This repository contains the Matlab codes to reproduce our numerical experiments comparing different subproblem solver implementations of Riemannian Adaptive Regularization with Cubics (Riemannian ARC).
The main script is compare_solvers_on_problems.m.
It is necessary to download Manopt (and to add it to Matlab's path) to run these experiments. Preferably obtain the latest version from https://github.com/NicolasBoumal/manopt. Please run with a version of Manopt dating from Oct. 4, 2018 or more recent.
The code pdf_print_code.m produces PDFs from Matlab figures. It relies on the utility pdfcrop being installed. If this is an issue, simply remove the call to pdf_print_code in the main script.
Authors: Bryan Zhu and Nicolas Boumal (based on code authored by Naman Agarwal, Nicolas Boumal, Brian Bullins, Coralia Cartis at https://github.com/NicolasBoumal/arc)
May 2, 2019