Proposition A.1(iii), p. 31 — Q_α(y) → 0 whenever y_α − y_β → −∞ for some β ≠ α
ProvedReinfRegGames.StrictStable.proposition_A_1_iiichoice-maplearning-in-gamesp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1penalty-function
Let be a penalty function on the simplex of a finite set (Definition 2.1, strong convexity constant ), and let be its choice map, . Let in . Then whenever ; uniformly:
A strategy whose score falls arbitrarily far behind that of another strategy is played with vanishing probability. In the proof of Theorem 5.2 this turns the divergence of the relative scores into convergence .
Formalization Note The page's " whenever " is stated in the uniform form that the proof (A.3)–(A.4) gives; it is equivalent to the sequential form. is the predicate " maximizes over ".
Preamble
import Mathlib import Definitions.Def_agt_games import Definitions.Def_ReinfRegGames_Extinction_Model
Formal statement
namespace ReinfRegGames.StrictStable
theorem proposition_A_1_iii {B : Type*} [Fintype B] (h : (B → ℝ) → ℝ) (K : ℝ) (hpen : ReinfRegGames.Extinction.IsPenalty h K)
(α β : B) (hβα : β ≠ α) :
∀ ε > 0, ∃ M : ℝ, ∀ y x : B → ℝ, ReinfRegGames.Extinction.IsChoice h y x → y α - y β < -M → x α < ε := by sorry
end ReinfRegGames.StrictStable
Source
Mertikopoulos & Sandholm, Learning in games via reinforcement and regularization, arXiv:1407.6267v2, p. 31, Proposition A.1(iii) (proof (A.3)–(A.4))
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.