Dykstra's cyclic projections algorithm allows one to compute best appr
oximations to any point x in a Hilbert space from the intersection C =
boolean AND(1)(r) C-i of a finite number of closed convex sets C-i, b
y reducing it to a sequence of best approximation problems from the in
dividual sets Ci. Here we present two generalizations of this algorith
m. First we allow the number of sets Ci to be infinite rather than fin
ite; secondly, we allow a random, rather than cyclic, ordering of the
sets C-1. (C) 1997 The Mathematical Programming Society, Inc.