AFTD Auto-Formalizing Theoretical Domains

OnlineLearning.convex_compact_minimax

theoremverifiedformalizationimported Lean

Convex-compact minimax theorem on the line. Let $X, Y \subseteq \bbr$ be nonempty convex sets with $X$ compact, and let $f : \bbr \times \bbr \to \bbr$ be a payoff function that is bounded above and below on $X \times Y$. Suppose that for every $y \in Y$ the map $x \mapsto f(x,y)$ is continuous and convex on $X$, and that for every $x \in X$ the map $y \mapsto f(x,y)$ is concave on $Y$. Then the minimax identity holds: \[ \inf_{x \in X}\,\sup_{y \in Y}\, f(x,y) \;=\; \sup_{y \in Y}\,\inf_{x \in X}\, f(x,y). \]

Statement

set_option relaxedAutoImplicit false in
set_option autoImplicit false in
theorem OnlineLearning.convex_compact_minimax {X Y : Set ℝ} {f : ℝ → ℝ → ℝ}
    (h : ConvexCompactMinimaxHypotheses X Y f) :
    ConvexCompactMinimaxStatement X Y f
Formalizes
Convex-compact minimax theorem on the line, as formalized in TCSlib (`OnlineLearning.convex_compact_minimax`)
Lean proof
By Karim Abdel Sadek, Mark Bedaywi, from github.com/Shilun-Allan-Li/tcslib at c8a02591b1e5 (Copyright (c) 2026 Karim Abdel Sadek and Mark Bedaywi. All rights reserved, Apache-2.0). Compiled here against leanprover/lean4:v4.33.1, Mathlib v4.33.1; written for leanprover/lean4:v4.25.0.
Verified
07 Oct 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 3

Lean source view module on GitHub

set_option relaxedAutoImplicit false inset_option autoImplicit false in/-- The convex-compact minimax theorem: for a nonempty compact convex `X ⊆ ℝ`, a nonempty convex `Y ⊆ ℝ`, and a bounded payoff `f` that is convex and continuous in the row variable `x` for each `y ∈ Y` and concave in the column variable `y` for each `x ∈ X` (the hypotheses bundled in `ConvexCompactMinimaxHypotheses`), the upper value `inf_{x ∈ X} sup_{y ∈ Y} f x y` equals the lower value `sup_{y ∈ Y} inf_{x ∈ X} f x y` (the conclusion `ConvexCompactMinimaxStatement`). [CBL06, Thm 7.1]. Deviation: specialized to subsets of `ℝ`, and proved via the Hahn–Banach separation route of `ConvexMinimaxSeparation` rather than the source's no-regret argument, whose last step needs an extra uniformity hypothesis (see `ConvexMinimaxNoRegret`). This theorem is intentionally a thin wrapper. It hides the proof-route choice from downstream files and currently delegates to the completed separation proof. -/theorem OnlineLearning.convex_compact_minimax {X Y : Set ℝ} {f : ℝ → ℝ → ℝ}    (h : ConvexCompactMinimaxHypotheses X Y f) :    ConvexCompactMinimaxStatement X Y f := by  -- Keep the public theorem independent of proof-route details.  At present,  -- the separation proof is the strongest completed route.  exact convex_compact_minimax_by_separation h
See it in the graphDiscuss this resultChallenge it