Stability of Locally Optimal Solutions (Q4943944)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 1417166
Language Label Description Also known as
default for all languages
No label defined
    English
    Stability of Locally Optimal Solutions
    scientific article; zbMATH DE number 1417166

      Statements

      Stability of Locally Optimal Solutions (English)
      0 references
      0 references
      0 references
      0 references
      19 March 2000
      0 references
      parameterized optimization
      0 references
      Lipschitzian stability
      0 references
      tilt stability
      0 references
      coderivative Hessians
      0 references
      prox-regular functions
      0 references
      amenable functions
      0 references
      The authors give a second-order characterization for the Lipschitz stability of local solutions to finite-dimensional parametrized optimization problems. The approach bases on the notions of proximal subgradients and coderivative Hessians of the essential objective function and especially on properties of continuous prox-regularity of this function. In detail, the authors regard the family of parametrized problems \[ P(u,v):\qquad\text{minimize}\quad f(x,u)-\langle v,x\rangle\quad\text{over }x\in\mathbb{R}^n, \] where \(f: \mathbb{R}^n\times \mathbb{R}^d\to \mathbb{R}\cup\{- \infty,+\infty\}\) is assumed to be lower-semicontinuous and proper, \(u\in\mathbb{R}^d\), \(v\in \mathbb{R}^n\). For \(x^0\in \mathbb{R}^n\) and \(\delta> 0\) let \[ m_\delta(u, v):= \inf_{|x-x^0|\leq\delta} \{f(x, u)- \langle v,x\rangle\},\;M_\delta(u,v):= \underset{|x-x^0|\leq\delta} {\text{argmin}} \{f(x, u)-\langle v,x\rangle\}. \] The point \(x^0\) is a (fully) stable optimal solution of \(P(u^0,v^0)\) if the mapping \(M_\delta\) is single-valued and both mappings \(M_\delta\) and \(m_\delta\) are Lipschitz continuous near \(x^0\). Obviously, a first-order necessary condition for the local solution of \(P(u,v)\) can be given in the form \(v\in\partial_x f(x,u)\) using the partially proximal subgradient mapping \(\partial_x f(x,u):= \partial f(.,u)(x)\). In the main theorem of the paper (Theorem 2.3) the authors provide a second-order characterization of full stability using the coderivative Hessian \(D^*(\partial_x f)(x,u|v)\). Assuming \(f\) to be continuously prox-regular, they point out that the point \(x^0\) -- for which the stationary condition \(v^0\in\partial_x f(x^0,u^0)\) and an additional constraint qualification hold -- is a locally optimal solution to \(P(u^0,v^0)\) that is fully stable if and only if the following conditions are fulfilled: \[ (x,u)\in D^*(\partial_xf)(x^0, u^0\mid v^0)(v),\;v\neq 0\Rightarrow \langle v,x\rangle>0,\tag{a} \] \[ (0,u)\in D^*(\partial_x f)(x^0, u^0\mid v^0)(0)\Rightarrow u= 0.\tag{b} \] Moreover, in that case the Lipschitz modulus of the mapping \(M_\delta\) can be estimated. In the classical smooth case, the second-order conditions of the theorem reduce to the positive definiteness of the Hessian \(\nabla^2_{xx}f(x^0, u^0)\).
      0 references

      Identifiers