Suprema and partial infima preserve convexity
ProvedQAlgorithms.Papers.Independent.Bridge50.P45.lemmaII3Let be a convex subset of a real vector space and an arbitrary set. If every slice of a real-valued function is convex on , then
is convex in the extended epigraph sense. If is also a convex subset of a real vector space and is jointly convex on , then
is convex in the same sense. Empty extrema have their extended values, and unbounded values are not replaced by real default values.
These closure properties justify convex optimization of families without finiteness assumptions on the indexing set.
import Definitions.Def_QAlgorithms_Papers_Independent_Bridge50_P45
noncomputable section
attribute [local instance] Classical.propDecidable Classical.decEq
open scoped Matrix ComplexConjugate ComplexOrder ENNReal Topology
open Filter Set
namespace QAlgorithms.Papers.Independent.Bridge50.P45
open QAlgorithms.Papers.Independent.Information100.P60 (Mat copies)
open QAlgorithms.Papers.Independent.Information100.P07 (support)
open QAlgorithms.Papers.Independent.Information100.P71 (supportPower SupportIncluded)
open QAlgorithms.Papers.Independent.Dependency50.P21 (relativeEntropy)
open QAlgorithms.Papers.Independent.Dependency50.P28 (pinch)
open QAlgorithms.Papers.Independent.Dependency50.P35 (lnOp)
open QAlgorithms.Papers.Independent.Next46.P23 (eigenvalues)
open QAlgorithms (spectralProjector)
/-- Lemma II.3, PDF p.4 (9)-(10). Extended epigraphs cover arbitrary, empty,
and unbounded indexing sets. The second clause requires joint convexity. -/
theorem lemmaII3 :
(∀ (V Y : Type) [AddCommGroup V] [Module ℝ V] (S : Set V) (f : V → Y → ℝ),
Convex ℝ S → (∀ y, ConvexOn ℝ S (fun x => f x y)) →
EConvexOn S (fun x => ⨆ y, (f x y : EReal))) ∧
(∀ (V W : Type) [AddCommGroup V] [Module ℝ V] [AddCommGroup W] [Module ℝ W]
(S : Set V) (T : Set W) (f : V × W → ℝ),
Convex ℝ S → Convex ℝ T → ConvexOn ℝ (S ×ˢ T) f →
EConvexOn S (fun x => ⨅ y : T, (f (x,y.1) : EReal))) := by sorry
end QAlgorithms.Papers.Independent.Bridge50.P45
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
Two statements hold. First, for every real module whose additive structure is an abelian group, every type , every convex set , and every function such that is convex on for each , the extended-real function has a convex real-height epigraph , and is convex. Second, for every pair of real modules with abelian additive groups, convex sets , , and real-valued function convex on , the extended-real function likewise has a convex real-height epigraph over . No topology or nonemptiness is assumed: an empty gives , an empty gives , and unbounded suprema or infima may also be infinite.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.