On the generation of ab initio potential energy surfaces using machine learning techniques - A thesis abstract

Aug 4, 2026

Upload by Anda

A potential energy surface is a key prerequisite for quantum molecular dynamics studies of chemical systems, linking nuclear geometry to energy via an n-dimensional model built from high-level ab initio calculations. Traditional fitting is slow and must be customized per system, motivating faster workflows. The thesis investigates three acceleration routes: simplifying the fitting process, reducing required fitting points, and lowering the computational cost of ab initio data generation, with neural networks and active learning methods guiding efficient construction.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages244

Reading Time10 h 15 min

AccessFree to download

Potential energy surfaces

The origin of the PES

Methods of surface fitting

Improving upon current methods

Machine learning

The types of machine learning

Neural networks

Machine learning in chemistry

Electronic structure theory

The Schrödinger equation

Correlation energy

Basis sets

Effective core potentials

The firefly algorithm

Active learning techniques

Nature-inspired optimisation

The firefly algorithm

Application to the water potential energy surface

Conclusions

Basis set development

Introduction

Computational details

Methods

Core polarization potentials

Results & discussion

Conclusions

The HSO2 potential energy surface

The history of the HSO2 surface

Method and basis set exploration

Building a potential energy surface

Conclusions

Conclusions & future work

On the generation of ab initio potential energy surfaces using machine learning techniques Adam N. Hill A thesis submitted in partial fulﬁlment of the requirements for the degree of Doctor of Philosophy August 31, 2023 Faculty of Science Department of Chemistry ABSTRACT A potential energy surface is a major pre-requisite to carrying out quantum molecular dynamics studies on chemical systems. These studies allow theoreticians to explore the behaviour of modern-day chemistry in ways that are not feasible in a lab, enabling predictions to be made about experiments performed at extreme temperatures and pressures, and even helping to reveal reaction pathways. To achieve this, a PES associates nuclear position and energy, constructing an n-dimensional surface from high-level ab initio calculations that are ﬁt using physically motivated functions. However, this ﬁtting is a painstakingly slow process, and must be tailored to individual systems. This thesis will explore three ways of speeding up the generation of ab initio PESs: simplifying the ﬁtting process, reducing the number of ﬁtting points, and reducing the computational cost of the ab initio calculations themselves. Machine learning (ML) algorithms oﬀer a number of potential advantages for the construction of PESs: ﬁrstly, they represent more of a “black-box” approach to the ﬁtting that promises an easier route to accurate surfaces; second, reducing the dimensionality of the problem holds the promise of constructing a surface from signiﬁcantly fewer points. These algorithms also have access to active learning techniques that aim to reduce the size of machine learning datasets. As such, a particular subset of machine learning model, the neural network, will be used along side a novel application of a ﬁreﬂy inspired optimisation algorithm to speed up PES generation. While the development of new basis sets paired with correlation consistent eﬀective core potentials will aim to speed up data generation. To Kendra v ACKNOWLEDGMENTS A lot has happened in four and a half years. I got married, witnessed a global pandemic, had a baby, got a job, and ﬁnally ﬁnished this PhDin that order. It has been tough, and I would not have made it through without the incredible support of those around me. First and foremost, I must thank both Grant and Anthony for your guidance and support throughout everything that has transpired over the last few years. You have been incredibly kind and patient with me, and for that I will forever be thankful. Harry, Jaz, thank you for keeping me sane and letting me distract you while we all deﬁnitely should have been working, and Heather, thank you for your friendship and support while I navigated postgraduate life. To my parents, Dai and Lynda, you consistently support me in everything I do, and encourage me to be the best I can be. I could not ask for better role models (I’m just glad that I can say I submitted my thesis before you Mum) . To my little sister and baby bro, you are the best family I could ask for. Keep being you, and don’t let anyone tell you otherwise. Finally, to my wonderful wife, Kendra, and my joy in the last twelve months, Teddy. I love you both so much, I’m really not sure I would have got through this without either of you. Kendra, you have supported and encouraged me, you have held and comforted me, and most importantly you are my best friend. Thank you, for everything. Contents 1 POTENTIAL ENERGY SURFACES 1 1.1 The origin of the PES ...................... 1 1.2 Methods of surface ﬁtting ................... 4 1.3 Improving upon current methods ............... 16 2 MACHINE LEARNING 19 2.1 The types of machine learning ................. 20 2.2 Neural networks . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.3 Machine learning in chemistry ................ 39 3 ELECTRONIC STRUCTURE THEORY 45 3.1 The Schrödinger equation ................... 45 3.2 Correlation energy .................

FAQs

Why are potential energy surfaces (PES) needed for quantum molecular dynamics studies?

PES provide the energy as a function of nuclear geometry, enabling quantum molecular dynamics simulations and helping study reaction pathways that are difficult to access experimentally.

What main bottlenecks slow down generating ab initio PESs?

Fitting the ab initio results to physically motivated functional forms is painstakingly slow and must be tailored to each specific chemical system.

Which acceleration strategies are explored in the thesis?

The thesis examines three approaches: simplifying the fitting process, reducing the number of fitting points, and reducing the computational cost of the underlying ab initio calculations.

Aug 4, 2026

Upload by Anda