La soutenance de thèse de Liang Qin aura lieu
le mercredi 21 octobre à 14h
Sa thèse intitulée "application of event-chain Monte Carlo in realistic long-range systems" a été réalisée au Laboratoire de Physique de l’Ecole normale supérieure sous la direction de Werner Krauth.
— > Lien de la visioconférence : https://global.gotomeeting.com/join/476160029.
La soutenance aura lieu également en présentiel en salle L361 - Département de Physique de l’ENS - 24 rue Lhomond 75005 PARIS
My thesis studies the behavior of event-chain Monte Carlo (ECMC) in long-range particle systems. Most realistic systems of chemical and biochemical interests feature long-range Coulomb interaction, which hinders the computational achievement in large-scale simulations. ECMC is a novel irreversible method that best exploits the stochastic nature of Monte Carlo.
The complexity of sampling a homogeneous system is the product of the long-range computation complexity and relaxation time. First, we find out suitable formulations for electrostatics under periodic boundary conditions. ECMC favors pairwise factors, while commonly used molecular dynamics always employ N-body Ewald summation. On the other hand, dipole systems can significantly benefit from the factorization. Dipole factors group all Coulomb interactions between two dipoles, resemble a "four-body" interaction, and achieve 1/r^3 event rate. This leads to an O(NlogN) algorithm for homogeneous 3D dipole systems.
For the relaxation time, ECMC can achieve fast relaxation of density modes, while challenges appear for the water system’s rotation. We test schemes, including molecular translation, and find that the naive ECMC scheme advancing particles along axes is the reason for slow rotation. The study of a simple 2D dipole shows that sequential Monte Carlo with gradually changing directions can bring faster mixing in dipole orientation.
I will also present our experience of developing JeLLyFysh, a universal particle simulation application via ECMC. Unlike long documentation, I will show how to install and run JeLLyFysh, as well as an animation displaying its event processing inside.