Documentation

Books.ConvexAnalysis_Rockafellar_1970.Chap06.section30_part5

Theorem 6.30.8: for a concave bifunction G : ℝ^m → ℝ^n with perturbation function h = sup G, a vector u* ∈ ℝ^m is a Kuhn--Tucker vector for (Q) if and only if h(0) is finite and -u* ∈ ∂ h(0). In Lean, finiteness of h(0) is expressed by h 0 ≠ ⊤ and h 0 ≠ ⊥.

Corollary 30.1.30

Corollary 30.1.31

Proposition 30.1.32

Corollary 30.1.33

theorem fenchelYoung_inequality_and_eq_iff_mem_subdifferential {n : } (g : (Fin n)EReal) (hg : ProperConvexFunctionOn Set.univ g) (x xStar : Fin n) :
↑(x ⬝ᵥ xStar) g x + fenchelConjugate n g xStar (g x + fenchelConjugate n g xStar = ↑(x ⬝ᵥ xStar) IsEuclideanSubgradientAt g x xStar)

Theorem 6.30.5: if g : ℝ^n → (-∞, +∞] is proper and convex, then for every x, x* ∈ ℝ^n one has the Fenchel--Young inequality ⟪x, x*⟫ ≤ g x + g*(x*). Moreover, equality holds if and only if x* is a Euclidean subgradient of g at x, equivalently x* ∈ ∂ g(x) under the standard identification of ℝ^n with its dual.

Theorem 6.30.6: if g : ℝ^n → (-∞, +∞] is a proper closed convex function and g* is its Fenchel conjugate, then for every x, x* ∈ ℝ^n one has x* ∈ ∂ g(x) if and only if x ∈ ∂ g*(x*), where both subdifferentials are interpreted in the Euclidean sense on Fin n → ℝ.

Corollary 30.5.2