Are there elementarily equivalent models $\mathcal{M}, \mathcal{N}$ of the same cardinality s.t. neither can be elementarily embedded into the other?

Do there exist elementarily equivalent models $\mathcal{M}, \mathcal{N}$ of the cardinality $\kappa=\omega$ such that neither can be elementarily embedded into the other?

If the models were not required to be of the cardinality $\omega$, then one could consider models of the theory of 3 equivalence classes where each class has to be infinite: $\mathcal{M}$ would have its domain $M = M_1 \sqcup M_2 \sqcup M_3$ and $dom(\mathcal{N})=N_1 \sqcup N_2 \sqcup N_3$ where $M_1, M_2, M_3, N_1, N_2, N_3$ are equivalence classes of their respective cardinalities $\omega_1, \omega_1, \omega_2, \omega_0, \omega_2, \omega_2$. Because of the incompatible cardinalities, there is no embedding from $\mathcal{M}$ to $\mathcal{N}$ or vice versa. Therefore this is the answer to my question for the cardinality $\kappa \gt \omega_1$. For $\kappa=\omega_1$ one could use a similar trick with equivalence classes over two relations and the incompatibility of the cardinalities $\omega_0, \omega_1$. But what about the cardinality $\kappa=\omega_0$?

• You probably wanna change "elementary substructure" to "can be elementarily embedded into" before the pedantic folks come and point out that you can easily arrange two copies of $\Bbb N$ which are not subsets of one another and therefore neither is a substructure of the other. – Asaf Karagila Jan 18 '15 at 20:50

The language contains a relation symbol $<$ and the constants $a_i,b_i$ for $i\in\omega$. Consider two models, both with domain $\mathbb Q$ and the natural interpretation of $<$. In both models $b_{i+1}<b_i<a_i<a_{i+1}$. In the first model $a_i\to+\infty$ and $b_i\to-1$ in the second model $a_i\to+1$ and $b_i\to-\infty$.
• I do not understand how you can apply back-and-forth property: let $h:\mathcal{M} \rightharpoonup \mathcal{N}$ be a finite partial isomorphism between your first and the second model. Then there exists $a_N \in L$ such that $a_N^{\mathcal{M}}>1$ and $a_N^{\mathcal{N}}<1$ so you cannot extend $h$ by $a_N^{\mathcal{M}}$ as any such extended bijection would not preserve the interpretation of the constant $a_N$ and hence would not be an isomorphism. – Dávid Natingga Jan 23 '15 at 7:32
• @david Indeed to mention back-and-forth is misleading (I erased it). Reason as follows: ${\mathbb Q},<$ has quantifier elimination and this is preserved if after naming some elements. So, naming a different set of elements, if they are order-isomorphic, results in elementary equivalent models. – Primo Petri Jan 23 '15 at 9:12
Hint: consider models of the theory of linear orders where every element has a successor and a predecessor and their quotient by the (undefinable) equivalence relation $x\sim y$ iff there are finitely many elements betwen $x$ and $y$.