Relation (18) — Pᵢ(x, s, M) = ∏_{j≠i} ∂φⱼ/∂M is a distribution function in M, equal to 1 for M ≥ M₍ᵢ₎
ProvedWhittleGittins.Retirement.relation18_P_distributionIn Whittle's bandit process with per-project horizons , let be the value of the single-project -process for project (14), let be its index, the infimal with , and define
Regarded as a function of , is non-negative, non-decreasing and at most , and
the maximum being over the other projects that can still be operated (). Moreover each such index satisfies . Hence is a distribution function in with all its support in .
This is the step of the proof of Theorem 1 that allows of (13) to be integrated by parts against .
Formalization Note is the right derivative. Projects with have , derivative and index ; they are excluded from the maximum defining (if no other project is active, and ). Projects are indexed by Fin N.
import Mathlib import Definitions.Def_WhittleGittins_Retirement_BanditProcess
namespace WhittleGittins.Retirement
open MeasureTheory ProbabilityTheory
/-- Relation (18) (p. 147): `Pᵢ(x, s, M) = ∏_{j≠i} ∂φⱼ(xⱼ, sⱼ, M)/∂M` is, as a function of `M`,
non-negative, non-decreasing, at most one, and equal to one for `M ≥ M₍ᵢ₎ = max_{j≠i} Mⱼ` (the
maximum over the other active projects); moreover every index of an active project is `≤ K`. -/
theorem relation18_P_distribution {N : ℕ} {X : Fin N → Type*} [∀ i, MeasurableSpace (X i)]
(B : BanditProcess N X) (x : ∀ j, X j) (s : Fin N → ℕ) (i : Fin N) :
(∀ M, 0 ≤ Pdist B x s i M) ∧ Monotone (Pdist B x s i) ∧ (∀ M, Pdist B x s i M ≤ 1) ∧
(∀ M, (∀ j, j ≠ i → 0 < s j → index B j (x j) (s j) ≤ M) → Pdist B x s i M = 1) ∧
(∀ j, 0 < s j → index B j (x j) (s j) ≤ B.K) := by sorry
end WhittleGittins.Retirement
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.