Objective acceleration and nonlinear GMRES for unconstrained optimization

This repository contains implementations of two optimization acceleration algorithms discussed in A.N. Riseth. Objective acceleration for unconstrained optimization, 2017

The code contains modifications of the original Matlab code provided by Hans De Sterck on his website.

All references to N-GMRES-O in the code correspond to the O-ACCEL algorithm in A. N. Riseth. Objective acceleration for unconstrained optimization

Requirements

These files require the Poblano toolbox for MATLAB to run.

List of relevant files

ngmres.m The N-GMRES algorithm as implemented by De Sterck

ngmres_o.m The Objective Acceleration algorithm by Riseth, following the same structure as ngmres.o.

ngmres_test_general.m Provides convergence plots for the different algorithms. To show convergence plots for an instance of Problem A, with n=200, call

ngmres_test_general(0,1,200,400,true)

runme_writestats.m Runs 1000 instances of each test problem and writes the statistics to file in the directory data.

ngmres_test_tensor_CP.m Provides convergence plots for the different algorithms on the tensor CP decomposition problem. To show convergence plots for an instance of the problem, call

ngmres_test_tensor_CP(0,600,true)

runme_writestats_tensor_CP.m Runs 1000 instances the tensor CP problem and writes the statistics to file in the directory data.

BibTeX reference for article:

@article{riseth2017objective,
   author = {Riseth, Asbj{\o}rn N.},
    title = "{Objective acceleration for unconstrained optimization}",
  journal = {ArXiv e-prints},
archivePrefix = "arXiv",
   eprint = {1710.05200},
 primaryClass = "math.OC",
 keywords = {Mathematics - Optimization and Control, Mathematics - Numerical Analysis, 49M05, 65B99, 65K10},
     year = 2017,
    month = oct,
   adsurl = {http://adsabs.harvard.edu/abs/2017arXiv171005200N},
  adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}