Let $M$ be a saturated model of a theory $T$ in a first-order language $\mathcal{L}$, and let $N$ be a submodel of $M$.

Is it possible to have a substructure $A\neq N$ of $M$ such that $N \subset A \subset M$ and every element of $A$ is definable by a formula in $L$ with parameters from $N$?


Yes. Let $T$ be the theory of an endless discrete order, which is a complete theory. Let $M=\mathbb{Z}\cdot\mathbb{Q}$ consist of $\mathbb{Q}$ copies of the $\mathbb{Z}$ order, which is a countable saturated model of $T$, and let $N$ consist of only the even elements in each copy of $\mathbb{Z}$, which still forms a model of $T$ (though not an elementary substructure of $M$), with $A$ adding in some or all of the odd elements. Each of these new elements in $A$ is definable in $M$ from their immediate predecessors, which are in $N$.

  • $\begingroup$ Wouldn't it work with taking any $N$ which is not an elementary submodel, and considering $A$ to be the elementary submodel generated by $N$? $\endgroup$ – Asaf Karagila Oct 14 '14 at 12:14
  • $\begingroup$ Yes, that's right. The non-elementary submodels must miss some elements in one of their $\mathbb{Z}$ chains, and then $A$ would fill in those missing elements, which would be definable from the other elements of that $\mathbb{Z}$ chain. $\endgroup$ – Joel David Hamkins Oct 14 '14 at 12:27
  • $\begingroup$ (I was actually talking about arbitrary languages and models, not just this one. For example $\Bbb N$ with $\leq$ and $\Bbb N\setminus\{6\}$ with $\leq$.) $\endgroup$ – Asaf Karagila Oct 14 '14 at 12:33
  • $\begingroup$ Well, in general there might not be "the" elementary submodel generated by a given submodel, unless you have definable Skolem functions, which are not always available. But when they are, as here, then yes, that is what is going on with this example. $\endgroup$ – Joel David Hamkins Oct 14 '14 at 12:36
  • $\begingroup$ For example, consider DLO and $\mathbb{Z}$ sitting inside $\mathbb{Q}$. This is not an elementary substructure, but we cannot add any new points definably from integers. $\endgroup$ – Joel David Hamkins Oct 14 '14 at 12:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.