Memetic Phase Retrieval and HPC for the Imaging of Matter at Atomic Resolution

Memetic Algorithms represent one of the most promising implementation of Evolutionary Algorithms. Their strength resides in the ability to exploit stochastic and deterministic optimization methods at the same time. A Memetic Algorithm has been applied to the phase retrieval problem in the field of Coherent Diffraction Imaging, called Memetic Phase Retrieval; it represents a significant improvement in the imaging of matter via coherent diffraction experiments. Memetic Phase Retrieval requires the latest High Performance Computing resources, due to the high dimensionality of the problem and the involvement of the Fourier Transform: an efficient parallel implementation, able to fully exploit multi-core and multi-node hardware, is needed. The implementation of Memetic Phase Retrieval, which exploits the hybrid OpenMP/MPI parallel programming paradigm, along with MPI Remote Memory Access communications, is presented. Its scaling performances on the recent Intel's Knights Landing hardware are shown.