Conway's cosmological theorem and automata theory
Pierre Lairez, Aleksandr Storozhenko
TL;DR
The paper presents an automata-theoretic, computational proof of Conway's cosmological theorem for the audioactive (look-and-say) sequence. By modeling the derivation with a small set of transducers (Audio, Multi, Mark, Scissors) and constructing split- and atom-focused automata (Splitting, Atom, AtomicF), the authors prove stabilization of the atom set at day $n\ge 24$ via a transducer composition argument and minimization. The approach yields explicit finite-state encodings (e.g., 21-state Splitting, 26-state Atom, and a manageable AtomicF ∘ Audio^n ∘ Src family) and demonstrates that $E_{24}=E_{25}$, hence $E_n=E_{24}$ for all $n\ge 24$, with the full atom list detailed in the appendix. This work illustrates how automata theory and experimental mathematics can provide a transparent, computer-assisted pathway to deep combinatorial results in number theory.
Abstract
John Conway proved that every audioactive sequence (a.k.a. look-and-say) decays into a compound of 94~elements, a statement he termed the cosmological theorem. The underlying audioactive process can be modeled by a finite-state machine, mapping one sequence of integers to another. Leveraging automata theory, we propose a new proof of Conway's theorem based on a few simple machines, using a computer to compose and minimize them.
