Computation of the Density Matrix in Electronic Structure Theory in Parallel on Multiple Graphics Processing Units

The algorithm developed in Cawkwell, M. J. et al. J. Chem. Theory Comput. 2012, 8, 4094 for the computation of the density matrix in electronic structure theory on a graphics processing unit (GPU) using the second-order spectral projection (SP2) method [Niklasson, A. M. N. Phys. Rev. B 2002, 66, 155115] has been efficiently parallelized over multiple GPUs on a single compute node. The parallel implementation provides significant speed-ups with respect to the single GPU version with no loss of accuracy. The performance and accuracy of the parallel GPU-based algorithm is compared with the performance of the SP2 algorithm and traditional matrix diagonalization methods on a multicore central processing unit (CPU).