# nLab diagram of a first-order structure

model theory

## Dimension, ranks, forking

• forking and dividing?

• Morley rank?

• Shelah 2-rank?

• Lascar U-rank?

• Vapnik–Chervonenkis dimension?

# Contents

## Idea

Given a model $M$ of a (not necessarily complete) first-order theory $T$, one can canonically associate theories $T_{\mathsf{Diag}(M)}$ and $T_{\mathsf{EDiag}(M)}$ whose models are precisely the models $N \models T$ into which $M$ embeds as a substructure and elementary substructure, respectively.

In the case of the latter, this gives a theory whose category of models is precisely the co-slice category $M/\mathbf{Mod}(T)$ of models under $M$ in $\mathbf{Mod}(T)$.

## Definition

Let $M$ be a first-order structure in the language $\mathcal{L}$. We obtain an expanded language $\mathcal{L}(M)$ by adding to $\mathcal{L}$ new constant symbols $c_m$ for each $m \in M$, and $M$ is naturally an $\mathcal{L}(M)$-structure by interpreting each new constant as its namesake. The elementary diagram of $M$, written $\mathsf{EDiag}(M)$, is the set of all $\mathcal{L}$-sentences possibly using constants $c_m$ which are true in $M$, i.e. the $\mathcal{L}(M)$-theory of $M$.

The quantifier-free diagram of $M$, written $\mathsf{Diag}(M)$, is obtained the same way as $\mathsf{EDiag}(M)$, but only allowing quantifier-free $\mathcal{L}(M)$-sentences.

If $N$ models $\mathsf{EDiag}(M)$, then $N$ contains $M$ as an elementary substructure. If $N$ models $\mathsf{Diag}(M)$, then $N$ contains $M$ as an induced substructure.

If we are given $M$ and $T$ as above, we simply obtain $T_{\mathsf{Diag}(M)}$ and $T_{\mathsf{EDiag}(M)}$ as the union of $T$ (viewed as an $\mathcal{L}(M)$-theory) with $\mathsf{Diag}(M)$ and $\mathsf{EDiag}(M)$, respectively.

## Examples

• A trivial example is ACF$_0$, the theory of algebraically closed fields of characteristic zero (in the language of rings). Since $\mathbb{Q}$ is the prime field of characteristic zero, any algebraically closed field models $\mathsf{Diag}(\mathbb{Q})$, and in fact since each element of $\mathbb{Q}$ is already definable in ACF$_0$, $\mathsf{Diag}(\mathbb{Q})$ is just the quantifier-free part of ACF$_0$.

• Let $R$ be the countable random graph. Since it is an omega-categorical structure, any countable model of $\mathsf{EDiag}(R)$ will again be isomorphic to $R$. This is not true if we replace $\mathsf{EDiag}(R)$ with $\mathsf{Diag}(R)$, since there are all sorts of ways to extend $R$ while ensuring it no longer satisfies the almost-sure theory of finite graphs. (For example, we could add a new vertex and connect it to all the vertices from $R$.)

## Remarks

• For $T' = T_{\mathsf{Diag}(M)}$ or $T_{\mathsf{EDiag}(M)}$ there is an obvious interpretation $T \to T'$ which induces for every $N \models T'$ a map of automorphism groups $\operatorname{Aut}_{\mathcal{L}(M)}(N) \to \operatorname{Aut}_{\mathcal{L}}(N)$, corresponding to the inclusion
$\operatorname{Aut}_{\mathcal{L}}(N/M) \hookrightarrow \operatorname{Aut}_{\mathcal{L}}(N)$

of the pointwise stabilizer of $M$ in $N$ into the full automorphism group of $N$.

• To add a distinct constant symbol to a theory is to adjoin a new global point to its syntactic category. This doesn’t do very much unless if you additionally specify its type, i.e. the ultrafilter of subobjects above it. When we pass to the quantifier-free diagram of a model, we specify constants named after the model up to quantifier-free types, and when we pass to the elementary diagram of a model, we specify constants named after the model up to complete types.

• A first-order theory T eliminates quantifiers if and only if it is “substructure-complete”: given any model $M$ of $T$ and any substructure $N \subseteq M$, $T_{\mathsf{Diag}(N)}$ is complete.

###### Remark

The process of passing from $T$ to $T_{\mathsf{Diag}(M)}$ (resp. $\mathsf{EDiag}$) is functorial in the way you would expect the process of passing from a category of models to a co-slice category of models to be on corepresenting objects.

That is (now eliminating imaginaries and working with the pretopos completions of syntactic categories): if $T'$ is an $\mathcal{L}'$-theory and $M$ is an $\mathcal{L}'$-structure, and $T$ is an $\mathcal{L}$-theory over $T'$ via an interpretation $T \overset{F}{\to} T'$, then there are naturally-induced interpretations

$T'_{\mathsf{Diag}(F^*M)} \to T_{\mathsf{Diag}(M)}$

and

$T'_{\mathsf{EDiag}(F^*M)} \to T_{\mathsf{EDiag}(M)}.$
###### Proof

The interpretation $F$ induces a “taking reducts” functor

$F^* \overset{\operatorname{df}}{=} \mathbf{Mod}(F) = \mathbf{Pretop}(F, \mathbf{Set}) : \mathbf{Mod}(T) \to \mathbf{Mod}(T').$

We restrict $F^*$ to the full subcategory consisting of those models of $T$ embedding (resp. elementarily embedding) the structure $M$. These are elementary classes, and so those full subcategories are sub-ultracategories of $\mathbf{Mod}(T)$. The restrictions of $F^*$ are ultrafunctors

$\underline{\mathbf{Mod}}(T_{\mathsf{Diag}(M)}) \to \underline{\mathbf{Mod}}(T'_{\mathsf{Diag}(F^*M)})$

and

$\underline{\mathbf{Mod}}(T_{\mathsf{EDiag}(M)}) \to \underline{\mathbf{Mod}}(T'_{\mathsf{EDiag}(F^*M)})$

because $F^*$ already was, and so by Makkai’s strong conceptual completeness, must be reflected by the desired interpretations.

(That the latter functor is well-defined just follows from the fact that specifying an object $c \in \mathbf{C}$ and a functor $G : \mathbf{C} \to \mathbf{D}$ naturally induces a functor on the co-slice categories $c/\mathbf{C} \to G(c)/\mathbf{D}$. That the former functor is well-defined is less automatic but still trivial to check.)

• Dave Marker, (2002), Model theory: an introduction, section 2.3