Seminar

Seminar

Computational experiments on approximate mixed quantum-classical approaches based on mapping formalisms

  • POSTED DATE : 2019-11-13
  • WRITER : 화학과
  • HIT : 94
  • DATE : 2019년 11월 14일(목) 오후 4시 30분
  • PLACE : 화학관 2층 서병인강의실 (330226호실)

세미나가 이번주 목요일(11월 14일)에 개최됩니다.
많은 참석 부탁드립니다.
감사합니다.


===============================================================================

제  목 : Computational experiments on approximate mixed quantum-classical approaches based on mapping formalisms
연  사 : 김현우 박사(KRICT)
일  시 : 2019년 11월 14일(목) 오후 4시 30분
장  소 : 화학관 2층 서병인강의실 (330226호실)

================================================================================ 

 

Computational experiments on approximate mixed quantum-classical approaches based on mapping formalisms

 

Hyun Woo Kim

Center for Molecular Modeling and Simulation, Korea Research Institute of Chemical Technology (KRICT), Daejeon, Korea.

 

Corresponding Author: ahwk@krict.re.kr

 

Molecular dynamics (MD) simulations are applicable to study interesting chemical phenomena in multiple excited states with the help of approximate quantum dynamics methods. Representative examples are mixed quantum-classical (MQC) approaches such as surface hopping algorithms and mean-field approaches. As it is well-known that some MQC approaches are not accurate in the long-time limit because of their approximations, several methods are developed so far. Here, I will present computational analysis on approximate MQC approaches based on mapping formalisms including the application of machine learning algorithms. I tried to modify equations of motion by utilizing the quantum-classical Liouville equation in the mapping basis and its approximation which is called Poisson bracket mapping equation (PBME). I found several techniques such as trajectory-branching improve the performance of PBME. For applying machine learning algorithms, I generated a large amount of data from MD simulations with PBME and then applied machine learning algorithms to improve its performance. Machine learning models can predict corrections to PBME by including the information from a set of trajectories. I will discuss some issues such as the energy conservation along MQC trajectories.