Tilt Stability of a Local Minimum

amenable functions coderivatives generalized Hessians monotone mappings perturbations prox-regularity proximal subgradient second-order optimality sensitivity subdifferential continuity tilt stability variational analysis

Mathematics Subject Classification ID

Nonsmooth analysis (49J52) Sensitivity, stability, well-posedness (49K40) Differentiation theory (Gateaux, Fréchet, etc.) on manifolds (58C20) Nonlinear programming (90C30) Sensitivity, stability, parametric optimization (90C31)

Review text

The authors discuss the behaviour of local minimizers of a function \(f:\mathbb{R}^n\to \mathbb{R}\) when this function is tilted by adding a linear term. In this sense, a point \(x^0\in \mathbb{R}^n\) is said to give a tilt-stable minimum of \(f\) if there exists a neighborhood \(N(x^0)\) such that the mapping \[ M(v):= \underset{x\in N(x^0)}{\text{argmin}} \{f(x)- f(x^0)- \langle v,x- x^0\rangle\} \] is single-valued and Lipschitzian near \(v= 0\) with \(M(0)= x^0\). If the function \(f\) is assumed to be \(C^2\) then tilt-stability of \(x^0\) is equivalent to the positive definiteness of the Hessian \(\nabla^2f(x^0)\), i.e., it is equivalent to the strict convexity of \(f\) near \(x^0\). In this case, the mapping \(M(v)\) turns out to be the inverse to the gradient mapping: \(M(v)= (\nabla f)^{-1}(v)\). This result is extended for nonsmooth functions using the notion of proximal subgradients \(\partial f(x)\) and the generalized Hessian \(\partial^zf(x| v)\) (with \(v\in\partial f(x)\)) of Mordukhovich. In the main theorem of the paper it is shown that for a function \(f\) having \(0\in\partial f(x^0)\) and which is prox-regular and subdifferentially continuous at \(x^0\) the following conditions are equivalent: (a) \(x^0\) gives a tilt-stable local minimum of \(f\); (b) the mapping \(\partial f^2(x^0| 0)\) is positive definite; (c) the mapping \(\partial f\) has strict monotonicity properties near \(x^0\); (d) there exists a proper, l.s.c., strongly convex function \(h\) with \(h(x^0)= f(x^0)\), \(0\in \partial h(x^0)\) and \(f\) and \(h\) have nearly the same subgradients around \(x^0\). Analogous to the smooth case, all the conditions imply the existence of a neighborhood \(N(x^0)\) such that \(M(v)\in (\partial f)^{-1}(v)\cap N(x^0)\).

Recommendations

Cited in

(only showing first 100 items - show all)

This page was built for publication: Tilt Stability of a Local Minimum

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4389192)