Theorem 18 — Jensen form of tilted secant convexity
OpenTiltedCvx.Supplement.theorem_18convex-analysisgeneralized-convexitynonsmooth-analysisoptimizationp2o-gran-per-chapterp2o-plan-bookp2o-strengthenedp2o-v1
Let and . Then is –tilted secant convex if and only if
where is the binary tilted mean (14)–(15) (, the arithmetic mean).
This weakens Jensen's inequality by down-weighting the smaller endpoint value by ; at it is convexity, at quasiconvexity.
Formalization Note is encoded by its proper extension, a function that never takes .
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 : K → ℝ ∪ {+∞} on a convex K → f : E → EReal with f x ≠ ⊥ for all x (its proper extension); reason: both sides read f only on dom f; every such function is the extension of itself on K = E, and conversely
- encoding: M_{γ,λ}(a, b) of (14)–(15) → binMean γ λ a b on the real values f(y), f(z); reason: the paper's formula, including the γ = 0 convention
Preamble
import Mathlib import Definitions.Def_TiltedCvx_effDom import Definitions.Def_TiltedCvx_TiltedSecant import Definitions.Def_TiltedCvx_Supplement_binMean
Formal statement
namespace TiltedCvx.Supplement
/-- **Theorem 18 (Jensen form).** Pedramfar–Khanna–Quinn–Aggarwal, arXiv:2611.03472, Theorem 18,
p.25. Let `γ ∈ [0, 1]` and `f : K → ℝ ∪ {+∞}`. Then `f` is γ–tilted secant convex if and only if
`f(λy + (1 − λ)z) ≤ M_{γ,λ}(f(y), f(z))` for all `y, z ∈ dom(f)` and `λ ∈ [0, 1]` (16).
**Formalization Note** `f : K → ℝ ∪ {+∞}` (with `K` the convex set of §4) is encoded by its
proper extension, a function `E → EReal` that never takes `−∞` (`+∞` off `K`); only `dom f`
enters both sides. `M_{γ,λ}` is `binMean`, with the `γ = 0` convention (15); its arguments are the
real values `f(y), f(z)`. -/
theorem theorem_18 {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
[FiniteDimensional ℝ E] (γ : ℝ) (hγ0 : 0 ≤ γ) (hγ1 : γ ≤ 1) (f : E → EReal)
(hbot : ∀ x, f x ≠ ⊥) :
TiltedCvx.TiltedSecant γ f ↔
∀ y ∈ TiltedCvx.effDom f, ∀ z ∈ TiltedCvx.effDom f, ∀ l : ℝ, 0 ≤ l → l ≤ 1 →
f (l • y + (1 - l) • z) ≤ ((binMean γ l (f y).toReal (f z).toReal : ℝ) : EReal) := by sorry
end TiltedCvx.Supplement
Source
Pedramfar–Khanna–Quinn–Aggarwal, Tilted Convexity and an Exact Offline–Online Separation, arXiv:2611.03472, Theorem 18, p.25