In the current version of Phoebe, the scattering matrix is symmetrized before the relaxons solver is run as a default. In many cases, this is unnecessary, and for large matrices, is slower than the diagonalization of the matrix -- therefore, to avoid users running into this, we set the symmetrizeMatrix input option to false by default (and make a a further note about this in the input docs).
This is because, of course, dong A + A^T/ 2 for a distributed matrix is not really a sensible operation. We leave the option there, to be used for production runs. Tests show that in copper, the difference between the solve with and without matrix symmetrization is < 1 W/m/K.
In the current version of Phoebe, the scattering matrix is symmetrized before the relaxons solver is run as a default. In many cases, this is unnecessary, and for large matrices, is slower than the diagonalization of the matrix -- therefore, to avoid users running into this, we set the symmetrizeMatrix input option to false by default (and make a a further note about this in the input docs).
This is because, of course, dong A + A^T/ 2 for a distributed matrix is not really a sensible operation. We leave the option there, to be used for production runs. Tests show that in copper, the difference between the solve with and without matrix symmetrization is < 1 W/m/K.