Convex Optimization · Problem 9.20

Equality constrained minimization — Boyd & Vandenberghe

Mathematics — Optimization · conjugate-self-concordant-function

Problem

Conjugate of a self-concordant function. Suppose f:Rn→Rf : \mathbf{R}^n \to \mathbf{R} is closed, strictly convex, and self-concordant. We show that its conjugate (or Legendre transform) f∗f^* is self-concordant. ((a)) Show that for each y∈dom f∗y \in \mathbf{dom}\, f^*, there is a unique x∈dom fx \in \mathbf{dom}\, f that satisfies y=∇f(x)y = \nabla f(x). Hint. Refer to the result of exercise 9.19. ((b)) Suppose yˉ=∇f(xˉ)\bar{y} = \nabla f(\bar{x}). Define g(t)=f(xˉ+tv),h(t)=f∗(yˉ+tw)g(t) = f(\bar{x} + tv), \qquad h(t) = f^*(\bar{y} + tw) where v∈Rnv \in \mathbf{R}^n and w=∇2f(xˉ)vw = \nabla^2 f(\bar{x}) v. Show that g′′(0)=h′′(0),g′′′(0)=−h′′′(0).g''(0) = h''(0), \qquad g'''(0) = -h'''(0). Use these identities to show that f∗f^* is self-concordant.

Browse the problem library