On primal regularity estimates for single-valued mappings (Q497975)

Let \(X,Y\) be metric spaces and \(F: X\rightrightarrows Y\) a set-valued mapping with domain \(\mathrm{dom}(F)=\{x\in X : F(x)\neq\emptyset\}\) and graph \(\mathrm{gph}(F)=\{(x,y)\in X\times Y : y\in F(x)\}\). One says that \(F: X\rightrightarrows Y\) is open with a linear rate near \((\bar x,\bar y) \in \mathrm{gph}(F)\) if there are \(c>0\) and \(\varepsilon>0\) such that \[ (*)\qquad B(y,ct)\cap B(\bar y,\varepsilon)\subset F(B(x,t)), \] for all \((x,y)\in B((\bar x,\bar y),\varepsilon)\cap\mathrm{gph}(F)\) and all \(t\in (0,\varepsilon)\). Here \(B(z,r)\) denotes the closed ball of center \(z\) and radius \(r>0\). The upper bound of all \(c>0\) such that \((*)\) holds for some \(\varepsilon>0\) is called the rate of openness (or modulus of surjection) of \(F\) near \((\bar x,\bar y )\) and is denoted by \(\mathrm{sur}F(\bar x,\bar y)\). An equivalent notion is that of metric regularity, meaning the existence of \(k>0\) and of neighborhoods \(U\) of \(\bar x\) and \(V\) of \(\bar y\) such that \[ (**)\qquad d(x,F^{-1}(y))\leq k d(y,F(x)), \] for all \((x,y)\in U\times V\). The lower bound of all \(k>0\) for which \((**)\) holds for some neighborhoods \(U,V\) is called the modulus of metric regularity of \(F\) near \((\bar x,\bar y)\) and is denoted by \(\mathrm{reg}F(\bar x,\bar y)\). These two quantities are related by the equality \(\mathrm{reg}F(\bar x,\bar y)\cdot\mathrm{sur}F(\bar x,\bar y)=1\). The aim of the present paper is to provide lower bounds for the rate of openness for set-valued mappings that are single-valued on their domains, which are supposed to be closed convex subsets of a Banach space \(X\). The space \(Y\) is also supposed to be Banach. The authors use three methods to approximate \(F\) around a reference point: tangential approximation by set-valued mappings associated with the Bouligand tangent cone to the graph of \(F\), approximations by positively homogeneous set-valued mappings whose graphs contain the graph of \(F\) (e.g. strict prederivatives), and by bunches of continuous linear operators. One shows, by examples, that these three approaches are independent, unless the Banach spaces \(X,Y\) are finite dimensional.