Merge branch 'StateMatrixUpdate' into 'master'
avoid expensive matrix update profiling showed that update_matrices has a huge impact on performance. It's better to avoid updating the forward and backward transformations and computing them on the fly instead. See merge request !30