Mahler's Problem on Liouville Numbers I. Analytic Rigidity
Diego Marques
math.NT
Sep 13, 2026 · v5
TL;DR
Theorems 1.1 and 1.3, the analytic rigidity results answering Mahler's question, are formalized and machine-checked in Lean 4; the code is in a public GitHub repository.
Abstract
In 1984, Mahler asked whether a transcendental entire function can map the set of Liouville numbers $\mathscr{L}$ into itself. We answer this question in the negative by proving a local rigidity theorem for real-analytic functions. More precisely, if $U\subset\mathbb{R}$ is an open interval and $f:U\to\mathbb{R}$ is real-analytic with $f(\mathscr{L}\cap U)\subseteq\mathscr{L}$, then $f$ is the restriction to $U$ of a rational function in $\mathbb{R}(x)$. Quantitatively, there exists an absolute constant $τ>2$ such that, for every nonrational real-analytic function $f$ and every nonempty open subinterval $V\subset U$, the set of $ξ\in V\cap\mathscr{L}$ satisfying $μ(f(ξ))\leqτ$ contains a Cantor set. The proof combines a two-height counting estimate near analytic graphs with a nested-interval construction.
Problem
Mahler asked in 1984 whether a transcendental entire function can map the set of Liouville numbers into itself. The question was recorded by Waldschmidt and Bugeaud and stayed open for four decades.
Approach
The author proves a local rigidity theorem: a real-analytic function on an open interval that preserves Liouville numbers must be a rational function. The proof uses Wronskian localization to obtain uniform sublevel bounds and anisotropic determinant estimates. These combine into a two-height counting estimate for rational points near analytic graphs, followed by a nested-interval construction that produces a Cantor set. The main theorems are formalized in Lean 4.
Results
Mahler's question has a negative answer. For any nonrational real-analytic function f and any subinterval V, there is a Cantor set of Liouville numbers in V on which the irrationality exponent of f(ξ) is at most τ = 100.