Constrained Expressions and their Derivatives

This paper proposes an extension to classical regular expressions by the addition of two operators allowing the inclusion of boolean formulae from the zeroth order logic. These expressions are called constrained expressions. The associated language is defined thanks to the notion of interpretation and of realization. We show that the language associated when both interpretation and realization are fixed is stricly regular and can be not regular otherwise. Furthermore, we use an extension of Antimirov partial derivatives in order to solve the membership test in the general case. Finally, we show that once the interpretation is fixed, the membership test of a word in the language denoted by a constrained expression can be undecidable whereas it is always decidable when the interpretation is not fixed.

💡 Research Summary

The paper introduces “constrained expressions”, an extension of classical regular expressions obtained by adding two new operators that embed zeroth‑order logical formulae (first‑order logic without quantifiers) into the pattern language. The first operator, written as E | φ, attaches a Boolean formula φ to a regular expression E and yields the language L(E) only when φ evaluates to true under a given interpretation and variable realization; otherwise it yields the empty set. The second operator, written as α ⊣ E, prefixes a word α to the expression and returns α only if α belongs to L(E), otherwise the empty set. By allowing function symbols, predicate symbols and variables from a logical signature, constrained expressions can describe non‑regular languages such as {aⁿbⁿcⁿ | n ≥ 0} by enforcing a length‑equality predicate over three sub‑words.

The authors formalize the notions of interpretation (a domain together with concrete meanings for function and predicate symbols) and realization (an assignment of domain elements to variables). When both interpretation and realization are fixed, the language of a constrained expression is shown to be regular; a finite‑state automaton can be constructed, and the expression can be compiled into a deterministic machine. If either component is left free, the resulting language may be non‑regular.

To solve the membership problem for arbitrary constrained expressions, the paper extends Antimirov’s partial derivatives. The derivative of a constrained expression with respect to a symbol is defined so that it simultaneously updates the regular‑expression part, the variable bindings, and the logical constraints. By iteratively computing derivatives for each symbol of the input word, the problem reduces to checking whether the empty word belongs to the derivative set. This approach generalizes the classic residual method for regular expressions while handling logical predicates and variable assignments.

A key theoretical contribution is the reduction of membership to a logical satisfiability problem. When the interpretation is fixed, the membership test becomes equivalent to checking the satisfiability of a Boolean formula that may involve arbitrarily complex predicates. The authors prove that, for sufficiently expressive interpretations (e.g., where predicates can encode Turing‑machine computations), the membership problem is undecidable. Conversely, when the interpretation is not fixed, the satisfiability problem remains decidable because the logical fragment is essentially propositional (zeroth‑order) and can be solved by exhaustive search or SAT‑solvers.

The paper concludes that constrained expressions dramatically increase expressive power beyond regular languages while preserving regularity under fixed semantics. However, this power comes at the cost of potential undecidability of the membership test, highlighting a trade‑off between flexibility of logical constraints and algorithmic tractability. The results provide a rigorous foundation for extending regex‑based tools with logical predicates and suggest careful design of the interpretation layer to maintain decidability.