Yes, it is just an external library. If you have installed the toolbox you can either call it using the "solvelin" wrapper function mimicking Matlab's mldivide/backslash call, for example:
x = solvelin( A_sparse_matrix, b_rhs, 'mumps' );
The solvelin.m function code is open so you can see (and change) how it works by opening it ">> edit solvelin.m" if the toolbox paths are loaded.
Or you can use the mumps mex file and wrappers "initmumps/dmumps" directly, found in the "/lib/mumps/" subfolder where the toolbox is installed, which would end up looking something like this:
% Initialization of a MATLAB Mumps structure.
id = initmumps();
id.SYM = 0;
id = dmumps(id, [], mumpsfunc);
id.JOB = 6; % Set analysis + factorization + solve.
id.ICNTL(1:4) = -1; % suppress output.
% id.ICNTL(1:4) = [6,0,6,2]; % standard output.
% Mumps reordering:
% 0 - Approximate Minimum Degree (AMD)
% 2 - Approximate Minimuim Fill (AMF)
% 3 - SCOTCH
% 4 - PORD
% 5 - METIS
% 6 - Approximate Minimum Degree with automatic quasi row-detection (QAMD)
% 7 - Automatic choice by MUMPS
id.ICNTL(7) = 7;
id.ICNTL(14) = 25; % Percentage increase in estimated working space.
id.RHS = b; % Set RHS/load vector.
% Call Mumps.
id = dmumps(id, A, mumpsfunc);
flag = id.INFOG(1);
if( flag==0 )
x = id.SOL;
else
warning( ['MUMPS linear solver failed with error INFO(1:2) = ', ...
num2str(flag),':',num2str(id.INFOG(2))] )
x = zeros(size(b));
end
% Release memory.
id.JOB = -2;
id = dmumps(id, [], mumpsfunc);
Note that the mumps mex lib was compiled for/with Matlab R2019b as it seems there were some compatibility issues with R202x and later. Mumps typically works for R202x however sometimes, especially Windows, triggers some unresolved issue and can crash the session (and is therefore only used as default for <= R2019b). Unfortunately, it's not quite clear how to fix this, possibly the Mumps Fortran > C wrappers have to be rewritten somehow.
Locked
