Convex \(L^ p\) approximation (Q5903786)

Let (T,\(\Sigma\),\(\mu)\) be a finite positive measure space and let K be a finite dimensional compact convex set of \(L^ p=L^ p(T,\Sigma,\mu)\), such that \(0\not\in K\). The paper is concerned with the following algorithm for finding the best approximation \(g^*\) from K to 0: given \(g_ n\in K\) define \(g_{n+1}\) as the solution to the problem min W(g\({}_ n,g)\), \(g\in K\), where \(W(g_ n,g)=\int_{T}| g_ n|^{p-2} g^ 2 d\mu.\) The proof of the convergence relies essentially on the remark that \(X=\{g\in L^ p:\) \(W(g_ n,g)<\infty \}\) is a linear subspace of \(L^ p\) and \(W(g_ n,g)^{1/2}\) is a weighted \(L^ 2\)-norm on X. The author proves the convergence of the algorithm without using any linear search procedure which was an essentialy ingredient in two previous papers [\textit{L. Karlowitz}, J. Approximation Theory 3, 123-127 (1970; Zbl 0222.41016)] and \textit{B. L. Chalmers}, \textit{A. G. Egger} and \textit{G. D. Taylor} [J. Approximation Theory 37, 326-334 (1983; Zbl 0517.41032)].