Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition C.1, p. 33 — the choice map Q is Lipschitz and equals the derivative dh* of the convex conjugate

Proved
ReinfRegGames.TimeAvg.proposition_C_1

by mikedeng1 · Oct 9, 2026 · Mathlib 0df444a (Lean v4.33.1)

choice-mapconvex-conjugatedanskinlipschitzp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1

Let BBB be a nonempty finite set, Δ=Δ(B)\Delta = \Delta(B)Δ=Δ(B) the simplex of probability vectors on BBB, and hhh a penalty function on Δ\DeltaΔ with strong convexity constant K>0K > 0K>0 (Definition 2.1). For a score vector y∈RBy \in \mathbb R^By∈RB, the choice map Q(y)Q(y)Q(y) is the maximizer of ⟨y∣x⟩−h(x)\langle y | x \rangle - h(x)⟨y∣x⟩−h(x) over x∈Δx \in \Deltax∈Δ, and h∗(y)=max⁡x∈Δ{⟨y∣x⟩−h(x)}h^*(y) = \max_{x\in\Delta}\{\langle y | x\rangle - h(x)\}h∗(y)=maxx∈Δ​{⟨y∣x⟩−h(x)} is the convex conjugate (C.1). Then:

  1. for every yyy a maximizer Q(y)Q(y)Q(y) exists;
  2. QQQ is Lipschitz: there is a constant LLL such that ∥Q(y)−Q(y′)∥≤L ∥y−y′∥\|Q(y) - Q(y')\| \le L\,\|y - y'\|∥Q(y)−Q(y′)∥≤L∥y−y′∥ for all y,y′y, y'y,y′;
  3. h∗h^*h∗ is (Fréchet) differentiable at every y∈RBy \in \mathbb R^By∈RB, with
dh∗(y)=Q(y),i.e.h∗(y+z)=h∗(y)+⟨z∣Q(y)⟩+o(∥z∥).dh^*(y) = Q(y), \qquad\text{i.e.}\qquad h^*(y + z) = h^*(y) + \langle z | Q(y)\rangle + o(\|z\|).dh∗(y)=Q(y),i.e.h∗(y+z)=h∗(y)+⟨z∣Q(y)⟩+o(∥z∥).

The identity Q=dh∗Q = dh^*Q=dh∗ is what turns the Fenchel coupling into a Lyapunov-type function along the reinforcement learning dynamics (Lemma C.6); the Lipschitz property makes the trajectory x(t)=Q(y(t))x(t) = Q(y(t))x(t)=Q(y(t)) continuous.

Formalization Note The Lipschitz constant is existential, as on the page (which gives none). Distances are those of Lean's sup metric on RB\mathbb R^BRB; Lipschitz continuity does not depend on the choice of norm. Item 2 implies that the maximizer is unique. The derivative is stated as HasFDerivAt with the linear functional z↦∑βxβzβz \mapsto \sum_\beta x_\beta z_\betaz↦∑β​xβ​zβ​. The hypothesis that BBB is nonempty is the page's standing assumption that Δ\DeltaΔ is the simplex of Rn\mathbb R^nRn, n≥1n \ge 1n≥1 (otherwise no maximizer exists).

Preamble
import Mathlib
import Definitions.Def_agt_games
import Definitions.Def_ReinfRegGames_Extinction_Model
Formal statement
namespace ReinfRegGames.TimeAvg

theorem proposition_C_1 {B : Type*} [Fintype B] [DecidableEq B] [Nonempty B]
    (h : (B → ℝ) → ℝ) (K : ℝ) (hpen : ReinfRegGames.Extinction.IsPenalty h K) :
    (∀ y : B → ℝ, ∃ x, ReinfRegGames.Extinction.IsChoice h y x) ∧
    (∃ L : ℝ, ∀ y y' x x' : B → ℝ, ReinfRegGames.Extinction.IsChoice h y x → ReinfRegGames.Extinction.IsChoice h y' x' →
      dist x x' ≤ L * dist y y') ∧
    (∀ y x : B → ℝ, ReinfRegGames.Extinction.IsChoice h y x →
      HasFDerivAt (ReinfRegGames.Extinction.conj h) (∑ β, x β • ContinuousLinearMap.proj (R := ℝ) (φ := fun _ : B => ℝ) β) y) := by sorry

end ReinfRegGames.TimeAvg
Source
Mertikopoulos & Sandholm, Learning in games via reinforcement and regularization, arXiv:1407.6267v2, p. 33, Proposition C.1
Human review
  • Endorsed by Shuze Chen · Oct 9, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 9, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me