[SY-N5] Localised on-the-fly Kinetic Monte Carlo
Various on-the-fly Kinetic Monte Carlo algorithms have been used to simulate the long-term evolution of atomistic systems. Unlike classical Kinetic Monte Carlo methods, these algorithms search for possible transitions/reactions only when necessary (hence on-the-fly), rather than relying on a previously-known set of transitions. Unfortunately, their performance deteriorates when large systems or systems with many similar transition mechanisms are investigated.
We will present a localised on-the-fly Kinetic Monte Carlo variant that is based on the k-ART algorithm by El-Mellouhi, Lewis, and Mousseau. This algorithm is capable of localising transition searches, enabling the simulation of significantly larger systems, and can also recycle known transition mechanisms. Unlike k-ART, which uses a graph-based technique to identify similar local environments, we use an RMSD (root mean square deviation) measure that is invariant under index permutations and rotations of local environments. This allows our algorithm to deal not only with highly-structured materials like crystals, but also with amorphous materials, which are difficult to classify with graphs.
We will conclude by demonstrating computational results of our algorithm, and discuss potential areas of application.
We will present a localised on-the-fly Kinetic Monte Carlo variant that is based on the k-ART algorithm by El-Mellouhi, Lewis, and Mousseau. This algorithm is capable of localising transition searches, enabling the simulation of significantly larger systems, and can also recycle known transition mechanisms. Unlike k-ART, which uses a graph-based technique to identify similar local environments, we use an RMSD (root mean square deviation) measure that is invariant under index permutations and rotations of local environments. This allows our algorithm to deal not only with highly-structured materials like crystals, but also with amorphous materials, which are difficult to classify with graphs.
We will conclude by demonstrating computational results of our algorithm, and discuss potential areas of application.