Lecture V (§5) — Proposition 5.4 (Scott 1981, PRG-19) #
Let x, y, and τ(x, y) have the same type 𝒟, and let g range over (𝒟 → 𝒟). Then the
equation
λx. !y. τ(x, y) = !g. λx. τ(x, g(x))
is true.
In this framework τ : 𝒟 × 𝒟 → 𝒟 is an approximable map. We model the two sides
directly:
- the left side
λx. !y. τ(x, y)ispfix τ := fix ∘ curry(τ), the approximable map sendingxto the least fixed point of the sectiony ↦ τ(x, y)(this is!y.τ(x,y), manifestly approximable by Theorem 4.2 — exactly Scott's appeal to 5.1 that "fis a function"); - the right side
!g. λx. τ(x, g(x))is the least fixed point(recOp τ).fixElementof the approximable operatorrecOp τ := curry(τ ∘ ⟨p₁, eval⟩)on the function space, which sends a mapgtoλx. τ(x, g(x)).
The proof is Scott's: pfix τ is a fixed point of recOp τ (so !g.… ⊑ pfix τ),
and conversely
the value (recOp τ).fixElement (x) is a fixed point of the section y ↦ τ(x, y)
(so
pfix τ (x) ⊑ (recOp τ).fixElement (x) for every x). Everything is at the level
of |𝒟| and the
function space, so the data (pfix, recOp) is choice-free
(#print axioms ⊆ {propext, Quot.sound}); the final map equality goes through the
permitted
ext_of_toElementMap.
The section y ↦ τ(x, y) of τ : 𝒟 × 𝒟 → 𝒟 at a fixed x, as an
approximable endomap of
𝒟 (the curried τ applied to x).
Equations
Instances For
The left-hand side λx. !y. τ(x, y) of Proposition 5.4: fix ∘ curry(τ).
Equations
Instances For
pfix τ (x) = !y. τ(x, y), the least fixed point of the section at x.
The operator λg. λx. τ(x, g(x)) of Proposition 5.4, as an approximable
endomap of the
function space (𝒟 → 𝒟): recOp τ := curry(τ ∘ ⟨p₁, eval⟩).
Equations
Instances For
recOp τ (g) (x) = τ(x, g(x)).
pfix τ is a fixed point of recOp τ (pointwise: τ(x, pfix τ x) = pfix τ x).
Proposition 5.4 (Scott 1981, PRG-19). λx. !y. τ(x, y) = !g. λx. τ(x, g(x)).