Short Synchronizing Words for Random Automata

TL;DR

This paper proves that a random automaton with n states on a 2-letter alphabet almost surely has a short synchronizing word of length proportional to the square root of n times a logarithmic factor, improving previous bounds.

Contribution

It introduces the concept of w-trees and shows that random automata are w-trees for some short word, leading to a new upper bound on synchronizing word length.

Findings

Short synchronizing words exist with high probability.

Random automata are w-trees for some short word.

Improves the upper bound from O(n log^3 n) to O(√n log n).

Abstract

We prove that a uniformly random automaton with $n$ states on a 2-letter alphabet has a synchronizing word of length $O(n^{1/2}\log n)$ with high probability (w.h.p.). That is to say, w.h.p. there exists a word $\omega$ of such length, and a state $v_0$, such that $\omega$ sends all states to $v_0$. Prior to this work, the best upper bound was the quasilinear bound $O(n\log^3n)$ due to Nicaud (2016). The correct scaling exponent had been subject to various estimates by other authors between $0.5$ and $0.56$ based on numerical simulations, and our result confirms that the smallest one indeed gives a valid upper bound (with a log factor). Our proof introduces the concept of $w$-trees, for a word $w$, that is, automata in which the $w$-transitions induce a (loop-rooted) tree. We prove a strong structure result that says that, w.h.p., a random automaton on $n$ states is a $w$-tree for some word $w$ of length at most $(1+\epsilon)\log_2(n)$, for any $\epsilon>0$. The existence of the (random) word $w$ is proved by the probabilistic method. This structure result is key to proving that a short synchronizing word exists.