Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Main Theorem and converse — c.e.d.s are exactly the action distributions of Bayes-rational information systems

Proved
Aumann1987.BayesRational.ced_iff_bayes_rational_system

by mikedeng1 · Oct 4, 2026 · Mathlib 0df444a (Lean v4.33.1)

bayesian-rationalitycorrelated-equilibriumgame-theoryp2o-batch-pfp1ap2o-gran-per-chapterp2o-plan-paperp2o-v1

Let GGG be an nnn-person game with action sets SiS^iSi and payoffs hih^ihi, and let QQQ be a real function on action nnn-tuples S=S1×⋯×SnS=S^1\times\dots\times S^nS=S1×⋯×Sn. Then QQQ is a correlated equilibrium distribution of GGG if and only if there is an information system — a finite set Ω\OmegaΩ of states, a common prior ppp on Ω\OmegaΩ, an information partition Pi\mathcal P^iPi of Ω\OmegaΩ for each player, and action functions si:Ω→Si\mathbf s^i:\Omega\to S^isi:Ω→Si measurable with respect to Pi\mathcal P^iPi — in which every player is Bayes rational at every state and

Q(a)=p{ω∈Ω: s(ω)=a}for every a∈S.Q(a)=p\{\omega\in\Omega:\ \mathbf s(\omega)=a\}\qquad\text{for every }a\in S.Q(a)=p{ω∈Ω: s(ω)=a}for every a∈S.

This is the two-sided form announced in the paper's introduction: if each player maximizes expected utility given his information at every state, under a common prior, the chosen actions form a correlated equilibrium, and conversely every correlated equilibrium arises in this way.

Formalization Note. State spaces and probability spaces range over Type (universe 0), which loses nothing for finite sets. Bayes rationality at a state whose information cell has probability 000 is vacuous (0/0=00/0=00/0=0 in Lean); the prior need not have full support.

Preamble
import Mathlib
import Definitions.Def_agt_games
import Definitions.Def_Aumann1987_BayesRational_CorrelatedEquilibrium
import Definitions.Def_Aumann1987_BayesRational_InformationSystem
Formal statement
namespace Aumann1987.BayesRational

/-- **Main Theorem and its converse** (Aumann 1987, Econometrica 55, Main Theorem, Sect. 3, p. 7,
PDF p. 8, and its converse, Sect. 4d, p. 11, PDF p. 12; announced in Sect. 1, p. 2, PDF p. 3):
"Thus under Bayesian rationality, the set of all information systems corresponds precisely to the
set of all correlated equilibria."

A function `Q` on action `n`-tuples is a correlated equilibrium distribution of the game with
payoffs `h` if and only if there is an information system — a finite set `Ω` of states, a common
prior `p` on `Ω`, an information partition `𝒫ⁱ` for each player, and action functions `sⁱ`
measurable w.r.t. `𝒫ⁱ` — in which every player is Bayes rational at every state and the
distribution of the action `n`-tuple `s` is `Q`.

**Formalization Note.** Finite sets of states and finite probability spaces range over `Type`
(universe 0), which loses nothing for finite sets. Bayes rationality at a state whose cell has
probability `0` is vacuous (`condExp` divides by `0` and returns `0`); the paper's argument
never uses such states. The prior is not required to have full support. -/
theorem ced_iff_bayes_rational_system {ι : Type*} [Fintype ι] [DecidableEq ι] {S : ι → Type*}
    (h : ι → (∀ i, S i) → ℝ) (Q : (∀ i, S i) → ℝ) :
    IsCED h Q ↔
      ∃ (Ω : Type) (_ : Fintype Ω) (I : InformationSystem Ω S),
        (∀ i ω, IsBayesRationalAt h I.p I.P I.s i ω) ∧ distr I.p I.s = Q := by sorry

end Aumann1987.BayesRational
Source
Aumann, Correlated Equilibrium as an Expression of Bayesian Rationality, Econometrica 55 (1987), DOI 10.2307/1911154, p. 7 (PDF p. 8), Main Theorem, and p. 11 (PDF p. 12), Sect. 4d; announced p. 2 (PDF p. 3)
Human review
  • Endorsed by Shuze Chen · Oct 4, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 4, 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