Documentation

LeanPool.Erdos111.Probability.Minimax

A finite minimax lemma #

Let finitely many pure strategies x have payoffs f i x against finitely many constraints i. If every nonnegative combination of the constraints is met by some pure strategy, then one mixed strategy meets all the constraints at once. The proof separates the convex hull of the payoff vectors from the closed lower orthant at c.

theorem Erdos111.exists_mixed_le {ι : Type u_1} {X : Type u_2} [Fintype ι] [Fintype X] (f : ι → X → ℝ) (c : ι → ℝ) (h : ∀ (ρ : ι → ℝ), (∀ (i : ι), 0 ≤ ρ i) → ∃ (x : X), ∑ i : ι, ρ i * f i x ≤ ∑ i : ι, ρ i * c i) :
∃ (p : X → ℝ), (∀ (x : X), 0 ≤ p x) ∧ ∑ x : X, p x = 1 ∧ ∀ (i : ι), ∑ x : X, p x * f i x ≤ c i

If every nonnegative combination of the constraints f i ≤ c i is met by a pure strategy, a single mixed strategy meets every constraint.