Optimal parallel acceleration of flat-histogram Monte Carlo sampling methods for atomistic simulation
Date:
Contributed talk at the 2026 APS Global Summit conference.
Abstract
Flat histogram methods such as Wang-Landau sampling [1] provide an efficient route for high-throughput calculation of phase diagrams in atomistic and lattice-model systems. Numerous parallelisation schemes have been proposed to improve sampling performance across distributed architectures [2-4]. In this study, these schemes are systematically benchmarked-both in isolation and in combination-to establish best practice for scalable flat-histogram simulations [5]. The schemes examined include energy-domain decomposition with both static sub-domains and a dynamic sub-domain sizing approach which we propose. We also assess the benefit of replica exchange and multiple random walkers per sub-domain to determine which factors most strongly influence parallel efficiency and load balance. The influence of sub-domain overlap is likewise discussed. As an illustrative test case, we implement [6] and apply [7] these strategies to a lattice-based model describing the internal energies of the AlTiCrMo refractory high-entropy superalloy, which is known to crystallographically order into a B2 (CsCl) structure with decreasing temperature. Our results demonstrate the superlinear speedup available from energy domain decomposition, and that implementation of non-uniform energy windows has a greater benefit than adding additional random walkers per window.
References
[1] F. Wang, D. P. Landau, Phys. Rev. Lett. 86, 2050 (2001).
[2] T. Vogel et al., Phys. Rev. Lett. 110, 210603 (2013).
[3] J. Zierenberg et al., Comput. Phys. Commun. 184, 1155-1162 (2013).
[4] J. Gross et al., Comput. Phys. Commun. 229, 57-67 (2018).
[5] H. J. Naguszewski et al. arXiv:2510.11562.
[6] H. J. Naguszewski et al., arXiv:2505.05393.
[7] C. D. Woodgate, H. J. Naguszewski et al., J. Phys.: Mater. 8, 045002 (2025).