Section 2, Eq. (6) — both E(S_n/n) and (p₁+⋯+p_n)/n tend to γ + δ²/(1−γ) under R₁
ProvedRobbinsSeqDesign.StayOrSwitch.expected_average_tendstobandit-algorithmsmarkov-chainsp2o-batch-pfp2bp2o-gran-per-chapterp2o-plan-paperp2o-v1
Let coins and have head probabilities , not both and not both ; let be the payoff of the -th toss under the rule , , the probability of heads on the -th toss, and as in (5). Then
This is the long-run expected payoff per toss of a person who uses .
Formalization Note The Lean states both limits explicitly. The expectation is a Bochner integral of the average payoff over the trajectory law.
Preamble
import Mathlib import Definitions.Def_RobbinsSeqDesign_StayOrSwitch_Rules open MeasureTheory ProbabilityTheory Filter Topology
Formal statement
namespace RobbinsSeqDesign.StayOrSwitch
/-- Section 2, Eq. (6), p. 531: if `α, β` are not both `0` and not both `1`, then under rule
`R₁`, both `E(S_n/n)` and `(p₁ + ⋯ + p_n)/n` tend to
`γ + δ²/(1 − γ)`. -/
theorem expected_average_tendsto (α β : unitInterval)
(hnot0 : ¬((α : ℝ) = 0 ∧ (β : ℝ) = 0)) (hnot1 : ¬((α : ℝ) = 1 ∧ (β : ℝ) = 1)) :
Tendsto (fun n => expectedAverage ruleR1 α β n) atTop
(𝓝 (gamma α β + delta α β ^ 2 / (1 - gamma α β))) ∧
Tendsto (fun n => (∑ i ∈ Finset.Icc 1 n, headProb ruleR1 α β i) / n) atTop
(𝓝 (gamma α β + delta α β ^ 2 / (1 - gamma α β))) := by sorry
end RobbinsSeqDesign.StayOrSwitch
Source
Robbins, Some aspects of the sequential design of experiments, Bull. Amer. Math. Soc. 58 (1952), p. 531, Section 2, Eq. (6)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.