\documentclass[11pt]{amsart} \usepackage{amsmath} \usepackage{amssymb} \usepackage{amsthm} \usepackage{hyperref} \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \theoremstyle{definition} \newtheorem{definition}[theorem]{Definition} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \numberwithin{equation}{section} \setlength{\emergencystretch}{2em} \newcommand{\NN}{\mathbb{N}} \newcommand{\ZZ}{\mathbb{Z}} \newcommand{\QQ}{\mathbb{Q}} \newcommand{\CC}{\mathbb{C}} \newcommand{\FF}{\mathbb{F}} \newcommand{\PP}{\mathbb{P}} \newcommand{\Qbar}{\overline{\QQ}} \newcommand{\cL}{\mathcal{L}} \newcommand{\tL}{\widetilde{\mathcal{L}}} \newcommand{\cA}{\mathcal{A}} \newcommand{\Sym}{\operatorname{Sym}} \newcommand{\supp}{\operatorname{supp}} \newcommand{\spn}{\operatorname{span}} % Labels and scope follow cycle2/RESIDUES.md, R-S1-16..21, 27 (whole-chain refereed, REFEREE_CHAIN.md): % strong-six input = Roy 1992 Cor. 2 (= DALAG Cor. 11.16), constants in Qbar, real logarithms (other branches: Remark % 4.4, proof written out in the 2026-11-19 referee pass, papers/thmB/REFEREE.md). \begin{document} \hypersetup{pdftitle={t to the power log t is transcendental for almost all integers t},pdfauthor={Shawn Krisman}} \title[$t^{\log t}$ for almost all integers]{$t^{\log t}$ is transcendental\\ for almost all integers $t$} \author{Shawn Krisman} \subjclass[2020]{Primary 11J81; Secondary 11J86, 11B75, 11N25} \keywords{six exponentials theorem, linear forms in logarithms, multiplicative Sidon sets, transcendence, $\alpha^{\log\alpha}$} \begin{abstract} Let $\tL=\Qbar+\Qbar\cL$ be the $\Qbar$-vector space spanned by $1$ and the logarithms of nonzero algebraic numbers. We prove that \[ \#\{2\le t\le X:\ (\log t)^2\in\tL\}\le\pi(X)+O(X^{3/4}). \] Hence $t^{\log t}=e^{(\log t)^2}$ is transcendental for all integers $t\ge2$ outside a set of density zero, and no relation $(\log t-B)(\log t-A)=C'+\gamma\nu$ with $A,B,C',\gamma$ algebraic and $\nu$ a logarithm of an algebraic number holds outside that set. The proof uses Roy's strong six exponentials theorem only through configurations of exponent vectors in dimensions $2$ and $3$, where it makes the primitive exceptional integers a multiplicative Sidon set, and then Erd\H{o}s's bound for such sets. The prime numbers show that the bound $\pi(X)$ is the limit of the method. We also show that the configurations of exponent vectors allowed by the strong six exponentials theorem can be much larger than in dimensions $2$ and $3$: there are such sets of $2^r$ points in $\PP^{n-1}$ with $n=1+r(r+1)/2$. No individual number such as $2^{\log2}$ is decided. \end{abstract} \maketitle % =========================================================================== \section{Introduction} % =========================================================================== Let $\cL=\{\lambda\in\CC:e^\lambda\in\Qbar^\times\}$ be the set of logarithms of nonzero algebraic numbers, and let \[ \tL=\Qbar+\Qbar\cL \] be the $\Qbar$-vector space spanned by $1$ and $\cL$; its elements are the \emph{linear forms in logarithms} of Roy \cite{Roy92}. For an integer $t\ge2$, $\log t$ denotes the real logarithm. Whether $2^{\log 2}=e^{(\log2)^2}$, or more generally $\alpha^{\log\alpha}$, is transcendental is open; Waldschmidt lists it as a consequence of the strong five exponentials conjecture \cite[\S3, after Conj.~3.5]{WODP}. Diaz conjectured that $\lambda^2\notin\tL$ for every $\lambda\in\tL\setminus\Qbar$ (Conjecture (P3) of \cite[\S1]{Di07}, a consequence of the strong four exponentials conjecture) and remarked that he knew no $\lambda$ for which this is proved \cite[\S1]{Di07}. We show that the conjecture holds for $\lambda=\log t$ for almost all integers $t$. \begin{theorem}\label{thm:main} For $X\ge2$, \[ E(X):=\#\{2\le t\le X:\ (\log t)^2\in\tL\}\le\pi(X)+O(X^{3/4}). \] \end{theorem} The theorem is unconditional. Its inputs are Roy's strong six exponentials theorem \cite[Cor.~2]{Roy92}, Baker's theorem \cite{Ba66} and Erd\H{o}s's bound for multiplicative Sidon sets \cite{Erd38}. With Erd\H{o}s's later sharpening \cite{Erd69}, \cite[(4)]{Erd69b} the error term becomes $O(X^{3/4}(\log X)^{-3/2})$ (Remark~\ref{rem:error}). \begin{corollary}\label{cor:main} Let $\mathcal E\subset\ZZ_{\ge2}$ be the set of $t$ with $(\log t)^2\in\tL$, so that $\#(\mathcal E\cap[2,X])\le(1+o(1))X/\log X$. For every integer $t\ge2$ with $t\notin\mathcal E$: \begin{enumerate} \item $t^{\log t}=e^{(\log t)^2}$ is transcendental; \item $e^{(\log t)^2/\gamma}\notin\Qbar^\times e^{\beta+\delta\log t}$ for all $\gamma\in\Qbar^\times$ and $\beta,\delta\in\Qbar$; \item there are no $A,B,C',\gamma\in\Qbar$ and $\nu\in\cL$ with $(\log t-B)(\log t-A)=C'+\gamma\nu$. \end{enumerate} \end{corollary} \begin{proof} If $t^{\log t}=\alpha\in\Qbar$, then $\alpha\ne0$ and $(\log t)^2\in\cL$. If $e^{\lambda^2/\gamma}=\alpha e^{\beta+\delta\lambda}$ with $\lambda=\log t$, then $\lambda^2/\gamma-\beta-\delta\lambda\in\cL$, so $\lambda^2\in\tL$. In (3), $\lambda^2=C'-AB+(A+B)\lambda+\gamma\nu\in\tL$. In each case $t\in\mathcal E$. The bound on $\mathcal E$ is Theorem~\ref{thm:main} with the prime number theorem. \end{proof} In (3) the exceptional set $\mathcal E$ does not depend on $A,B,C',\gamma,\nu$: the relation fails off $\mathcal E$ for all admissible constants at once, and $\nu$ is not restricted in height. The hypothesis that the constants are algebraic cannot be dropped: with complex constants, (3) holds at every $t$ (take $\nu=\log2$ and $\gamma=((\log t-B)(\log t-A)-C')/\log2$). \subsection*{The method} Write $\log t=\sum_ia_i\log p_i$ with the exponent vector $a(t)\in\ZZ_{\ge0}^n$. The strong six exponentials theorem, together with Baker's theorem, puts linear-algebra constraints on the set of exponent rays $[a]$ with $\lambda_a^2\in\tL$ (Section~\ref{sec:consistency}): no three of them lie in a plane, and no four in a three-dimensional subspace. These two facts say exactly that the integers with primitive exponent vector and $(\log t)^2\in\tL$ form a multiplicative Sidon set: all products $t_it_j$ are distinct (Section~\ref{sec:proof}). Erd\H{o}s \cite{Erd38} proved that such a set in $[1,X]$ has at most $\pi(X)+O(X^{3/4})$ elements. The primes themselves are compatible with all the constraints the method can see (Section~\ref{sec:limits}), so $\pi(X)$ is the limit of the method: an improvement has to use more than these constraints, for instance information about individual numbers such as $p^{\log p}$. In dimension $\ge4$ the constraints allow much larger configurations: Proposition~\ref{prop:toric} gives $2^r$ admissible rays in dimension $1+r(r+1)/2$. This is why the proof counts through the multiplicative structure of the integers rather than by bounding the number of exceptional rays among $y$-smooth integers. \subsection*{Related work} The six exponentials theorem was stated explicitly and proved independently by Lang \cite[Ch.~II]{Lan66} (see also \cite{Lan66b}) and Ramachandra \cite{Ram68}; for its earlier history see \cite[\S1.3]{W00}. Roy \cite{Roy92} proved the strong form used here, after Waldschmidt's version with coefficients in $\Qbar+\cL$ \cite[Cor.~2.1]{W88}; see \cite[\S\S1.3, 11.6]{W00}. The six exponentials theorem already says something about infinitely many integers at once: if $\beta$ is irrational, then the integers $t\ge2$ with $t^\beta$ algebraic generate a multiplicative group of rank at most $2$, and of the numbers $2^\pi,3^\pi,5^\pi,\dots$ at most two are algebraic \cite[Cor.~1, Cor.~3]{Lan66b}. There the exponent $\beta$ is fixed; in $t^{\log t}$ it moves with $t$. The consequences of the strong six exponentials theorem that we know of are statements about a fixed small number of numbers. The closest to ours is Waldschmidt's \cite[Cor.~2.4]{Wa05var}, restated by Diaz \cite[Cor.~2\,(P), point~1]{Di07}: if $\lambda_1,\lambda_2,\lambda_3\in\tL\setminus\Qbar$ and $1,\lambda_2,\lambda_3$ are linearly independent over $\Qbar$, then $\lambda_1\lambda_2$ and $\lambda_1\lambda_3$ do not both lie in $\tL$. Lemma~\ref{lem:roy} below is this corollary written in our notation, with Waldschmidt's proof. With $\lambda_1=\lambda_2=\log s$, $\lambda_3=\log t$ and $(\log st)^2=(\log s)^2+2\log s\log t+(\log t)^2$, it shows that $(\log s)^2$, $(\log t)^2$ and $(\log st)^2$ do not all lie in $\tL$ when $s,t\ge2$ are multiplicatively independent integers; this is the dimension-$2$ case of Lemma~\ref{lem:small} for three integers. The dimension-$3$ case and the passage to a count over all integers do not appear in \cite{Wa05var} or \cite{Di07}. Counts of exceptions along an integer parameter, obtained by combining the six exponentials theorem with a combinatorial lemma, are also known for the numbers $2^{\pi^k}$. By the six exponentials theorem one of $2^\pi,2^{\pi^2},2^{\pi^3}$ is transcendental. As reported in Waldschmidt's survey \cite[\S4(a)]{W13}, Srinivasan \cite{Sri74,Sri79} showed that at most $O(\sqrt N)$ of the numbers $2^{\pi^k}$, $1\le k\le N$, are algebraic, and the implied constant was bounded by $2$ by Ramachandra and Srinivasan \cite{RS83} and by $\sqrt2$ by Balasubramanian and Ramachandra \cite{BR82} (announced in \cite[\S2]{RS83}). In \cite[\S2]{RS83} the six exponentials theorem shows that every nonzero difference occurs at most twice among the exceptional exponents $k$, and an elementary counting lemma gives the bound. Our argument has the same structure: a transcendence theorem forces a Sidon-type property of the exceptional set, and an extremal bound for such sets finishes the proof; here the property is multiplicative and the bound is Erd\H{o}s's. We could not find Theorem~\ref{thm:main}, or a density statement for $t^{\log t}$, in the sources we checked. % =========================================================================== \section{The inputs}\label{sec:inputs} % =========================================================================== \begin{theorem}[Roy {\cite[Cor.~2]{Roy92}}; see {\cite[Cor.~11.16]{W00}}]\label{thm:roy} Let $M$ be a $2\times3$ matrix with entries in $\tL$. If the rows of $M$ are linearly independent over $\Qbar$ and the columns of $M$ are linearly independent over $\Qbar$, then $M$ has rank $2$. \end{theorem} Equivalently \cite[p.~23]{Roy92}: if $x_1,x_2\in\CC$ and $y_1,y_2,y_3\in\CC$ are each linearly independent over $\Qbar$, then one of the six products $x_iy_j$ is not in $\tL$. Waldschmidt's earlier form \cite[Cor.~2.1]{W88}, for matrices with entries in $\Qbar+\cL$, does not suffice below: the hypothesis $(\log t)^2\in\tL$ of Theorem~\ref{thm:main} puts the entries $\lambda_v\lambda_u$ of the matrix in the proof of Lemma~\ref{lem:roy} in $\tL$, not in $\Qbar+\cL$. \begin{theorem}[Baker \cite{Ba66}]\label{thm:baker} If $\lambda_1,\dots,\lambda_n\in\cL$ are linearly independent over $\QQ$, then $1,\lambda_1,\dots,\lambda_n$ are linearly independent over $\Qbar$. \end{theorem} \begin{theorem}[Erd\H{o}s {\cite[\S2]{Erd38}}]\label{thm:erdos} If $a_1<\dotsm$ or $j>m$. If $v_i\ne0$ for some $i>m$, then taking $j=i$ gives $u_i=0$, and then $v_iu_j=-u_iv_j=0$ for all $j$, so $u=0$. Hence if $K_S(v)\neq0$, then $v\in W'$. In that case pick $j\le m$ with $v_j\ne0$; for $i>m$ we get $u_iv_j=-v_iu_j=0$, so $u\in W'$. \end{proof} In particular $c(n)$ is nondecreasing in $n$, and the consistency of a set of points depends only on the points, not on the ambient space. \begin{lemma}\label{lem:small} Let $T$ be a consistent set. \begin{enumerate} \item No three points of $T$ lie on a projective line. \item No four points of $T$ lie in a projective plane. \end{enumerate} Hence $c(2)=2$ and $c(3)=3$. \end{lemma} \begin{proof} (1) Let $[w_1],[w_2],[w_3]$ be distinct points of $\PP(W')$, $\dim W'=2$. In a basis of $W'$, write $w_i=(1,s_i)$ or $(0,1)$; the matrices $w_iw_i^{\mathsf T}$ are linearly independent (Vandermonde), so they span $\Sym^2W'$, which contains $w_1\circ W'$. (2) Let $[w_1],\dots,[w_4]$ be distinct points of $\PP(W')$, $\dim W'\le3$. By (1) we may assume that $\dim W'=3$ and that no three are collinear. Then $w_1,w_2,w_3$ is a basis of $W'$ and $w_4=aw_1+bw_2+cw_3$ with $abc\ne0$; replacing $w_1,w_2,w_3$ by $aw_1,bw_2,cw_3$ (the same points), in this basis the points are $e_1,e_2,e_3$ and $(1,1,1)$. The span of $e_1e_1^{\mathsf T},e_2e_2^{\mathsf T},e_3e_3^{\mathsf T}$ and the all-ones matrix is \[ S_4=\{B\in\Sym_3:B_{12}=B_{13}=B_{23}\}. \] Take $v=(0,1,1)$. For $u\in\Qbar^3$ the off-diagonal entries of $v\circ u$ are $\frac12u_1$, $\frac12u_1$ and $\frac12(u_2+u_3)$, so $v\circ u\in S_4$ for all $u$ in the plane $U=\{u_1=u_2+u_3\}$. Thus $S_4\supset v\circ U$ with $\dim U=2$. For the values of $c$: frames are consistent (Lemma~\ref{lem:frame}), so $c(n)\ge n$, and (1), (2) give the upper bounds for $n=2,3$. \end{proof} The proof of (2) is the general one in coordinates: four points in general position in $\PP^2$ lie on a pencil of conics, the pencil contains a singular member, and its vertex $v$ gives $v\circ U\subset S$ with $\dim U=2$. \begin{corollary}\label{cor:explicit} At least one of the numbers $(\log2)^2,(\log3)^2,(\log6)^2$, and at least one of the numbers $(\log2)^2,(\log3)^2,(\log5)^2,(\log30)^2$, is not in $\tL$. In particular one of $2^{\log2},3^{\log3},6^{\log6}$ and one of $2^{\log2},3^{\log3},5^{\log5},30^{\log30}$ is transcendental. \end{corollary} \begin{proof} Take $\ell=(\log2,\log3,\log5)$. The exponent vectors are $e_1,e_2,e_1+e_2$ (three points on a line) and $e_1,e_2,e_3,e_1+e_2+e_3$ (four points in a plane). If all numbers in one group were in $\tL$, these points would lie in $Z$, so they would form a consistent set by Lemma~\ref{lem:roy}, against Lemma~\ref{lem:small}. \end{proof} % =========================================================================== \section{Proof of Theorem~\ref{thm:main}}\label{sec:proof} % =========================================================================== Let $X\ge2$ and $n=\pi(X)$. For $1\le t\le X$ let $a(t)\in\ZZ_{\ge0}^n$ be its exponent vector, so that $\log t=\lambda_{a(t)}$. Call $t\ge2$ \emph{exceptional} if $(\log t)^2\in\tL$, that is, $[a(t)]\in Z$. This is a property of the ray, since $\lambda_{ka}^2=k^2\lambda_a^2$. Call $t$ \emph{primitive} if the gcd of the entries of $a(t)$ is $1$, and let $F_0$ be the set of primitive exceptional $t\le X$. Two distinct primitive vectors in $\ZZ_{\ge0}^n\setminus\{0\}$ are not proportional over $\Qbar$ (a proportionality factor would be a positive rational, hence $1$), so distinct elements of $F_0$ give distinct points of $Z$. \begin{lemma}\label{lem:sidon} If $t_1,t_2,t_3,t_4\in F_0$ and $t_1t_2=t_3t_4$, then $\{t_1,t_2\}=\{t_3,t_4\}$ as multisets. \end{lemma} \begin{proof} If the two sides share an element, cancelling it gives the claim. So assume $\{t_1,t_2\}\cap\{t_3,t_4\}=\emptyset$ and write $a_i=a(t_i)$. If $t_1=t_2$ and $t_3=t_4$, then $t_1=t_3$, which is excluded. If $t_1=t_2$ and $t_3\ne t_4$, then $2a_1=a_3+a_4$, so the three distinct points $[a_1],[a_3],[a_4]$ of $Z$ lie on a line, against Lemma~\ref{lem:small}(1). (The case $t_3=t_4$, $t_1\ne t_2$ is symmetric.) Otherwise the four points $[a_i]$ are distinct and $a_1+a_2=a_3+a_4$, so they lie in a projective plane, against Lemma~\ref{lem:small}. \end{proof} \begin{proof}[Proof of Theorem~\ref{thm:main}] By Lemma~\ref{lem:sidon} the products $t_it_j$ of elements of $F_0\subset[2,X]$ are pairwise distinct, so $|F_0|\le\pi(X)+O(X^{3/4})$ by Theorem~\ref{thm:erdos}. An exceptional $t\le X$ that is not primitive is a perfect power $t_0^k$ with $k\ge2$, and there are at most $\sum_{2\le k\le\log_2X}X^{1/k}=O(\sqrt X)$ of these. Hence $E(X)\le\pi(X)+O(X^{3/4})$. \end{proof} \begin{remark}\label{rem:error} With the upper bound of \cite{Erd69} (see \cite[(4)]{Erd69b}) in place of Theorem~\ref{thm:erdos}, the same proof gives $E(X)\le\pi(X)+O(X^{3/4}(\log X)^{-3/2})$, since the perfect powers contribute only $O(\sqrt X)$. \end{remark} \begin{remark} Only Lemma~\ref{lem:small}, that is, consistency in dimensions $2$ and $3$, enters the proof. Part (1) alone does not give a bound $\pi(X)+o(\pi(X))$. It says that no three elements of $F_0$ have linearly dependent exponent vectors, and the integers $pq\le X$ with primes $p