A Brief History of History-Determinism (Invited Talk)

Most nondeterministic automata models are more expressive, or at least more succinct, than their deterministic counterparts; however, this comes at a cost, as deterministic automata tend to have better algorithmic properties. History-deterministic automata are an intermediate model that allows a restricted form of nondeterminism: all nondeterministic choices must be resolvable on-the-fly, with only the knowledge of the word prefix read so far - as opposed to general nondeterminism, which allows for guessing the future of the word. History-deterministic automata combine some of the algorithmic benefits of determinism with some of the increased power of nondeterminism, thus enjoying (some of) the best of both worlds. History-determinism, as it is understood today, has its roots in several independently invented notions: Kupferman, Safra and Vardi’s automata recognising tree languages derived from word languages [Kupferman et al., 2006] (a notion that has been later referred to as automata that are good-for-trees [Udi Boker et al., 2013]), Henzinger and Piterman’s good-for-games automata [Thomas Henzinger and Nir Piterman, 2006], and Colcombet’s history-deterministic automata, introduced in his work on regular cost-automata [Colcombet, 2009]. In the ω-regular setting, where they were initially most studied, the notions of good-for-trees, good-for-games and history-determinism are equivalent, despite differences in their definitions. The key algorithmic appeal of these automata is that like deterministic automata, they have good compositional properties. This makes them particularly useful for applications such as reactive synthesis, where composition of games and automata is at the heart of effective solutions. Since then, history-determinism has received its fair share of attention, not least because of its relevance to synthesis. Indeed it turns out to be a natural and useful form of nondeterminism more broadly, and can be generalised to all sorts of different automata models: alternating automata [Udi Boker and Karoliina Lehtinen, 2019], pushdown automata [Enzo Erlich et al., 2022; Enzo Erlich et al., 2022], timed automata [Thomas A. Henzinger et al., 2022; Sougata Bose et al., 2022], Parikh automata [Enzo Erlich et al., 2022], and quantiative automata [Udi Boker and Karoliina Lehtinen, 2021], to name a few. In each of these models, history-determinism offers some trade-offs between the power of nondeterminism and the algorithmic properties of determinism. In particular, depending on the model, they can be either more expressive or more succinct than their deterministic counterparts, while retaining better algorithmic properties - in particular with respect to deciding universality, language inclusion and games - than fully nondeterministic automata. The drive to extend history-determinism to more powerful automata models has also lead to a better understanding of the properties of these automata, of how they compare to related notions (such as good-for-games automata and determinisability by pruning), and of the various games and tools used to study them. This talk aims to give a broad introduction to the notion of history determinism as well as an overview of some of the recent developements on the topic. It will also highlight some of the many problems that remain open. It is loosely based on a recent survey, written jointly with Udi Boker, which gives an informal presentation of what are, in our view, the key aspects of history-determinism [Udi Boker and Karoliina Lehtinen, 2023].