Building potential energy surfaces of heteroaromatic compounds using neural networks

Benninga, Joël (2021) Building potential energy surfaces of heteroaromatic compounds using neural networks. Research Project, Chemistry.

Preview	Text mCHEM_2021_BenningaJ.pdf Download (1MB) | Preview
	Text Toestemming.pdf Restricted to Registered users only Download (126kB)

Abstract

Machine learning is on its way to revolutionize the field of computational chemistry. In recent years it has made remarkable progress in many areas due to advances in the development of machine learning algorithms and improved hardware resources. Applications of machine learning in the field of computational chemistry include molecular dynamics, Monte Carlo simulations, and more specifically in drug design and material screening. The goal of this research project is to explore the application of machine learning for the building of potential energy surfaces of heteroaromatic compounds, as prerequisite for excited-state quantum dynamics. The potential energy surface is the most basic quantity to describe a chemical system, and determines all of its properties. Although quantum mechanical methods exist to accurately calculate potential energy surfaces, these are computationally expensive. Therefore, neural networks – a class of machine learning algorithms – have recently emerged as a promising alternative for the construction of potential energy surfaces. In this project, three neural networks are optimized by grid search, trained on surface hopping dynamics simulations, and finally applied to predict potential energy surfaces of pyrazine, pyrrole and furan.

Item Type:	Thesis (Research Project)
Supervisor name:	Faraji, S.S. and Thie, A.S.
Degree programme:	Chemistry
Thesis type:	Research Project
Language:	English
Date Deposited:	16 Nov 2021 12:27
Last Modified:	16 Nov 2021 12:27
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/26289

Actions (login required)

View Item