§2 p. 3 (T*2.4.4) and Proposition 3.1, p. 8 — K* is proper and F* is a ν-self-scaled barrier for K*
OpenSelfScaledLongStep.PrimalDual.dual_isSelfScaledBarrierLet be a proper cone (closed, convex, pointed, with nonempty interior), let be a -self-scaled barrier for (Definition 2.1), and let
be its conjugate barrier on the dual cone (2.6). Then:
- is again a proper cone (closed, convex, pointed, with nonempty interior);
- is a -self-scaled barrier for : it is a -self-concordant logarithmically homogeneous (nondegenerate) barrier for , and for all ,
This is the primal–dual symmetry of self-scaled barriers stated on p. 3 (via T*2.4.4 of Nesterov–Nemirovskii: is a -self-concordant logarithmically homogeneous barrier for , and is pointed with nonempty interior) and in Proposition 3.1, p. 8, (3.3): ; the conjugate of over equals on . It lets every result stated for the primal pair be applied verbatim to the dual pair , which the paper does throughout ("the corresponding statements for the dual case have an absolutely symmetric form").
Formalization Note The conclusion is IsProperCone (dualCone K) ∧ IsSelfScaledBarrier (dualCone K) (conj K F) ν with the same definitions as for the primal side, so the conjugate of the conjugate is conj (dualCone K) (conj K F); is identified with through the Euclidean inner product.
import Mathlib import Definitions.Def_dualCone import Definitions.Def_SelfScaledIPM_ShortStep_Setting open scoped InnerProductSpace open SelfScaledIPM.ShortStep
namespace SelfScaledLongStep.PrimalDual
/-- **Dual symmetry of self-scaled barriers** (p. 3, via T*2.4.4 of Nesterov–Nemirovskii, and
Proposition 3.1, p. 8). If `F` is a `ν`-self-scaled barrier for the proper cone `K`, then the dual
cone `K*` is proper and the conjugate barrier `F* = conj K F` is a `ν`-self-scaled barrier for
`K*`. -/
theorem dual_isSelfScaledBarrier {n : ℕ} (K : Set (EuclideanSpace ℝ (Fin n))) (hK : IsProperCone K)
(F : EuclideanSpace ℝ (Fin n) → ℝ) (ν : ℝ) (hν : 1 ≤ ν) (hF : IsSelfScaledBarrier K F ν) :
IsProperCone (ConvexOptimization.dualCone K) ∧
IsSelfScaledBarrier (ConvexOptimization.dualCone K) (conj K F) ν := by sorry
end SelfScaledLongStep.PrimalDual