Hoffman's error bounds and uniform Lipschitz continuity of best \(l_ p\)-approximations (Q1378404)

\textit{R. Holmes} and \textit{B. Kripke} [Mich. Math. J. 15, 225-248 (1968; Zbl 0177.16201)] proved that on \((\mathbb R^n, |\cdot|)\) endowed with the \(l_p\)-norm, \(1 < p < \infty\), the metric projection \(P_M\) onto the linear subspace \(M\) of \(\mathbb R^n\) is Lipschitz continuous with the constant depending of p. The main result of this paper is Theorem 1. Let \(M\) be a linear subspace of \((\mathbb R^n, |\cdot|)\). Then there exists a constant \(\lambda \) independent of p such that \[ |P_M(x) - P_M(y)|_p \leq \lambda |x-y|_p, \qquad \forall\quad x,y\in \mathbb R^n, 1 < p < \infty. \] Using a variation of Hoffman's error bound for approximate solutions of systems of linear inequalities the authors obtain matrix inequalities and finally the uniform Lipschitz constant \(\lambda\). As a consequence it follows the Lipschitz continuity for a selection of \(P_M\) in the case \(p=1\) and \(p=\infty\).