:main-proof} use the first paper's approximation result to obtain simultaneous approximation and complete the proof. The appendices supply the detailed arguments for the trace pairing, square-free reduction, and primitive-element construction. \section{Finite setting and statement}\label{sec:finite-setting} We write $x\mathrel{:}T$ to mean that $x$ is of type $T$. For an explicit finite list or set $E$, $x\in E$ denotes membership in $E$. For an explicit finite list or set $E$, the notation $E\subseteq T$ (also $E\subset T$) asserts that each listed entry is of type $T$. % For finite sets $E,D$, the relation $E\subseteq D$ is checked by comparing % each entry of $E$ with the entries of $D$. Both conventions involve % only finitely many tests. A \emph{rational complex number} (\emph{Gaussian rational}) is a pair $a,b\mathrel{:}\Q$, written $a+ib$, with $i^2=-1$. We denote this type by $K$. For $u=a+ib\mathrel{:} K$, write \[ |a|:=\max\{a,-a\}\quad(a\mathrel{:}\Q),\qquad \bar u:=a-ib, \qquad \N(u):=u\bar u=a^2+b^2, \qquad \sig(u):=|a|+|b|. \] The notation $j\mathrel{:}\mathbb N$ means that $j$ is a nonnegative integer. For a type $T$ and a specified nonnegative integer $\ell$, $T^\ell$ denotes the type of $\ell$-tuples with entries of type $T$; $T^0$ has the single empty tuple. Subscripts $>0$ and $\ge0$ impose the indicated rational or integer inequalities. The notation $u\mathrel{:} K^\times$ means $(u\mathrel{:} K)\land(u\ne0)$. A polynomial of type $K[X]=K[X_1,\ldots,X_n]$ is specified by its finite list of nonzero terms with coefficients of type $K$. For $\alpha=(\alpha_1,\ldots,\alpha_n)\mathrel{:}\mathbb N^n$, define \[ X^\alpha:=X_1^{\alpha_1}\cdots X_n^{\alpha_n}, \qquad |\alpha|:=\alpha_1+\cdots+\alpha_n. \] For nonzero $f=\sum_\alpha c_\alpha X^\alpha\mathrel{:} K[X]$, define \[ \deg f:=\max\{|\alpha|:c_\alpha\ne0\}. \] % Here, $\max$ denotes the largest element of a finite nonempty set of integers. % Polynomial sums and maxima range over the supplied finite term list. A bound such as $\deg f0}$, $f_1,\ldots,f_m\mathrel{:} K[X_1,\ldots,X_n]$, and $\varepsilon\mathrel{:}\Q_{>0}$. Then at least one of the following holds: \begin{enumerate}[label=\textup{(\roman*)}] \item there exist polynomials $h_1,\ldots,h_m\mathrel{:} K[X_1,\ldots,X_n]$ satisfying \eqref{2.1}; \item there exists a point $z\mathrel{:} K^n$ satisfying $\mathcal N_F(z)<\varepsilon^2$. \end{enumerate} In particular, if $I=\langle F\rangle$ is proper, then \textup{(ii)} holds. \end{theorem} \section{Elementary finite bounds}\label{sec:finite-bounds} \begin{lemma}[Rational size inequalities]\label{lem:size} For $u,v\mathrel{:} K$, \[ \N(u)\le \sig(u)^2, \qquad \sig(u+v)\le\sig(u)+\sig(v), \qquad \sig(uv)\le\sig(u)\sig(v). \tag{3.1}\label{3.1} \] For $b_{\rm bd}\mathrel{:}\Q_{\ge0}$ and a one variable polynomial $f(X)=\sum_{j=0}^{d}c_jX^j\mathrel{:} K[X]$, define \[ \|f\|_{b_{\rm bd}}:=\sum_{j=0}^{d}\sig(c_j)b_{\rm bd}^j. \tag{3.2}\label{3.2} \] For $t\mathrel{:} K$ with $\sig(t)\le b_{\rm bd}$, one has \[ \sig(f(t))\le\|f\|_{b_{\rm bd}}, \qquad \N(f(t))\le\|f\|_{b_{\rm bd}}^2. \tag{3.3}\label{3.3} \] For $z,w\mathrel{:} K$ with $\sig(z),\sig(w)\le b_{\rm bd}$ and an integer $j\ge1$, \[ \sig(z^j-w^j)\le jb_{\rm bd}^{j-1}\sig(z-w), \tag{3.4}\label{3.4} \] and \[ \sig(f(z)-f(w)) \le \sig(z-w)\sum_{j=1}^{d}j\sig(c_j)b_{\rm bd}^{j-1}. \tag{3.5}\label{3.5} \] \end{lemma} \begin{proof} See \cite[Lemmas~3.2--3.4]{FiniteFTA}. \end{proof} \begin{lemma}[Finite nonvanishing grid]\label{lem:grid} Let $n,d\mathrel{:}\mathbb N$. If \[ f\mathrel{:} K[X_1,\ldots,X_n],\qquad f\ne0,\qquad \deg f\le d, \] then there is some \[ a\in\{0,1,\ldots,d\}^n \tag{3.6}\label{3.6} \] satisfying $f(a)\ne0$. \end{lemma} \begin{proof} For $n=0$, evaluation at the empty tuple leaves the nonzero constant unchanged. For $n=1$, if $f(0)=\cdots=f(d)=0$, repeated Euclidean division gives \[ X(X-1)\cdots(X-d)\mid f, \] contrary to $\deg f\le d$ and $f\ne0$. For $n>1$, write \[ f=\sum_{j=0}^{d}f_j(X_1,\ldots,X_{n-1})X_n^j \] and choose $j_0$ such that $f_{j_0}\ne0$. Since $\deg f_{j_0}\le d$, the induction hypothesis gives \[ a'=(a_1,\ldots,a_{n-1})\in\{0,1,\ldots,d\}^{n-1} \qquad\text{with}\qquad f_{j_0}(a')\ne0. \] Consequently \[ g(X_n):=f(a',X_n) =\sum_{j=0}^{d}f_j(a')X_n^j \] is a nonzero polynomial of degree at most $d$. By the one-variable case, there exists $a_n\in\{0,1,\ldots,d\}$ such that \[ f(a',a_n)=g(a_n)\ne0. \] Thus $a=(a',a_n)\in\{0,1,\ldots,d\}^n$ has the required property. \end{proof} \section{Polynomial reduction and Gr\"obner bases}\label{sec:groebner} \subsection{Term orders and polynomial reduction} \begin{definition}[Degree--lexicographic order] \label{def:deglex} Fix a variable order \[ X_1\succ X_2\succ\cdots\succ X_n . \] For \(\alpha,\beta\mathrel{:}\mathbb N^n\), define \[ X^\alpha\prec_{\mathrm{dlex}}X^\beta \] if \[ |\alpha|<|\beta|, \] or if \(|\alpha|=|\beta|\) and, at the first coordinate \(k\) for which \(\alpha_k\ne\beta_k\), one has \(\alpha_k<\beta_k\). We write \(\prec\) for \(\prec_{\mathrm{dlex}}\), and \(M\preceq N\) means \(M\prec N\) or \(M=N\). Then \[ M\prec N\Longrightarrow UM\prec UN \tag{4.0}\label{4.0} \] for all monomials \(M,N,U\). Moreover, for a monomial \(M\) with \(\deg M=d\), \[ \{N:N\preceq M\} \subseteq \{X^\alpha:|\alpha|\le d\},\qquad \#\{X^\alpha:|\alpha|\le d\}=\binom{d+n}{n}. \tag{4.0a}\label{4.0a} \] % Thus every initial segment of the degree--lexicographic order is finite. The order $\prec$ is total, so every nonempty finite set of monomials has a unique maximum. % Maxima of monomials are taken with respect to $\prec$, % and numerical maxima with respect to the usual order. \end{definition} \begin{definition}[Leading data] For nonzero $p=\sum c_\alpha X^\alpha$, define its leading monomial by \[ \LM(p):=\max\{X^\alpha:c_\alpha\ne0\}. \] The leading coefficient $\LC(p)$ is the coefficient of $\LM(p)$ in $p$. Define its leading term by \[ \LT(p):=\LC(p)\LM(p). \] The polynomial is monic if $\LC(p)=1$. \end{definition} \begin{definition}[Reduction] If $p,g\ne0$ and $\LM(g)\mid\LM(p)$, one reduction step is \[ p\longmapsto p-\frac{\LC(p)}{\LC(g)}\frac{\LM(p)}{\LM(g)}g. \tag{4.1}\label{4.1} \] Given an ordered finite tuple $G=(g_1,\ldots,g_s)$ of nonzero polynomials (possibly the empty tuple), complete division starts with \[ f_{\mathrm{cur}}:=p,\qquad r:=0,\qquad \mathfrak q_1=\cdots=\mathfrak q_s:=0 \] and preserves \[ p=\sum_{j=1}^{s}\mathfrak q_jg_j+f_{\mathrm{cur}}+r. \tag{4.1a}\label{4.1a} \] If $\LT(f_{\mathrm{cur}})=cM$ and $g_j$ is the first reducer with $\LM(g_j)\mid M$, define \[ U:=\frac{M}{\LM(g_j)},\qquad \lambda:=\frac{c}{\LC(g_j)}, \] and update \[ f_{\mathrm{cur}}\leftarrow f_{\mathrm{cur}}-\lambda Ug_j,\qquad \mathfrak q_j\leftarrow \mathfrak q_j+\lambda U. \tag{4.1b}\label{4.1b} \] If no such reducer exists, update \[ f_{\mathrm{cur}}\leftarrow f_{\mathrm{cur}}-cM,\qquad r\leftarrow r+cM. \tag{4.1c}\label{4.1c} \] Only a monomial satisfying \[ \LM(g_j)\nmid M\quad(1\le j\le s) \tag{4.1d}\label{4.1d} \] enters $r$. Thus, when $f_{\mathrm{cur}}=0$, complete division records \[ p=\mathfrak q_1g_1+\cdots+\mathfrak q_sg_s+r, \tag{4.2}\label{4.2} \] where every monomial of $r$ satisfies \eqref{4.1d}. % Terms created by \eqref{4.1b} remain in $f_{\mathrm{cur}}$ % and are tested when they become leading. \end{definition} \begin{lemma}[Finite degree--lexicographic division] \label{lem:deglex-division} Let nonzero \(p\mathrel{:} L[X_1,\ldots,X_n]\) have total degree \(d\), and let \(G\) be any finite reducer tuple. Complete division \eqref{4.1a}--\eqref{4.1c} performs at most \[ \binom{d+n}{n} \tag{4.3}\label{4.3} \] steps. Every monomial occurring in \(f_{\mathrm{cur}}\) or \(r\) during the division has total degree at most \(d\). For \(p=0\), division returns the zero remainder without any steps. \end{lemma} \begin{proof} Let \(M=\LM(f_{\mathrm{cur}})\). In the reduction case \eqref{4.1b}, write \[ g_j=\LC(g_j)\LM(g_j) +\sum_{N\prec\LM(g_j)}c_NN. \tag{4.4}\label{4.4} \] With \(U=M/\LM(g_j)\), equation \eqref{4.0} gives \[ UN\prec U\LM(g_j)=M \qquad(N\prec\LM(g_j)). \tag{4.5}\label{4.5} \] Hence the term \(cM\) is cancelled and every newly created monomial is strictly smaller than \(M\). Since the order first compares total degree, \[ UN\prec M\Longrightarrow \deg(UN)\le\deg M. \tag{4.6}\label{4.6} \] In the remainder case \eqref{4.1c}, \(M\) is removed from \(f_{\mathrm{cur}}\), no new active monomial is created, and \(M\) is moved to \(r\). Consequently the successive active leading monomials form a strictly decreasing sequence, and every one has total degree at most \(d=\deg p\). No leading monomial can occur twice. Equation \eqref{4.0a} gives the bound \eqref{4.3}. The same argument gives the asserted degree bound for every active and remainder monomial. \end{proof} \subsection{Critical pairs and the Gr\"obner property} \begin{definition}[S-polynomials and their representations] For monic $g_i,g_j$, define \[ \operatorname{lcm}(X^\alpha,X^\beta) :=X_1^{\max(\alpha_1,\beta_1)}\cdots X_n^{\max(\alpha_n,\beta_n)}, \] \[ M_{ij}:=\operatorname{lcm}(\LM(g_i),\LM(g_j)), \] \[ S(g_i,g_j):= \frac{M_{ij}}{\LM(g_i)}g_i- \frac{M_{ij}}{\LM(g_j)}g_j. \tag{4.7}\label{4.7} \] For $G=(g_1,\ldots,g_s)$, consider representations \[ S(g_i,g_j)=\sum_{k=1}^{s}\mathfrak q_{ijk}g_k \tag{4.8}\label{4.8} \] for $i0), \tag{4.15b}\label{4.15b} \] and, for \(1\le i\le k\), \(c_i>0\), and \(h_{\rm ctr}\in\{0,\ldots,\delta(q)\}\), by \[ (\mathbf c;q) \longrightarrow_{\delta,k} (\mathbf c-\mathbf e_i+h_{\rm ctr}\mathbf e_{i-1};q+1). \tag{4.15c}\label{4.15c} \] A \emph{\(k\)-counter sequence} is a finite sequence \[ \sigma=(\mathcal C^{(0)},\ldots,\mathcal C^{(t)}),\qquad \mathcal C^{(u)}\longrightarrow_{\delta,k}\mathcal C^{(u+1)} \quad(0\le u0\), exactly \(h_q\) slices of dimension \(k-1\) are added. If \(k=0\), the family \(\mathcal A_q\) is empty. Thus \[ \mathbf c(q+1)= \begin{cases} \mathbf c(q)-\mathbf e_0,&k=0,\\ \mathbf c(q)-\mathbf e_k+h_q\mathbf e_{k-1},&k>0. \end{cases} \] Since \(c_k(q)>0\) and \(0\le h_q\le\delta(q)\), \[ (\mathbf c(q);q)\longrightarrow_{\delta,n} (\mathbf c(q+1);q+1). \] Thus \[ ((\mathbf c(0);0),\ldots,(\mathbf c(\ell);\ell)) \] is an \(n\)-counter sequence of length \(\ell\) with initial state \((1,0,\ldots,0;0)\). By Lemma~\ref{lem:gamma-finite}, \[ \ell\le\gamma_{\delta,n}(1,0,\ldots,0;0)=R_{\delta,n}. \] \end{proof} \begin{theorem}[Finite Gr\"obner basis construction] \label{thm:finiteGB} Let \[ F=(f_1,\ldots,f_m)^{\mathsf t},\qquad f_j\mathrel{:} L[X_1,\ldots,X_n]. \] A finite procedure produces a monic Gr\"obner basis \(G\) of \(\langle F\rangle\), representations \eqref{4.8} with the leading monomial bound \eqref{4.8a}, and polynomial matrices \(\mathsf A,\mathsf B\) satisfying \[ G=\mathsf A F,\qquad F=\mathsf B G. \tag{4.17}\label{4.17} \] It decides whether \(1\mathrel{:}\langle F\rangle\) and, in that case, produces an identity \(1=\sum_jh_jf_j\). \end{theorem} \begin{proof} Start with the monic normalizations of the nonzero entries of \(F\). Successively reduce each unexamined S-polynomial by the current tuple, adjoining the monic normalization of every nonzero remainder and retaining the expressions in terms of \(F\). Let \(s_0\) be the number of nonzero entries of \(F\), and let \(d\) be their maximum degree, with \(d=0\) if \(s_0=0\). Let \[ d_0:=\max(1,d),\qquad \delta(q):=2^{q+1}d_0,\qquad R:=R_{\delta,n}, \] and \[ n_{\rm row}:=s_0+R,\qquad n_{\rm pair}:=\binom{n_{\rm row}}2. \tag{4.18}\label{4.18} \] These bounds depend only on the initial data. For the successively adjoined polynomials \(g^{(q)}\), write \(\LM(g^{(q)})=X^{a^{(q)}}\). Lemma~\ref{lem:buch-degree} and complete reduction give \[ |a^{(q)}|\le\delta(q),\qquad a^{(i)}\nleq a^{(q)}\quad(i0$. For $1\le k\le \ell$, maximality of $S$ gives \[ \neg\operatorname{Ind}_I(S\cup\{Y_k\}). \tag{5.5}\label{5.5} \] Apply Lemma~\ref{lem:elim} with retained variables \((S,Y_k)\) and eliminated variables \(Y\setminus\{Y_k\}\). It supplies a first homogeneous row \[ \widetilde h_k(X_0,S,Y_k)\mathrel{:} K[X_0,S,Y_k],\qquad \widetilde h_k\ne0,\qquad \widetilde h_k\mathrel{:}\langle\widetilde F\rangle. \] Dehomogenize its representation at \(X_0=1\). Equation \eqref{4.24} gives \[ \begin{aligned} h_k(S,Y_k) &:=\widetilde h_k(1,S,Y_k) =\sum_{i=0}^{d_k}c_{ki}(S)Y_k^i\\ &=\sum_{j=1}^{m}u_{kj}(S,Y)f_j(S,Y),\\ c_{k,d_k}(S)&\ne0,\qquad d_k\ge1. \end{aligned} \tag{5.6}\label{5.6} \] Indeed, $d_k=0$ would give a nonzero $h_k\mathrel{:} K[S]$ with $h_k\mathrel{:} I$, contradicting \eqref{5.4}. Let \[ L:=K(S), \qquad I_L:=\langle f_1,\ldots,f_m\rangle_{L[Y]}. \tag{5.7}\label{5.7} \] % Here $L$ is the rational-function type: a supplied object is represented % by $f(S)/g(S)$ with $f,g\mathrel{:} K[S]$ and $g\ne0$, with the equality and % arithmetic specified in the finite-setting conventions. Division of \eqref{5.6} by $c_{k,d_k}(S)$ gives \[ \widehat h_k :=c_{k,d_k}(S)^{-1}h_k =Y_k^{d_k}+\sum_{i=0}^{d_k-1}a_{ki}(S)Y_k^i \mathrel{:} I_L. \tag{5.8}\label{5.8} \] These are the finite dimensionality data required by Definition~\ref{def:finite-quotient-type}. In polynomial congruence notation, \[ Y_k^{d_k}\equiv-\sum_{i=0}^{d_k-1}a_{ki}(S)Y_k^i\pmod{I_L}. \tag{5.9}\label{5.9} \] Induction on the exponent of $Y_k$, followed by induction on $k$, reduces every monomial, modulo $I_L$, to an $L$-linear combination of elements of the finite set \[ \mathcal M:= \left\{Y_1^{\alpha_1}\cdots Y_\ell^{\alpha_\ell}: 0\le\alpha_k0$, then $x$ is nilpotent if and only if $x^s=0$. Compute bases for $\ker M_x^j$ for $0\le j\le s$. Each basis is expressible in the next, and any change of dimension increases it by at least one. Either the last kernel has dimension $s$, or two successive kernels already have the same solution type. In the latter case, $M_x^{j+2}v=0$ implies $M_xv\mathrel{:}\ker M_x^{j+1} =\ker M_x^j$, so the next kernel also agrees; finite induction repeats this conclusion up to any supplied exponent $q$. Thus $M_x^q=0$ implies $M_x^s=0$, and applying this to $1$ gives $x^s=0$. For the zero algebra the test is immediate. \end{remark} \section{A primitive element by finite search}\label{sec:primitive} \begin{lemma}[Finite square-free reduction]\label{lem:square-free} Let $u\mathrel{:} L[X_0]$ be nonzero of degree $d$. Finite Euclidean divisions compute a monic polynomial $u_{\mathrm{sf}}$ such that \[ u_{\mathrm{sf}}\mid u, \qquad u\mid u_{\mathrm{sf}}^{d}, \qquad \gcd(u_{\mathrm{sf}},u_{\mathrm{sf}}')=1. \] If $\gcd(u,u')\ne1$, then $\deg u_{\mathrm{sf}}<\deg u$. \end{lemma} \medskip \noindent See Appendix~\ref{app:square-free-proof} for the proof. \begin{theorem}[Finite primitive-element construction] \label{thm:primitive} Let $B\ne0$ be a finite reduced commutative $K$-algebra of dimension $s$, generated by $\beta_1,\ldots,\beta_\ell$. Here the generating data include polynomials $H_i\mathrel{:} K[Y_1,\ldots,Y_\ell]$ expressing the supplied basis vectors as $e_i=H_i(\beta_1,\ldots,\beta_\ell)$. Define \[ E_s:=\frac{s(s-1)}2. \tag{7.1}\label{7.1} \] A search through the finite grid $\{0,1,\ldots,E_s\}^\ell$ finds $c=(c_1,\ldots,c_\ell)$ such that \[ \theta:=\sum_{k=1}^{\ell}c_k\beta_k, \qquad B=K[\theta]. \tag{7.2}\label{7.2} \] % Here $K[\theta]$ denotes the type of values $\varphi(\theta)$ represented in % the coordinates of $B$. Reduction by the minimal polynomial of $\theta$ % allows $\deg\varphi0}$, a finite rational procedure returns $t\mathrel{:} K$ such that \[ \N(p(t))<\eta^2. \tag{8.1}\label{8.1} \] Before receiving $\eta$, the argument determines from the coefficients of $p$ a positive integer $R_p$ such that every resulting point satisfies \[ \sig(t)\le2R_p. \tag{8.2}\label{8.2} \] \end{theorem} \begin{proof} This is the approximation theorem of the first paper~\cite{FiniteFTA}. Its construction chooses $R_p$ by a strict rational leading-term inequality and a mesh by rational variation bounds, then searches a finite rational square grid. The proof uses integer crossing numbers of finite rational polygons and a finite triangulation. The radius depends only on $p$; only the mesh depends on $\eta$. \end{proof} \section{Simultaneous approximation in one parameter} \begin{theorem}[Exact one-parameter divisibility] \label{thm:simultaneous} Let $a\mathrel{:} K^e$ and let $J_a$ be the proper zero-dimensional ideal type in $K[Y_1,\ldots,Y_\ell]$ produced by Theorem~\ref{thm:specialize}. Let \[ A=K[Y]/J_a,\qquad B=A/\Nil(A), \] and let $p,\varphi_1,\ldots,\varphi_\ell$ be produced by Theorem~\ref{thm:primitive}. Then finite univariate division computes $s_j(X_0)\mathrel{:} K[X_0]$ such that \[ f_j(a,\varphi_1(X_0),\ldots,\varphi_\ell(X_0))=p(X_0)s_j(X_0) \qquad(1\le j\le m). \tag{9.1}\label{9.1} \] For every $\varepsilon\mathrel{:}\Q_{>0}$, one call to Theorem~\ref{thm:univariate} produces $t\mathrel{:} K$ satisfying \[ \sum_{j=1}^{m} \N\bigl(f_j(a,\varphi_1(t),\ldots,\varphi_\ell(t))\bigr) <\varepsilon^2. \tag{9.2}\label{9.2} \] \end{theorem} \begin{proof} Let \[ \alpha_k:=[Y_k]\mathrel{:} A, \qquad \beta_k:=[\alpha_k]_{\Nil(A)}\mathrel{:} B. \tag{9.2a}\label{9.2a} \] The definitions of the two quotients give \[ f_j(a,\alpha_1,\ldots,\alpha_\ell)=0\text{ in }A, \qquad f_j(a,\beta_1,\ldots,\beta_\ell)=0\text{ in }B. \tag{9.2b}\label{9.2b} \] Let $\theta$ be the element of Theorem~\ref{thm:primitive}. Then \[ \beta_k=\varphi_k(\theta), \qquad B\cong K[X_0]/\langle p\rangle,\quad [X_0]_{\langle p\rangle}\longmapsto\theta. \tag{9.2c}\label{9.2c} \] Define \[ g_j(X_0):=f_j(a,\varphi_1(X_0),\ldots,\varphi_\ell(X_0))\mathrel{:} K[X_0]. \tag{9.2d}\label{9.2d} \] Equations \eqref{9.2b}--\eqref{9.2c} give $g_j(\theta)=0$. Finite univariate division by the monic polynomial $p$ gives \[ g_j(X_0)=p(X_0)s_j(X_0)+r_j(X_0), \qquad \deg r_j<\deg p=s. \tag{9.2e}\label{9.2e} \] At $X_0=\theta$, one has $r_j(\theta)=0$. Write \[ r_j(X_0)=\sum_{i=0}^{s-1}r_{ji}X_0^i. \] Since $(1,\theta,\ldots,\theta^{s-1})$ is a basis of $B$, \[ \sum_{i=0}^{s-1}r_{ji}\theta^i=0 \Longrightarrow r_{j0}=\cdots=r_{j,s-1}=0. \tag{9.2f}\label{9.2f} \] Thus $r_j=0$, and \eqref{9.2e} is exactly \eqref{9.1}. Compute $R_p$ from Theorem~\ref{thm:univariate}, let $b_{\rm bd,p}:=2R_p$, and define \[ C_j:=\|s_j\|_{b_{\rm bd,p}},\qquad C:=\sum_{j=1}^{m}C_j^2,\qquad \eta_{\rm app}:=\frac{\varepsilon}{1+C}. \tag{9.3}\label{9.3} \] These are explicit nonnegative rationals and $\eta_{\rm app}>0$. Apply Theorem~\ref{thm:univariate} to $(p,\eta_{\rm app})$; the polynomial $p$ is nonconstant because $\deg p=\dim_KB\ge1$. It returns $t$ with $\sig(t)\le b_{\rm bd,p}$ and $\N(p(t))<\eta_{\rm app}^2$. Write \[ s_j(X_0)=\sum_{i=0}^{v_j}u_{ji}X_0^i. \] Then \[ \sig(s_j(t)) \le\sum_{i=0}^{v_j}\sig(u_{ji})\sig(t)^i \le\sum_{i=0}^{v_j}\sig(u_{ji})b_{\rm bd,p}^i =C_j, \qquad \N(s_j(t))\le C_j^2. \tag{9.3a}\label{9.3a} \] Let $\Phi(t):=(\varphi_1(t),\ldots,\varphi_\ell(t))$. Multiplicativity of $\N$ and \eqref{9.1} give \begin{align*} \sum_{j=1}^{m}\N(f_j(a,\Phi(t))) &=\N(p(t))\sum_{j=1}^{m}\N(s_j(t))\cr &\le\eta_{\rm app}^2C =\varepsilon^2\frac{C}{(1+C)^2} <\varepsilon^2, \tag{9.4}\label{9.4} \end{align*} because $(1+C)^2-C=1+C+C^2>0$. \end{proof} \section{Proof of the main theorem}\label{sec:main-proof} \begin{proof}[Proof of Theorem~\ref{thm:main}] Fix the variable order, use degree--lexicographic monomial order, and examine each finite list of critical pairs in lexicographic order. Theorem~\ref{thm:finiteGB}, with $L=K$, computes \[ G=\mathsf A F,\qquad F=\mathsf B G,\qquad \langle F\rangle=\langle G\rangle. \tag{10.1}\label{10.1} \] If $g_\kappa=c\mathrel{:} K^\times$, then \[ 1=c^{-1}g_\kappa =\sum_{j=1}^{m}c^{-1}\mathsf A_{\kappa j}f_j, \tag{10.2}\label{10.2} \] which is alternative \textup{(i)}. If no such $g_\kappa$ occurs, the last assertion of Theorem~\ref{thm:finiteGB} gives \[ \neg(1\mathrel{:}I),\qquad I:=\langle F\rangle. \tag{10.3}\label{10.3} \] Apply Theorem~\ref{thm:specialize}; if no $Y$-variables remain, then all the $f_j$ vanish and $z=0$ gives alternative \textup{(ii)}. Otherwise, Lemma~\ref{lem:standard-basis} and Theorems~\ref{thm:trace-radical}, \ref{thm:primitive}, and \ref{thm:simultaneous} give an approximate common zero in the reordered coordinates; restoring the original variable order yields $z\mathrel{:} K^n$ with $\mathcal N_F(z)<\varepsilon^2$, proving alternative \textup{(ii)}. \end{proof} \appendix \section{The trace pairing and the nilradical} \label{app:trace-pairing-proof} \begin{lemma}[Nilpotent endomorphisms have zero trace] \label{lem:app-nilpotent-trace} If \(\mathsf T^q=0\) on a finite-dimensional \(L\)-vector space \(\mathcal V\), then \[ \Tr(\mathsf T)=0. \tag{A.1}\label{A.1} \] \end{lemma} \begin{proof} The zero-dimensional case is immediate. Otherwise take $q\ge1$ and let $\mathcal K_j$ denote the coordinate solution type of $\mathsf T^jv=0$. Starting with the empty basis $\mathcal B_0$, row reduction successively constructs bases \[ \mathcal B_j=(\mathcal B_{j-1},w_{j,1},\ldots,w_{j,d_j}), \qquad 1\le j\le q. \tag{A.2}\label{A.2} \] Indeed, each vector already in $\mathcal B_{j-1}$ satisfies $\mathsf T^jv=0$; append independent columns from a computed basis of $\ker \mathsf T^j$ until a basis of that solution type is obtained. Since $\mathsf T^q=0$, $\mathcal B_q$ is a basis of $\mathcal V$. For each newly appended vector, the identity $\mathsf T^{j-1}(\mathsf T w_{j,a})=\mathsf T^jw_{j,a}=0$ gives, by row reduction, coefficients such that \[ \mathsf T w_{j,a}=\sum_{b\in\mathcal B_{j-1}}c_{b,j,a}b. \tag{A.3}\label{A.3} \] Thus the image of every vector in the ordered basis $\mathcal B_q$ is a linear combination of earlier basis vectors. Its matrix has zero diagonal. Since trace is the sum of diagonal entries and is invariant under change of basis, this gives \eqref{A.1}. \end{proof} \begin{proof}[Proof of Theorem~\ref{thm:trace-radical}] For $u,v\mathrel{:} A$, associativity gives \[ M_uM_v=M_{uv}, \qquad M_u^q=M_{u^q}. \tag{6.2a}\label{6.2a} \] Suppose $x^q=0$. For every $y\mathrel{:} A$, commutativity and \eqref{6.2a} give \[ (xy)^q=x^qy^q=0, \qquad M_{xy}^q=0. \tag{6.2b}\label{6.2b} \] Lemma~\ref{lem:app-nilpotent-trace}, applied to \(\mathsf T=M_{xy}\), gives \[ \langle x,y\rangle_A=\Tr(M_{xy})=0 \quad(y\mathrel{:} A). \tag{6.2e}\label{6.2e} \] Thus nilpotent elements satisfy the kernel condition in \eqref{6.3}. Conversely, let $x$ be nonnilpotent, and let $n_0:=\dim_KA$. The $n_0+1$ vectors \[ 1,x,x^2,\ldots,x^{n_0} \] are linearly dependent. Let $m_x(X_0)$ be their least-degree monic relation. For every $g\mathrel{:} K[X_0]$, Euclidean division gives \[ g=\mathfrak q_gm_x+r_g,\qquad \deg r_g<\deg m_x, \] and therefore \[ g(x)=0\quad\Longleftrightarrow\quad m_x\mid g. \tag{6.5}\label{6.5} \] Indeed, $g(x)=0$ implies $r_g(x)=0$, and minimality forces $r_g=0$. Extract the highest power of $X_0$ dividing $m_x$: \[ m_x(X_0)=X_0^qh(X_0), \qquad q\ge0,\quad h(0)\ne0,\quad\deg h>0. \tag{6.6}\label{6.6} \] The last inequality follows because $h\mathrel{:} K^\times$ would give $x^q=0$. Since $\gcd(X_0^q,h)=1$, the Euclidean algorithm computes \[ u_0(X_0)X_0^q+v_0(X_0)h(X_0)=1. \tag{6.7}\label{6.7} \] Define \[ g(X_0):=u_0(X_0)X_0^q, \qquad \mathfrak e:=g(x). \tag{6.8}\label{6.8} \] Equation \eqref{6.7} gives \[ X_0^q\mid g,\qquad h\mid(g-1). \] Hence \[ X_0^q\mid(g^2-g),\qquad h\mid(g^2-g), \qquad m_x=X_0^qh\mid(g^2-g), \tag{6.9}\label{6.9} \] where the last divisibility follows from $\gcd(X_0^q,h)=1$. Equation \eqref{6.5} yields \[ \mathfrak e^2=\mathfrak e, \qquad h(x)\mathfrak e=u_0(x)m_x(x)=0. \tag{6.10}\label{6.10} \] If $\mathfrak e=0$, then \eqref{6.5} gives $m_x\mid u_0X_0^q$. Coprimality gives $h\mid u_0$: if \(a_0X_0^q+b_0h=1\), then \[ u_0=a_0u_0X_0^q+b_0u_0h, \] and both terms on the right are divisible by \(h\). Hence both terms on the left of \eqref{6.7} are divisible by $h$. This would give $h\mid1$, contrary to $\deg h>0$. Therefore \[ \mathfrak e^2=\mathfrak e,\qquad \mathfrak e\ne0,\qquad h(x)\mathfrak e=0. \tag{6.11}\label{6.11} \] Since $h(0)\ne0$, one has $\gcd(X_0,h)=1$. Compute \[ w(X_0)X_0+z(X_0)h(X_0)=1 \tag{6.12}\label{6.12} \] and let $b:=w(x)\mathfrak e$. Equations \eqref{6.11}--\eqref{6.12} give \[ xb=w(x)x\mathfrak e=(1-z(x)h(x))\mathfrak e=\mathfrak e. \tag{6.13}\label{6.13} \] The idempotence of $\mathfrak e$ gives, for every $a\mathrel{:} A$, \[ a=\mathfrak e a+(a-\mathfrak e a), \qquad \mathfrak e a\mathrel{:}\operatorname{im}M_{\mathfrak e},\quad a-\mathfrak e a\mathrel{:}\ker M_{\mathfrak e}. \] Moreover, \[ v=\mathfrak e u,\quad M_{\mathfrak e} v=0 \quad\Longrightarrow\quad v=\mathfrak e u=\mathfrak e^2u=\mathfrak e v=0. \] Thus \[ A=\operatorname{im}M_{\mathfrak e}\oplus\ker M_{\mathfrak e}. \tag{6.14}\label{6.14} \] In an adapted basis, \[ M_{\mathfrak e}=\operatorname{diag}(\mathsf I_m,0), \qquad m:=\dim_K\operatorname{im}M_{\mathfrak e}\ge1. \tag{6.15}\label{6.15} \] Here $m\ge1$ because $\mathfrak e=M_{\mathfrak e}(1)\ne0$. Characteristic zero gives \[ \langle x,b\rangle_A =\Tr(M_{xb})=\Tr(M_{\mathfrak e})=m\ne0. \tag{6.16}\label{6.16} \] Writing $b$ in the supplied basis shows that at least one $\langle x,e_j\rangle_A$ is nonzero. Thus a nonnilpotent element does not satisfy the left-hand condition of \eqref{6.3}. Together with \eqref{6.2e}, this proves \eqref{6.3}. For a basis $(e_1,\ldots,e_{n_0})$, define the Gram matrix \[ \Gamma_{ij}:=\Tr(M_{e_ie_j}). \tag{6.17}\label{6.17} \] For $x=\sum_i x_ie_i$, \[ \langle x,e_j\rangle_A =\sum_{i=1}^{n_0}x_i\Gamma_{ij}. \tag{6.18}\label{6.18} \] Hence finite row reduction of $\Gamma$ computes the coordinate kernel in \eqref{6.3}. The nilpotents form an ideal. If $x^q=0$, then $(ax)^q=0$; if additionally $y^\ell=0$, then \[ (x+y)^{q+\ell-1} =\sum_{j=0}^{q+\ell-1}\binom{q+\ell-1}{j}x^jy^{q+\ell-1-j}=0, \tag{6.19}\label{6.19} \] because every summand contains either $x^q$ or $y^\ell$. Its computed $K$-basis therefore also generates it as an ideal type in $A$. Row reduction expresses each product of a basis vector of $A$ with one of these generators in the same ideal basis. Thus \eqref{6.4} is defined by the finite coordinate construction of Proposition~\ref{prop:coordinate-quotient}, whose hypotheses have now been established. Its unit is nonzero because $1$ is not nilpotent. Finally, \[ [x]_{\Nil(A)}^q=0 \Longrightarrow (x^q\mathrel{:}\Nil(A)) \Longrightarrow x^{qv}=0\text{ for some }v \Longrightarrow [x]_{\Nil(A)}=0. \tag{6.20}\label{6.20} \] Therefore $B$ is reduced. For an explicit multiplication table, let $(n_1,\ldots,n_{\ell_{\rm nil}})$ be the basis of $\Nil(A)$ returned by row reduction of $\Gamma$. Extend it, again by row reduction, to a basis \[ (n_1,\ldots,n_{\ell_{\rm nil}},b_1,\ldots,b_s) \quad\text{of }A. \tag{6.21}\label{6.21} \] For each $i,j$, multiplication in $A$ and coordinate reduction in \eqref{6.21} give unique scalars \[ b_ib_j=\sum_{\lambda=1}^{\ell_{\rm nil}}u_{ij}^{\,\lambda}n_\lambda +\sum_{k=1}^{s}d_{ij}^{\,k}b_k. \tag{6.22}\label{6.22} \] Writing $\overline b_k:=[b_k]_{\Nil(A)}$ for the quotient coordinate vector represented by $b_k$ gives \[ B=\bigoplus_{k=1}^{s}K\cdot\overline b_k,\qquad \overline b_i\,\overline b_j =\sum_{k=1}^{s}d_{ij}^{\,k}\overline b_k. \tag{6.23}\label{6.23} \] Thus \eqref{6.17}, \eqref{6.21}, and \eqref{6.22} compute $B$ solely by finite matrix operations. \end{proof} \section{Square-free reduction and the primitive element} \label{app:primitive-details} \subsection{Square-free reduction} \label{app:square-free-proof} \begin{proof}[Proof of Lemma~\ref{lem:square-free}] Normalize $u$ to be monic; this preserves its divisibility relations up to nonzero constant factors. We construct the required polynomial by induction on $d=\deg u$. If $d=0$, take $u_{\mathrm{sf}}=1$. For $d>0$, compute $g=\gcd(u,u')$. If $g=1$, take $u_{\mathrm{sf}}=u$; then $u\mid u^d$. Suppose $g\ne1$. Characteristic zero gives $u'\ne0$, so \[ u=gv,\qquad 0<\deg g