Being polite is not enough (and other limits of theory combination)
Guilherme V. Toledo, Benjamin Przybocki, Yoni Zohar
TL;DR
This paper demonstrates that politeness is not sufficient for theory combination by constructing decidable theories whose joint theory is undecidable, thereby clarifying limits of prominent combination methods such as Nelson–Oppen, gentle, polite, and shiny. It introduces finite-signature variants of key theories and provides finitized proofs, showing that the same limitation results hold under finite signatures. The authors also present two new combination theorems—one based on gentle with computable finite spectra and another on shininess without finite models—extending the toolkit for combining theories while preserving decidability. Collectively, the work sharpens our understanding of the necessary conditions for compositional decidability and offers practical avenues for enhancing SMT solvers with more flexible theory integration strategies. The results have significant implications for automated reasoning, formal verification, and the design of theory solvers in systems like $\textsf{cvc5}$ and related tools.
Abstract
In the Nelson-Oppen combination method for satisfiability modulo theories, the combined theories must be stably infinite; in gentle combination, one theory has to be gentle, and the other has to satisfy a similar yet weaker property; in shiny combination, only one has to be shiny (smooth, with a computable minimal model function and the finite model property); and for polite combination, only one has to be strongly polite (smooth and strongly finitely witnessable). For each combination method, we prove that if any of its assumptions are removed, then there is no general method to combine an arbitrary pair of theories satisfying the remaining assumptions. We also prove new theory combination results that weaken the assumptions of gentle and shiny combination.
