p. 23 — for a finite strategy set, conditions (3) and (4) are equivalent
ProvedHartSchmeidler.Compact.cond3_iff_cond4_of_finiteLet be a set of players, each strategy space a topological space with its Borel σ-algebra, and let carry the product topology and its Borel σ-algebra . Fix a player whose strategy set is finite and discrete, assume is bounded and -measurable, and let be a probability measure on . Then condition (3) for player ,
holds if and only if condition (4) for player ,
holds.
This shows that (4), the definition of a correlated equilibrium used for infinite strategy sets, reduces to the earlier condition (3) whenever a player's strategy set is finite, so §4 extends §3.
Formalization Note is the set of profiles with . The finite strategy set carries the discrete topology, so every subset of and every map is measurable. With bounded and measurable all the integrands are integrable, so the integrability requirement of the definition is not repeated here.
import Definitions.Def_HartSchmeidler_Compact_Game
namespace HartSchmeidler.Compact
open MeasureTheory
/-- p. 23: when the strategy set `Sⁱ` of player `i` is finite, condition (3) for player `i`
(one inequality per recommendation `r` and deviation `t`) is equivalent to condition (4) for
player `i` (one inequality per measurable deviation map `ζ : Sⁱ → Sⁱ`). -/
theorem cond3_iff_cond4_of_finite {ι : Type*} [DecidableEq ι] {S : ι → Type*}
[∀ j, TopologicalSpace (S j)] [∀ j, MeasurableSpace (S j)] [∀ j, BorelSpace (S j)]
(h : ι → Profile S → ℝ) (i : ι) [Fintype (S i)] [DiscreteTopology (S i)]
(hmeas : Measurable (h i)) (hbdd : ∃ M : ℝ, ∀ s, |h i s| ≤ M)
(p : Measure (Profile S)) [IsProbabilityMeasure p] :
(∀ r t : S i,
0 ≤ ∫ s in {s : Profile S | s i = r}, (h i s - h i (Function.update s i t)) ∂p) ↔
(∀ ζ : S i → S i, Measurable ζ →
0 ≤ ∫ s : Profile S, (h i s - h i (Function.update s i (ζ (s i)))) ∂p) := by sorry
end HartSchmeidler.Compact
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.