Retirement is optimal above the Gittins fair charge
ProvedBanditAlgorithm.gittinsRetirementValue_eq_zero_of_index_lebanditsgittins-indexoptimal-stopping
If the charge is at least the Gittins index , every admissible play-then-stop strategy has nonpositive expected discounted net reward. Since immediate retirement gives zero, the retirement-game value satisfies .
Preamble
import Definitions.Def_GittinsRetirementValue open MeasureTheory ProbabilityTheory ENNReal
Formal statement
theorem BanditAlgorithm.gittinsRetirementValue_eq_zero_of_index_le
{S : Type*} [MeasurableSpace S]
(P : Kernel S S) [IsMarkovKernel P]
{r : S → ℝ} (hr : Measurable r) {α : ℝ}
(hα0 : 0 < α) (hα1 : α < 1)
(hint : DiscountedRewardIntegrable P r α)
(x : S) (γ : ℝ) (hg : gittinsIndex P r α x ≤ γ) :
gittinsRetirementValue P r α γ x = 0 := by
sorrySource
Lattimore and Szepesvári, Bandit Algorithms (CUP 2020), fair-charge characterization Eq. (35.8), ratio characterization Eq. (35.9), and Lemma 35.7, printed pp.448--449.