\documentclass[11pt]{amsart} \pdfsuppressptexinfo=-1 \usepackage[margin=1.1in]{geometry} \usepackage[T1]{fontenc} \usepackage{lmodern} \usepackage{amsmath,amssymb,amsthm,mathtools} \usepackage{booktabs,enumitem,array} \usepackage{placeins} \usepackage[hidelinks]{hyperref} \hypersetup{pdftitle={A lower bound for the smallest limit point of the absolute Mahler measure of totally positive algebraic integers}, pdfauthor={Tejasvi Singh Tomar}} \newtheorem{theorem}{Theorem}[section] \newtheorem{lemma}[theorem]{Lemma} \newtheorem{proposition}[theorem]{Proposition} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \numberwithin{equation}{section} \setlength{\emergencystretch}{2em} \raggedbottom \newcommand{\R}{\mathbb{R}} \newcommand{\Z}{\mathbb{Z}} \newcommand{\C}{\mathbb{C}} \newcommand{\Qbar}{\overline{\mathbb{Q}}} \newcommand{\logp}{\log^{+}} \newcommand{\Res}{\operatorname{Res}} \newcommand{\disc}{\operatorname{disc}} \newcommand{\tr}{\operatorname{tr}} % --------------------------------------------------------------------------------------------- % All values of the certificate in one place (fixed at the freeze). \newcommand{\Mzero}{\tfrac{19152077587359311629}{35184372088832000000}} \newcommand{\MzeroDec}{0.5443347841} \newcommand{\EMzero}{1.7234615} \newcommand{\EMzeroLong}{1.72346152} \newcommand{\HalfM}{0.2721673} \newcommand{\EHalfM}{1.3128067} \newcommand{\NPoly}{19} \newcommand{\NNodes}{5{,}473} \newcommand{\Sval}{0.05288541} \newcommand{\DegSum}{0.70100096} \newcommand{\Budget}{0.80677179} \newcommand{\Kappa}{0.19322820} \newcommand{\Kconst}{4.88} \newcommand{\GmaxRatio}{2.44} \newcommand{\EffLam}{0.0529} \newcommand{\EffC}{3.28} \newcommand{\EffOmega}{1.72334} \newcommand{\Ival}{2.8781306519\ldots\cdot10^{-11}} \newcommand{\SuppLo}{0.08525} \newcommand{\SuppHi}{11.2} \newcommand{\NZero}{401} \newcommand{\XLO}{1/1000} \newcommand{\XHI}{881} \newcommand{\LTail}{0.790} \newcommand{\RTail}{0.3866} \newcommand{\NWinA}{199} \newcommand{\NInitA}{5{,}568} \newcommand{\NChkA}{28{,}408} \newcommand{\MinEndA}{2.884\cdot10^{-6}} \newcommand{\TimeA}{474} \newcommand{\ProcA}{64} \newcommand{\XLOB}{1/32} \newcommand{\XHIB}{112/5} \newcommand{\LTailB}{0.1253} \newcommand{\RTailB}{0.0564} \newcommand{\NWinB}{198} \newcommand{\NInitB}{5{,}708} \newcommand{\NChkB}{8{,}424} \newcommand{\MinCertB}{1.57\cdot10^{-9}} \newcommand{\TimeB}{18} \newcommand{\ProcB}{32} \newcommand{\Margin}{3\cdot10^{-6}} \newcommand{\MDual}{0.54433778} \newcommand{\PrimalVal}{0.5443536} \newcommand{\NoEnergyVal}{0.543251} \newcommand{\NoEnergyOmega}{1.72159} % Earlier bounds. \newcommand{\SmyVal}{1.7171} \newcommand{\FlamNinetySix}{1.720566} \newcommand{\MuWuVal}{1.721916} \newcommand{\FlamVal}{1.722069} \newcommand{\FlamHalf}{0.271763} \newcommand{\SmyEll}{1.31427} \newcommand{\SmyEllSq}{1.7273} \newcommand{\SmyEllUp}{1.7274} \newcommand{\SchinzelHalf}{0.2406059} % Certificate file and archive (fixed at the freeze). \newcommand{\CertFile}{tp\_certificate.json} \newcommand{\CertSha}{9f1e51157fbe30ce8ab436c8552d7d6a9b5aaaa5483449de00ec68716eac3864} \newcommand{\ShaChkA}{de4434350b21bbe8e831ced9729a8592f6496f671a256d18285004a28b4cd4ec} \newcommand{\ShaChkB}{33bce343517e35399626348d0e07b690215576cd81bdc0d58b98b01f97e0f5ad} \newcommand{\SuppDOI}{10.5281/zenodo.23715959} \def\RepoCommit{} \IfFileExists{commit.tex}{\input{commit.tex}}{} \newcommand{\RepoLink}{\url{https://github.com/tejasvi/totally-positive-mahler-measure/tree/v4}\ifx\RepoCommit\empty\else\ (commit \texttt{\RepoCommit})\fi} % Lean formalization (native track). \newcommand{\lean}[1]{\texttt{\detokenize{#1}}} \newcommand{\NLeanAux}{47} \newcommand{\NLeanLeaves}{8{,}815} \newcommand{\NLeanBlocks}{43} \newcommand{\NLeanEnergy}{19{,}445} \newcommand{\KCoreHours}{240} \newcommand{\LeanVersion}{v4.35.0-rc3} \newcommand{\MathlibCommit}{c55e6e786f49} % --------------------------------------------------------------------------------------------- \hyphenation{name-spaces} \begin{document} \title[Totally positive algebraic integers]{A lower bound for the smallest limit point of the absolute Mahler measure of totally positive algebraic integers} \author{Tejasvi Singh Tomar} \renewcommand{\urladdrname}{{\itshape ORCID}} \urladdr{\href{https://orcid.org/0000-0003-4668-3639}{https://orcid.org/0000-0003-4668-3639}} \subjclass[2020]{11R06, 11R80; 11G50, 31A15, 65G20, 68V20} \keywords{Mahler measure, Weil height, totally positive algebraic integers, auxiliary functions, logarithmic energy, discriminant, formal verification} \begin{abstract} Let $\ell_+$ be the smallest limit point of the absolute Mahler measures $\Omega(\alpha)=M(\alpha)^{1/\deg\alpha}$ of totally positive algebraic integers~$\alpha$. By work of Smyth, $\ell_+\le\SmyEllUp$, and the best lower bound was Flammang's $\ell_+\ge\FlamVal$. We prove that all but finitely many totally positive algebraic numbers $\alpha$ satisfy $\Omega(\alpha)>\EMzero$, and that $\ell_+>\EMzero$. We also prove that all but finitely many totally real algebraic numbers $\beta$ have absolute logarithmic Weil height $\log\Omega(\beta)>\HalfM$. The proof adds to Smyth's method of auxiliary functions the logarithmic energy term that Orloski, Talebizadeh Sardari and Smith derived from the integrality of discriminants for the Schur--Siegel--Smyth trace problem. The resulting inequality is verified rigorously by computer, and the lower bounds are also proved in the Lean proof assistant. \end{abstract} \maketitle \section{Introduction}\label{sec:intro} Let $\alpha$ be an algebraic number of degree $n$, with minimal polynomial $P(x)=a\prod_{i=1}^n(x-\alpha_i)$ in $\Z[x]$, primitive and with leading coefficient $a\ge1$. The Mahler measure and the absolute Mahler measure of $\alpha$ are \[ M(\alpha)=a\prod_{i=1}^n\max\bigl(1,|\alpha_i|\bigr),\qquad \Omega(\alpha)=M(\alpha)^{1/n}, \] and $h(\alpha)=\log\Omega(\alpha)$ is the absolute logarithmic Weil height of $\alpha$. We call $\alpha$ totally positive if all its conjugates $\alpha_i$ are real and positive. Let $\mathcal L_+$ be the set of the values $\Omega(\alpha)$, where $\alpha$ runs through the totally positive algebraic integers, and let $\ell_+$ be the smallest limit point of $\mathcal L_+$. Schinzel \cite{Sch73} proved that $\Omega(\alpha)\ge(1+\sqrt5)/2$ for every totally positive algebraic integer $\alpha\ne0,1$, with equality for the roots of $x^2-3x+1$. Smyth \cite[Theorems~1 and~2]{Smy80} proved that the values $\Omega(\beta)$, for totally real algebraic integers $\beta$, have $\ell=\SmyEll\ldots$ as a limit point and are dense in $(\ell,\infty)$. If $\beta\ne0$ is a totally real algebraic integer, then $\beta^2$ is a totally positive algebraic integer and $\Omega(\beta^2)=\Omega(\beta)^2$. Hence every number in $[\ell^2,\infty)$ is a limit point of $\mathcal L_+$, and $\ell_+\le\ell^2=\SmyEllSq\ldots$. Lower bounds for $\ell_+$ come from the method of auxiliary functions. Suppose that \begin{equation}\label{eq:smyth} \logp x\ \ge\ m+\sum_{j}c_j\log|Q_j(x)|\qquad\text{for all }x>0, \end{equation} where $\logp x=\log\max(1,x)$, the $Q_j$ are finitely many nonzero integer polynomials and $c_j\ge0$. If $\alpha$ is a totally positive algebraic integer with $Q_j(\alpha)\ne0$ for all $j$, then each product $\prod_iQ_j(\alpha_i)$ is a nonzero integer, since it is a resultant, and averaging \eqref{eq:smyth} over the conjugates of $\alpha$ gives $\log\Omega(\alpha)\ge m$. Smyth \cite{Smy81} found such an inequality by linear programming and checked it in interval arithmetic. He stated the result for totally real algebraic integers $\beta$, as $\Omega(\beta)>1.31040$ when $\beta^2$ is not in an explicit finite set, and his inequality gives $\Omega(\alpha)>1.31040^2>\SmyVal$ for all totally positive algebraic integers $\alpha$ outside that set \cite[(6) and (9)]{Smy81}. Flammang raised the bound to $\FlamNinetySix$ \cite{Fla96}, Mu and Wu to $\MuWuVal$ \cite{MW13}, and Flammang to $\FlamVal$ \cite[Theorem~1]{Fla15}. In each case the bound holds for all but finitely many totally positive algebraic integers, so it is a lower bound for $\ell_+$. In \cite{Fla15} the exceptions are the roots of seven listed polynomials, two of which, $Q_6$ and $Q_9$ in our Table~\ref{tab:polys}, are reducible. Flammang conjectures that the next point of $\mathcal L_+$ is $1.722325$, attained at the roots of $x^6-12x^5+44x^4-67x^3+44x^2-12x+1$. \begin{theorem}\label{thm:main} All but finitely many totally positive algebraic numbers $\alpha$ satisfy $\Omega(\alpha)>\EMzero$. More precisely, for every $\varepsilon>0$, all but finitely many totally positive algebraic numbers $\alpha$ satisfy $\log\Omega(\alpha)\ge m_0-\varepsilon$, and $\ell_+\ge e^{m_0}>\EMzero$, where $m_0=\MzeroDec\ldots$. \end{theorem} Here $m_0$ is the rational number~\eqref{eq:m0}, the constant $m$ of the auxiliary inequality that we verify in Section~\ref{sec:cert}. With Smyth's upper bound, Theorem~\ref{thm:main} gives $\EMzero<\ell_+\le\ell^2=\SmyEllSq\ldots$. The bound for every $\varepsilon>0$ in Theorem~\ref{thm:main} is equivalent to saying that the essential minimum $\sup\{H:\{\alpha\text{ totally positive}:\Omega(\alpha)\le H\}\text{ is finite}\}$ is at least $e^{m_0}$. Applied to the squares of totally real algebraic numbers, Theorem~\ref{thm:main} gives the following. \begin{corollary}\label{cor:real} For every $\varepsilon>0$, all but finitely many totally real algebraic numbers $\beta$ satisfy $h(\beta)\ge m_0/2-\varepsilon$. In particular, all but finitely many totally real algebraic numbers $\beta$ satisfy $h(\beta)>\HalfM$ and $\Omega(\beta)>\EHalfM$. \end{corollary} \begin{proof} Let $\varepsilon>0$, and let $\beta\ne0$ be a totally real algebraic number with $h(\beta)0 \] for some $\lambda\ge0$ and some $\nu$ that gives mass at most $Kr$ to every disc of radius $r\le1$, for a constant $K$. A measure on $\R$ with a bounded density has this property, and for $\lambda=0$ the inequality is \eqref{eq:smyth}. Lemma~\ref{lem:aux} below proves that this suffices, with an explicit error term for each degree. The leading coefficient of $\alpha$ enters through $M(\alpha)$, the resultants and the discriminant, and it is absorbed because $\sum_jc_j\deg Q_j+2\lambda\le1$, a condition that the auxiliary inequality itself forces. This is why Theorem~\ref{thm:main} holds for all totally positive algebraic numbers, not only for integers. Orloski, Talebizadeh Sardari and Smith \cite{OSS25} introduced this energy condition for the Schur--Siegel--Smyth trace problem, the lower limit of $\tr(\alpha)/\deg(\alpha)$ over totally positive algebraic integers, where the left side is $x$ instead of $\logp x$. For that problem, Smyth's method \cite{Smy84} and its refinements gave a sequence of improved lower bounds. Serre showed, in a letter published as an appendix to \cite{AP08}, that inequalities with $x$ on the left and only polynomial terms on the right cannot hold with a constant larger than $1.89830\ldots$, and Smith \cite{Smi24} later proved that the trace constant is less than $1.89831$. Discriminants entered the trace problem already in Siegel's work \cite{Sie45}. Orloski and Talebizadeh Sardari \cite{OT23} proved that every weak* limit $\mu$ of the uniform measures on the conjugates of a sequence of distinct algebraic integers, all of whose conjugates lie in a fixed compact set, satisfies $\int\log|Q(x_1,\dots,x_n)|\,d\mu(x_1)\cdots d\mu(x_n)\ge0$ for every nonzero $Q\in\Z[x_1,\dots,x_n]$ \cite[Section~1.1]{OSS25}. For $Q=x_1-x_2$ this says that the energy of $\mu$ is nonnegative. For the Mahler measure of totally positive algebraic integers, Rhin \cite{Rhi04} proved a version of Schinzel's inequality that improves as the discriminant grows \cite[Section~9.2]{Smy08}. Our auxiliary function uses the polynomial $x$ and the reciprocal polynomials $x^dU_j(x+1/x-2)$, where $d=\deg U_j$, built from eighteen of the twenty-seven polynomials $U_j$ of Flammang's table \cite[Table~1]{Fla15}. The reference measure $\nu$ has a continuous piecewise linear density with rational nodes, so that its potential and energy have closed forms (Lemma~\ref{lem:closed}) and the auxiliary inequality can be verified on $(0,\infty)$ in ball arithmetic \cite{Joh17} (Section~\ref{sec:cert}); the certificate, the programs and the Lean files are archived at \href{https://doi.org/\SuppDOI}{doi:\SuppDOI}. Numerically, without the energy term the best auxiliary function with these nineteen polynomials falls short of Flammang's bound (Section~\ref{sec:energy}). Section~\ref{sec:aux} proves the lower bound from an auxiliary function with an energy term. Section~\ref{sec:cert} describes the certificate and its verification and proves Theorem~\ref{thm:main}. Section~\ref{sec:lean} describes a proof in Lean of Theorem~\ref{thm:main} and Corollary~\ref{cor:real}. Section~\ref{sec:open} lists open problems, and Appendix~\ref{sec:obs} explains how the certificate was found, and collects numerical observations. \section{Auxiliary functions with an energy term}\label{sec:aux} For finite positive Borel measures $\theta,\eta$ on $\C$ with compact support we write \[ U_\eta(z)=\int\log|z-y|\,d\eta(y),\qquad I(\theta,\eta)=\int U_\eta\,d\theta,\qquad I(\theta)=I(\theta,\theta). \] Since $\log|z-y|$ is bounded above on compact sets, these integrals exist in $[-\infty,\infty)$, and $I(\theta,\eta)=I(\eta,\theta)$ by Tonelli's theorem. We write $D(y,r)=\{z\in\C:|z-y|0$ we have $-2\log t=\int_0^\infty(e^{-ut^2}-e^{-u})\,\frac{du}u$, and the integrand has constant sign, so its absolute value integrates to $2|\log t|$. A measure with an atom has energy $-\infty$, so $\theta$ and $\eta$ have no atoms, and the diagonal $\{z=w\}$ is null for $|\sigma|\times|\sigma|$. By Fubini's theorem and $\sigma(\C)=0$, \[ -2\iint\log|z-w|\,d\sigma(z)\,d\sigma(w)=\int_0^\infty\Bigl(\iint e^{-u|z-w|^2}\,d\sigma(z)\,d\sigma(w)\Bigr)\frac{du}u\ \ge\ 0, \] since the Gaussian kernel $e^{-u|z-w|^2}$ is positive definite on $\C=\R^2$ (its Fourier transform is positive). The left side is $-2\bigl(I(\theta)-2I(\theta,\eta)+I(\eta)\bigr)$ by bilinearity. \end{proof} \begin{lemma}[Auxiliary function with an energy term]\label{lem:aux} Let $Q_1,\dots,Q_J\in\Z[x]$ be nonzero, $c_1,\dots,c_J>0$, $\lambda\ge0$ and $m\in\R$. Let $\nu$ be a probability measure on $\C$ with compact support, and let $K>0$ be such that $\nu(D(y,r))\le Kr$ for all $y\in\C$ and $00$ only finitely many totally positive algebraic numbers $\alpha$ satisfy $\log\Omega(\alpha)\le m-\varepsilon$. \end{corollary} \begin{proof} Choose $n_0\ge1/(2K)$ such that $\lambda(1+\log(2Kn))/n<\varepsilon$ for all $n\ge n_0$. If $\alpha$ is totally positive and $\log\Omega(\alpha)\le m-\varepsilon$, then either $\alpha$ is one of the finitely many roots of the polynomials $Q_j$, or $\alpha$ has degree less than $n_0$ by \eqref{eq:eff}. In the second case $h(\alpha)\le m$, and by Northcott's theorem \cite[Theorem~1.6.8]{BG06} there are only finitely many such $\alpha$. \end{proof} \begin{remark}\label{rem:flammang} For $\lambda=0$, Lemma~\ref{lem:aux} is the classical argument of \cite{Smy81}, and its conclusion is then $\log\Omega(\alpha)\ge m$ for every totally positive algebraic number $\alpha$ that is not a root of one of the $Q_j$, because the measure $\nu$ plays no role. For instance, the uniform measure on $[0,1]$ satisfies the hypothesis with $K=2$, and then every $n\ge1$ is admissible. Flammang \cite[Section~2.1]{Fla15} uses auxiliary functions $\logp x-c_0\log x-\sum_jc_j\log|Q_j(x)|$ with reciprocal $Q_j$ and $2c_0+\sum_jc_j\deg Q_j=1$, which are invariant under $x\mapsto1/x$. Lemma~\ref{lem:aux} with $\lambda=0$ applies to them, so bounds obtained with such functions, which have no energy term, also hold for all totally positive algebraic numbers outside a finite set. \end{remark} \begin{remark}\label{rem:dual} For a fixed finite set of polynomials, the best bound that Lemma~\ref{lem:aux} can give is limited by a problem on measures. Let $\theta$ be a probability measure on $(0,\infty)$ with compact support, with $I(\theta)\ge0$ and $\int\log|Q_k|\,d\theta\ge0$ for all $k$. The condition $I(\theta)\ge0$ excludes atoms. If \eqref{eq:F} holds, then integrating it against $\theta$ and using Lemma~\ref{lem:neg} gives \[ \int\logp x\,d\theta\ \ge\ m+\lambda\bigl(2I(\theta,\nu)-I(\nu)\bigr)\ \ge\ m+\lambda I(\theta)\ \ge\ m. \] So every value $m$ that Lemma~\ref{lem:aux} gives with these polynomials is at most the infimum of $\int\logp x\,d\theta$ over such measures. For heights defined by a continuous Green function on $\C$, Burgos Gil, Menares, Qu and Sombra \cite[Theorem~D]{BMQS26} proved that the best bound obtainable from auxiliary functions with arbitrary finite sets of integer polynomials equals the corresponding infimum over measures, which is the essential minimum of the height. Their setting includes neither the restriction to totally positive numbers nor an energy term (see Section~\ref{sec:open}). \end{remark} \section{The certificate}\label{sec:cert} \subsection{The polynomials} The certificate uses the set $\mathcal Q$ of the $\NPoly$ polynomials $Q_0,\dots,Q_{18}$ of Table~\ref{tab:polys}. For $1\le k\le18$, the polynomial $Q_k$ is the reciprocal polynomial $Q_k(x)=x^{d}U(x+1/x-2)$ of degree $2d$, where $U$ is the polynomial of degree $d$ listed in \cite[Table~1]{Fla15} under the index $j$ given in the table. The weights $c_k$ are dyadic rationals. The polynomial $Q_1=(x-1)^2$ vanishes at the corner $x=1$ of $\logp x$, so the auxiliary function tends to $+\infty$ there. The polynomials $Q_6$, $Q_9$ and $Q_{15}$ are products of a polynomial and its reciprocal, and $Q_{15}$ is the only $Q_k$ whose leading coefficient is not $1$. All roots of the $Q_k$ are real and nonnegative, except for two pairs of complex roots of $Q_{16}$. \begin{table}[ht] \centering \caption{The polynomials and their weights, rounded to four significant digits; the certificate in the archive gives the exact weights. For $1\le k\le18$, $Q_k(x)=x^{\deg U}U(x+1/x-2)$, where $U$ is the polynomial $U_j$ of \cite[Table~1]{Fla15}, given by its coefficients from the leading one down to the constant term.}\label{tab:polys} \small \begin{tabular}{@{}rrp{0.58\textwidth}rl@{}} \toprule $k$ & $j$ & $U_j$, or $Q_0$ & $\deg Q_k$ & $c_k$\\ \midrule 0 & -- & $Q_0=x$ & 1 & $1.932\cdot10^{-1}$\\ 1 & 1 & $1,0$ & 2 & $1.558\cdot10^{-1}$\\ 2 & 2 & $1,-1$ & 2 & $3.860\cdot10^{-2}$\\ 3 & 3 & $1,-2$ & 2 & $4.143\cdot10^{-7}$\\ 4 & 4 & $1,-3,1$ & 4 & $1.295\cdot10^{-2}$\\ 5 & 5 & $1,-4,1$ & 4 & $2.372\cdot10^{-4}$\\ 6 & 6 & $1,-5,6,-1$ & 6 & $1.686\cdot10^{-3}$\\ 7 & 7 & $1,-6,5,-1$ & 6 & $5.191\cdot10^{-4}$\\ 8 & 8 & $1,-7,13,-7,1$ & 8 & $4.143\cdot10^{-3}$\\ 9 & 10 & $1,-7,14,-8,1$ & 8 & $7.517\cdot10^{-4}$\\ 10 & 11 & $1,-8,14,-7,1$ & 8 & $9.505\cdot10^{-5}$\\ 11 & 13 & $1,-13,61,-131,136,-66,14,-1$ & 14 & $1.237\cdot10^{-5}$\\ 12 & 14 & $1,-13,61,-132,138,-67,14,-1$ & 14 & $1.946\cdot10^{-6}$\\ 13 & 16 & $1,-13,62,-137,147,-73,15,-1$ & 14 & $8.026\cdot10^{-7}$\\ 14 & 18 & $1,-15,83,-220,303,-220,83,-15,1$ & 16 & $8.049\cdot10^{-4}$\\ 15 & 20 & $2,-26,128,-308,391,-265,94,-16,1$ & 16 & $3.204\cdot10^{-7}$\\ 16 & 21 & $1,-21,182,-853,2386,-4151,4545,-3109,$ $1298,-315,40,-2$ & 22 & $4.052\cdot10^{-7}$\\ 17 & 24 & $1,-25,266,-1582,5817,-13835,21682,$ $-22449,15233,-6667,1833,-302,27,-1$ & 26 & $3.650\cdot10^{-7}$\\ 18 & 25 & $1,-31,413,-3141,15261,-50187,115410,$ $-189036,222621,-189036,115410,-50187,$ $15261,-3141,413,-31,1$ & 32 & $5.251\cdot10^{-8}$\\ \bottomrule \end{tabular} \end{table} The roots of the $Q_k$, which Lemma~\ref{lem:aux} excludes, include totally positive algebraic integers with $\Omega(\alpha)0$ that is not a root of one of the polynomials $Q_k$. \end{proposition} Both programs use Arb ball arithmetic \cite{Joh17} through python-flint \cite{pyflint}. In ball arithmetic every computed quantity is an interval, given by a midpoint and a radius, that is guaranteed to contain the exact value. Both programs read the data as exact rationals, check that they have the required form (distinct polynomials with nonzero leading coefficients, among them $x$ with a positive weight, nonnegative weights and values $G_k$, increasing nodes, $G_0=G_N=0$ and $s>0$), and compute $\sum_kc_k\deg Q_k+2s$ exactly. Its value is $\Budget\ldots<1$. All nineteen weights are positive, as Lemma~\ref{lem:aux} requires. Apart from the certified root isolation and factorization of FLINT \cite{FLINT}, acceptance depends only on exact comparisons of rationals and on the positivity of lower ends of balls. Heuristic choices, such as window widths, cut points and limits on bisection, therefore affect the running time but not the validity of an accepted bound. The first program checks $\sum_iC_i=\sum_iC_ie_i=0$ exactly, evaluates $U_\rho$ and $I(\rho)$ by Lemma~\ref{lem:closed}, and uses three kinds of bound. On an interval $(0,x_{\mathrm{lo}}]$ it bounds $F$ from below by a positive constant. This bound uses $\logp x=0$, the inequality $-c_0\log x\ge-c_0\log x_{\mathrm{lo}}$, upper bounds for $|Q_k|$ with $k\ge1$, and the inequality $U_\rho(x)\le U_\rho(0)$, which holds because $U_\rho$ decreases to the left of the support of $G$. A Sturm sequence shows that no $Q_k$ with $k\ge1$ vanishes on $(0,x_{\mathrm{lo}}]$. On an interval $[x_{\mathrm{hi}},\infty)$ the bounds $|Q_k(x)|\le\|Q_k\|_1x^{\deg Q_k}$, where $\|Q_k\|_1$ is the sum of the absolute values of the coefficients, and $U_\rho(x)\le s\log x$ give $F(x)\ge\kappa\log x-b$. Here $\kappa=1-\sum_kc_k\deg Q_k-2s>0$ and $b=m_0-I(\rho)/s+\sum_kc_k\log\|Q_k\|_1$, and $\kappa\log x_{\mathrm{hi}}-b>0$. On $[x_{\mathrm{lo}},x_{\mathrm{hi}}]$ the program puts each root of a $Q_k$ in a short window $W$. The roots are located by the certified root isolation of FLINT and confirmed by Sturm counts and exact signs. Let $x_c$ be the rational centre of $W$. On $W$, each $\log|Q_k|$ is bounded from above, roots included, by the logarithm of $|Q_k(x_c)|+\frac{|W|}2\max_W|Q_k'|$, and $U_\rho$ is bounded from above by $U_\rho(x_c)+\frac{|W|}2\max_W|U_\rho'|$. The function $\logp x$ is bounded from below on $W$ by its value at the left end of $W$. The nodes, the windows and $1$ cut the rest into intervals $[p,q]$. On each of them the program applies Lemma~\ref{lem:chord} below to $F$, with a rigorous upper bound for $\int_p^q(F'')_+$, and it checks that the windows and the intervals cover $[x_{\mathrm{lo}},x_{\mathrm{hi}}]$. \begin{lemma}\label{lem:chord} Let $f$ be continuously differentiable on $[p,q]$ with $f'$ absolutely continuous, and let $\delta=q-p$. Then for $p\le x\le q$, \[ f(x)\ \ge\ \min\bigl(f(p),f(q)\bigr)-\frac\delta4\int_p^q\bigl(f''(t)\bigr)_+\,dt. \] \end{lemma} \begin{proof} Let $\Lambda$ be the linear function with $\Lambda(p)=f(p)$ and $\Lambda(q)=f(q)$. Then $f(x)=\Lambda(x)-\int_p^qk(x,t)f''(t)\,dt$ with $k(x,t)=(\min(x,t)-p)(q-\max(x,t))/\delta$. Since $0\le k\le\delta/4$, the integral is at most $\frac\delta4\int_p^q(f'')_+$, and $\Lambda(x)\ge\min(f(p),f(q))$. \end{proof} On each interval $[p,q]$ of the first program, $F$ satisfies the hypotheses of Lemma~\ref{lem:chord}. Indeed, $[p,q]$ contains no root of a $Q_k$, and $1$ and the nodes are not interior points, so $\logp x$ and the polynomial terms are smooth on $[p,q]$. The function $U_\rho'(x)=-\sum_iC_iH'(e_i-x)$ is absolutely continuous, since $H'(t)=t\log|t|-t$ is an indefinite integral of the locally integrable function $\log|t|$. A second program checks Proposition~\ref{prop:check} in a different way, with direct bounds near the irrational roots of the $Q_k$ and second-order Taylor bounds instead of Lemma~\ref{lem:chord}; it shares no code with the first and was written from a description of the certificate format, which also gives the closed forms of Lemma~\ref{lem:closed} and the tail bounds of the first program. Table~\ref{tab:checks} lists the data of the certificate and the smallest lower bounds for $F$ that the two programs obtain, and the archive describes both programs in more detail, with their cut points and other parameters. \begin{table}[ht] \centering \caption{The certificate and the smallest lower bounds for $F$ in the two checks.}\label{tab:checks} \small \begin{tabular}{@{}p{0.27\textwidth}p{0.33\textwidth}p{0.33\textwidth}@{}} \toprule Bound $m_0$ & \multicolumn{2}{l}{\eqref{eq:m0}}\\ $\sum_kc_k\deg Q_k$; $s$ & \multicolumn{2}{l}{$\DegSum\ldots$; $\Sval\ldots$}\\ $\sum_kc_k\deg Q_k+2s$ & \multicolumn{2}{l}{$\Budget\ldots<1$}\\ Nodes; $G=0$ outside; $K$ & \multicolumn{2}{l}{$\NNodes$ in $[0,\SuppHi]$; $[\SuppLo,\SuppHi]$; $\Kconst$}\\ $I(\rho)$ & \multicolumn{2}{l}{$\Ival$}\\ \midrule & first program & second program\\ \midrule Smallest bound & $\MinEndA$ (lower bound for $F$ at the ends of the intervals) & $\MinCertB$ (lower bound on an interval)\\ \bottomrule \end{tabular} \end{table} \begin{proof}[Proof of Theorem~\ref{thm:main}] By Proposition~\ref{prop:check}, the function \eqref{eq:Fcert} is nonnegative on $(0,\infty)$ outside the roots of the $Q_k$. It is the function \eqref{eq:F} with $\nu=\rho/s$, $\lambda=s$ and $m=m_0$, and $\nu$ satisfies the hypothesis of Lemma~\ref{lem:aux} with $K=\Kconst$. The bound for every $\varepsilon>0$ follows from Corollary~\ref{cor:fin}. Since $e^{m_0}=\EMzeroLong\ldots$, Corollary~\ref{cor:fin} with $\varepsilon=m_0-\log\EMzero>0$ shows that only finitely many totally positive algebraic numbers satisfy $\Omega(\alpha)\le\EMzero$. Finally, by Corollary~\ref{cor:fin}, for every $\varepsilon>0$ only finitely many values $\Omega(\alpha)$ with $\alpha$ a totally positive algebraic integer lie below $e^{m_0-\varepsilon}$, so $\ell_+\ge e^{m_0}$. \end{proof} \begin{remark}\label{rem:eff} The proof is effective. With $\lambda=s<\EffLam$ and $K=\Kconst$, inequality \eqref{eq:eff} gives \[ \log\Omega(\alpha)\ \ge\ m_0-\frac{\EffLam\,(\EffC+\log n)}n \] for every totally positive algebraic number $\alpha$ of degree $n$ that is not a root of one of the $Q_k$. For instance $\Omega(\alpha)>\EffOmega$ for $n\ge10^4$. \end{remark} \section{Formal verification}\label{sec:lean} We also prove Theorem~\ref{thm:main} and Corollary~\ref{cor:real} in the Lean~4 proof assistant \cite{Lean4} with the Mathlib library \cite{Mathlib}. The formal statements are built from Mathlib's notions of irreducibility, integrality, complex roots, Mahler measure and accumulation point. An integer polynomial $P$ is a minimal polynomial of $\alpha\in\C$ if it is irreducible in $\Z[x]$, has positive degree and vanishes at $\alpha$; then $\Omega(\alpha)=M(P)^{1/\deg P}$, where $M(P)$ is Mathlib's Mahler measure, which equals $|a|\prod_i\max(1,|\alpha_i|)$. The number $\alpha$ is totally positive if all complex roots of $P$ are real and positive, and totally real if they are all real. Formally, we prove the first two assertions of Theorem~\ref{thm:main} as stated, as the propositions \lean{TpOmega} and \lean{TpMain}, with $m_0$ the rational number~\eqref{eq:m0}. We prove the third as \lean{TpLimitPoint} and \lean{ExpM0Gt}: every accumulation point $y$ of $\mathcal L_+$, in Mathlib's sense, satisfies $y\ge e^{m_0}$, and $e^{m_0}>\EMzero$; this form does not presuppose that $\ell_+$ exists. The three assertions of Corollary~\ref{cor:real} are \lean{CorReal}, \lean{CorRealDec} and \lean{CorRealOmega}. These propositions can be read without the proofs, and a check after the build confirms that each formal theorem proves the proposition of the same name. The formal theorems have no hypotheses. Their proofs cover the resultant and discriminant bounds \eqref{eq:res} and \eqref{eq:disc}, the smoothing of the conjugates over small circles and the energy inequality of the proof of Lemma~\ref{lem:aux}, with the instance of Lemma~\ref{lem:neg} that it uses, the averaging over the conjugates with $r=1/(2Kn)$, Corollary~\ref{cor:fin} with Northcott's theorem in the form of Mathlib's finiteness of integer polynomials of bounded degree and Mahler measure, Lemma~\ref{lem:closed}, the reduction of Theorem~\ref{thm:main} to Proposition~\ref{prop:check}, and Proposition~\ref{prop:check} itself, with the help of the checks described next. Corollary~\ref{cor:real} is derived from Theorem~\ref{thm:main} applied to $\beta^2$, as in Section~\ref{sec:intro}. For Proposition~\ref{prop:check}, a checking program written in Lean's core language works with integers only and computes logarithms with proved error bounds. It checks that a list of intervals, computed in advance, tiles $[2^{-30},21]$, and it bounds $F$ from below on most of them by a variant of Lemma~\ref{lem:chord} and on short windows around the roots of the $Q_k$ term by term. It also checks the two tails $(0,2^{-30}]$ and $[21,\infty)$, a lower bound for $I(\rho)$ and the inequality $\sum_kc_k\deg Q_k+2s<1$. A Lean proof shows that whenever the program accepts, the conclusion of Proposition~\ref{prop:check} holds. Every check is decided by Lean's kernel, so that the trusted base is Lean's kernel, including its built-in arithmetic on natural numbers, and all the formal theorems depend only on the standard axioms. The archive prints the propositions, gives the versions of Lean and Mathlib, the build commands and further checks of the proof, and describes a second version of the proof of Theorem~\ref{thm:main}, in which the checks are evaluated as compiled code. \section{Open problems}\label{sec:open} We end with three questions that the results leave open. \begin{enumerate}[leftmargin=2em] \item Is $\ell_+=\ell^2$, the upper bound that Smyth's construction gives? \item Which totally positive algebraic integers have $\Omega(\alpha)