Elementary equivalences and accessible functors (Q2636525)

The article under review introduces a generalization of (infinitary) elementary equivalence to a very wide family of categories which the authors call mono-generated. Roughly they are categories where every object can be resolved as a directed colimit of objects of small size. This includes in particular accessible categories (see [\textit{M. Makkai} and \textit{R. Paré}, Accessible categories: The foundations of categorical model theory. Providence, RI: American Mathematical Society (1989; Zbl 0703.03042); \textit{J. Adámek} and \textit{J. Rosický}, Locally presentable and accessible categories. Cambridge: Cambridge University Press (1994; Zbl 0795.18007)]) but is more general: for example no closure under colimits is required, and the ``small'' objects are allowed to form a proper class. Specifically, the authors introduce for a regular cardinal \(\lambda\), a relation, \(\lambda\)-equivalence, between objects of a \(\lambda\)-mono-generated category. Morally, \(M\) is \(\lambda\)-equivalent to \(N\) whenever there is a \(\lambda\)-back and forth system between \(M\) and \(N\). In this case, however, there may not be any meaningful notion of cardinality, so the authors use a notion of size intrinsic to the category in question: an object is called \(\lambda\)-generated if (essentially) any inclusion of the object into a \(\lambda\)-directed colimit of monomorphisms factors through one of the components of the system. It is then proven that \(\lambda\)-equivalence has the expected basic properties: it is an equivalence relation (transitivity is not so easy to prove), two isomorphic objects are \(\lambda\)-equivalent, and if \(M\) is \(\lambda\)-equivalent to a \(\lambda\)-generated object \(N\), then \(M\) is isomorphic to \(N\). Moreover, it is shown that \(\lambda\)-equivalence specializes to \(\mathbb{L}_{\infty, \lambda}\)-elementary equivalence in categories of structures. The authors also study \(\lambda\)-embeddings: the category-theoretic notion corresponding to \(\mathbb{L}_{\infty, \lambda}\)-elementary embeddings. Interestingly, some of the results proven there seem to be new even for the syntactic case. One of the authors' motivation for the present paper is related to formalizations of the Lefschetz-Weil principle in algebraic geometry. Previous work of \textit{P. Eklof} [Proc. Am. Math. Soc. 37, 333--339 (1973; Zbl 0254.14004)] had given a syntactic formalization of this principle using \(\mathbb{L}_{\infty, \omega}\)-equivalence: a functor between appropriate categories which preserves directed colimits preserves \(\mathbb{L}_{\infty, \omega}\)-equivalence. Since any two algebraically closed fields of infinite transcendence degree (universal domains) which have the same characteristic are \(\mathbb{L}_{\infty, \omega}\)-equivalent, this is meant to formalize the fact that ``algebraic geometry is the same'' inside any two universal domain of the same characteristic. The present article generalizes Eklof's result as follows: If \(F\) is a functor between \(\lambda\)-mono-generated categories which preserves \(\lambda\)-directed colimits and monomorphisms, then \(F\) preserves \(\lambda\)-equivalence. Since many functors of interest do not preserve monomorphisms, the authors ask whether this last condition can be removed. A positive answer is established when the codomain of \(F\) is the category of models axiomatized by certain (infinitary) universal sentences.