Categorical Logic
Table of Contents
Source: https://www.math.ias.edu/~lurie/278x.html
1. Lecture 1: Overview
\(T \) theory. \( \tilde{M} \) picked subset for each model \( M \) of \( T \) st
- for \( f: M \rightarrow M' \) elementary embedding, \( \tilde{M} = f^{-1}( \tilde{M}) \).
- for models \( \{M_j\}_{j \in J} \), ultrafilter \( \mathcal{U} \) on J, \( M := (\prod_{j \in J} M_j)/ \mathcal{U} \) : \((a_j)_{j \in J} \in \tilde{M} \) iff \(a_j \in \tilde{M_j}\) for almost all \(j \in J\).
Then: ex formula \( \varphi \), \( \tilde{M} = \varphi(M) \) for every model \(M \).
This is a natural companion to Gödel’s Completeness Theorem: it describes the conditions under which there exists a formula with a certain semantic interpretation. The Completeness Theorem ensures that this formula is unique up to provable equivalence. They work to together into the direction of recovering the syntax of a theory from its semantics.
\(T \) theory. \( \operatorname{Mod}(T)\) category of models of \(T \) (morphisms are elementary embeddings). Then:
- \(T\) can be recovered up to equivalence from the structure of \( \operatorname{Mod}(T)\) as a category with ultraproducts.
- Can I reformulate this ❓
- More precisely: have canonical equivalence \( \operatorname{Syn}(T) \simeq \operatorname{UFun}( \operatorname{Mod}(T),\Set)\)
\(T \) := theory of abelian groups. Then \( \operatorname{Mod}(T)\) only contains elementary embeddings, ie not all group homomorphisms. But there is a “nonclassical” theory \(T'\) st \( \operatorname{Mod}(T')\) is the category of abelian groups. Therefore we want to study this “nonclassical” logic, the logic of pretopoi. What is \(T'\) ❓
2. Lecture 2:
\(T\) first-order theory. The weak syntactic category \( \operatorname{Syn}_0(T)\) of \(T\) consists of
- objects: \([ \varphi(x)]\) for every formula \( \varphi\).
- morphisms: \(f: [ \varphi(x)] \rightarrow [ \psi(y)]\) is a collection of definable maps \((f_M: M[ \varphi] \rightarrow M[ \psi])_{M \text{ model of } T}\).
- composition: in every model.
- identity: in every model the identity.
Alternatively we could define the morphisms to be equivalence classes of formulas that define the same graph. \(M \) model of \(T \). \( \operatorname{Syn}_0(T) \rightarrow \Set\), \([ \varphi] \mapsto M[ \varphi]\) is a functor.
\(T \) first-order theory. \(M\), \(N\) models of \(T \). TFAE:
- elementary embedding \(f_0: M \rightarrow N\).
- natural transformation \(f: M[-] \rightarrow N[-]\).
We can identify a model \(M\) of \(T\) and functor its functor \(M[-]: \operatorname{Syn}_0(T) \rightarrow \Set\). The functor \(M[-]\) is “unbiased”, ie it doesn’t consider one underlying set on which we interpret the language, but all sets that can be constructed form the model.
Idea: \(f_0\) restricts to morphisms between the def sets. \(f\) gives \(f_{x=x}\).
It is enough to check \(M \models \varphi(x) \Rightarrow N \models \varphi(f(x))\) to verify that \(f\) is an elementary embedding. For the other direction, use the negation of \( \varphi\)!
A typed language \( \mathcal{L}\) consists of data:
- a set of types \( \mathfrak{T}\).
- a set of predicate symbols \(\{P_i\}\).
- for predicate symbol \(P_i\), the arity is a finite sequence \((t_1, \dots, t_n)\) of types.
Each variable is assigned a type st each type has infinitely many variables. A (typed) formula is just a normal formula with three caveats:
- add a new formula \(\bot\) for “false”.
- \(x = y\) only if \(x\),\(y\) have the same type.
- \(P_i(x_1, \dots, x_n)\) only if \(x_j\) has type \(t_j\) where \((t_1, \dots, t_n)\) is the arity of \(P_i\).
This is a proper generalization if there are infinitely many types. Why ❓
Propositional logic is the typed first-order logic with zero types. Let \(T\) be a propositional theory. Then: \( \operatorname{Syn}_0(T)\) is equivalent to the partially ordered set of equivalence classes of sentences. This is a Boolean algebra via:
- order: \( \varphi \le \psi\) iff \( \varphi \Rightarrow \psi\)
- join: \( \varphi \lor \psi\)
- complement: \( \neg \varphi\)
3. Lecture 3: The Structure of \( \operatorname{Syn}_0(T)\)
\(T\) typed first-order theory. Then:
- \( \operatorname{Syn}_0(T)\) has finite limits.
- For \(M\) model of \(T\), \(M[-]\) preserves finite limits..
Idea: Pullback + finial object. Pullback is \( \exists z: \theta(x,z) \land \theta'(y,z)\) where \( \theta\), \( \theta'\) are the morphisms (think of sets). Final object is \( \forall x: x = x\) (gets interpreted in \(M^0 = \{pt\}\)).
\(T\) typed first-order theory. \(f: X \rightarrow Y\) in \( \operatorname{Syn}_0(T)\). TFAE:
- \(f\) is an isom.
- For \(M\) model of \(T\), \(f_M\) is an isom.
\(T\) typed first-order theory. \(f: X \rightarrow Y\) in \( \operatorname{Syn}_0(T)\). TFAE:
- \(f\) is a mono.
- For \(M\) model of \(T\), \(f_M\) is a mono.
\(T\) typed first-order theory. \( \varphi(x)\), \( \psi(x)\) formulas with the same free variables.
- If \(T \models \forall x: \varphi(x) \Rightarrow \psi(x)\), then for every model \(M\) of \(T\), we identify \(M[ \varphi]\) with a subset of \(M[ \psi]\).
- This identification determines a mono \([ \varphi] \rightarrow [ \psi]\) in \( \operatorname{Syn}_0(T)\).
- A mono is called special iff it arises from the above construction.
Idea: A mono is special iff it comes from inclusions.
\(T\) typed first-order theory. \(f: X \rightarrow Y\) in \( \operatorname{Syn}_0(T)\). Then:
- Ex factorization \( X \xrightarrow{g} Y \xrightarrow{h} Z\) of \(f\) st:
- \(g\) is an effective epi.
- \(h\) is a special mono.
- Can pick factorization st:
- For \(M\) model of \(T\), \(M[-]\) preserves the factorization.
- Ie \(g_M\) is surjective, \(h_M\) is an inclusion.
Idea: \(Y := \exists x: \theta(x,y)\), where \( \theta\) defines \(f\) (image).
\( \mathcal{C}\) category with fiber products. \(g: X \rightarrow Y\) in \( \mathcal{C}\).
- \(X \times_Y X\) is the fiber product of \(X\) with itself over \(Y\) via \(g\).
- \( \pi, \pi': X \times_Y X \rightarrow X\) the two projections.
- \(g\) is an effective epimorphisms iff \(g\) is the coequalizer of \( \pi\), \( \pi'\). Ie for \(T \in \mathcal{C}\), \( \operatorname{Hom}_{ \mathcal{C}}(Y,T) \simeq \{ f \in \operatorname{Hom}_{ \mathcal{C}}(X,T) \mid f \circ \pi = f \circ \pi' \}\) (UP of coeq).
Every effective epi is an epi.
- In \( \Set \): eff epi iff it’s an epi iff it’s surjective.
- In \( \Ring \): eff epi iff surjective. But there are non-surjective epis.
\( \mathcal{C}\) category with fiber products. \(f: X \rightarrow Y\) in \( \mathcal{C}\). Then: \(X \rightarrow Y \rightarrow Z\) factorization of \(f\) with \(h\) mono, \(g\) eff epi, then this factorization is unique up to unique isom.
Idea: Use UP of eff epi and compute it by using mono.
\(T\) typed first-order theory. \(f: X \rightarrow Y\) mono in \( \operatorname{Syn}_0(T)\). Then:
- Ex isom \( \phi: X \rightarrow X'\), special mono \(g: X' \rightarrow Y\), \(f = g \circ \phi\).
Proof.
\( \mathcal{C}\) category. \(X\) object.
- \( \operatorname{Sub}(X)\) is the set of equivalence classes of subobjects of \(X\).
- \(A \subset B\) in \( \operatorname{Sub}(X)\) :iff there exists a triangle from \(A\) to \(B\) over \(X\).
\(T\) typed first-order theory. \(X = [ \varphi(x)]\) object in \( \operatorname{Syn}_0(T)\). Then:
- For \(A \in \operatorname{Sub}(X)\), ex \( \psi(x)\) formula, \(T \models \forall x: \psi(x) \Rightarrow \varphi(x)\) and \(A = [ \psi(x)]\) with the special mono to \(X\).
- For \( A = [\psi_1(x)], B = [\psi_2(x)] \in \operatorname{Sub}(X)\), \(A \subset B\) iff \(T \models \forall x: \psi_1(x) \Rightarrow \psi_2(x)\).
- For \( A = [\psi_1(x)], B = [\psi_2(x)] \in \operatorname{Sub}(X)\), \(A = B\) iff \(T \models \forall x: \psi_1(x) \Leftrightarrow \psi_2(x)\).
\(T\) typed first-order theory. \(X \in \operatorname{Syn}_0(T)\). Then:
- The partially ordered set \( \operatorname{Sub}(X)\) is a Boolean algebra.
- \(M[-]: \operatorname{Sub}(X) \rightarrow M[X]\) is a homo of Boolean algebras, where \(M[X]\) is the lattice of all subsets.
\(T\) typed first-order theory. \(F: \operatorname{Syn}_0(T) \rightarrow \Set\) functor. TFAE:
- Ex \(M\) model of \(T\), \(F \cong M[-]\)
- \(F\) satisfies:
- \(F\) preserves finite limits.
- \(F\) sends eff epis to surjections.
- For \(X \in \operatorname{Syn}_0(T)\), the map \( \operatorname{Sub}(X) \rightarrow \operatorname{Sub}(F(X))\) is a homo of upper semilattices, ie it preserves least upper bounds of finite subsets.
- We can replace 3. by: \( \operatorname{Sub}(X) \rightarrow \operatorname{Sub}(FX)\) is a Boolean algebra homo.
In words: \(F\) has to respect finite limits (pullbacks and terminal object), eff epis and least upper bounds of finitely many subobjects.
Proof.
4. Lecture 4: Coherent Categories
\(T\) typed first-order theory. \(f: X \rightarrow Y\) in \( \operatorname{Syn}_0(T)\). TFAE:
- \(f\) is eff epi.
- For \(M\) model of \(T\), \(f_M\) is surjective.
Proof.
\( \mathcal{C}\) category.
- \( \mathcal{C}\) is coherent iff
- \( \mathcal{C}\) has finite limits.
- For \(f: X \rightarrow Z\), ex factorization \(X \xrightarrow{g} Y \xrightarrow{h} Z\) of \(f\), \(g\) is eff epi and \(h\) is mono.
- For \(X \in \mathcal{C}\), \( \operatorname{Sub}(X)\) is an upper semilattice, ie it has a least element and two subobjects \(A, B \subset X\) habe a least upper bound \(A \lor B\).
- The collection of eff epis is stable under pullbacks.
- For \(f: X \rightarrow Y\), the map \(f^{-1}: \operatorname{Sub}(Y) \rightarrow \operatorname{Sub}(X)\) is a homo of upper semilattices.