Proposition C.2(ii), p. 34 — , with iff
ProvedReinfRegGames.Extinction.proposition_C_2_iibregman-divergenceconvex-analysisp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1strong-convexity
Let be a -strongly convex penalty function on the simplex and let . Suppose the one-sided derivative is a real number , so that the Bregman divergence of (C.4) is finite. Then
- ;
- if and only if ;
Together with (C.11), this makes and the Fenchel coupling quantitative measures of the distance between and .
Formalization Note When the page sets , and all three claims hold trivially in the extended reals; the statement therefore quantifies over the points at which the one-sided derivative is a real number. The norm is the Euclidean norm of the definition of penalty function.
Preamble
import Mathlib import Definitions.Def_agt_games import Definitions.Def_ReinfRegGames_Extinction_Model import Definitions.Def_ReinfRegGames_Extinction_Bregman
Formal statement
namespace ReinfRegGames.Extinction
theorem proposition_C_2_ii {B : Type*} [Fintype B] (h : (B → ℝ) → ℝ) (K : ℝ) (hpen : IsPenalty h K)
(p : B → ℝ) (hp : p ∈ stdSimplex ℝ B) (x : B → ℝ) (hx : x ∈ stdSimplex ℝ B) (d : ℝ)
(hd : HasOneSidedDeriv h x (p - x) d) :
0 ≤ h p - h x - d ∧ (h p - h x - d = 0 ↔ p = x) ∧
1 / 2 * K * sqDist x p ≤ h p - h x - d := by sorry
end ReinfRegGames.Extinction
Source
Mertikopoulos & Sandholm, Learning in games via reinforcement and regularization, arXiv:1407.6267v2, p. 34, Proposition C.2(ii), (C.6)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.