\documentclass[11pt,letterpaper]{article} \usepackage[T1]{fontenc} \usepackage{lmodern} \usepackage[left=.96in,right=.96in,top=.90in,bottom=.91in]{geometry} \usepackage{amsmath,amssymb,amsthm,microtype} \usepackage{xcolor} \usepackage[colorlinks=true,linkcolor=blue!40!black,citecolor=blue!40!black, urlcolor=blue!40!black]{hyperref} \hypersetup{pdftitle={A height criterion for irrational weight-three Lambert sums}, pdfauthor={twinheterodox},pdfsubject={Number theory}} \newtheorem{theorem}{Theorem}[section] \newtheorem{corollary}[theorem]{Corollary} \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \numberwithin{equation}{section} \newcommand{\Q}{\mathbb Q} \newcommand{\Z}{\mathbb Z} \newcommand{\C}{\mathbb C} \newcommand{\lp}{\log^{+}} \title{\fontsize{16}{20}\selectfont A height criterion for irrational weight-three Lambert sums} \author{twinheterodox} \date{18 November 2026} \begin{document} \maketitle \begin{abstract} For an algebraic number \(b\) with \(|b|>1\), the weight-three Lambert sum is \(Z(b)=\sum_{n\ge1}n^2/(b^n-1)\). We prove that \(Z(b)\) lies outside \(\Q(b)\) whenever the Mahler measure \(M(b)\) satisfies \(\log M(b)<(352\pi^2/2889)\,r\log|b|\), where \(r=1\) for a real base and \(r=2\) otherwise. The criterion allows several conjugates outside the unit disk and applies to nonintegral bases. In particular, \(Z(3+\sqrt3)\notin\Q(\sqrt3)\), although both conjugates of the base exceed one. \end{abstract} \section{Two expanding conjugates} For \(|b|>1\), the weight-three Lambert sum \[ Z(b)=\sum_{n=1}^{\infty}\frac{n^2}{b^n-1} \] converges absolutely. With \(q=1/b\), this is a \(q\)-analogue of \(\zeta(3)\). It is irrational at every integer base \(b\ge2\) \cite[Theorem~1.1]{Uplift}. For an algebraic base \(b\), every partial sum lies in \(\Q(b)\). Can we show that the limit lies outside that field? At \(b=2+\sqrt3\) the answer is yes. The other conjugate, \(2-\sqrt3\), lies inside the unit disk, so the earlier algebraic-base theorem gives \(Z(2+\sqrt3)\notin\Q(\sqrt3)\) \cite[Theorem~1.2]{Previous}. Now increase the base by one. The field stays \(\Q(\sqrt3)\) and the other conjugate crosses the unit circle to \(3-\sqrt3>1\). We prove that \(Z(3+\sqrt3)\) still lies outside \(\Q(\sqrt3)\) by quantifying how large the other conjugates can be. For a specified algebraic base \(b\), write its primitive integer minimal polynomial as \(a_0\prod_{j=1}^d(x-b_j)\), with \(a_0>0\). Its \emph{Mahler measure} is \[ M(b)=a_0\prod_{j=1}^d\max(1,|b_j|). \] The product counts each conjugate separately. In particular, a nonreal base and its complex conjugate contribute two factors of \(|b|\). \begin{theorem}\label{thm:height} Let \(b\) be algebraic with \(|b|>1\). Put \(r=1\) for real \(b\) and \(r=2\) for nonreal \(b\). If \begin{equation}\label{eq:height} \frac{\log M(b)}{r\log|b|}<\frac{352\pi^2}{2889}, \end{equation} then \(Z(b)\notin\Q(b)\). \end{theorem} Both roots \(3\pm\sqrt3\) have Mahler measure \(6\). To five decimal places the ratio in \eqref{eq:height} is \(1.15273\) at \(3+\sqrt3\) and \(7.54740\) at \(3-\sqrt3\), against a cutoff of \(1.20253\). When the larger root is the base, the smaller root contributes about \(15.27\%\) as much logarithmic growth as the larger one, within the roughly \(20.25\%\) allowed by the criterion. \begin{corollary}\label{cor:sqrt} At \(b=3+\sqrt3\), the real number \(Z(b)\) lies outside \(\Q(\sqrt3)\). \end{corollary} \begin{proof} The roots of \(x^2-6x+6\) are \(b\) and \(b'=3-\sqrt3\), so \(M(b)=bb'=6\). Since \(\sqrt3>12/7\), we have \(b>33/7\) and \(b'<9/7\). The inequality \(9^6<33\cdot7^5\) gives \((b')^631/10\). \end{proof} The same \(7/6\) bound gives a family of nonintegral examples. Let \(b\) be the larger root of \(2x^2-mx+1\) for an integer \(m\ge129\). Its discriminant \(m^2-8\) lies strictly between \((m-1)^2\) and \(m^2\). The lower bound on \(m\) makes the polynomial negative at \(2^6=64\), so \(2u^2\) and \(v\ge2\). \section{The moments and their determinant}\label{sec:inputs} We recall the polynomials whose values at conjugate bases will be multiplied. Their two arguments have different roles. The first is the base \(t\). The second is a variable \(X\) that can be set equal to the Lambert sum at one embedding and left arbitrary at another. First take \(t>1\) real. The Euler derivative \(\theta=x\,d/dx\) sends \(x^j\) to \(jx^j\), so applying it twice supplies the weight \(j^2\). The denominator comes from summing on a geometric sequence. Put \[ L_t(f)=\sum_{m\ge1}(\theta^2 f)(t^{-m}),\qquad L_t(x^j)=j^2\sum_{m\ge1}t^{-mj}=\frac{j^2}{t^j-1}\quad(j\ge1). \] Also \(L_t(1)=0\). For \(f\) holomorphic near \([0,t^{-1}]\) the sum converges since \(\theta^2f(x)=xf'(x)+x^2f''(x)=O_f(x)\) near zero. To see what happens at a pole, write \(G(y)=\sum_{j\ge1}j^2y^j=y(1+y)/(1-y)^3\) for \(|y|<1\). Summing the geometric expansions of the Lambert series gives \(Z(t)=\sum_{m\ge1}G(t^{-m})\) by absolute convergence. For the first pole \(f(x)=(1-x/t)^{-1}\), differentiating its power series gives \(\theta^2 f(x)=G(x/t)\). The \(m\)th summand of \(L_t(f)\) is therefore \(G(t^{-(m+1)})\), skipping the first term \(G(t^{-1})\). The pole at \(t^a\) skips the first \(a\) terms in the same way, giving \[ L_t\big((1-x/t^a)^{-1}\big)=Z(t)-S_a(t),\qquad S_a(t)=\sum_{m=1}^aG(t^{-m}). \] Now treat \(t\) as an indeterminate. For \(K\ge1\) put \(r_a=t^a\) for \(1\le a\le K\) and \(D(x)=\prod_{a=1}^K(1-x/r_a)\). Division and partial fractions express \(p(x)/D(x)\) over \(\Q(t)\) as a polynomial plus a linear combination of \((1-x/r_a)^{-1}\). Define \(\mathcal L_{t,X}\) by giving the polynomial part the monomial values above and each simple fraction the value \(X-S_a(t)\). For real \(t>1\), setting \(X=Z(t)\) recovers \(L_t\). These moments form the determinant \[ \Delta_{K,t}(X)= \det\left[\mathcal L_{t,X}\!\left(\frac{x^{i+j}}{D(x)}\right)\right]_{0\le i,j0\). The following hold. \begin{enumerate} \item \(P_K\in\Z[t,X]\), \(\deg_XP_K\le K\), and \(d_K\le\gamma K^3+O(K^2\log(2K))\). \item For fixed \(|b|>1\), \(\log|P_K(b,Z(b))|\le-\varepsilon K^3\log|b|+ O_b(K^2\log(2K))\), with \(\log 0=-\infty\). \item For each fixed \(X_0\in\C\), \(\max_{|t|\le1}\lp|P_K(t,X_0)|=O_{X_0}(K^2\log^3(2K))\). \item For each fixed \(|b|>1\) there are arbitrarily large \(K\) for which \(P_K(b,Z(b))\ne0\). \end{enumerate} \end{proposition} The construction and the cyclotomic exponents come from the integer-base weight-three determinant of uplift-universal-inc \cite[\S\S1--5 and Appendix~A]{Uplift}. The cubic scale of decay for Hankel determinants on geometric nodes already occurs in Zudilin's work on the generalized \(q\)-logarithm \cite[\S4, (8)]{Zudilin}. The estimates in Proposition~\ref{prop:inputs} use the complex-base form of the weight-three determinant, including its bound on the closed disk for every fixed second argument. \section{An integer from all the conjugates}\label{sec:places} Take \(b=3+\sqrt3\) and \(b'=3-\sqrt3\). If \(Z(b)=u+v\sqrt3\) for rational \(u,v\), its field conjugate would be \(Y=u-v\sqrt3\). The two factors in the norm of \(P_K(b,Z(b))\) would then be \(P_K(b,Z(b))\) and \(P_K(b',Y)\). The series converges at both roots, yet conjugation can change the behavior of a limit. For example, it sends the sequence \((2-\sqrt3)^n\to0\) to \((2+\sqrt3)^n\to\infty\). To cover the unknown \(Y\), we need a bound at \(b'\) for every fixed second argument. Proposition~\ref{prop:inputs}(2) gives decay for the first factor. For the second, the degree in the \emph{base} controls growth from the unit circle out to \(b'\). The degree bound \(\deg_tP_K\le d_K\) from Lemma~\ref{lem:degree} suffices. For fixed \(Y\), the reversed polynomial \(u^{d_K}P_K(1/u,Y)\) is then holomorphic at the origin. Apply the maximum modulus principle to it at \(u=1/t\) for \(|t|>1\). Its modulus on the unit circle is bounded by Proposition~\ref{prop:inputs}(3). Together with that same estimate inside the disk this gives, for every \(t\in\C\), \begin{equation}\label{eq:arch} \lp|P_K(t,Y)|\le d_K\lp |t|+O_Y(K^2\log^3(2K)). \end{equation} Thus the logarithmic cost at \(b'\) is at most \(d_K\log b'+o(K^3)\). For a general base, the same degree controls the denominators. \begin{proof}[Proof of Theorem~\ref{thm:height}, assuming Lemma~\ref{lem:degree}] Suppose \(\xi=Z(b)\) belongs to \(\Q(b)\). With \(d=[\Q(b):\Q]\), we may write \(\xi=g(b)/h\) for a fixed \(g\in\Z[t]\) of degree at most \(d-1\) and a positive integer \(h\). This representation specifies the conjugate value \(g(b_j)/h\) at \emph{every} root \(b_j\) of the minimal polynomial \(f\). Since \(\deg_X P_K\le K\), \[ Q_K(t)=h^K P_K\big(t,g(t)/h\big)\in\Z[t],\qquad \deg Q_K\le d_K+(d-1)K . \] Choose the unbounded sequence on which \(Q_K(b)=h^KP_K(b,\xi)\ne0\). The irreducible \(f\) then has no common root with \(Q_K\). Their resultant is a nonzero integer, as it is the determinant of their integer Sylvester matrix. Its product expression is \begin{equation}\label{eq:res} R_K=\operatorname{Res}(f,Q_K) =a_0^{\deg Q_K}h^{dK}\prod_{j=1}^d P_K\big(b_j,g(b_j)/h\big). \end{equation} Thus \(a_0\), which may exceed one for a nonintegral base, is raised only to the degree of \(Q_K\). The fixed denominator \(h\) contributes only \(O_{b,\xi}(K)\) to \(\log|R_K|\). At \(b\) apply Proposition~\ref{prop:inputs}(2). If \(b\) is nonreal, \(g(\bar b)/h=\bar\xi=Z(\bar b)\), giving a second factor with the same decay. At each of the other \(d-r\) roots use \eqref{eq:arch} with the fixed value \(Y=g(b_j)/h\). In \eqref{eq:res} the leading coefficient \(a_0\) and these remaining roots together contribute at most \(d_K(\log M(b)-r\log|b|)+O_{b,\xi}(K^2\log^3(2K))\) to the logarithm. Consequently \begin{align*} 0\le\log|R_K| &\le -r\varepsilon K^3\log|b| +d_K\big(\log M(b)-r\log|b|\big)+O_{b,\xi}(K^2\log^3(2K))\\ &\le \left(\gamma\log M(b)-\tfrac{11}{18}r\log|b|\right)K^3+o(K^3). \end{align*} Here \(\log M(b)-r\log|b|\ge0\), so the upper bound for \(d_K\) applies. Since \((11/18)/\gamma=352\pi^2/2889\), \eqref{eq:height} makes the last coefficient strictly negative. \end{proof} \section{The degree in the base}\label{sec:degree} \begin{lemma}\label{lem:degree} For every \(K\ge1\), the degree of \(P_K(t,X)\) in \(t\) is \(d_K\), and its coefficient of \(t^{d_K}\) is \((-1)^{K(K-1)/2}X^K\). \end{lemma} The remaining task is to see why the determinant contributes no additional power of \(t\) after division by \(t^{e_K}\). We let \(t\) tend to \(+\infty\) at fixed \(K\). The nodes \(r_a=t^a\) separate, and the polynomials that interpolate at them make the dependence on \(X\) diagonal. \begin{proof} Take real \(t\ge2\) and write \(L=L_t=\mathcal L_{t,Z(t)}\) as in Section~\ref{sec:inputs}. Changing \(Z(t)\) to \(Y\) adds \(Y-Z(t)\) at each simple fraction and leaves the polynomial moments fixed. The coefficient of the simple fraction at \(r_a\) in \(p/D\) is \(w_a p(r_a)\), where \[ w_a=\prod_{c\ne a}(1-r_a/r_c)^{-1}. \] This follows by multiplying by \(1-x/r_a\) and evaluating at \(r_a\). Thus the change in the functional on \(p/D\) is \((Y-Z(t))\sum_a w_a p(r_a)\), even when division gives a polynomial quotient. Choose the degree-\((K-1)\) cardinal polynomials \(\ell_a(x)=w_aD(x)/(1-x/r_a)\), characterized by \(\ell_a(r_c)=\delta_{ac}\). In the matrix with entries \(\mathcal L_{t,Y}(\ell_a\ell_c/D)\) the preceding change contributes only to the diagonal. The full matrix is \[ (Y-Z(t))W+WHW,\qquad W=\operatorname{diag}(w_1,\ldots,w_K),\qquad H_{ac}=L\!\left(\frac{D(x)}{(1-x/r_a)(1-x/r_c)}\right). \] This also covers \(a=c\), when the fraction retains a simple pole. The evaluation matrix from monomials to node values is \(V=(r_a^i)_{1\le a\le K,\,0\le i