Theorem 4.3: under Condition D, the index policy that continues a superprocess of maximal ν(S_i, x_i) with its Condition-D control is optimal for a SFAS
ProvedAllocationIndices.superprocess_index_theoremTheorem 4.3 (Index theorem for a SFAS). A policy for a simple family of alternative superprocesses comprised of , which all satisfy Condition D, is optimal if it is an index policy with respect to .
Formally, in discrete time: a decision process on a countable state space with finite control type and bounded rewards, , and a Condition-D control for ( feasible, and whenever it is optimal to select in state in it is optimal to apply ; the superprocesses of the family are copies of , a family of distinct superprocesses being the disjoint-union case). If a policy for the family of superprocesses at every decision time almost surely continues a superprocess of maximal index and applies the control , then is optimal: it is feasible and its payoff equals the supremum of the payoffs of all feasible policies from every initial state-vector. Since (Note 4.2 at ), is an index policy with respect to .
Why the control is pinned to . The book's proof runs the index policy with the controls of Condition D. An index policy that breaks a tie among controls differently need not be optimal. Take , and a state with two controls: pays and leads to a state paying forever, and pays and leads to a state paying forever. Both have index , Condition D holds with , and the index policy that applies earns against .
With singleton control sets this is Theorem 2.1 (the proved platform theorem). The book's proof reduces the semi-Markov case to stochastic discounting and runs a prevailing-stake argument; here the family is Markov with unit decision intervals.
import Definitions.Def_AllocationIndices_Superprocess open MeasureTheory ProbabilityTheory BanditAlgorithm
namespace AllocationIndices
theorem superprocess_index_theorem {S U : Type*} [MeasurableSpace S] [Countable S]
[MeasurableSingletonClass S] [MeasurableSpace U] [Fintype U] [Nonempty U]
[MeasurableSingletonClass U] (D : DecisionProcess S U) (hD : D.BoundedRewards)
{a : ℝ} (ha0 : 0 < a) (ha1 : a < 1) {g : S → U} (hg : IsConditionDControl D a g)
{n : ℕ} {π : SFASPolicy n S U} (hπ : IsSuperIndexPolicy D a g π) :
IsOptimalSFASPolicy D a π := by sorry
end AllocationIndices
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.