Theorem 3.7 — sufficient conditions for three-region and no-inspection policies
ProvedRossQC.Sufficient.theorem_3_7Consider the two-state model with , , and . Write for the discounted value of equation (3).
- If
then there are and an optimal stationary rule that produces for , inspects for , and revises for . When revision is optimal for at least one admissible belief, . 2. If
then there is and an optimal stationary rule that produces for and revises for , with no inspection interval.
These conditions distinguish parameter regimes in which a single inspection interval can be selected from those admitting a rule that never inspects.
Formalization Note “Optimal” means that the rule attains the minimum in equation (3) at every , including threshold points. Part (a)'s printed is false when revision is never optimal; the paper's proof allows an infinite threshold, represented here by . When , Lean's quotient with denominator is zero; the implications are still valid in that boundary case.
import Definitions.Def_RossQC_Sufficient_Model
namespace RossQC.Sufficient
/-- Ross, Theorem 3.7(a,b), pp. 592–593. A β-optimal stationary rule
selects a minimum of (3) at every admissible belief, including threshold
points. The printed bound `P₂ ≤ 1` in (a) is imposed when the optimal revise
region is nonempty; otherwise `P₂ > 1` allows that interval to be empty, as
the paper's proof explicitly permits. `0 < C` makes the displayed quotient
well-defined. Lean's zero-denominator convention applies when `π = 1`,
where `V_β(1) - V_β(π) = 0`; the two implications retain their intended
conclusions in that case. -/
theorem theorem_3_7 (M : Model)
(hβ0 : 0 < M.β) (hβ1 : M.β < 1)
(hπ0 : 0 ≤ M.π) (hπ1 : M.π ≤ 1)
(hC0 : 0 < M.C) (hCI : M.C < M.I) (hIR : M.I < M.R) :
((1 - M.β) * (M.R + M.β * M.value M.π) / M.C ≤
(M.R - M.I) / (M.β * (M.value 1 - M.value M.π)) →
∃ P₁ P₂ : ℝ, M.π ≤ P₁ ∧ P₁ ≤ P₂ ∧
((∃ P ∈ Set.Icc (0 : ℝ) 1,
M.rhs3 M.value P .revise = M.value P) → P₂ ≤ 1) ∧
∀ P ∈ Set.Icc (0 : ℝ) 1,
M.rhs3 M.value P (M.threeRegion P₁ P₂ P) = M.value P) ∧
((M.I + M.β * (M.value 1 - M.value M.π) ≥
M.C / (1 - M.β * (1 - M.π)) ∨
(M.R - M.I) / (M.β * (M.value 1 - M.value M.π)) ≤
(M.R / M.C) * (1 - M.β * (1 - M.π))) →
∃ P₁ : ℝ, M.π ≤ P₁ ∧
∀ P ∈ Set.Icc (0 : ℝ) 1,
M.rhs3 M.value P (M.twoRegion P₁ P) = M.value P) := by sorry
end RossQC.Sufficient
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.