default_solver_settings.m

%% Set Manopt options (if desired)
Manopt_opts.tolgradnorm = 1e-2;  % Stopping tolerance for norm of Riemannian gradient
Manopt_opts.rel_func_tol = 1e-5;  % Additional stopping criterion for Manopt: stop if the relative function decrease between two successive accepted iterates is less than this value
Manopt_opts.miniter = 1;  % Minimum number of outer iterations (i.e. accepted update steps) to perform
Manopt_opts.maxiter = 300;  % Maximum number of outer iterations (i.e. accepted update steps) to perform
Manopt_opts.maxinner = 500;  % Maximum number of iterations for the conjugate-gradient method used to compute approximate Newton steps
%manopt_options.maxtime = 60*60;  % Maximum computation time to allow, in seconds
%manopt_options.solver = @steepestdescent;  % Select Manopt solver to use: {trustregions (default), conjugategradient, steepestdescent}


%% Set SE-Sync options (if desired)
SE_Sync_opts.r0 = 5;  % Initial maximum-rank parameter at which to start the Riemannian Staircase
SE_Sync_opts.rmax = 10;  % Maximum maximum-rank parameter at which to terminate the Riemannian Staircase
SE_Sync_opts.eig_comp_rel_tol = 1e-4;  % Relative tolerance for the minimum-eigenvalue computation used to test for second-order optimality with MATLAB's eigs() function
SE_Sync_opts.min_eig_lower_bound = -1e-3;  % Minimum eigenvalue threshold for accepting a maxtrix as numerically positive-semidefinite
SE_Sync_opts.Cholesky = false;  % Select whether to use Cholesky or QR decomposition to compute orthogonal projections