\(\lambda\)-satisfiability, \(\lambda\)-consistency property, the downward Löwenheim-Skolem theorem, and the failure of the interpolation theorem for \(L_{\kappa,\kappa}\) with \(\kappa\) a strong limit cardinal of cofinality \(\lambda\) (Q582286)

It is well-known that interpolation fails in the infinitary languages \(L_{\kappa \lambda}\) beyond \(L_{\omega_ 1\omega}\) as shown by J. Malitz and others. One line of research to overcome this difficulty is the study of the chain models of C. Karp. Every usual structure is a chain model and so validity in all chain models is a strengthening of ordinary validity. Karp already showed that this strengthening is enough to give interpolants for sentences valid in the strong sense. Chain models of length \(\omega\), i.e. \(\omega\)-chain models, have usually been applied for \(L_{\kappa \kappa}\) where \(\kappa\) is a singular strong limit cardinal of cofinality \(\omega\). In the present paper the author considers \(\lambda\)-chain models in the analogous case where \(cf(\kappa)=\lambda >\omega\). He presents a version of the chain consistency properties (the main tool in the area) for this case and shows that a sentence has a chain model if and only if it belongs to a chain consistency property. The result is actually proved for certain good \(\lambda\)-sequences of sentences. One part of the argument gives a downward Löwenheim-Skolem theorem. The most interesting ingredient of the paper is, however, the observation that interpolation (even Beth's theorem) fails for the semantics given by \(\lambda\)-chain models, and that this even follows from Malitz's original example. (There are several misprints in the paper, most of which are of a harmless linguistic nature. In connection with this example there is a disturbing one: the sentence \(\sigma\) used is in \(L'_{\omega_ 1\omega_ 1}\), not in \(L'_{\omega_ 1\omega}\). This is the reason why the example does not work for \(\omega\)-chain models, which the reviewer would have liked to be mentioned in the paper.)