Mathematical foundations · PhysLib theorem
Sion's Minimax Theorem
Under compactness, convexity, semicontinuity, quasiconvexity, quasiconcavity, and boundedness hypotheses, PhysLib proves equality of the two optimization orders.
What it says
What the theorem says
PhysLib first establishes the easy weak-duality direction, then uses compactness to reduce an infinite closed-set family to a finite one and combines quasiconvexity and quasiconcavity to obtain the reverse inequality.
Context
Why it matters
Sion's theorem is a widely used minimax principle underlying optimization arguments in information theory, game theory, and variational analysis.
Inside the proof
Proof architecture
The proof combines order-theoretic weak duality with a compact finite-subcover argument that upgrades pointwise separating choices into one global response.
Step 1
Establish weak duality
For any function with bounded sections, the supremum of pointwise infima cannot exceed the infimum of pointwise suprema.
sup_y inf_x f(x,y) <= inf_x sup_y f(x,y)
Show the Lean ingredients
ciSup_ciInf_le_ciInf_ciSuph_bdd_1h_bdd_0Step 2
Use compactness to select finitely many tests
Assuming a strict gap, closed sublevel sets cover the compact first domain; PhysLib extracts a finite family that preserves the contradiction setup.
compact S implies a finite closed subfamily suffices
Show the Lean ingredients
exists_betweenhS₁.elim_finite_subfamily_closedu.NonemptyStep 3
Collapse the assumed gap
The finite minimax lemma and semicontinuity bounds produce a response whose value contradicts the two levels chosen inside the hypothetical gap.
inf_x sup_y f(x,y) = sup_y inf_x f(x,y)
Show the Lean ingredients
sion_exists_min_finlowerSemicontinuousOn_iff_isClosed_preimagehb₂.not_geAuthoritative formal type
Exact PhysLib statement
sion_minimax.{u_4, u_5} {M : Type u_4} [NormedAddCommGroup M] {N : Type u_5} {f : M → N → ℝ} {S : Set M} {T : Set N} (hfc₂ : ∀ y ∈ T, LowerSemicontinuousOn (fun x => f x y) S) (hS₁ : IsCompact S) (hS₃ : S.Nonempty) (hT₃ : T.Nonempty) [Module ℝ M] [ContinuousSMul ℝ M] [AddCommGroup N] [TopologicalSpace N] [SequentialSpace N] [T2Space N] [ContinuousAdd N] [Module ℝ N] [ContinuousSMul ℝ N] (hfc₁ : ∀ x ∈ S, UpperSemicontinuousOn (f x) T) (hfq₂ : ∀ y ∈ T, QuasiconvexOn ℝ S fun x => f x y) (hfq₁ : ∀ x ∈ S, QuasiconcaveOn ℝ T (f x)) (hT₂ : Convex ℝ T) (hS₂ : Convex ℝ S) (h_bddA : BddAbove (Set.image2 f S T)) (h_bddB : BddBelow (Set.image2 f S T)) : ⨅ x, ⨆ y, f ↑x ↑y = ⨆ y, ⨅ x, f ↑x ↑yThis is the theorem's exact Lean type. The diagrams and commentary are explanatory; the pinned statement and source are authoritative.
Provenance and verification
Sources and technical checks
Show declaration provenance and Lean checks
sion_minimaxQuantumInfo.ForMathlib.SionMinimax6bd0b4b99d3dsorryAx reportedThe declaration footprint describes source layout, not proof difficulty or the size of its dependencies. The pinned checks verify the cited Lean theorem; semantic review remains a separate PhysAtlas judgment.



