%! TEX root = MT.tex % vim: tw=50 % 15/03/2025 11AM \begin{fcprop}[] \label{prop:15.11} Assuming: - $T$ a complete $\mathcal{L}$-theory with infinite models - $\mathcal{M} \equiv \mathcal{N}$ are both countable and saturated Then: $\mathcal{M} \cong \mathcal{N}$. \end{fcprop} \begin{proof} Exercise (use back and forth argument). \end{proof} \newpage \section{Omitting Types} Let $\mathcal{M}$ be an $\mathcal{L}$-structure. Which types must be realised? \begin{fcdefn}[] \glsadjdefn{isol}{isolated}{$n$-type}% \label{defn:16.1} We say $p \in \Snm_n^{\mathcal{M}}(A)$ is \emph{isolated} if it is an isolated point with respect to topology on $\Snm_n^{\mathcal{M}}(A)$ (i.e. $\{p\}$ is open). \end{fcdefn} \begin{example*} For $a \in A \subseteq \mathcal{M}$, $\tp^{\mathcal{M}}(a / A)$ is isolated by $x = a$ ($\{\tp^{\mathcal{M}}(a / M)\} = [x = a]$). \end{example*} \begin{fcprop}[] \label{prop:16.2} Assuming: - $p \in \Snm_n^{\mathcal{M}}(A)$. Then: the following are equivalent: \begin{cenum}[(i)] \item $p$ is \gls{isol} \item $\{p\} = [\varphi(z)]$ for some $\mathcal{L}_A$-formula $\varphi(\ol{x})$. In this case we say $\varphi(\ol{x})$ \emph{isolates} $p$. \item There is an $\mathcal{L}_A$-formula $\varphi(\ol{x}) \in p$ such that for any $\psi(\ol{x}) \in p$, \[ \Th_A(\mathcal{M}) \models \forall x (\varphi(\ol{x}) \to \psi(\ol{x})) .\] \end{cenum} \end{fcprop} \begin{center} \includegraphics[width=0.6\linewidth]{images/bbbcfe1cbef84a5e.png} \end{center} \begin{proof} (i) $\iff$ (ii): Obvious. (ii) $\implies$ (iii): Assume $\varphi(\ol{x})$ isolates $p$. Fix an $\mathcal{L}_A$-formula $\psi(\ol{x})$. We want to show $\mathcal{M} \models \forall \ol{x} (\varphi(\ol{x}) \to \psi(\ol{x}))$. So suppose $\mathcal{M} \models \varphi(\ol{a})$. Then $\tp^{\mathcal{M}}(\ol{a} / A) \in [\varphi(\ol{x})] = \{p\}$. So $\tp^{\mathcal{M}}(\ol{a} / A) = p$, hence $M \models \psi(\ol{a})$. (iii) $\implies$ (ii): By assumption, for every $\mathcal{L}_A$-formula we have $\psi(\ol{x}) \in p$, $[\varphi(\ol{x})] \subseteq [\psi(\ol{x})]$. Thus if $q \sin [\varphi(\ol{x})]$, $a \in [\psi(\ol{x})]$. So $\psi(\ol{x}) \in q$, so $p \subseteq q$, so $p = q$. \end{proof} \begin{fcprop}[] \label{prop:16.3} Assuming: - $T$ is complete and consistent - $p \in S_n(T)$ is \gls{isol} Then: $p$ is realised in every $M \models T$. \end{fcprop} \begin{proof} Fix $p \in S_n(T)$, \gls{isol} by $\varphi(\ol{x})$. Fix $M \models T$. By \cref{prop:13.4}, there is some $\mathcal{N} \esup \mathcal{M}$ realising $p$. So $\mathcal{N} \models \exists \ol{x} \varphi(\ol{x})$, so $\mathcal{M} \models \exists \ol{x} \varphi(\ol{x})$. Fix $\ol{a} \in \mathcal{M}^n$ such that $\mathcal{M} \models \varphi(\ol{a})$. Then $\ol{a} \models p$ as for any $\varphi(\ol{x}) \in p$ we have \[ \mathcal{M} \models \forall \ol{x}(\varphi(\ol{x}) \to \psi(\ol{x})) .\] So $\mathcal{M} \models \varphi(\ol{a})$. \end{proof} \begin{fcthm}[Omitting Types Theorem] \label{thm:16.4} Assuming: - $\mathcal{L}$ is countable - $p \in S_n(T)$ is non-\gls{isol} Then: there is a countable $M \models T$ such that $p$ is not realised in $\mathcal{M}$ ($\mathcal{M}$ \emph{omits} $p$). \end{fcthm} \begin{proof} Let $\mathcal{L}^* = \mathcal{L} \cup C$, with $C$ a countably infinite set of new constants. An $\mathcal{L}^*$-theory has the \emph{witness property} if for any $\mathcal{L}^*$-formula $\varphi(\ol{x})$ there is a constant $c \in C$ such that $T^* \models \exists x \varphi(x) \to \psi(c)$. \textbf{Fact:} Suppose $T^*$ is a complete, satisfiable $\mathcal{L}^*$-theory with the witness property. Define $\sim$ on $C$ such that $c \sim d$ if and only if $T^* \models c = d$. Let $M = C / \sim$, and define an $\mathcal{L}^*$-structure on $M$ such that: \begin{itemize} \item $c^M = [c]$ \item $f^M([c_1], \ldots, [c_n]) = [d]$ if and only if \[ T^* \models f(c_1, \ldots, c_n) = d .\] \item $R^M = \{([c_1], \ldots, [c_n]) \in \mathcal{M}^n : T^* \models R(c_1, \ldots, c_n)\}$. \end{itemize} Then $\mathcal{M}$ is a well-defined $\mathcal{L}^*$-structure and $\mathcal{M} \models T^*$. Note we have $\mathcal{M} \models \varphi([c_1], \ldots, [c_n])$ if and only if $T^* \models \varphi(c_1, \ldots, c_n)$. We call $\mathcal{M}$ the Henkin model of $T^*$. Fix $p \in S_n(T)$ non-\gls{isol}. Aim: build a complete, satisfiable $\mathcal{L}^*$-theory $T^* \supseteq T$, with the witness property, . Such that for all $c_1, \ldots, c_n \in C$ there is \emph{some} $\varphi(\ol{x}) \in p$ such that $T^* \models \neg \varphi(c_1, \ldots, c_n)$. Then the Henkin model of $T^*$ omits $p$. Enumerate all the $\mathcal{L}^*$-sentences $\varphi_0, \varphi_1, \ldots$ and all $c^(n = \{\ol{c}_1, \ol{c}_2, \ldots\}$. We build a satisfiable $\mathcal{L}^*$-theory $T \cup \{\theta_1, \theta_2, \ldots\}$ such that \begin{enumerate}[(0)] \item $\models \theta_i \to \theta_j$ for all $i > j$. \item (Completeness): Either $\models \theta_{3i + 1} \to \varphi_i$ or $\models \theta_{3i + 1} \to \neg \varphi_i$. \item (Witnessing property): If $\varphi_i$ is $\exists v \psi(v)$ for some $\psi$ and $\models \theta_{3i + 1} \to \varphi_i$ then $\models \theta_{3i + 2} \to \psi(c)$ for some $c$. (check this does ensure the witness property). \item (Omit $p$): $\models \theta_{3i + 3} \to \neg \psi(c_i)$ for some $\psi(\ol{x}) \in p$. \end{enumerate} Let $\theta_0$ be $\forall v(v =v)$, and suppose we have $\theta_0, \ldots, \theta_m$. \textbf{Case 1:} $m + 1 = 3i + 1$ for some $i$. If $T \cup \{\theta_m, \varphi_i\}$ is satisfiable then $\theta_{m + 1} = \theta_m \wedge \varphi_i$. Otherwise $\theta_{m + 1} = \theta_m \wedge \neg \varphi$. So $T \cup \{\theta_{m + 1}\}$ is satisfiable by construction. \textbf{Case 2:} $m + 1 = 3i + 2$ for some $i$. Suppose $\varphi_i$ is $\exists v, \psi(v)$ for some $\psi$ an $\mathcal{L}^*$-formula, and $\models \theta_i \to \varphi_i$ (otherwise, let $\theta_{m + 1} = \theta_m$). Choose a $c \in C$ not used in $\theta_m$. Let $\theta_{m + 1}$ be $\theta_m \wedge \psi(i)$. Exercise: check $T \cup \{\theta_{m + 1}\}$ is satisfiable. \textbf{Case 3:} $m + 1 = 3i + 3$ for some $i$. Let $\ol{c}_i = (c_1, \ldots, c_n)$. Without loss of generality assume $x_1, \ldots, x_n$ not used in $\theta_m$. We build an $\mathcal{L}$-formula as follows: \begin{itemize} \item Replace $c_t$ by $x_t$ ($t \in \{1, \ldots, n\}$). \item Replace any $c \in C \setminus \{c_1, \ldots, c_n\}$ by new variables $v_c$ and add $\exists v_c$ to the front. \end{itemize} Then $\varphi(\ol{x})$ doesn't \gls{isol} $p$. By \cref{prop:16.2}, there is some $\psi(\ol{x}) \in p$ with \[ T \not\models \forall x(\varphi(\ol{x}) \to \psi(\ol{x})) .\] Let $\theta_{m + 1}$ be $\theta_m \wedge \neg \psi(c_1, \ldots, c_n)$. Check $\theta_{m + 1}$ is satisfiable. TODO \end{proof}