Ordering regular languages: a danger zone

TL;DR

This paper explores Wheeler automata and languages, showing their computational advantages and limitations, and establishing complexity bounds for related decision problems in automata theory.

Contribution

It provides upper bounds for problem complexity in Wheeler automata and demonstrates hardness results when generalizing beyond Wheeler constraints.

Findings

Polynomial-time algorithms for Wheeler automata problems

Deciding Wheeler automaton membership is polynomial

Generalizations lead to NP-complete or PSPACE-complete problems

Abstract

Ordering the collection of states of a given automaton starting from an order of the underlying alphabet is a natural move towards a computational treatment of the language accepted by the automaton. Along this path, Wheeler \emph{graphs} have been recently introduced as an extension/adaptation of the Burrows-Wheeler Transform (the now famous BWT, originally defined on strings) to graphs. These graphs constitute an important data-structure for languages, since they allow a very efficient storage mechanism for the transition function of an automaton, while providing a fast support to all sorts of substring queries. This is possible as a consequence of a property -- the so-called \emph{path coherence} -- valid on Wheeler graphs and consisting in an ordering on nodes that "propagates" to (collections of) strings. By looking at a Wheeler graph as an automaton, the ordering on strings corresponds to the co-lexicographic order of the words entering each state. This leads naturally to consider the class of regular languages accepted by Wheeler automata, i.e. the Wheeler languages. It has been shown that, as opposed to the general case, the classic determinization by powerset construction is polynomial on Wheeler languages. As a consequence, most of the classical problems turn out to be "easy" -- that is, solvable in polynomial time -- on Wheeler languages. Moreover, deciding whether a DFA is Wheeler and deciding whether a DFA accepts a Wheeler language is polynomial. Our contribution here is to put an upper bound to easy problems. For instance, whenever we generalize by switching to general NFAs or by not fixing an order of the underlying alphabet, the above mentioned problems become "hard" -- that is NP-complete or even PSPACE-complete.