On the von Neumann alternating algorithm in Hilbert space (Q1822209)

The performance of an algorithm is one of the most interesting problems concerning this one. In the work of \textit{J. von Neumann} [Functional Operators. II. (1950; Zbl 0039.117)] one finds an algorithm for computing best approximations to a given x in a Hilbert space X. More precisely, X being a Hilbert space and U and V any pair of closed subspaces with proximity maps \(P_ U\), \(P_ V\) respectively, the sequence \(\{[(I-P_ V)(I-P_ U)]^ nx\}\) converges to x-w, where w is a best approximation to x from \(\overline{U+V}.\) In this paper the authors study the performance of this algorithm in Hilbert space, particularly in the case when \(U+V\) is not closed. They establish that the convergence can be very slow, in opposite to the fact that if \(U+V\) is closed, the convergence rate is a geometric one. Showing off that the algorithm goes under the name of Diliberto-Strauss, when \(X=C(T\times S)\), \(U=C(T)\), \(U=C(S)\), with the usual uniform norm and S, T compact Hausdorff and the fact that in recent years many authors payed attention to it, the authors of this paper show that a detailed study of the Hilbert space can be expected to give some insight into the performance of the algorithm in these other cases.