2000 character limit reached
Convergence Analysis of the Fast Subspace Descent Methods for Convex Optimization Problems (1810.04116v2)
Published 9 Oct 2018 in math.NA and cs.NA
Abstract: The full approximation storage (FAS) scheme is a widely used multigrid method for nonlinear problems. In this paper, a new framework to design and analyze FAS-like schemes for convex optimization problems is developed. The new method, the Fast Subspace Descent (FASD) scheme, which generalizes classical FAS, can be recast as an inexact version of nonlinear multigrid methods based on space decomposition and subspace correction. The local problem in each subspace can be simplified to be linear and one gradient descent iteration (with an appropriate step size) is enough to ensure a global linear (geometric) convergence of FASD.