A GPU enhanced approach to identify atomic vacancies in solid materials

Identification of vacancies in atomic structures plays a crucial role in the characterization of a material, from structural to dynamical properties. In this work we introduce a computationally improved vacancy recognition technique, based in a previous developed search algorithm. The procedure is highly parallel, based in the use of Graphics Processing Unit (GPU), taking advantage of parallel random number generation as well as the use of a large amount of simultaneous threads as available in GPU architecture. This increases the spatial resolution in the sample and the speed during the process of identification of atomic vacancies. The results show an improvement of efficiency up to two orders of magnitude compared to a single CPU. Along with the above a reduction of required parameters with respect to the original algorithm is presented. We show that only the lattice constant and a tunable overlap parameter are enough as input parameters, and that they are also highly related. A study of those parameters is presented, suggesting how the parameter choice must be addressed.