The alternating direction method is an attractive approach for large proble
ms. The convergence proof of the method is based on the exact solutions of
the subproblems. Computing the solution of the subproblems exactly can be e
xpensive if the number of unknowns is large. In this paper, for convex quad
ratic minimization problems, we propose a modified alternating direction me
thod which can overcome the above mentioned disadvantage. (C) 2000 Elsevier
Science Ltd. All rights reserved.