A new efficient distributed parallel direct eigenvalue solver for symmetric matrices. It contains both an improved one-step ScaLAPACK type solver (ELPA1) and the two-step solver ELPA2.
ELPA uses the same matrix layout as ScaLAPACK. The actual parallel linear algebra routines are completely rewritten. ELPA1 implements the same linear algebra as traditional solutions (reduction to tridiagonal form by Householder transforms, divide & conquer solution, eigenvector backtransform). In ELPA2, the reduction to tridiagonal form and the corresponding backtransform are
replaced by a two-step version, giving an additional significant performance improvement.
ELPA has demonstrated good scalability for large matrices on up to 294.000 cores of a BlueGene/P system.