Presidents and fellows of harvard university the discovery of new, high-performance permanent magnet materials is critical for modern technologies but challenging due to the complex interplay of…

The Department of Energy Advanced Research Projects Agency awarded a $2.590951 million cooperative agreement to President and Fellows of Harvard College on August 4, 2026, under the Advanced Research Projects Agency–Energy program (CFDA 81.135).

Harvard will develop computational methods and machine learning models to accelerate the discovery and design of next-generation permanent magnet materials. The project addresses a critical gap: existing computational methods, including semilocal density functional theory, cannot accurately capture magnetic properties in materials with localized d and f electrons. Harvard will enhance the computational pipeline by accurately describing electron correlations, develop foundational machine learning energy models that capture quantum effects across broad material ranges, and use accelerated models for large-scale screening to identify candidate materials for experimental validation. The work integrates computational development with experimental characterization to validate predictions and enable computationally driven magnet discovery.

Performance occurs in Cambridge, Delaware, and New Jersey through August 3, 2029. The award funds research at Harvard's Office for Sponsored Programs, leveraging the university's expertise in quantum science and advanced materials research to support the national objective of developing high-potential, high-impact energy technologies before they reach private-sector investment readiness.

Presidents and fellows of harvard university the discovery of new, high-performance permanent magnet materials is critical for modern technologies but challenging due to the complex interplay of quantum effects governing their essential properties and the vast unknown space of candidate materials. The performance of magnets is governed by bulk properties, which depend sensitively on structure and composition. Therefore, the rapid rational discovery of next-generation materials requires accurate and efficient computational methods that guide and supplement experimental investigations. However, even a quantitative description of known permanent magnets is currently very difficult with existing computational methods. Idealized physical models do not describe realistic materials, and popular first principles methods like semilocal density functional theory (DFT) struggle to capture key magnetic properties due to strong correlations and self-interactions in systems with localized d and f ELECTRONS1,2. The lack of adequate methods has been the primary obstacle to the computationally driven discovery of magnets, forcing experimental research to rely on heuristics and chance. Development of new methods can only proceed in close interaction with experimental characterization that provides validation of predictions. Once high accuracy and reliability are achieved on the quantum level, machine learning methods will be key to large-scale simulations and fast screening, enabling the identification of new materials to be made and studied experimentally. We will pursue the following major objectives in this proposed project: 1. Significantly enhance the computational pipeline for understanding and designing next-generation magnets by accurately describing electron correlations. 2. Develop foundational machine learning energy models that faithfully capture quantum effects and describe complex magnetic interactions across a broad range of materials. 3. Use accelerated models for screening and detailed calculations of realistic compounds and perform multi-modal experimental characterization to verify the computational predictions. 4. Engage in a close experimental-computational collaboration to iteratively optimize, synthesize, and study new magnetic compounds identified by screening