Convergence of an algorithm for finding the distance between a ball in a subspace and a sum of subspaces

Y. Ritov*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

Let S0,S1,...,Sk be all subspaces of a Hilbert space, and let S = S1⊕,..., ⊕Sk. An algorithm is investigated for finding members of S0 and S with the minimal angle between them. The algorithm is a modification of the alternating projection algorithm of von Neumann [Ann. of Math., 50(1949), 401-485; Functional Operators, The Geometry of Orthogonal Spaces, Ann. of Math Stud., 1950]. It is similar to the algorithm suggested by Brieman and Friedman [J. Amer. Statist. Assoc., 17(1985), pp. 580-598] without a proof. The convergence of the algorithm is proved to be exponentially fast.

Original languageEnglish
Pages (from-to)1355-1367
Number of pages13
JournalSIAM Journal on Numerical Analysis
Volume27
Issue number5
DOIs
StatePublished - 1990

Fingerprint

Dive into the research topics of 'Convergence of an algorithm for finding the distance between a ball in a subspace and a sum of subspaces'. Together they form a unique fingerprint.

Cite this