guest ::
| Home > Publications database > Acceleration of MDMC simulations by introducing a multipole coefficient based energy prediction algorithm |
| Home > Publications database > Acceleration of MDMC simulations by introducing a multipole coefficient based energy prediction algorithm |
| Master Thesis | FZJ-2025-03601 |
2025
This record in other databases:
Please use a persistent id in citations: doi:10.34734/FZJ-2025-03601
Abstract: This work focuses on improving the coupled MD-MC simulation framework by enhancing the MC++ code by H. Ganesan that is used to model strain aging processes in e.g. steel. Key improvements include the addition of a LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator) interface to replace the discontinued IMD (ITAP Molecular Dynamics) library, the conversion of the EAM-potential (Embedded Atom Method), and the translation of the concept of ’virtual atoms’ to LAMMPS which represent dislocation sites in body-centered cubic (bcc) ferrite. To further accelerate simulations, a novel multi-pole coefficient based algorithm for predicting the potential energy of atom configurations is proposed, based on database searches for similar configurations. This approximation aims to reduce the computational cost of energy calculations, although its accuracy and reliability are critically evaluated. Additionally, to address the assumed increased data throughput required by the new algorithm, a modification of a parallel replica method is introduced to the MC++ code, aimed at reducing communication overhead while maintaining the accuracy of simulation results. These advancements are expected to significantly enhance the efficiency and scalability of the MDMC simulation framework for modeling strain aging processes in steel. Steel has long been a vital material in engineering, with recent advancements in high-strength steels addressing the automotive industry’s need for enhanced fuel efficiency and reduced environmental impact. One key strengthening mechanism in steel, particularly low-carbon steel, is strain aging, where carbon atoms segregate at dislocation sites, impeding dislocation motion and enhancing yield strength. However, simulating strain aging processes—such as carbon diffusion and dislocation evolution—presents significant challenges due to the disparity in timescales involved. Molecular Dynamics (MD) and Monte Carlo (MC) simulations offer complementary approaches to model these processes, but limitations in computational efficiency persist. Thus the coupled MDMC framework was developed which is further improved and accelerated in the course of this work.
|
The record appears in these collections: |
PDF