Theorem 3.4 — exactly when always producing is optimal
ProvedRossQC.AlwaysProduce.theorem_3_4Consider the two-state deterioration model with , , and . Write for the policy that always produces without inspection, for its discounted cost, and for the optimal discounted value. Then is optimal from every initial belief exactly when
If , there is a one-sided neighborhood of within on which revision is strictly cheaper in the Bellman equation than either production or inspection. Thus every discounted-optimal policy revises there.
This gives an exact parameter boundary for when inspection and revision can both be omitted, and it identifies forced revision when the boundary fails.
Formalization Note “ is optimal” means for every . “Every optimal policy revises” is stated as revision being the unique minimizing action. The positive bad-state cost is implicit in the model and explicit here.
import Mathlib import Definitions.Def_RossQC_AlwaysProduce_Model
namespace RossQC.AlwaysProduce
/-- Ross, *Quality Control under Markovian Deterioration*, Theorem 3.4(a,b),
p. 592. Formalization Note: `psi` is the discounted series cost of `R⁰`;
`R⁰` is β-optimal exactly when it equals `value` throughout `[0,1]`.
“Every β-optimal policy revises” is represented by revision being the unique
minimizer of equation (3). “Near 1” is a one-sided neighborhood in `[0,1]`.
`0 < C` makes explicit the positive bad-state cost implicit in the model. -/
theorem theorem_3_4 (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) :
((∀ P ∈ Set.Icc (0 : ℝ) 1, M.psi P = M.value P) ↔
M.C / (1 - M.β * (1 - M.π)) ≤ M.R) ∧
(M.R < M.C / (1 - M.β * (1 - M.π)) →
∃ ε : ℝ, 0 < ε ∧
∀ P ∈ Set.Icc (0 : ℝ) 1, 1 - ε < P →
M.rhs3 M.value P .revise < M.rhs3 M.value P .produce ∧
M.rhs3 M.value P .revise < M.rhs3 M.value P .inspect) := by sorry
end RossQC.AlwaysProduce
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.