Regensburg 2025 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

DY: Fachverband Dynamik und Statistische Physik

DY 32: Nonlinear Stochastic Systems

DY 32.4: Vortrag

Donnerstag, 20. März 2025, 10:30–10:45, H43

A Framework for Sparse Kinetic Monte-Carlo Models — •Bat-Amgalan Bat-Erdene, Roya Ebrahimi Viand, Karsten Reuter, and Sebastian Matera — Fritz-Haber-Institut der MPG, Berlin

The long-time dynamics of many problems in condensed matter physics are controlled by the interplay of rare events, e.g. chemical kinetics or crystal growth. Such problems are typically formulated as discrete-state Markov jump processes and can be simulated by kinetic Monte Carlo (kMC) methods. We are developing a software framework for implementing efficient kMC simulation models for arbitrary such processes. The key ingredients are i) a code generator for an optimized C++ skeleton where the user specifies the problem via a Python interface, and ii) the possible formulation as a sparse kMC model. Prototypical examples for sparsity appear in spatially extended models, where in each step the state changes only locally and interactions are only short ranged. This can then be exploited to achieve near-constant computational complexity per kMC time step. We evaluate the framework's efficiency on a dynamical Ising and a CO oxidation model on regular lattices. We find that our framework achieves a similar performance as specialized state-of-the-art kMC software for lattice kMC. Moreover, our framework offers a much larger flexibility, which we demonstrate on an implementation of Coupled Finite Differences for parameter sensitivity.

Keywords: Kinetic Monte Carlo; Software; High Performance; Simulation