Increasing bounded continuous approximations preserving convexity or concavity
OpenQAlgorithms.Papers.Independent.Support50.P33.monotoneShapeApproximationLet be a real Hausdorff locally convex topological vector space, and let be convex and von Neumann bounded. Assume the relative topology of is generated by a countable subfamily of the ambient defining seminorms, and that is complete and separable in the resulting metrizable additive uniformity. The set is allowed to be empty. All regularity and convergence below are relative to .
Write for the Borel probability measures on with the weak convergence topology. The barycenter is the Pettis integral of inclusion: every continuous real linear functional is integrable against , and . The set is -compact when the barycenter preimage of every compact subset of is compact in ; it is pointwise -compact when every barycenter fiber is compact.
Write for the bounded continuous real functions on . For an extended-valued cost , let be its greatest convex minorant and its greatest lower semicontinuous convex minorant. A finite lower bound means for all for some , so positive infinity is permitted but negative infinity is excluded on .
Assume is -compact and stable: the midpoint map , , is open in the relative topologies. Let be lower semicontinuous with a finite lower bound. If is convex, there is a sequence of bounded continuous real convex functions on satisfying
If instead is concave, the same assertion holds with every concave. Extended concavity means convexity of , and the convention for both signs of infinity at zero weight is . The finite bound on each may depend on ; the limit can equal positive infinity. No uniform convergence is asserted.
The result preserves the convex or concave shape during monotone continuous approximation.
import Definitions.Def_QAlgorithms_Papers_Independent_Support50_P33
noncomputable section
open Set Filter MeasureTheory TopologicalSpace
open scoped Topology ENNReal
namespace QAlgorithms.Papers.Independent.Support50.P33
open QAlgorithms.Papers.Independent.Frontier18.P10
open QAlgorithms.Papers.Independent.Dependency50.P25
attribute [local instance] QAlgorithms.Papers.Independent.Support50.P33.paperMeasurable
attribute [local instance] QAlgorithms.Papers.Independent.Support50.P33.paperBorel
section LocallyConvexResults
variable {E : Type*} [UniformSpace E] [AddCommGroup E] [Module ℝ E]
[hgroup : IsUniformAddGroup E] [hsmul : ContinuousSMul ℝ E]
[hlc : LocallyConvexSpace ℝ E] [hhaus : T2Space E]
include hgroup hsmul hlc hhaus
/-- Corollary, arXiv:1004.3792v1, PDF p.5. Both the convex and concave
alternatives are retained. Stability is relative openness of midpoint
mixing. Limits may be positive infinite; approximants are real, bounded
and continuous, with a bound for each term and pointwise convergence. -/
theorem monotoneShapeApproximation (A : Set E)
(hA : GeneralBarycenterSpace A) (hc : Convex ℝ A)
(hstable : StableConvexSet A) (hmu : MuCompact A) :
(∀ f : E → EReal, LowerSemicontinuous (fun x : A => f x) →
(∃ c : ℝ, ∀ x ∈ A, (c : EReal) ≤ f x) → extendedConvexOn A f →
∃ g : ℕ → E → ℝ,
(∀ n, BoundedContinuousOn A (g n) ∧ ConvexOn ℝ A (g n)) ∧
(∀ x ∈ A, Monotone (fun n => g n x)) ∧
∀ x ∈ A, Tendsto (fun n => (g n x : EReal)) atTop (𝓝 (f x))) ∧
(∀ f : E → EReal, LowerSemicontinuous (fun x : A => f x) →
(∃ c : ℝ, ∀ x ∈ A, (c : EReal) ≤ f x) →
extendedConvexOn A (fun x => -f x) →
∃ g : ℕ → E → ℝ,
(∀ n, BoundedContinuousOn A (g n) ∧ ConcaveOn ℝ A (g n)) ∧
(∀ x ∈ A, Monotone (fun n => g n x)) ∧
∀ x ∈ A, Tendsto (fun n => (g n x : EReal)) atTop (𝓝 (f x))) := by sorry
end LocallyConvexResults
end QAlgorithms.Papers.Independent.Support50.P33
end
Read-back
What the Lean code literally says, in plain math · gpt-6.1-sol
Let be the ambient real vector space with its uniform structure and induced topology, and let be closed, convex, and von Neumann bounded, with complete, separable, and metrizable. Assume that, for every open , the set is open in . For a Borel probability measure on , define to be a chosen point for which every continuous real linear functional is -integrable on and satisfies , if such a point exists, and zero otherwise; assume also that is compact for every compact . Then two assertions hold. First, for every whose restriction to is lower semicontinuous, which has a real lower bound on , and which satisfies for all and all real with , there exists a sequence such that each is continuous and convex on , each has an absolute bound on for some real , and, for every , is nondecreasing and converges to in the extended-real topology. Second, the same assertion holds with each concave on when the convexity hypothesis on is replaced by for all such . These inequalities use extended-real arithmetic, including zero coefficients; the lower bound permits , and the bounds need not be uniform in . The set may be empty, making all conditions and conclusions quantified over its points vacuous.
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.