Physics

AI predicts optical properties to accelerate energy and quantum materials discovery

An AI tool called GNNOpt accurately predicts light spectra based solely on crystal structure, accelerating the development of photovoltaics and quantum materials. Credit: Nguyen Tuan Hung et al.

Researchers at Tohoku University and the Massachusetts Institute of Technology (MIT) have announced a new AI tool that captures high-quality light spectra that is as accurate as quantum simulations but runs a million times faster, allowing researchers to explore photovoltaic materials and photovoltaic materials. It has the potential to accelerate the development of quantum materials.

Understanding the optical properties of materials is essential to the development of optoelectronic devices such as LEDs, solar cells, photodetectors, and photonic integrated circuits. These devices are critical to the current resurgence of the semiconductor industry.

Traditional calculation methods using basic physical laws require complex mathematical calculations and vast amounts of computing power, making it difficult to test large numbers of materials quickly. Overcoming this challenge could lead to the discovery of new photovoltaic materials for energy conversion and a deeper understanding of the fundamental physics of materials across the optical spectrum.

A team led by Assistant Professor Nguyen Tuan Hung from Tohoku University’s Frontier Research Institute for Interdisciplinary Sciences (FRIS) and Associate Professor Minda Li from the Massachusetts Institute of Technology’s School of Nuclear Engineering (NSE) did just that, using only the material’s crystal structure as input. A new AI model that predicts optical properties over a wide range of optical frequencies.

Lead author Nguyen and his colleagues recently published their findings in an open-access paper in the journal Advanced Materials.

“Optics is an interesting aspect of condensed matter physics that is governed by a causal relationship known as the Kramers-Kroenig (KK) relationship,” Nguyen says. “Once one optical property is known, all other optical properties can be derived using the KK relationship. It is interesting to observe how the AI ​​model can capture physical concepts through this relationship. .”

Due to laser wavelength limitations, it is difficult to obtain the complete frequency range of the optical spectrum in experiments. Simulations are also complex, require high convergence criteria, and are computationally expensive. As a result, the scientific community has long sought more efficient ways to predict the optical spectra of various materials.

“The machine learning model used for optical prediction is called a graph neural network (GNN),” points out Ryotaro Okabe, an MIT chemistry graduate student. “GNNs provide a natural representation of molecules and materials by representing atoms as graph nodes and bonds between atoms as graph edges.”

However, although GNNs have shown promise in predicting material properties, they lack universality, especially in representing crystal structures. To circumvent this challenge, Nguyen and colleagues devised universal ensemble embedding, which creates multiple models or algorithms to unify data representation.

“While this ensemble embedding is beyond human intuition, it can be broadly applied to improve prediction accuracy without affecting the structure of neural networks,” said MIT graduate student in electrical engineering and computer science. explains Abhijatmedhi Chotrattanapituk.

Ensemble embedding methods are universal layers that can be applied seamlessly to any neural network model without changing the structure of the neural network. “This means that universal embedding can be easily integrated into any machine learning architecture and has the potential to have a significant impact on data science,” said Mingda Li.

This method enables highly accurate optical predictions based solely on crystal structure, and is suitable for a wide range of applications such as material screening for high-performance solar cells and detection of quantum materials.

Looking to the future, the researchers aim to develop new databases of various material properties, such as mechanical and magnetic properties, to enhance the ability of AI models to predict material properties based solely on crystal structure. Masu.

Further information: Nguyen Tuan Hung et al., Universal Ensemble-Embedding Graph Neural Network for Direct Prediction of Optical Spectra from Crystal Structures, Advanced Materials (2024). DOI: 10.1002/adma.202409175

Provided by Tohoku University

Citation: AI predicts optical properties to accelerate energy and quantum materials discovery (October 7, 2024) https://phys.org/news/2024-10-ai-optical-properties-discovery-energy Retrieved October 7, 2024 from.html

This document is subject to copyright. No part may be reproduced without written permission, except in fair dealing for personal study or research purposes. Content is provided for informational purposes only.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button