content/model-theory/interpolation/interpolation.tex
1% Part: model-theory2% Chapter: interpolation34\documentclass[../../../include/open-logic-chapter]{subfiles}56\begin{document}78\olchapter{mod}{int}{The Interpolation Theorem}910\olimport{introduction}1112\olimport{separation}1314\olimport{interpolation-proof}1516\olimport{definability}1718\OLEndChapterHook1920\end{document}
content/model-theory/interpolation/introduction.tex
1% Part: model-theory2% Chapter: interpolation3% Section: introduction45\documentclass[../../../include/open-logic-section]{subfiles}67\begin{document}89\olfileid{mod}{int}{int}1011\olsection{Introduction}1213The interpolation theorem is the following result: Suppose14$\Entails !A \lif !B$. Then there is !!a{sentence} $!C$ such that15$\Entails !A \lif !C$ and $\Entails !C \lif !B$. Moreover, every16!!{constant}, !!{function}, and !!{predicate} (other than $\eq$) in17$!C$ occurs both in $!A$ and~$!B$. The !!{sentence} $!C$ is called an18\emph{interpolant} of $!A$ and~$!B$.1920The interpolation theorem is interesting in its own right, but its21main importance lies in the fact that it can be used to prove results22about definability in a theory, and the conditions under which23combining two consistent theories results in a consistent theory. The24first result is known as the Beth definability theorem; the second,25Robinson's joint consistency theorem.2627\end{document}
content/model-theory/interpolation/separation.tex
1% Part: model-theory2% Chapter: interpolation3% Section: separation45\documentclass[../../../include/open-logic-section]{subfiles}67\begin{document}89\olfileid{mod}{int}{sep}1011\olsection{Separation of \printtoken{P}{sentence}}1213A bit of groundwork is needed before we can proceed with the proof of14the interpolation theorem. An interpolant for $!A$ and $!B$ is15!!a{sentence}~$!C$ such that $!A \Entails !C$ and $!C \Entails !B$.16By contraposition, the latter is true iff $\lnot !B \Entails \lnot17!C$. !!^a{sentence}~$!C$ with this property is said to \emph{separate}18$!A$ and $\lnot !B$. So finding an interpolant for $!A$ and $!B$19amounts to finding !!a{sentence} that separates $!A$ and $\lnot !B$.20As so often, it will be useful to consider a generalization: a21sentence that separates two \emph{sets} of !!{sentence}s.2223\begin{defn}24A sentence $!C$ \emph{separates} sets of sentences $\Gamma$ and25$\Delta$ if and only if $\Gamma \Entails !C$ and $\Delta \Entails26\lnot !C$. If no such !!{sentence} exists, then $\Gamma$ and $\Delta$27are \emph{inseparable}.28\end{defn}2930The inclusion relations between the classes of31models of $\Gamma$, $\Delta$ and $!C$ are represented below:32\begin{figure}[h]33 \centering34 \begin{tikzpicture}[node distance=2cm, auto, thick]35 \draw [rounded corners] (0,0) -- (6,0) -- (6,3) -- (0,3) -- cycle;36 \draw (1.5,1.5) circle (0.9cm);37 \draw (4.5,1.5) circle (0.9cm);38 \path node at (1.5,1.5) {\Large $\Gamma$}; 39 \path node at (4.5,1.5) {\Large $\Delta$}; 40 \path node at (0.4,2.6) {\Large $\formula{C}$};41 \path node at (3.4,0.4) {\Large $\lnot \formula{C}$};42 \draw (2.5,0) .. controls (2.5,1.5) and (3.5,1.5) .. (3.5,3);43 \end{tikzpicture}44 \caption{$!C$ separates $\Gamma$ and $\Delta$}45 \ollabel{fig:sep}46\end{figure}474849\begin{lem}\ollabel{lem:sep1}50Suppose $\Lang{L}_0$ is the language containing every !!{constant},51!!{function} and !!{predicate} (other than $\doteq$) that occurs in52\emph{both} $\Gamma$ and $\Delta$, and let $\Lang{L}'_0$ be obtained53by the addition of infinitely many new !!{constant}s $\Obj c_n$ for $n54\ge 0$. Then if $\Gamma$ and $\Delta$ are inseparable in $\Lang{L}_0$,55they are also inseparable in $\Lang{L}'_0$.56\end{lem}5758\begin{proof}59We proceed indirectly: suppose by way of contradiction that $\Gamma$60and $\Delta$ are separated in $\Lang{L}'_0$. Then $\Gamma \Entails61\Subst{!C}{c}{x}$ and $\Delta \Entails \lnot \Subst{!C}{c}{x}$ for some $!C \in62\Lang{L}_0$ (where $c$ is a new !!{constant}---the case where $!C$63contains more than one such new !!{constant} is similar). By64compactness, there are \emph{finite} subsets $\Gamma_0$ of $\Gamma$65and $\Delta_0$ of $\Delta$ such that $\Gamma_0 \Entails \Subst{!C}{c}{x}$66and $\Delta_0 \Entails \lnot \Subst{!C}{c}{x}$. Let $!G$ be the67conjunction of all !!{formula}s in $\Gamma_0$ and $!H$ the68conjunction of all !!{formula}s in $\Delta_0$. Then69\begin{align*}70 !G & \Entails \Subst{!C}{c}{x}, & !H \Entails \lnot \Subst{!C}{c}{x}.71\end{align*}72From the former, by Generalization, we have $!G \Entails73\lforall[x][!C]$, and from the latter by contraposition,74$\Subst{!C}{c}{x} \Entails \lnot !H$, whence also $\lforall[x][!C]75\Entails \lnot \delta$. Contraposition again gives $!H \Entails76\lnot \lforall[x][!C]$. By monotonicity,77\begin{align*}78 \Gamma &\Entails \lforall[x][!C], & 79 \Delta & \Entails \lnot \lforall[x][!C],80\end{align*}81so that $\lforall[x][!C]$ separates $\Gamma$ and $\Delta$ in82$\Lang{L}_0$. 83\end{proof}8485\begin{lem}\ollabel{lem:sep2}86Suppose that $\Gamma \cup \{ \lexists[x][!S] \}$ and $\Delta$ are87inseparable, and $c$ is a new !!{constant} not in $\Gamma$, $\Delta$,88or $!S$. Then $\Gamma \cup \{ \lexists[x][!S], \Subst{!S}{c}{x} \}$89and $\Delta$ are also inseparable.90\end{lem}9192\begin{proof}93Suppose for contradiction that $!C$ separates $\Gamma \cup \{94\lexists[x][!S], \Subst{!S}{c}{x}\}$ and $\Delta$, while at the same95time $\Gamma \cup \{\lexists[x]{!S} \}$ and $\Delta$ are96inseparable. We distinguish two cases:97\begin{enumerate}98\item $c$ does not occur in $!C$: in this case $\Gamma \cup99 \{\lexists[x][!S], \lnot!C \}$ is satisfiable (otherwise $!C$100 separates $\Gamma \cup \{\lexists[x][!S] \}$ and $\Delta$). It101 remains so if $\Subst{!S}{c}{x}$ is added, so $!C$ does not separate102 $\Gamma \cup \{ \lexists[x][!S], \Subst{!S}{c}{x} \}$ and $\Delta$103 after all.104\item $c$ does occur in $!C$ so that $!C$ has the form105 $\Subst{!C}{c}{x}$. Then we have that106 \[107 \Gamma \cup \{ \lexists[x][!S], \Subst{!S}{c}{x}\} \Entails \Subst{!C}{c}{x},108 \]109 whence $\Gamma, \lexists[x][!S] \Entails \lforall[x][(!S \lif !C)]$110 by the Deduction Theorem and Generalization, and finally $\Gamma111 \cup \{ \lexists[x][!S] \} \Entails \lexists[x][!C]$. On the other112 hand, $\Delta \Entails \lnot \Subst{!C}{c}{x}$ and hence by113 Generalization $\Delta \Entails \lnot \lexists[x][!C]$. So $\Gamma114 \cup \{\lexists[x][!S] \}$ and $\Delta$ are separable, a115 contradiction.116\end{enumerate}117\end{proof}118119\end{document}
content/model-theory/interpolation/interpolation-proof.tex
1% Part: model-theory2% Chapter: interpolation3% Section: interpolation-proof45\documentclass[../../../include/open-logic-section]{subfiles}67\begin{document}89\olfileid{mod}{int}{prf}1011\olsection{Craig's Interpolation Theorem}1213\begin{thm}[Craig's Interpolation Theorem]14\ollabel{thm:interpol} 15If $\Entails !A \lif !B$, then there is !!a{sentence} $!C$ such that16$\Entails !A \lif !C$ and $\Entails !C \lif !B$, and every17!!{constant}, !!{function}, and !!{predicate} (other than $\eq$) in18$!C$ occurs both in $!A$ and~$!B$. The !!{sentence} $!C$ is called an19\emph{interpolant} of $!A$ and~$!B$.20\end{thm}2122\begin{proof}23Suppose $\Lang{L}_1$ is the language of $!A$ and $\Lang{L}_2$ is the24language of $!B$. Let $\Lang{L}_0 = \Lang{L}_1 \cap \Lang{L}_2$. For25each $i \in \{0, 1, 2 \}$, let $\Lang{L}'_i$ be obtained from26$\Lang{L}_i$ by adding the infinitely many new !!{constant}s $\Obj c_0,27\Obj c_1, \Obj c_2, \dots$. 2829If $!A$ is unsatisfiable, $\lexists[x][\eq/[x][x]]$ is an30interpolant. If $\lnot !B$ is unsatisfiable (and hence $!B$ is valid),31$\lexists[x][\eq[x][x]]$ is an interpolant. So we may assume also that32both $!A$ and $\lnot !B$ are satisfiable.3334In order to prove the contrapositive of the Interpolation Theorem,35assume that there is no interpolant for $!A$ and $!B$. In other words,36assume that $\{!A\}$ and $\{\lnot !B\}$ are inseparable in37$\Lang{L}_0$.3839Our goal is to extend the pair $(\{ !A \}, \{\lnot!B\})$ to a40maximally inseparable pair $(\Gamma^*, \Delta^*)$. Let $!A_0$,41$!A_1$, $!A_2$, \dots enumerate the !!{sentence}s of $\Lang{L}_1$, and42$!B_0$, $!B_1$, $!B_2$, \dots enumerate the !!{sentence}s43of~$\Lang{L}_2$. We define two increasing sequences of sets of44!!{sentence}s $(\Gamma_n, \Delta_n)$, for $n \ge 0$, as follows. Put45$\Gamma_0 = \{ !A\}$ and $\Delta_0 = \{\lnot !B \}$. Assuming46$(\Gamma_n, \Delta_n)$ are already defined, define $\Gamma_{n+1}$ and47$\Delta_{n+1}$ by:48\begin{enumerate}49\item If $\Gamma_n \cup \{!A_n \}$ and $\Delta_n$ are inseparable in50 $\Lang{L}'_0$, put $!A_n$ in $\Gamma_{n+1}$. Moreover, if $!A_n$ is51 an existential !!{formula} $\lexists[x][!S]$ then pick a new52 !!{constant} $c$ not occurring in $\Gamma_n$, $\Delta_n$, $!A_n$ or53 $!B_n$, and put $\Subst{!S}{c}{x}$ in $\Gamma_{n+1}$.54\item If $\Gamma_{n+1}$ and $\Delta_n \cup \{!B_n \}$ are inseparable55 in $\Lang{L}'_0$, put $!B_n$ in $\Delta_{n+1}$. Moreover, if $!B_n$56 is an existential !!{formula} $\lexists[x][!S]$, then pick a new57 !!{constant} $c$ not occurring in $\Gamma_{n+1}$, $\Delta_n$, $!A_n$58 or $!B_n$, and put $\Subst{!S}{c}{x}$ in $\Delta_{n+1}$.59\end{enumerate}60Finally, define:61\begin{align*}62 \Gamma^* & = \bigcup_{n\ge 0} \Gamma_n, & 63 \Delta^* & = \bigcup_{n\ge 0} \Delta_n.64\end{align*}65By simultaneous induction on $n$ we can now prove:66\begin{enumerate}67\item\ollabel{part-a} $\Gamma_n$ and $\Delta_n$ are inseparable in68 $\Lang{L}'_0$;69\item\ollabel{part-b} $\Gamma_{n+1}$ and $\Delta_n$ are inseparable in70 $\Lang{L}'_0$.71\end{enumerate}72The basis for \olref{part-a} is given by \olref[sep]{lem:sep1}. For73part \olref{part-b}, we need to distinguish three cases:74\begin{enumerate}75\item If $\Gamma_0 \cup \{!A_0 \}$ and $\Delta_0$ are separable, then76 $\Gamma_1 = \Gamma_0$ and \olref{part-b} is just \olref{part-a};77\item If $\Gamma_1 = \Gamma_0 \cup\{ !A_0\}$, then $\Gamma_1$ and78 $\Delta_0$ are inseparable by construction.79\item It remains to consider the case where $!A_0$ is existential, so80 that $\Gamma_1 = \Gamma_0 \cup \{ \lexists[x][!S], \Subst{!S}{c}{x}81 \}$. By construction, $\Gamma_0 \cup \{ \lexists[x][!S]\}$ and82 $\Delta_0$ are inseparable, so that by \olref[sep]{lem:sep2} also83 $\Gamma_0 \cup \{ \lexists[x][!S], \Subst{!S}{c}{x} \}$ and84 $\Delta_0$ are inseparable.85\end{enumerate}86This completes the basis of the induction for \olref{part-a} and87\olref{part-b} above. Now for the inductive step. For \olref{part-a}, if88$\Delta_{n+1} = \Delta_n \cup \{ !B_n \}$ then $\Gamma_{n+1}$ and89$\Delta_{n+1}$ are inseparable by construction (even when $!B_n$ is90existential, by \olref[sep]{lem:sep2}); if $\Delta_{n+1} = \Delta_n$91(because $\Gamma_{n+1}$ and $\Delta_n \cup \{!B_n\}$ are separable),92then we use the induction hypothesis on \olref{part-b}. For the93inductive step for \olref{part-b}, if $\Gamma_{n+2} = \Gamma_{n+1} \cup94\{!A_{n+1} \}$ then $\Gamma_{n+2}$ and $\Delta_{n+1}$ are95inseparable by construction (even when $!A_{n+1}$ is existential,96by \olref[sep]{lem:sep2}); and if $\Gamma_{n+2} = \Gamma_{n+1}$ then97we use the inductive case for \olref{part-a} just proved. This98concludes the induction on \olref{part-a} and \olref{part-b}. 99100It follows that $\Gamma^*$ and $\Delta^*$ are inseparable; if not, by101compactness, there is $n \ge 0$ that separates $\Gamma_n$ and102$\Delta_n$, against \olref{part-a}. In particular, $\Gamma^*$ and103$\Delta^*$ are consistent: for if the former or the latter is104inconsistent, then they are separated by $\lexists[x][\eq/[x][x]]$ or105 $\lforall[x][\eq[x][x]]$, respectively.106107We now show that $\Gamma^*$ is maximally consistent in108$\Lang{L}'_1$ and likewise $\Delta^*$ in $\Lang{L}'_2$. For the109former, suppose that $!A_n \notin \Gamma^*$ and $\lnot !A_n110\notin \Gamma^*$, for some $n \ge 0$. If $!A_n \notin \Gamma^*$111then $\Gamma_n \cup \{!A_n \}$ is separable from $\Delta_n$, and112so there is $!C \in \Lang{L}'_0$ such that both:113\begin{align*}114 \Gamma^* & \Entails !A_n \lif !C, & 115 \Delta^* & \Entails \lnot !C.116\end{align*}117Likewise, if $\lnot !A_n \notin \Gamma^*$, there is $!C' \in118\Lang{L}'_0$ such that both:119\begin{align*}120 \Gamma^* & \Entails \lnot !A_n \lif !C', & 121 \Delta^* & \Entails \lnot !C'.122\end{align*}123By propositional logic, $\Gamma^* \Entails !C \lor !C'$ and124$\Delta^* \Entails \lnot (!C \lor !C')$, so $!C \lor125!C'$ separates $\Gamma^*$and $\Delta^*$. A similar argument126establishes that $\Delta^*$ is maximal. 127128Finally, we show that $\Gamma^* \cap \Delta^*$ is maximally consistent129in $\Lang{L}'_0$. It is obviously consistent, since it is the130intersection of consistent sets. To show maximality, let $!S \in131\Lang{L}'_0$. Now, $\Gamma^*$ is maximal in $\Lang{L'_1}132\supseteq \Lang{L'_0}$, and similarly $\Delta^*$ is maximal in133$\Lang{L'_2} \supseteq \Lang{L'_0}$. It follows that either134$!S \in \Gamma^*$ or $\lnot !S \in \Gamma^*$, and either135$!S \in \Delta^*$ or $\lnot !S \in \Delta^*$. If $!S \in136\Gamma^*$ and $\lnot !S \in \Delta^*$ then $!S$ would137separate $\Gamma^*$ and $\Delta^*$; and if $\lnot !S \in138\Gamma^*$ and $!S \in \Delta^*$ then $\Gamma^*$ and $\Delta^*$139would be separated by $\lnot !S$. Hence, either $!S \in140\Gamma^* \cap \Delta^*$ or $\lnot !S \in \Gamma^* \cap \Delta^*$,141and $\Gamma^* \cap \Delta^*$ is maximal. 142143Since $\Gamma^*$ is maximally consistent, it has a model144$\Struct{M}'_1$ whose !!{domain} $\Domain{M'_1}$ comprises all and145only the elements $\Assign{c}{M'_1}$ interpreting the146!!{constant}s---just like in the proof of the completeness theorem147(\olref[fol][com][cth]{thm:completeness}). Similarly, $\Delta^*$ has a148model $\Struct{M}'_2$ whose !!{domain} $\Domain{M'_2}$ is given by the149interpretations $\Assign{c}{M'_2}$ of the !!{constant}s.150151Let $\Struct{M_1}$ be obtained from $\Struct{M'_1}$ by dropping152interpretations for !!{constant}s, !!{function}s, and !!{predicate}s in153$\Lang{L'_1} \setminus \Lang{L'_0}$, and similarly for154$\Struct{M_2}$. Then the map $h \colon M_1 \to M_2$ defined by155$h(\Assign{c}{M'_1}) = \Assign{c}{M'_2}$ is an156isomorphism in $\Lang{L}'_0$, because $\Gamma^* \cap \Delta^*$ is157maximally consistent in $\Lang{L}'_0$, as shown. This follows158because any $\Lang{L}'_0$-!!{sentence} either belongs to both159$\Gamma^*$ and $\Delta^*$, or to neither: so $\Assign{c}{M'_1} \in160\Assign{P}{M'_1}$ if and only if $\Atom{P}{c} \in \Gamma^*$ if and only if161$\Atom{P}{c} \in \Delta^*$ if and only if $\Assign{c}{M'_2} \in162\Assign{P}{M'_2}$. The other conditions satisfied by isomorphisms163can be established similarly.164165Let us now define a model $\Struct{M}$ for the !!{language}166$\Lang{L_1} \cup \Lang{L_2}$ as follows:167\begin{enumerate}168\item The !!{domain} $\Domain{M}$ is just $\Domain{M_2}$, i.e., the169 set of all elements $\Assign{c}{M'_2}$; 170\item If !!a{predicate}~$P$ is in $\Lang{L_2} \setminus171 \Lang{L_1}$ then $\Assign{P}{M} = \Assign{P}{M'_2}$;172\item If a predicate $P$ is in $\Lang{L}_1\setminus \Lang{L}_2$ then173 $\Assign{P}{M} = h(\Assign{P}{M'_2})$, i.e.,174 $\tuple{\Assign{c_1}{M'_2}, \dots, \Assign{c_n}{M'_2}} \in175 \Assign{P}{M}$ if and only if $\tuple{\Assign{c_1}{M'_1}, \dots,176 \Assign{c_n}{M'_1}} \in \Assign{P}{M'_1}$.177\item If !!a{predicate} $P$ is in $\Lang{L}_0$ then $\Assign{P}{M} =178 \Assign{P}{M'_2} = h(\Assign{P}{M'_1})$. 179\item !!^{function}s of $\Lang{L}_1 \cup \Lang{L}_2$, including180 !!{constant}s, are handled similarly.181\end{enumerate}182183Finally, one shows by induction on !!{formula}s that $\Struct{M}$ agrees184with $\Struct{M'_1}$ on all !!{formula}s of $\Lang{L'_1}$ and with185$\Struct{M'_2}$ on all !!{formula}s of $\Lang{L'_2}$. In particular,186$\Struct{M} \Entails \Gamma^* \cup \Delta^*$, whence $\Struct{M}187\Entails !A$ and $\Struct{M} \Entails \lnot!B$, and188$\not\Entails !A \lif !B$. This concludes the proof of189Craig's Interpolation Theorem.190\end{proof}191192\end{document}
content/model-theory/interpolation/definability.tex
1% Part: model-theory2% Chapter: interpolation3% Section: definability45\documentclass[../../../include/open-logic-section]{subfiles}67\begin{document}89\olfileid{mod}{int}{def}1011\olsection{The Definability Theorem}1213One important application of the interpolation theorem is Beth's14definability theorem. To define an $n$-place relation~$R$ we can give15!!a{formula}~$!C$ with $n$ free !!{variable}s which does not16involve~$R$. This would be an \emph{explicit} definition of~$R$ in17terms of~$!C$. We can then say also that a theory~$\Sigma(P)$ in a18!!{language} containing the $n$-place !!{predicate}~$P$ explicitly19defines~$P$ if it contains (or at least entails) a formalized explicit20definition, i.e.,21\[22\Sigma(P) \Entails \lforall[x_1][\dots23 \lforall[x_n][(\Atom{P}{x_1,\dots, x_n} \liff !C(x_1, \dots,24 x_n))]].25\]26But an explicit definition is only one way of defining---in the sense27of determining completely---a relation. A theory may also be such28that the interpretation of~$P$ is fixed by the interpretation of the29rest of the !!{language} in any model. The definability theorem30states that whenever a theory fixes the interpretation of~$P$ in this31way---whenever it \emph{implicitly defines}~$P$---then it also32explicitly defines it.3334\begin{defn}35Suppose $\Lang{L}$ is !!a{language} not containing the36!!{predicate}~$P$. A set $\Sigma(P)$ of !!{sentence}s of $\Lang{L}37\cup \{P\}$ \emph{explicitly defines}~$P$ if and only if there is38!!a{formula}~$!C(x_1, \dots, x_n)$ of $\Lang{L}$ such that39\[40\Sigma(P) \Entails \lforall[x_1][\dots41 \lforall[x_n][(\Atom{P}{x_1,\dots, x_n} \liff !C(x_1, \dots,42 x_n))]].43\]44\end{defn}4546\begin{defn}47Suppose $\Lang{L}$ is !!a{language} not containing the48!!{predicate}s~$P$ and~$P'$. A set $\Sigma(P)$ of !!{sentence}s of49$\Lang{L} \cup \{P\}$ \emph{implicitly defines} $P$ if and only if50\[51\Sigma(P) \cup \Sigma(P') \Entails \lforall[x_1][\dots52 \lforall[x_n][(\Atom{P}{x_1,\dots, x_n} \liff \Atom{P'}{x_1,\dots,53 x_n})]],54\]55where $\Sigma(P')$ is the result of uniformly replacing $P$ with $P'$56in $\Sigma(P)$.57\end{defn}5859In other words, for any model $\Struct{M}$ and $R, R' \subseteq60\Domain{M}^n$, if both $\Expan{M}{R} \Entails \Sigma(P)$ and61$\Expan{M}{R'} \Entails \Sigma(P')$, then $R=R'$; where62$\Expan{M}{R}$ is the !!{structure}~$\Struct{M'}$ for the63expansion of $\Lang{L}$ to $\Lang{L} \cup \{P\}$ such that64$\Assign{P}{M'} = R$, and similarly for $\Expan{M}{R'}$.6566\begin{thm}[Beth Definability Theorem] A set $\Sigma(P)$ of $\Lang{L}67 \cup\{P\}$-!!{formula}s implicitly defines $P$ if and only $\Sigma(P)$68 explicitly defines $P$.69\end{thm}7071\begin{proof}72If $\Sigma(P)$ explicitly defines $P$ then both73\begin{align*}74 \Sigma(P) & \Entails & \lforall[x_1][\dots \lforall[x_n]75 [(\Atom{P}{x_1,\dots, x_n} \liff !C(x_1,\dots,x_n))]]\\76 \Sigma(P') & \Entails & \lforall[x_1][\dots \lforall[x_n]77 [(\Atom{P'}{x_1,\dots, x_n} \liff !C(x_1,\dots,x_n))]]78\end{align*}79and the conclusion follows. For the converse: assume that $\Sigma(P)$80implicitly defines $P$. First, we add !!{constant}s $c_1$, \dots,~$c_n$ to81$\Lang{L}$. Then82\[83\Sigma(P) \cup \Sigma(P') \Entails84\Atom{P}{c_1, \dots, c_n} \to \Atom{P'}{c_1, \dots, c_n}.85\]86By compactness, there are finite sets $\Delta_0 \subseteq \Sigma(P)$87and $\Delta_1 \subseteq \Sigma(P')$ such that88\[89\Delta_0 \cup \Delta_1 \Entails90\Atom{P}{c_1, \dots, c_n} \to \Atom{P'}{c_1, \dots, c_n}.91\]92Let $!D(P)$ be the conjunction of all !!{sentence}s $!A(P)$ such that93either $!A(P) \in \Delta_0$ or $!A(P') \in \Delta_1$ and let $!D(P')$94be the conjunction of all !!{sentence}s $!A(P')$ such that either95$!A(P) \in \Delta_0$ or $!A(P') \in \Delta_1$. Then $!D(P) \land96!D(P') \Entails \Atom{P}{c_1, \dots, c_n} \to P'c_1\dots c_n$. We can97re-arrange this so that each !!{predicate} occurs on one side of98$\Entails$:99\[100!D(P) \land \Atom{P}{c_1, \dots, c_n} \Entails101!D(P') \to \Atom{P'}{c_1, \dots, c_n}.102\]103By Craig's Interpolation Theorem there is !!a{sentence} $!C(c_1,\dots, c_n)$104not containing $P$ or $P'$ such that:105\begin{align*}106 !D(P) \land \Atom{P}{c_1, \dots, c_n} & \Entails !C(c_1,\dots, c_n); \\107 !C(c_1,\dots, c_n) & \Entails !D(P') \to \Atom{P'}{c_1, \dots, c_n}.108\end{align*}109From the former of these two entailments we have: $!D(P) \Entails110\Atom{P}{c_1,\dots, c_n} \lif !C(c_1,\dots, c_n)$. And from the111latter, since an $\Lang{L} \cup \{P\}$-model $\Expan{M}{R}112\Entails !A(P)$ if and only if the corresponding $\Lang{L} \cup113\{P'\}$-model $\Expan{M}{R} \models !A(P')$, we have114$!C(c_1,\dots, c_n) \Entails !D(P) \lif \Atom{P}{c_1,\dots, c_n}$,115from which:116\[117!D(P) \Entails !C(c_1,\dots,c_n) \to \Atom{P}{c_1,\dots, c_n}.118\]119Putting the two together, $!D(P) \Entails \Atom{P}{c_1,\dots, c_n}120\liff !C(c_1, \dots, c_n)$, and by monotonicity and generalization also121\[122\Sigma(P) \Entails123\lforall[x_1][\dots\lforall[x_n][(\Atom{P}{x_1,\dots, x_n} \liff124 !C(x_1,\dots, x_n))]].125\]126\end{proof}127128\end{document}129