Proposition C.2(iii), p. 34 — whenever in
ProvedReinfRegGames.Extinction.proposition_C_2_iiibregman-divergencecontinuityconvex-analysisp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be a -strongly convex penalty function on the simplex and . Let be a sequence in converging to a point . Let and be the (real) one-sided derivatives of (C.3). Then the Bregman divergences converge:
Continuity of on is the second ingredient, with (C.11), of Proposition C.4.
Formalization Note By Proposition C.2(i) the one-sided derivatives at points of are real numbers; the statement takes them as data satisfying the predicate HasOneSidedDeriv.
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
open Filter Topology
theorem proposition_C_2_iii {B : Type*} [Fintype B] (h : (B → ℝ) → ℝ) (K : ℝ) (hpen : IsPenalty h K)
(p : B → ℝ) (hp : p ∈ stdSimplex ℝ B) (xs : ℕ → B → ℝ) (x : B → ℝ)
(hxs : ∀ j, xs j ∈ deltaP p) (hx : x ∈ deltaP p) (hlim : Tendsto xs atTop (𝓝 x))
(ds : ℕ → ℝ) (d : ℝ) (hds : ∀ j, HasOneSidedDeriv h (xs j) (p - xs j) (ds j))
(hd : HasOneSidedDeriv h x (p - x) d) :
Tendsto (fun j => h p - h (xs j) - ds j) atTop (𝓝 (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(iii)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.