Large matrices with more than a few thousand rows and columns usually do
not fit into the memory of one processor anymore. With this option, the
Casida matrix is distributed in block-cyclic fashion over all cores in the
ParOther group. The diagonalization is done in parallel using ScaLAPACK
or ELPA, if available. For very large matrices (>100000), only the
ParOther strategy should be used because the diagonalization dominates
the run time of the computation.