Gittins index as the fair retirement charge
ProvedBanditAlgorithm.gittinsRetirementValue_eq_zero_iff_index_lebanditsgittins-indexoptimal-stopping
The retirement-game value at state is zero exactly when the charge is at least the Gittins index:
Thus the stopping-ratio definition of agrees with its fair-charge characterization.
Preamble
import Theorems.Thm_BanditAlgorithm_gittinsRetirementValue_eq_zero_of_index_le import Theorems.Thm_BanditAlgorithm_gittinsRetirementValue_pos_of_lt_index open MeasureTheory ProbabilityTheory ENNReal
Formal statement
theorem BanditAlgorithm.gittinsRetirementValue_eq_zero_iff_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) :
gittinsRetirementValue P r α γ x = 0 ↔
gittinsIndex P r α x ≤ γ := by
sorrySource
Lattimore and Szepesvári, Bandit Algorithms (CUP 2020), Eqs. (35.8)--(35.9), printed pp.448--449.