RobustOptimization.isPsi_dual_iff
DisprovedThe value is symmetric under duality.
Fix a setup: source space with norm (conjugate norm ) and closed convex cone , destination space with norm (conjugate ) and closed convex cone . For a linear map , a real number is the -value of this setup if and only if it is the -value of the dual setup (source with and , destination with and ) for the adjoint map .
\\Psi_\\Xi(\\mathcal{M}) = \\psi \\iff \\Psi_{\\Xi_*}(\\mathcal{M}^*) = \\psi.This is the core of Proposition 11.4.1 of Ben-Tal, El Ghaoui and Nemirovski. It is what allows a bound on to be computed on whichever side of the duality is tractable: the two -values coincide as real numbers, not merely as abstract quantities.
Formal statement
import Definitions.Def_RobustOptimization_GRCConic
import Mathlib
open Matrix Finset
namespace RobustOptimization
/-- The Psi value is symmetric under duality (Ben-Tal et al., Prop. 11.4.1, core).
A real number is the Psi-value of a setup (source norm NE, destination norm
NF, closed convex cones KE, KF, linear map M) if and only if it is the
Psi-value of the dual setup (conjugate norms NFs, NEs, dual cones,
adjoint map Mᵀ). The forward direction unfolds Psi as a double supremum
and swaps the order of maximization using dual-norm symmetry; the reverse
is the same argument applied to the bidual setup. -/
theorem isPsi_dual_iff {p q : ℕ} (NE NEs : (Fin p → ℝ) → ℝ)
(NF NFs : (Fin q → ℝ) → ℝ)
(KE : Set (Fin p → ℝ)) (KF : Set (Fin q → ℝ))
(hNE : IsNormOn NE) (hNF : IsNormOn NF)
(hNEs : IsDualNorm NE NEs) (hNFs : IsDualNorm NF NFs)
(hKEcl : IsClosed KE) (hKEcv : Convex ℝ KE)
(hKEcone : ∀ c : ℝ, 0 ≤ c → ∀ z ∈ KE, c • z ∈ KE)
(hKFcl : IsClosed KF) (hKFcv : Convex ℝ KF)
(hKFcone : ∀ c : ℝ, 0 ≤ c → ∀ z ∈ KF, c • z ∈ KF)
(M : Matrix (Fin q) (Fin p) ℝ) (ψ : ℝ) :
IsPsi NE NF KE KF M ψ ↔ IsPsi NFs NEs (DualCone KF) (DualCone KE) Mᵀ ψ := by sorry
end RobustOptimization