Theorem 15 — five equivalent forms of γ–tilted monotonicity
OpenTiltedCvx.Supplement.theorem_15convex-analysisgeneralized-convexitynonsmooth-analysisoptimizationp2o-gran-per-chapterp2o-plan-bookp2o-strengthenedp2o-v1
Let be proper and lower semicontinuous with , and . The following are equivalent:
- (Clarke) is –tilted monotone;
- (limiting) is –tilted monotone;
- (Fréchet) is –tilted monotone;
- is convex, is radially continuous, is locally Lipschitz, and is –tilted monotone;
- is convex, is radially continuous, is locally Lipschitz, and on is –tilted monotone:
for all where and exist, the gradient taken with respect to the affine hull of .
The class can therefore be tested with any of the three subdifferentials, or with relative gradients on the relative interior.
Formalization Note is the operator equal to for and empty elsewhere. is any with the published IsRelGrad f K x g.
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|ri(K) is locally Lipschitz → LocallyLipschitzOn (ri K) of x ↦ (f x).toReal, ri K = intrinsicInterior ℝ K; reason: f is finite on ri(K) ⊆ dom f, so toReal reads its values; same condition
- encoding: ∂̂f|ri(K) → x ↦ {v | x ∈ ri K ∧ v ∈ ∂̂f(x)}; reason: the restriction of the operator to ri(K), empty elsewhere
- encoding: ∇f(x) w.r.t. aff(K), where it exists → any g with the published IsRelGrad f K x g; reason: the published relative-gradient predicate; such g is unique
Preamble
import Mathlib import Definitions.Def_TiltedCvx_effDom import Definitions.Def_TiltedCvx_IsProper import Definitions.Def_TiltedCvx_psi import Definitions.Def_TiltedCvx_TiltedMonotone import Definitions.Def_TiltedCvx_IsRelGrad import Definitions.Def_ClarkeStrat_Proj_Setting import Definitions.Def_NonconvexSplitting_Shared_LimitingSubdiff import Definitions.Def_NAPPAL_VPEB_frechetSubdiff import Definitions.Def_TiltedCvx_Supplement_RadiallyContinuous open scoped InnerProductSpace open ClarkeStrat.Proj NonconvexSplitting.Shared NAPPAL.VPEB
Formal statement
namespace TiltedCvx.Supplement
/-- **Theorem 15 (five equivalent forms of γ–tilted monotonicity).**
Pedramfar–Khanna–Quinn–Aggarwal, arXiv:2611.03472, Theorem 15, p.23. Assume `f : ℝ^d → ℝ̄` is a
proper lsc function with effective domain `K := dom(f)` and `γ ∈ (0, 1]`. Then the following are
equivalent:
(i) `∂̄f` is γ–tilted monotone;
(ii) `∂f` is γ–tilted monotone;
(iii) `∂̂f` is γ–tilted monotone;
(iv) `K` is convex, `f` is radially continuous, `f|ri(K)` is locally Lipschitz, and `∂̂f|ri(K)`
is γ–tilted monotone;
(v) `K` is convex, `f` is radially continuous, `f|ri(K)` is locally Lipschitz, and
`x ↦ {∇f(x)} : ri(K) ⇒ ℝ^d` is γ–tilted monotone, i.e. `⟨∇f(y), y − x⟩ ≥ ψγ(⟨∇f(x), y − x⟩)` for
all `x, y ∈ ri(K)` where `∇f(x)` and `∇f(y)` exist, `∇f` taken with respect to `aff(K)`.
**Formalization Note** `∂̄f`, `∂f`, `∂̂f` are the published `ClarkeSubdiff`, `LimitingSubdiff`
and `frechetSubdiff`. `∂̂f|ri(K)` is the operator equal to `∂̂f(x)` for `x ∈ ri(K)` and `∅`
elsewhere. "`f|ri(K)` locally Lipschitz" is `LocallyLipschitzOn (ri K)` of `x ↦ (f x).toReal`
(finite there). `∇f(x)` is any `g` with the published `IsRelGrad f K x g`. For `γ > 0` every tilt
is defined. -/
theorem theorem_15 {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
[FiniteDimensional ℝ E] (γ : ℝ) (hγ0 : 0 < γ) (hγ1 : γ ≤ 1) (f : E → EReal)
(hf : TiltedCvx.IsProper f) (hlsc : LowerSemicontinuous f) :
List.TFAE
[TiltedCvx.TiltedMonotone γ (ClarkeSubdiff f),
TiltedCvx.TiltedMonotone γ (LimitingSubdiff f),
TiltedCvx.TiltedMonotone γ (frechetSubdiff f),
Convex ℝ (TiltedCvx.effDom f) ∧ RadiallyContinuous f ∧
LocallyLipschitzOn (intrinsicInterior ℝ (TiltedCvx.effDom f))
(fun x => (f x).toReal) ∧
TiltedCvx.TiltedMonotone γ
(fun x => {v | x ∈ intrinsicInterior ℝ (TiltedCvx.effDom f) ∧
v ∈ frechetSubdiff f x}),
Convex ℝ (TiltedCvx.effDom f) ∧ RadiallyContinuous f ∧
LocallyLipschitzOn (intrinsicInterior ℝ (TiltedCvx.effDom f))
(fun x => (f x).toReal) ∧
∀ x ∈ intrinsicInterior ℝ (TiltedCvx.effDom f),
∀ y ∈ intrinsicInterior ℝ (TiltedCvx.effDom f), ∀ gx gy : E,
TiltedCvx.IsRelGrad f (TiltedCvx.effDom f) x gx →
TiltedCvx.IsRelGrad f (TiltedCvx.effDom f) y gy →
TiltedCvx.psi γ ((⟪gx, y - x⟫_ℝ : ℝ) : EReal) ≤ ((⟪gy, y - x⟫_ℝ : ℝ) : EReal)]
:= by sorry
end TiltedCvx.Supplement
Source
Pedramfar–Khanna–Quinn–Aggarwal, Tilted Convexity and an Exact Offline–Online Separation, arXiv:2611.03472, Theorem 15, p.23