(11a) Efficient Implementation of Monte Carlo Algorithms on Graphical Processing Units for Simulation of Adsorption in Porous Materials

In this study, we present enhancements in Monte Carlo (MC) simulation speed and functionality within a new molecular simulation code, gRASPA, that runs on graphical processing units (GPUs). The code achieves significant performance improvements over serial versions of the RASPA code through Nvidia MPS utilization and an option for scalable high-throughput screening. The gRASPA code supports a wide range of MC simulations, including grand canonical Monte Carlo, NVT-Gibbs, Widom test particle insertions, and continuous-fractional component Monte Carlo, with plans for ongoing feature addition and performance optimization. The code can also perform grand canonical transition matrix Monte Carlo (GC-TMMC) simulations for rapid free energy calculations. Our work also integrates machine learning (ML) potentials within gRASPA, and as an example, we present results for the adsorption of CO₂ in Mg-MOF-74, where the ML potential provides more accurate results than a classical force field. A broader audience can also benefit from the translated gRASPA code from CUDA to SYCL that can run on non-Nvidia GPUs. The gRASPA code is open source aiming to give access to the broader simulation community, who may also make future contributions to the ongoing development of this simulation framework.