As CASTEP calculations are run with an ever greater number of bands, the memory
requirements per core, for storing the wavefunction data necessitate band-parallel calculations.
The systolic algorithm implemented in this work will serve to benefit the scaling of the largest contemporary
CASTEP calculations on HECToR and become ever more useful as the size of calculations increases into the future.
The triangular matrix work will however provide a useful speedup to all band-parallel calculations, no matter
These band-parallel improvements are available upon request to any HECToR user of CASTEP. They have been incorporated into the main CASTEP source code, and will be released worldwide in CASTEP 7.0.