Local collection and end-extensions of models of compositional truth (Q2662691)

This paper is one more contribution of Łełyk and Wcisło to the study of axiomatic truth theories, i.e., extensions of arithmetical theories by axioms expressing properties of the first-order satisfaction relation. A canonical such theory is \(\mathrm{CT}^-\) consisting of the Peano axioms (PA) and Tarski's compositional axioms for a truth predicate \(T\), but no induction axioms for \(T\). By a theorem of \textit{H. Kotlarski} et al. [Can. Math. Bull. 24, 283--293 (1981; Zbl 0471.03054)], \(\mathrm{CT}^-\) is conservative over PA, but many results by various authors show that adding a little bit of induction to \(\mathrm{CT}^-\) vastly increases its proof-theoretic strength. In a lecture given in 2014, Alex Jones and Richard Kaye conjectured that every countable recursively saturated model of PA is expandable to \(\mathrm{CT}^-+\mathrm{Coll}\), where \(\mathrm{Coll}\) is the full collection schema, and they suggested a way to prove it. The conjecture is still open. The main result of Łełyk and Wcisło is a confirmation of the conjecture for \(\mathrm{CT}^-\) extended by a local collection collection schema \(\mathrm{LocColl}\). A model \((M,T)\) satisfies \(\mathrm{LocColl}\) if for every \(c\) in \(M\), \((M,T_c)\) is a model of \(\mathrm{Coll}\), where \(T_c\) is a restriction of \(T\) to sentences of syntactic depth less than \(c\). In a series of lemmas, the authors show that for any countable recursively saturated model \(M\) and any cofinal in \(M\) sequence \((a_n)_{n\in\omega}\), there is a sequence of recursively saturated expansions \((M,T_n)\) that satisfy \(\mathrm{CT}^-\) for sentences of syntactic depth less than \(a_n\) and such that \(M\) expanded by \(\bigcup_{n\in\omega}T_n\) is a model of \(\mathrm{LocColl}\). In contrast, it is shown that \(\mathrm{CT}^-\) enhanced by the local version of the induction schema is equivalent to \(\mathrm{CT}^-\) with the induction schema for bounded formulas; hence it is not a conservative extension. \(\mathrm{CT}_n\) is \(\mathrm{CT}^-\) with the \(\Sigma_n(T)\)-induction schema, and \(B\Sigma_n(T)\) is \(\mathrm{CT}_0\) with the \(\Sigma_n(T)\)-collection schema. Following proofs of similar result for PA, in Section 5 of the paper the authors show that \(B\Sigma_{n+1}(T)\) is \(\Pi_{n+2}\) conservative over \(\mathrm{CT}_n\). An obstacle to proving that \(\mathrm{CT}^-+\mathrm{Coll}\) is conservative over \(\mathrm{CT}^-\) is a result by \textit{S. T. Smith} who showed that there are models of \(\mathrm{CT}^-\) with no end extension to a model of \(\mathrm{CT}^-\) [Ann. Pure Appl. Logic 42, No. 1, 21--43 (1989; Zbl 0701.03038)]. In the last section of the paper, the authors give a sketch of a proof of Albert Visser's related observation that if a model of PA is expandable to a model of \(\mathrm{CT}^-\), then it has an elementary end extension to a model that is expandable to a model of \(\mathrm{CT}^-\).