Conditioning convex and nonconvex problems (Q2565029): Difference between revisions

Given a metric space \((X,d)\) and \(f:X\rightarrow \overline{\mathbb{R}}\) such that \(m_f=\inf f(X)\in {\mathbb{R}}\) and \(S_f=\{x\in X\mid f(x)=m_f\}\) is nonempty, a function \(\varphi :{\mathbb{R}}_{+}\rightarrow \overline{\mathbb{R}} _{+}\) is said to be a conditioner for \(f\) if \(d(x,S_f)\leq \varphi (f(x)-m_f)\) \(\forall x\in X\), while \(\gamma :{\mathbb{R}}_{+}\rightarrow \overline{\mathbb{R}} _{+}\) such that \(\gamma (d(x,S_f))\leq f(x)-m_f\) \(\forall x\in X\) is called a growth function. The canonical growth function is defined by \(\gamma _f(r)=\inf \{f(x)-m_f\mid d(x,S_f)\geq r\}\), while the canonical conditioner is defined by \(\mu _f(r)=\sup \{d(x,S_f)\mid f(x)\leq m_f+r\}\). Properties of these functions and their relationships are studied. It is shown, for example, that if \(f\) is starshaped at each \(x\in S_f\) then \(\gamma _f(r)=\inf \{f(x)-m_f\mid d(x,S_f)=r\}\), and \(\gamma _f\) is starshaped. Then, the author obtains quantitative results for the minimum value function and for the solution set multifunction using local excesses and conditioners. In the sequel these results are applied for studying upper semi-continuity of subdifferentials in the convex case and to best approximation.