Parallel block Chebyshev subspace iteration algorithm optimized for sequences of correlated dense eigenproblems

In many material science applications simulations are made of dozens of sequences, where each sequence groups together eigenproblems with increasing self-consistent cycle outer-iteration index. Successive eigenproblems in a sequence possess a high degree of correlation. In particular it has been demonstrated that eigenvectors of adjacent eigenproblems become progressively more collinear to each other as the outer-iteration index increases. This result suggests one could use eigenvectors, computed at a certain outer-iteration, as approximate solutions to improve the performance of the eigensolver at the next one. In order to opti- mally exploit the approximate solution, we developed a block iterative eigensolver augmented with a Chebyshev polynomial accelerator (BChFSI). Numerical tests show that, when the sequential version of the solver is fed approximate solutions instead of random vectors, it achieves up to a 5X speedup. Moreover the parallel shared memory implementation of the algorithm obtains a high level of efficiency up to 80 \% of the theoretical peak performance. Despite the eigenproblems in the sequence being relatively large and dense, the parallel BChFSI fed with ap- proximate solutions performs substantially better than the corresponding direct eigensolver, even for a significant portion of the sought-after spectru