Dependent \(T\) and existence of limit models (Q482659)

It is well-known that saturated models of a complete first-order theory are unique up to isomorphism given their cardinality. The author generalized this to abstract elementary classes, considering the so-called \((\lambda,\kappa)\) limit models for pairs of cardinals \(\kappa<\lambda\). In this paper, he studies the question when such models exist in a complete first-order theory. Recall that a universal model \(M\models T\) of cardinality \(\lambda\) is a \textit{\((\lambda,\kappa)\) superlimit} model if for all ordinals \(\delta<\lambda^+\) of cofinality \(\kappa\) and for all \(\prec\)-continuous increasing sequences \((M_i:i\leq\delta)\) with \(M_i\cong M\) for all \(i<\delta\), one has \(M_\delta\cong M\). More generally, if there is a function \(F\) from the collection of models of \(T\) of cardinality \(\lambda\) to itself satisfying \(N\prec F(N)\) for all \(N\), such that {\parindent=6mm \begin{itemize} \item[(a)] for all ordinals \(\delta<\lambda^+\) of cofinality \(\kappa\) and all \(\prec\)-continuous increasing sequences \((M_i:i\leq\delta)\) with \(F(M_{i+1})\prec M_{i+2}\) for all \(i<\delta\) one has \(M\cong\bigcup_{i<\delta}M_i\), then \(M\) is a \textit{\((\lambda,\kappa)\) strong limit}; \item [(b)] for all \(\prec\)-continuous increasing sequences \((M_i:i<\lambda^+)\) with \(F(M_{i+1})\prec M_{i+2}\) for all \(i<\lambda^+\) there is a closed unbounded subset \(C\subseteq\lambda^+\) such that \(M\cong M_\delta\) for all \(\delta\in C\) of cofinality \(\kappa\), then \(M\) is a \textit{\((\lambda,\kappa)\) normal limit}. \end{itemize}} If \(F\) takes \(\prec\)-increasing continuous sequences as argument, and we replace the condition \(M_{i+1}\prec F(M_{i+1})\prec M_{i+2}\) by \(M_{i+1}\prec F(M_j:j\leq i+1)\prec M_{i+2}\), we obtain the notion of \textit{\((\lambda,\kappa)\) medium limit} from (a) and \textit{\((\lambda,\kappa)\) weak limit} from (b). If we replace \(M_{i+1}\prec F(M_{i+1})\prec M_{i+2}\) by \(M_{i+1}\prec N\prec M_{i+2}\) for some \(N\equiv_{M_{i+1}}F(M_{i+1})\) (and similarly for sequences), we obtain the notions of \textit{\((\lambda,\kappa)\) invariantly strong/normal/medium/weak limit}. Clearly, superlimit implies strong limit, which implies both medium and normal limit, either of which implies weak limit. There are also variants \textit{limit\(^+\)}, \textit{limit\(^-\)}, and for replacing the set of ordinals of cofinality \(\kappa\) by some arbitrary stationary subset of \(\lambda^+\), but this would go beyond the limits of this review. The author shows that {\parindent=6mm \begin{itemize} \item[(1)] The theory of dense linear orders has \((\lambda,\kappa)\) invariantly medium limits, but no \((\lambda,\kappa)\) superlimit. \item [(2)] Strongly independent theories have no \((\lambda,\kappa)\) invariantly medium limit; independent theories have no \((\lambda,\kappa)\) medium limit. \end{itemize}} He conjectures that dependent theories have \((\lambda,\kappa)\) invariantly medium limits.