Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 6.2, p. 26 — in 2-player zero-sum games with an interior equilibrium, time-averaged (RL) play converges to the Nash set

Proved
ReinfRegGames.TimeAvg.proposition_6_2

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

learning-in-gamesnash-equilibriump2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1reinforcement-learningtime-averagezero-sum-games

Let G\mathcal GG be a 2-player zero-sum game, u1=−u2u_1 = -u_2u1​=−u2​, with finite strategy sets A1,A2\mathcal A_1, \mathcal A_2A1​,A2​, and suppose G\mathcal GG has an interior Nash equilibrium x∗x^*x∗ (every pure strategy has positive probability under x∗x^*x∗). Let each player use a penalty function hkh_khk​ on Δ(Ak)\Delta(\mathcal A_k)Δ(Ak​) (Definition 2.1), and let (y(t),x(t))(y(t), x(t))(y(t),x(t)) be any orbit of the reinforcement learning dynamics (RL),

y˙k=vk(x),xk=Qk(yk)=arg⁡max⁡xk′∈Δ(Ak){⟨yk∣xk′⟩−hk(xk′)}.\dot y_k = v_k(x), \qquad x_k = Q_k(y_k) = \arg\max_{x_k' \in \Delta(\mathcal A_k)} \{\langle y_k | x_k'\rangle - h_k(x_k')\}.y˙​k​=vk​(x),xk​=Qk​(yk​)=argxk′​∈Δ(Ak​)max​{⟨yk​∣xk′​⟩−hk​(xk′​)}.

Then the time average xˉ(t)=t−1∫0tx(s) ds\bar x(t) = t^{-1}\int_0^t x(s)\,dsxˉ(t)=t−1∫0t​x(s)ds converges to the set NE(G)\mathrm{NE}(\mathcal G)NE(G) of mixed Nash equilibria of G\mathcal GG:

dist⁡(xˉ(t),NE(G))→0(t→∞).\operatorname{dist}\big(\bar x(t), \mathrm{NE}(\mathcal G)\big) \to 0 \qquad (t \to \infty).dist(xˉ(t),NE(G))→0(t→∞).

The trajectory x(t)x(t)x(t) itself need not converge (in Matching Pennies it cycles, possibly hitting the boundary of the strategy space infinitely often), but its time average does. The result holds for every penalty function, steep or not, and so covers both the replicator dynamics and the projection dynamics.

Formalization Note Players are indexed by Fin 2, and u1=−u2u_1 = -u_2u1​=−u2​ is u0(s)+u1(s)=0u_0(s) + u_1(s) = 0u0​(s)+u1​(s)=0 for every pure profile sss. Interiority is xkα∗>0x^*_{k\alpha} > 0xkα∗​>0 for all k,αk, \alphak,α. Distances are those of Lean's sup metric on profiles; the Nash set contains x∗x^*x∗, so the distance is to a nonempty set. (RL) is used in its differential form, equivalent to the integral form on p. 7. No steepness is assumed on the penalty functions, and strong convexity is stated for the Euclidean norm.

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

theorem proposition_6_2 {A : Fin 2 → Type*} [∀ k, Fintype (A k)] [∀ k, DecidableEq (A k)]
    (u : Fin 2 → (∀ k, A k) → ℝ) (hzs : ∀ s, u 0 s + u 1 s = 0)
    (xstar : ∀ k, A k → ℝ) (hNash : AGT.IsMixedNash u xstar) (hint : ∀ k α, 0 < xstar k α)
    (h : ∀ k, (A k → ℝ) → ℝ) (K : Fin 2 → ℝ) (hpen : ∀ k, ReinfRegGames.Extinction.IsPenalty (h k) (K k))
    (y x : ℝ → ∀ k, A k → ℝ) (horb : ReinfRegGames.Extinction.IsRLOrbit u h y x) :
    Filter.Tendsto (fun t => Metric.infDist (timeAvg x t) {z | AGT.IsMixedNash u z})
      Filter.atTop (nhds 0) := by sorry

end ReinfRegGames.TimeAvg
Source
Mertikopoulos & Sandholm, Learning in games via reinforcement and regularization, arXiv:1407.6267v2, p. 26, Proposition 6.2
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