Theorem 25 — strict hierarchy of the tilted classes
OpenTiltedCvx.Supplement.theorem_25convex-analysisgeneralized-convexitynonsmooth-analysisoptimizationp2o-gran-per-chapterp2o-plan-bookp2o-strengthenedp2o-v1
Let , and let stand for for every , for for every , or for for every . Then is the class of proper lower semicontinuous convex functions, and
Each family is strictly ordered in its parameter, and all positive-parameter classes lie strictly between the convex functions and the pseudoconvex class .
Formalization Note The inclusions and both equalities are stated for every finite-dimensional ; strictness is stated for nontrivial ().
Differences from the source:
- generalization: functions on ℝ^d → functions on a finite-dimensional real inner product space E; reason: every such E is isometric to ℝ^d and ℝ^d = EuclideanSpace ℝ (Fin d) is an instance, so the statement specializes to the paper's
- encoding: ℝ̄ = ℝ ∪ {±∞}, dom(f) = {f finite} → EReal and the published TiltedCvx.effDom; reason: Lean convention; same instances
- encoding: f : ℝ^d → ℝ̄ is convex → the published epigraph convexity WassersteinDRO.ConvexLossP1.IsConvexExt f (the epigraph {(x, t) : f x ≤ t} is convex); reason: the standard notion for extended-valued functions; Mathlib's ConvexOn needs an ordered-module codomain, which EReal is not
- reading: the strict inclusions ⊊ in ℝ^d → ⊂ under Nontrivial E (d ≥ 1); the inclusions and equalities for every E; reason: for d = 0 every proper function is convex and all the classes coincide, so ℝ^d with d ≥ 1 is the paper's setting; the published corollary_8 uses the same reading
Preamble
import Mathlib import Definitions.Def_TiltedCvx_IsProper import Definitions.Def_TiltedCvx_classF import Definitions.Def_TiltedCvx_classMono import Definitions.Def_TiltedCvx_classSec import Definitions.Def_WassersteinDRO_ConvexLossP1_IsConvexExt import Definitions.Def_TiltedCvx_Supplement_classZeroPlus
Formal statement
namespace TiltedCvx.Supplement
/-- **Theorem 25 (strict hierarchy).** Pedramfar–Khanna–Quinn–Aggarwal, arXiv:2611.03472,
Theorem 25, p.34. Let `0 < γ′ < γ < 1`, and let `C^θ` stand for `F^θ_mono` for every `θ`, for
`F^θ_sec` for every `θ`, or for `F^θ` for every `θ`. Then `C¹` is the class of proper lsc convex
functions, and
`C¹ ⊊ C^γ ⊊ C^γ′ ⊊ ⋃_{0<θ≤1} C^θ = ⋃_{0<θ≤1} F^θ ⊊ F^{0+} ⊊ F⁰` (29).
**Formalization Note** The three families are the published `classMono`, `classSec`, `classF`;
`F^{0+}` is `classZeroPlus` and `F⁰` the published `classF 0`. Convexity is the published
epigraph convexity `IsConvexExt`. The inclusions and the two equalities are stated for every
finite-dimensional `E`; strictness (`⊂`) is stated for nontrivial `E`, i.e. `d ≥ 1`: for `d = 0`
there is a single proper function, so all the classes coincide. -/
theorem theorem_25 {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
[FiniteDimensional ℝ E] (γ γ' : ℝ) (hγ'0 : 0 < γ') (hγ'γ : γ' < γ) (hγ1 : γ < 1)
(C : ℝ → Set (E → EReal))
(hC : C = TiltedCvx.classMono ∨ C = TiltedCvx.classSec ∨ C = TiltedCvx.classF) :
(C 1 = {f | TiltedCvx.IsProper f ∧ LowerSemicontinuous f ∧
WassersteinDRO.ConvexLossP1.IsConvexExt f}) ∧
(C 1 ⊆ C γ ∧ C γ ⊆ C γ' ∧ C γ' ⊆ ⋃ θ ∈ Set.Ioc (0 : ℝ) 1, C θ ∧
(⋃ θ ∈ Set.Ioc (0 : ℝ) 1, C θ) = ⋃ θ ∈ Set.Ioc (0 : ℝ) 1, TiltedCvx.classF θ ∧
(⋃ θ ∈ Set.Ioc (0 : ℝ) 1, (TiltedCvx.classF θ : Set (E → EReal))) ⊆ classZeroPlus ∧
classZeroPlus ⊆ (TiltedCvx.classF 0 : Set (E → EReal))) ∧
(Nontrivial E →
C 1 ⊂ C γ ∧ C γ ⊂ C γ' ∧ C γ' ⊂ ⋃ θ ∈ Set.Ioc (0 : ℝ) 1, C θ ∧
(⋃ θ ∈ Set.Ioc (0 : ℝ) 1, (TiltedCvx.classF θ : Set (E → EReal))) ⊂ classZeroPlus ∧
classZeroPlus ⊂ (TiltedCvx.classF 0 : Set (E → EReal))) := by sorry
end TiltedCvx.Supplement
Source
Pedramfar–Khanna–Quinn–Aggarwal, Tilted Convexity and an Exact Offline–Online Separation, arXiv:2611.03472, Theorem 25, p.34