Breakthrough algorithm expands the space of material discovery to a large extent
Nanoengineers at the University of California San Diego’s Jacobs School of Engineering have developed an AI algorithm that predicts the structure and dynamic properties of any material—existing or new—nearly. Called M3GNet, the algorithm was used to develop problem.ai, a database of more than 31 million unsynthesized materials with properties predicted by machine learning algorithms. Matterverse.ai facilitates the discovery of new technological materials with special properties.
The team behind M3GNet, led by UC San Diego nanoengineering professor Shyue Ping Ong, uses matverse.ai and M3GNet’s new capabilities to search for safer and energy-efficient electrodes and electrolytes. more capacity for rechargeable lithium-ion batteries. The project is explored in the November 28 issue of the magazine Natural Computational Science.
The properties of a material are determined by the arrangement of its atoms. However, existing methods for achieving such alignments are either too expensive or inefficient for many factors.
“Similar to proteins, we need to know the structure of the material to predict its properties.” Ong, associate director of the Center for Sustainable Energy and Energy at the Jacobs School of Engineering. “What we need is an AlphaFold for documents.”
AlphaFold is an AI algorithm developed by Google DeepMind to predict protein structure. To build the equivalent material, Ong and his team combined graph neural network with many-body interactions to build a deep learning architecture that works universally, with high accuracy, on all elements of the periodic table.
Chi Chen, a former senior project scientist in Ong’s lab and first author of the work, now a senior quantum architect at Microsoft Quantum, said: “Mathematical graphs are actually natural representations of a set of atoms. “Using graphs, we can represent the full complexity of materials without suffering the combinatorial explosion of terms in traditional forms.”
To train their model, the team used a huge database of materials’ energies, forces, and stresses collected in Materials Project In the century. The result is the M3GNet interaction potential (IAP), which can predict the energies and forces in any set of atoms. Matterverse.ai was created through combinatorial element substitution on more than 5,000 structural archetypes in the Inorganic Crystal Structure Database (ICSD). The M3GNet IAP was then used to obtain an equilibrium crystal structure—a process known as “relaxation”—for property prediction.
Of the 31 million documents in Mattverse.ai today, more than a million are predicted to be potentially stable. Ong and his team plan to greatly expand not only the number of materials but also the number of properties predicted by ML, including high-value properties with small data sizes, by using a multi-fidelity approach that they developed earlier.
In addition to structural loosening, the M3GNet IAP has wide applications in dynamic simulations of materials and property prediction.
“For example, we are often interested in how quickly lithium ions diffuse in the electrode or electrolyte of a lithium-ion battery. The faster the diffusion, the faster you can charge or discharge the battery.” Bee said. “We have shown that the M3GNet IAP can be used to predict the lithium conductivity of materials with great accuracy. We truly believe that the M3GNet architecture is an open-source transformative tool. greatly expands the possibility of discovering new structures and materials chemistry.”
To promote the use of M3GNet, the team released the framework as open source Python on Github. Since preprinting on Arxiv in February 2022, the team has received interest from academic researchers and industry insiders. There are plans to integrate M3GNet IAP as a tool in commerce material simulation package.
This piece was authored by Chi Chen and Shyue Ping Ong at UC San Diego.
Chi Chen, Universal Graph Deep Learning Interaction Potential for Periodic Tables, Natural Computational Science (2022). DOI: 10.1038/s43588-022-00349-3. www.nature.com/articles/s43588-022-00349-3
Provided by
University of California – San Diego
quote: Breakthrough algorithm expands the material discovery space by level (2022, Nov 28) retrieved Nov 28, 2022 from https://techxplore.com/news/2022-11-breakthrough- algorithm-exploration-space-materials.html
This document is the subject for the collection of authors. Other than any fair dealing for private learning or research purposes, no part may be reproduced without written permission. The content provided is for informational purposes only.