Transferable and scalable electronic structure simulations with the Materials Learning Algorithms package

Interactions between electrons and nuclei in matter determine all materials' properties. Understanding and modeling these interactions is of paramount importance, particularly in addressing critical scientific questions related to renewable energy solutions, sustainable materials, and semiconductor device modeling. However, simulations of the electronic structure face a common constraint—an accuracy-size tradeoff. While it is possible to simulate materials at the quantum level of accuracy, this is typically limited to a few thousand atoms, even with advanced tools like density functional theory (DFT). On the other hand, large-scale simulations often sacrifice predictive power due to necessary approximations.
The Materials Learning Algorithms (MALA) [1] package addresses these challenges by leveraging a combination of neural networks, physically constrained optimization algorithms [2], and efficient post-processing routines. Unlike existing approaches, MALA replaces DFT entirely, providing access to both scalar quantities of interest, such as energies, and volumetric information about the electronic structure, such as the electronic density. Our research has demonstrated that MALA can be applied to systems with arbitrary numbers of atoms (successfully testing up to 100,000 atoms) [3], across various temperature and pressure ranges [4]. We anticipate that MALA will have a significant impact, enabling unprecedented modeling capabilities in the fields of materials science and semiconductor device modeling.

[1] L. Fiedler, Z. A. Moldabekov, X. Shao, K. Jiang, T. Dornheim, M. Pavanello, A. Cangi, Phys. Rev. Res. 4, 043033 (2022).
[2] L. Fiedler, N. Hoffmann, P. Mohammed, G. A. Popoola, T. Yovell, V. Oles, J. A. Ellis, S. Rajamanickam, A. Cangi, Mach. Learn.: Sci. Technol. 3, 045008 (2022).
[3] L. Fiedler, N. A. Modine, S. Schmerler, D. J. Vogel, G. A. Popoola, A. P. Thompson, S. Rajamanickam, A. Cangi, Npj Comput. Mater. 9, 115 (2023).
[4] L. Fiedler, N. A. Modine, K. D. Miller, A. Cangi, arXiv:2306.06032 (2023).