Regularity modulus of arbitrarily perturbed linear inequality systems (Q2427270)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Regularity modulus of arbitrarily perturbed linear inequality systems
scientific article

    Statements

    Regularity modulus of arbitrarily perturbed linear inequality systems (English)
    0 references
    0 references
    0 references
    0 references
    8 May 2008
    0 references
    Let \(T\) be an arbitrary index set and \(\Theta\equiv (\mathbb R^n\times \mathbb R^T\) be the parameter space of all linear inequality systems \(\sigma=\{a_t'x\geq b_t\), \(t\in T\}\) in \(\mathbb R^n\). The space \(\Theta\) is endowed with the extended distance \(d\) given by \(d(\sigma_1,\sigma_2)= \sup_{t\in T}\|(a_t^{1'},b_t^1) (a_t^{2'},b_t^2)\|\), \(\|\cdot\|\) being any norm in \(\mathbb R^n\times\mathbb R\) satisfying \(\|(a',b)\|= \|(a',-b)\|\) for all \((a',b)\in\mathbb R^n\times\mathbb R\). The authors consider the mapping \({\mathcal G}:\mathbb R^n\rightrightarrows \Theta\) defined by \({\mathcal G}(x)= \{\sigma\in\Theta \mid a_t'x\geq b_t,\) \(\forall t\in T\}\). The modulus of the metric regularity \(\operatorname{reg}{\mathcal G}(\overline{x}\mid \overline{\sigma })\) of \({\mathcal G}\) at \((\overline{x},\overline{\sigma})\in \operatorname{gph}{\mathcal G}\) is defined as the infimum of all constants \(\kappa\geq 0\) for which there exist some neighborhoods \(U\) and \(V\) of \(\overline{x}\) and \(\overline{\sigma}=\{\overline{a}_t'x\geq \overline{b}_t\), \(t\in T\}\), respectively, such that \(d(x,{\mathcal G}^{-1}(\sigma))\leq \kappa d(\sigma,{\mathcal G}(x))\) for all \(x\in U\) and all \(\sigma \in V\). (This infimum is equal to \(+\infty \) if no such \(\kappa \) exists.) The main result of this paper computes \(\operatorname{reg}{\mathcal G}(\overline{x} \mid \overline{\sigma})\) in the case when the set \(\{\overline{a}_t'\), \(t\in T\}\) is bounded and \({\mathcal G}\) is metrically regular at \((\overline{x},\overline{\sigma})\) (that is, when \(\operatorname{reg}{\mathcal G}(\overline{x}\mid \overline{\sigma})+\infty\)). More specifically, it is proved that \(\operatorname{reg}{\mathcal G} (\overline{x}\mid \overline{\sigma})=0\) if \(\overline{x}\) is a strong Slater point of \(\overline{\sigma}\) (that is, if \(\inf_{t\in T}\{a_t'\overline{x}-b_t\}>0\)), and \(\operatorname{reg}{\mathcal G} (\overline{x}\mid \overline{\sigma})= \|(\overline{x},-1)\|_*\max \{\frac{1}{\|u\|_*}\mid (u,u'\overline{x})\in \text{cl(conv}(\{(\overline{a}_t',\overline{b}_t)\), \(t\in T\}))\}\) (the subscript \(_*\) is used to denote the dual norm) otherwise; in the latter case the set \(\{u\in \mathbb R^n\mid (u,u'\overline{x})\in \text{cl(conv}(\{(\overline{a}_t',\overline{b}_t)\), \(t\in T\}))\}\) is also proved to be nonempty and compact. As an application, the authors study the special case of an inequality system arising from a natural reformulation of the Lagrangian dual of an inequality constrained minimization problem.
    0 references
    metric regularity
    0 references
    Lagrangian duality
    0 references
    feasible set mapping
    0 references
    0 references

    Identifiers