lin-0084

6.2 Instance lift, schema expansion, and constraint stability

Three operations that sound similar must not be identified.

Pointwise instance lift.

The schema \(\mathcal S\) stays fixed and \(I\) is replaced by \(T_{\mathcal S}I\). This is the construction proved above.

Syntactic tangent expansion.

A new sketch \(\tau \mathcal S\) is presented with tangent objects, projections, zero sections, lifts, or vector field symbols. This changes the database schema.

Internal tangent schema.

If \(\mathcal S\) itself has tangent structure, one may apply its tangent functor internally. A general database schema does not have such structure automatically.

The second construction is the one most useful for a persistent infinitesimal CSQL atlas. It makes the added infinitesimal columns and foreign keys explicit. It should be called a tangent expansion, rather than written \(T\mathcal S\), unless a tangent structure on \(\mathcal S\) has actually been supplied.

Database schemas often present more than paths. They may declare products, pullbacks, limits, equations, or lifting constraints. Let \(\operatorname {Mod}(\mathbb S,\mathcal C)\) denote the full subcategory of \([\mathcal S,\mathcal C]\) satisfying a database sketch \(\mathbb S\).

Definition 6.2 Tangent-stable database sketch

A database sketch \(\mathbb S\) is tangent-stable in \(\mathcal C\) when \(I\models \mathbb S\) implies \(T_{\mathcal S}I\models \mathbb S\), with the declared comparison data preserved.

Path equations are preserved because \(T\) is a functor. A declared limit or pullback is preserved only when \(T\) preserves the relevant cone. The tangent category axioms guarantee selected pullbacks needed by tangent structure; they do not say that \(T\) preserves every database constraint.

Proposition 6.3 Restriction to constrained instances

If \(\mathbb S\) is tangent-stable, the pointwise tangent functor restricts to an endofunctor

\[ T_{\mathbb S}: \operatorname {Mod}(\mathbb S,\mathcal C) \longrightarrow \operatorname {Mod}(\mathbb S,\mathcal C). \]

If tangent stability fails, its missing comparison or failed cone is a typed obstruction rather than an infinitesimal database instance of the same declared type.

Proof

The first claim is the definition of tangent stability applied on objects; naturality of \(T\alpha \) supplies its action on instance morphisms. For the second, a lifted instance that violates a declared cone does not lie in the model subcategory. The failure is therefore a factorization or universal property obstruction of the kind introduced in Chapter 3.

Design principle

Do not silently coerce \(T_{\mathcal S}I\) back into the constrained database category. Either prove tangent stability, record the necessary comparison maps, or retain their failure as database-level obstruction data.