Descriptive complexity for pictures languages (extended abstract)

This paper deals with descriptive complexity of picture languages of any dimension by syntactical fragments of existential second-order logic. - We uniformly generalize to any dimension the characterization by Giammarresi et al. \cite{GRST96} of the class of \emph{recognizable} picture languages in existential monadic second-order logic. - We state several logical characterizations of the class of picture languages recognized in linear time on nondeterministic cellular automata of any dimension. They are the first machine-independent characterizations of complexity classes of cellular automata. Our characterizations are essentially deduced from normalization results we prove for first-order and existential second-order logics over pictures. They are obtained in a general and uniform framework that allows to extend them to other “regular” structures. Finally, we describe some hierarchy results that show the optimality of our logical characterizations and delineate their limits.

💡 Research Summary

The paper investigates the descriptive complexity of picture languages in arbitrary dimensions by means of syntactic fragments of existential second‑order logic (ESO). It begins by formalising d‑dimensional pictures as finite grids equipped with a finite set of colours (or symbols) and then develops normal‑form theorems for first‑order (FO) and ESO over such structures. The authors introduce three technical tools—region decomposition, global ordering, and coordinate compression—that allow any FO or ESO formula to be transformed into a canonical shape consisting of a bounded number of existentially quantified set variables followed by a universal FO part that only uses adjacency and equality predicates. Crucially, these transformations are uniform with respect to the dimension d, which means the same normal form works for 2‑D, 3‑D, or any higher‑dimensional picture.

Using this uniform normalisation, the paper generalises the classic result of Giammarresi, Rizzi, Schönhardt and Toruńczyk (1996), which identified the class of recognizable 2‑D picture languages with existential monadic second‑order logic (EMSO). The authors prove that for every d ≥ 1, the class of d‑dimensional recognizable picture languages coincides exactly with the languages definable in EMSO. This shows that recognizability, a purely automata‑theoretic notion, has a dimension‑independent logical characterisation.

The second major contribution is a machine‑independent characterisation of the class of picture languages that can be recognised in linear time by nondeterministic cellular automata (NCA) of any dimension. By analysing the parallel, local transition rule of an NCA, the authors show that the global computation performed in O(n) steps can be captured by an ESO formula of the form ∃X₁…∃X_k ∀y φ, where φ is a FO formula that only refers to the immediate neighbours of y and to the membership of y in the existentially quantified sets. Conversely, any language definable by such a formula can be recognised by an NCA in linear time. This is the first known logical characterisation of a cellular‑automaton complexity class, providing a bridge between parallel computation models and descriptive complexity.

Finally, the paper explores hierarchy results that stem from the normal‑form machinery. By restricting the number of existential set quantifiers or limiting the quantifier depth of the universal FO part, one obtains strict subclasses of recognizable languages and of linear‑time NCA languages, respectively. These hierarchies demonstrate the optimality of the presented logical characterisations and delineate their limits. Moreover, the authors argue that the same normal‑form framework can be adapted to other “regular” structures such as trees or bounded‑degree graphs, suggesting a broad applicability of their techniques.

In summary, the paper makes four key contributions: (1) uniform normal‑form theorems for FO and ESO over d‑dimensional pictures; (2) a dimension‑independent logical characterisation of recognizable picture languages via EMSO; (3) a novel logical characterisation of linear‑time nondeterministic cellular automata; and (4) hierarchy and optimality results that clarify the expressive power of the considered logical fragments. These results deepen the connection between automata theory, parallel computation, and descriptive complexity, and they provide a robust, machine‑independent framework for analysing regular structures beyond the traditional string setting.