Lifting facts between contexts, with an abnormality guard
DefinitionTranscendenceTowerLiftingMcCarthy's specializes axiom is a default rule for carrying facts from one context to another:
In words: what holds in a context still holds in a more specific one, unless it is abnormal for that particular transfer. Note the shape — an abnormality predicate guarding a conditional — which is the same apparatus circumscription is built on.
These definitions model such a transfer. Nothing here assumes the two contexts are stacked one above the other: the Sherlock Holmes context and the US-legal-history context are siblings rather than levels of a tower, and specializes is meant to relate contexts of that kind too.
A transfer is described by a blocking set: the assertions declared abnormal, and therefore not carried across. A blocking set is admissible when the resulting context is satisfiable, and minimal when no admissible blocking set is strictly smaller. Circumscribing the abnormality predicate is then exactly the preference for a minimal blocking set — block as little as possible.
Formalization note. Assertions are lifted, not consequences: a context transfers what it says, and the target draws its own conclusions. Blocking a consequence set would be ill-behaved in any case, since such a set is closed under logical equivalence, so excluding while retaining blocks nothing at all.
/-
Lifting facts between contexts, with an abnormality guard.
McCarthy's `specializes` axiom is a default rule for transferring facts between
contexts:
specializes(c₁,c₂) ∧ ¬ab(p,c₁,c₂) ∧ ist(c₁,p) → ist(c₂,p)
"what holds in a context still holds in a more specific one, unless it is
abnormal for that transfer". These definitions model such a transfer. Nothing
here presupposes that the two contexts are stacked: the Sherlock Holmes context
and the US-legal-history context are siblings, and `specializes` relates
contexts of that kind too.
A transfer is described by a **blocking set** -- the assertions declared
abnormal, hence not carried across. Circumscribing `ab` is then the preference
for a blocking set that is as small as possible.
Assertions are lifted, not consequences: a context transfers what it *says*,
and the target draws its own conclusions. Blocking a consequence set would be
ill-behaved anyway, since such a set is closed under equivalence and excluding
`p` while retaining `¬¬p` blocks nothing.
Definitions only -- every theorem about these lives in a separate submission.
-/
import Definitions.Def_TranscendenceTowerContextTheories
import Mathlib.Data.Set.Basic
set_option autoImplicit false
open TranscendenceTower TranscendenceTower.ContextTheories
namespace TranscendenceTower.Lifting
/-- Lift `src` into `tgt`, carrying across every assertion of `src` except
those in the blocking set `A`. -/
def Lift (src tgt A : Set L0) : Set L0 := tgt ∪ (src \ A)
/-- A blocking set is **admissible** when the lifted context is satisfiable.
Declaring assertions abnormal is how a conflict between the two contexts gets
resolved, so the question is which declarations suffice. -/
def Admissible (src tgt A : Set L0) : Prop := ∃ v : Val, Models (Lift src tgt A) v
/-- A **minimal** blocking set: admissible, with no admissible blocking set
strictly smaller. This is the circumscription of `ab` for a single transfer --
block as little as possible. -/
def MinBlock (src tgt A : Set L0) : Prop :=
Admissible src tgt A ∧ ∀ B : Set L0, Admissible src tgt B → B ⊆ A → A ⊆ B
end TranscendenceTower.Lifting