Towards the Development of Reactive Machine-Learned Potentials for Describing Chemistry in Solution

点击翻译按钮获取中文摘要

Year 1: General skills training and lab rotationsYears 2-4: MLPs are an exciting and current field of research. The benefits afforded by their use (particularly their ability to escape the fundamental trade-off between accuracy and computational efficiency) make them ideal for investigating the complex intricacies of liquid-phase reactions. There is now the opportunity to shine the 'computational microscope' on novel reaction phenomena.This project will functions as a proof-of-concept study for applying reactive MLPs to describing chemistry in solution. We will develop an MLP to accurately model the static, dynamical, and reactive properties of an aqueous system of molecules. Our methodology will employs NNPs (specifically, a committee of HD-NNPs), which we will use to provide a flexible and scalable model for simulating over different system types and sizes. We will develop this NNP towards modelling the reaction between CO2 and H2O (a fundamental process in a nature) across a range of conditions, environments, and media. In conducting this initial research, we hope to both gain valuable insight into our system of study while also improving our methodology for extensions to larger and more complex reactive systems. We anticipate that developing these methods will soon manifest a new, generalised framework for simulating reactions in liquids and at interfaces, which will prove important in synthetic chemistry for elucidating new pathways, optimal reaction conditions, and obtaining new insight into the fundamental principles governing reactive events.