Proposition C.2(i), p. 34 — D_h(p, x) < +∞ whenever x ∈ Δ_p
ProvedReinfRegGames.StrictStable.proposition_C_2_ibregman-divergenceconvex-analysisp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1penalty-function
Let be a -strongly convex penalty function on and . is the one-sided derivative (C.3) and is the Bregman divergence (C.4), which may equal when . is the union of the relative interiors of the faces of that contain (C.5).
If , then the one-sided derivative exists as a real number; in particular
Near a vertex , every point of lies in , so the Bregman divergence is finite there; the proof of Theorem 5.2 uses this to build the neighbourhood of .
Formalization Note The one-sided derivative is encoded as the predicate "the difference quotient tends to as " with real, and as ; no real-valued function with a junk value stands for .
Preamble
import Mathlib import Definitions.Def_agt_games import Definitions.Def_ReinfRegGames_Extinction_Model
Formal statement
namespace ReinfRegGames.StrictStable
theorem proposition_C_2_i {B : Type*} [Fintype B] (h : (B → ℝ) → ℝ) (K : ℝ) (hpen : ReinfRegGames.Extinction.IsPenalty h K)
(p : B → ℝ) (hp : p ∈ stdSimplex ℝ B) (x : B → ℝ) (hx : x ∈ stdSimplex ℝ B)
(hxp : ∀ α, 0 < p α → 0 < x α) :
∃ d : ℝ, Filter.Tendsto (fun t : ℝ => (h (x + t • (p - x)) - h x) / t) (nhdsWithin 0 (Set.Ioi 0)) (nhds d) := by sorry
end ReinfRegGames.StrictStable
Source
Mertikopoulos & Sandholm, Learning in games via reinforcement and regularization, arXiv:1407.6267v2, pp. 33–34, Proposition C.2(i), with (C.3)–(C.5)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.