Fig. 1: Schematics of the scalable Monte Carlo (SMC-X) method. | npj Computational Materials

Fig. 1: Schematics of the scalable Monte Carlo (SMC-X) method.

From: Revealing nanostructures in high-entropy alloys via machine-learning accelerated scalable Monte Carlo simulation

Fig. 1

a Illustrate the SMC-X method with a 2D square lattice, in which each site is denoted by two indices (iC, iA), where iC represents different link-cells, and iA represents the different atoms within the nA atom cell. b The whole chemical configuration is distributed on multiple GPUs, and the atoms near the boundary need to be communicated between GPUs nA times in every MC sweep. c A 2D view of the fcc or bcc lattice, site 1 (red) can swap with each of its nearest neighboring sites (yellow), and the green sites represent the ones that the local energies can be affected. The yellow and green sites, together with the centering atom, form the LIZ. The chemical environment with LIZ+ are needed to calculate the energy change due to a swap trial. d A log-log plot of computation time vs the system size to illustrate the speedup ratio of the SMC-X method as compared to DFT (MuST-KKR), linear-DFT (LSMS), and GNN (Allegro). The measured values for SMC-X are signified as stars and ideal scaling is assumed for all lines.

Back to article page