Materials Learning Algorithms (MALA) is a novel software framework that aims to transform materials science. The tool harnesses the power of machine learning to predict the electronic structure of materials, a fundamental aspect that underpins the understanding of various molecular and material properties. Addressing the limitations of current methods like density functional theory (DFT), MALA allows calculations at scales that were previously unattainable, enabling scientists to explore complex structures with remarkable efficiency and speed.
Materials Learning Algorithms (MALA) is a novel software framework that aims to transform materials science. The tool harnesses the power of machine learning to predict the electronic structure of materials, a fundamental aspect that underpins the understanding of various molecular and material properties. Addressing the limitations of current methods like density functional theory (DFT), MALA allows calculations at scales that were previously unattainable, enabling scientists to explore complex structures with remarkable efficiency and speed. Learn more about MALA and the contributions of one of the project’s scientific supervisors, Attila Cangi, a Helmholtz AI Associate and acting head of the Department Matter under Extreme Conditions at the Center for Advanced Systems Understanding, Helmholtz-Zentrum Dresden-Rossendorf.
MALA has emerged by the need to push the boundaries of computational methods in understanding electronic structures. Undoubtedly, density functional theory (DFT), a widely used approach in materials simulation which was awarded the Nobel Prize in 1998, has been instrumental so far. However, its computational demands pose significant constraints, especially as the system size increases. MALA steps in to revolutionize this process by developing machine learning models to replace DFT calculations, offering scalability and a significant boost in speed. This approach enables computations at previously unattainable scales and up to a 5,000-fold acceleration for systems tractable with DFT. This significant leap is a result of a concerted effort by the Sandia National Laboratories and the Center for Advanced Systems Understanding at Helmholtz-Zentrum Dresden-Rossendorf (HZDR).
Attila Cangi on supervising MALA:
"In collaboration with my former colleagues from Sandia National Laboratories, we initiated the MALA project in 2019 with a Laboratory Directed Research and Development grant from the United States Department of Energy. Upon my arrival at the Center for Advanced Systems Understanding in 2020, the MALA project became one of the core projects within my research group. Since then, we have been actively improving the software stack and converting it to a production-level code for electronic structure calculations driven by machine learning."
MALA integrates machine learning with physics-based approaches to predict the electronic structure of materials. It takes a hybrid approach, using deep learning to accurately predict local quantities, complemented by physics algorithms to compute global quantities of interest. By leveraging machine learning, MALA significantly speeds up conventional DFT calculations and enables calculations at previously unattainable length scales. The MALA team recently demonstrated this capability in npj Computational Materials (DOI: 10.1038/s41524-023-01070-z).
In addition to HZDR and Sandia National Laboratories, MALA is already being used by institutions and companies such as the Georgia Institute of Technology, North Carolina A&T State University, Sambanova Systems Inc., Nvidia Corp., and AMD Inc. MALA was awarded the prestigious R&D 100 Award 2023, which recognizes emerging technologies each year.
About Cangi’s work
Attila Cangi’s research group studies the structure and dynamics of matter under a wide range of conditions, from ambient to extreme states. His group develops innovative methods that integrate artificial intelligence, materials science, and high-performance computing to perform simulations based on the fundamental building blocks at the quantum level. His team’s research currently focuses on understanding phenomena arising from strong electromagnetic fields, high temperatures,and pressures.
For the past five years, Cangi has been working with artificial intelligence and machine learning in electronic structure theory, condensed matter physics, and materials science. In parallel with the development of the MALA project, he has also contributed to the construction of a novel class of machine-learning interatomic potentials for molecular dynamics simulations, combining lattice degrees of freedom with electronic spins. His group has also explored the overlap between machine learning and quantum computing. Additionally, his group has recently applied physics-informed neural networks to tackle time-dependent quantum mechanical problems. Moving forward, Cangi’s group intends to investigate the implementation ofgraph neural networks and machine learning to discover differential equations in the context of materials science.
*Other sources: AMP 06 September 2023, https://static.asminternational.org/amp/202309/.