Light switches molecules inside crystals to change colour, and this project will use machine learning to simulate that process far faster than current methods allow. The problem is that modelling how molecules behave when hit by light—especially inside solid crystals—requires enormous computing power. Current simulations are too slow to capture the full chain of events, limiting scientists’ ability to design better photochromic materials. These materials already underpin optical switches, data storage devices, energy-efficient coatings, and self-darkening eyewear. Without faster simulations, progress in optimising them remains stalled. This project will train graph neural networks on quantum chemistry data to accelerate non-adiabatic dynamics simulations in crystalline environments. The team will focus on diarylethene derivatives, a well-studied class of photochromic molecules, as a test case. If successful, the approach could extend simulation timescales by orders of magnitude, allowing researchers to predict how new materials behave before synthesising them. The work is fundamental science: it advances computational methods for excited-state chemistry. But faster, cheaper simulation of light-driven molecular switches could eventually speed up the development of smarter optical devices, more efficient energy storage, and responsive coatings—without the trial-and-error that currently slows materials discovery.
View original technical description
Light is a versatile stimulus, offering multiple adjustable parameters that allow for precise manipulation without physical contact or as an energy source. Recently, there has been a growing interest in light-controlled photochromic materials (PCMs). These materials have garnered attention due to their significance in fundamental research and diverse applications, including optical switches, optical data storage devices, energy-efficient coatings, energy storage, and eyewear. Photochromism, defined as the reversible transformation of a chemical species between two different isomers with distinct absorption spectra, plays a pivotal role in PCMs. This transformation is initiated by exposure to light and can occur in both forward and reverse directions. In this context, excited state modelling is essential to understand mechanisms in PCMs, aiding the design and optimization of new materials. However, computational limitations, particularly the high computational costs associated with non-adiabatic dynamics (NAMD) in crystalline environments, present significant challenges in studying these reactions. An alternative approach to address these computational challenges is to employ data-driven models alongside traditional quantum chemistry (QC) calculations. Machine learning (ML) models, trained using quantum chemical data, have shown great promise in predicting ground state energies and forces with remarkable precision. This data-driven approach has the potential to substantially accelerate simulations. This research project aims to integrate Graph Neural Networks (GNNs) with existing QC codes to enhance and expedite NAMD simulations within solid environments, extending the simulation timescales. Our research focuses around investigating the cyclization reaction of photochromic diarylethene derivatives. This will serve as a crucial case study to evaluate the efficacy of the suggested approaches in understanding the behaviour of known materials and aiding in the discovery.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know