Sion's minimax theorem
ProvedFamousTheorems.sion_minimax_theoremSion's minimax theorem. Let be a nonempty compact convex subset and a convex subset of real topological vector spaces, and with densely linearly ordered. Suppose that for each , is lower semicontinuous and quasiconvex on , and for each , is upper semicontinuous and quasiconcave on . Then
whenever these infima and suprema exist.
Sion's theorem (1958) generalizes von Neumann's minimax theorem for zero-sum games from bilinear to quasiconvex–quasiconcave functions. It is used in game theory, robust optimization, and statistical decision theory.
Formalization note. Mathlib's Sion.minimax. Since β is only a densely ordered linear order, suprema and infima are given as data with IsLUB/IsGLB hypotheses: sup_y x is , inf_sup is of these, inf_x y is , and sup_inf is of these. The conclusion is inf_sup = sup_inf.
import Mathlib
namespace FamousTheorems
theorem sion_minimax_theorem {E F β : Type*} [LinearOrder β] [DenselyOrdered β] [TopologicalSpace E] [AddCommGroup E] [Module ℝ E]
[IsTopologicalAddGroup E] [ContinuousSMul ℝ E] [TopologicalSpace F] [AddCommGroup F] [Module ℝ F]
[IsTopologicalAddGroup F] [ContinuousSMul ℝ F] {X : Set E} {Y : Set F} {f : E → F → β}
(hX_ne : X.Nonempty) (hX_cpt : IsCompact X) (hX : Convex ℝ X) (hY : Convex ℝ Y)
(hfy_lsc : ∀ y ∈ Y, LowerSemicontinuousOn (fun x => f x y) X) (hfy_qcvx : ∀ y ∈ Y, QuasiconvexOn ℝ X fun x => f x y)
(hfx_usc : ∀ x ∈ X, UpperSemicontinuousOn (fun y => f x y) Y) (hfx_qccv : ∀ x ∈ X, QuasiconcaveOn ℝ Y fun y => f x y)
(sup_y : E → β) (h_sup_y : ∀ x ∈ X, IsLUB {b | ∃ y ∈ Y, f x y = b} (sup_y x))
(inf_sup : β) (h_inf_sup : IsGLB {b | ∃ x ∈ X, sup_y x = b} inf_sup)
(inf_x : F → β) (h_inf_x : ∀ y ∈ Y, IsGLB {b | ∃ x ∈ X, f x y = b} (inf_x y))
(sup_inf : β) (h_sup_inf : IsLUB {b | ∃ y ∈ Y, inf_x y = b} sup_inf) : inf_sup = sup_inf := by sorry
end FamousTheoremsConfirmed by the mission captain (proposal self-audit).
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.