Navigating Material Space with ML-Generated Electronic Fingerprints
Ihor Neporozhnii a, Oleksandr Voznyy a
a University of Toronto, King's College Road, 10, Toronto, Canada
Materials for Sustainable Development Conference (MATSUS)
Proceedings of MATSUS Spring 2024 Conference (MATSUS24)
#AI - Automation and Nanomaterials (machine learning, artificial intelligence, robotics, accelerated discovery)
Barcelona, Spain, 2024 March 4th - 8th
Organizers: Ivan Infante and Oleksandr Voznyy
Poster, Ihor Neporozhnii, 504
Publication date: 18th December 2023

Finding materials with specific properties is one of the main challenges in the field of material science. The vast material space makes random exploration prohibitively expensive and requires more efficient methods. The viable approach is to search for new materials within the proximity of known materials that possess the desired property but may contain undesirable elements or have low stability making them unsuitable for a specific application.

The search for similar materials is commonly based on the correlation between the material structure and a property of interest and is conducted using structural fingerprints. Recent studies demonstrate the benefits of using fingerprints that include electronic properties, such as density of states (DOS), for clustering and navigating the material space [1], [2]. However, these electronic structure fingerprints are obtained through computationally intensive density functional theory (DFT) calculations, and therefore, are not applicable to high-throughput exploration of novel materials.

Machine learning (ML) demonstrated its ability to accurately predict material properties in significantly less time compared to classical techniques. In this study, we developed a method to find similar materials based on ML-generated projected DOS fingerprints. We demonstrate that using an ML model for fingerprint generation allows us to find materials with similar properties despite their significant structural or compositional differences. The efficiency of the ML model enables scaling this approach to material spaces containing over 100,000 materials, with a low computational cost.

© FUNDACIO DE LA COMUNITAT VALENCIANA SCITO
We use our own and third party cookies for analysing and measuring usage of our website to improve our services. If you continue browsing, we consider accepting its use. You can check our Cookies Policy in which you will also find how to configure your web browser for the use of cookies. More info