Given a fixed first order lexicon $\mathcal{L}$, suppose $\mathcal{K}$ is an axiomatizable class such that $\mathcal{K}\subseteq Mod(\varphi)$ for some sentence $\varphi$. If $Mod(\varphi)-\mathcal{K}$ is axiomatizable, Does $\mathcal{K}$ necessarily to be finitely axiomatizable?
I proved the converse but I don't know how to deal with this direction. I've already proved using compactness that if $\mathcal{K}$ and $\mathcal{K}^{c}$ are both axiomatizable classes, then $\mathcal{K}$ need to be finitely axiomatizable. At this point, I tried to use last fact but this is not the case because $Mod(\varphi)-\mathcal{K}$ is a relative complement, whereas $\mathcal{K}^{c}$ is the universal complement.
I would appreciate some hint, ¡Thanks in advance!
Best Answer
One way to do this is to shift attention away from $\mathcal{K}$ itself, so that we don't get stuck inside $Mod(\varphi)$.
Let $$\mathcal{K}_\varphi=\mathcal{K}\cup\{\mathfrak{A}:\mathfrak{A}\models\neg\varphi\}.$$
Now $Mod(\varphi)-\mathcal{K}=(\mathcal{K}_\varphi)^c$, and we're assuming that that's axiomatizable. Our natural next step is to show that $\mathcal{K}_\varphi$ is axiomatizable:
In fact, a minor tweak of this shows that the union of finitely many axiomatizable classes is axiomatizable:
OK, so we have that $\mathcal{K}_\varphi$ and $(\mathcal{K}_\varphi)^c$ are each axiomatizable. So we know that $\mathcal{K}_\varphi$ is in fact finitely axiomatizable. We now want to turn a finite axiomatization of $\mathcal{K}_\varphi$ into a finite axiomatization of $\mathcal{K}$: