Stationary logic and its friends. II (Q1820150)

This paper is the successor to Part I [reviewed above; see Zbl 0614.03032]. The three sections of the paper can be read independently. The first two sections assume some familiarity with stationary logic, denoted L(aa). The third section concerns a closure operation for abstract logic. In the first section we define, for regular \(\lambda\), the \(\lambda\)- interpretation of L(aa), denoted \(L(aa^{\lambda})\). In this notation, the standard interpretation is \(L(aa^{\omega})\). The most easily understandable case occurs when \(\lambda^{<\lambda}=\lambda\). Then for models with universe \(\lambda^+\), \(aa^{\lambda}\) expresses ''for all but a nonstationary set of ordinals of cofinality \(\lambda\) ''. We show if \(\lambda^{<\lambda}=\lambda\), then \(L(aa^{\lambda})\) has the same validities as \(L(aa^{\omega})\) and \(L(aa^{\lambda})\) is (\(\lambda\),\(\omega)\)-compact. The second section is devoted to the proof of the consistency of the following approximation to the \(\Delta\)-closure of L(Q) being contained in L(aa). Suppose \(L_ 1\cap L_ 2=L_ 0\), \(\psi_ 1\in L_ 1(Q)\) and \(\psi_ 2\in L_ 2(Q)\). Further suppose every finitely determinate \(L_ 0\)-structure either can be expanded to a model of exactly one of \(\psi_ 1\) of \(\psi_ 2\) or can be expanded to a finitely determinate model of exactly one of \(\psi_ 1\) or \(\psi_ 2\). Then there is a sentence \(\theta \in L_ 0(aa)\) such that every finitely determinate model of \(\psi_ 1\) satisfies \(\theta\) and no finitely determinate model of \(\psi_ 2\) satisfies \(\theta\). (So \(\theta\) separates the reducts of finitely determinate models of \(\psi_ 1\) from those of \(\psi_ 2.)\) The third section is a prolonged observation motivated by the algebraic intuition that homomorphic images are as fundamental as subalgebras. A common requirement on a logic is that one can talk about definable substructures. However, many common logics, such as L(Q), do not allow us to talk about definable homomorphic images; i.e., modeling out by definable congruences. We show that demanding a logic be congruence closed is innocuous. Every logic has an easily described congruence closure with inherits most of the good properties of the logic.