Search academic texts

Information × Registration Number 2123U011185, Article popup.category Препринт Title Molecular dynamics of charged dipole particles using machine learning (AI translated) popup.author Kukhar OleksandrKukhar Oleksandr popup.publication 01-01-2023 popup.source_user Український католицький університет popup.source https://hdl.handle.net/20.500.14570/4408 popup.publisher Description The method of molecular dynamics finds applications in various areas such as pharmacology, polymer science, nanotechnology, chemical catalysis, and drug dis- covery. An efficient and fast prediction of positions and dynamics of particles is of great importance in order to reduce computational efforts. This thesis focuses on extending the existing SE(3)-transformer-based graph neural network (GNN) approach proposed by Fuchs et al.[1], which successfully employs a self-attention mechanism for point clouds to describe dynamics of charged particles. The exten- sion developed in our study is aimed to improve an accuracy of molecular dynamics prediction for a more complex system consisting of particles with an orientation- dependent interaction and rotational degrees of freedom. As an example, a physical model presented as a fluid of charged particles bearing electric dipoles is examined. It is shown that our approach, which introduces a new attention mechanism, pro- vides better accuracy in describing such systems compared to the original approach. popup.nrat_date 2025-11-05 Close