Support vector at a point (§7, (33))
DefinitionTiltedCvx_Supplement_IsSupportVectorconvex-analysisgeneralized-convexitynonsmooth-analysisoptimizationp2o-gran-per-chapterp2o-plan-bookp2o-v1
Let , , and . A vector is a support vector of at if the tilted support inequality (4) holds at with against every point of :
The value-and-support oracle of Section 7 returns such a vector ; every Clarke subgradient of a –tilted convex function is one.
Definition code
import Mathlib
import Definitions.Def_TiltedCvx_psi
open scoped InnerProductSpace
namespace TiltedCvx.Supplement
/-- **Support vector (§7, (33)).** Pedramfar–Khanna–Quinn–Aggarwal, arXiv:2611.03472, §7, p.37:
for `f : K → ℝ` and `x ∈ K`, a vector `q` is a support vector at `x` if inequality (4) holds at
`x` with `v = q` for every `y ∈ K`, i.e. `f(y) − f(x) ≥ ψγ(⟨y − x, q⟩)` for all `y ∈ K`.
**Formalization Note** `f : E → ℝ`; only its values on `K` enter. -/
def IsSupportVector {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
(γ : ℝ) (K : Set E) (f : E → ℝ) (x q : E) : Prop :=
∀ y ∈ K, TiltedCvx.psi γ ((⟪y - x, q⟫_ℝ : ℝ) : EReal) ≤ ((f y - f x : ℝ) : EReal)
end TiltedCvx.Supplement
Source
Pedramfar–Khanna–Quinn–Aggarwal, Tilted Convexity and an Exact Offline–Online Separation, arXiv:2611.03472, §7, (33), p.37