Researchers from the Center for Advanced Systems Understanding (CASUS) at the Helmholtz-Zentrum Dresden-Rossendorf (HZDR) in Görlitz, Germany, and Sandia National Laboratories in Albuquerque, New Mexico, U.S., have made significant progress in overcoming the limitations of traditional electronic structure simulation techniques. Their innovative approach, called Materials Learning Algorithms (MALA), utilizes machine learning to achieve high fidelity and scalability across different time and length scales. This breakthrough has the potential to revolutionize fields such as drug design and energy storage, where understanding the arrangement of electrons in matter is crucial.
Electrons, as elementary particles, play a fundamental role in chemistry and materials science. Their interactions with each other and with atomic nuclei govern various phenomena observed in these fields. Gaining a comprehensive understanding of the electronic structure of matter provides valuable insights into molecular reactivity, energy transport within planets, and material failure mechanisms.
To tackle scientific challenges in these areas, computational modeling and simulation have become increasingly important, often leveraging the capabilities of high-performance computing. However, one significant hurdle in achieving accurate simulations with quantum precision has been the lack of a predictive modeling technique that combines high accuracy with scalability across different length and time scales.
Classical atomistic simulation methods are capable of handling large and complex systems but are limited by their omission of quantum electronic structure. On the other hand, simulation methods based on first principles, which do not rely on assumptions such as empirical modeling and parameter fitting, offer high fidelity but are computationally demanding. For example, density functional theory (DFT), a commonly used first principles method, exhibits cubic scaling with system size, limiting its predictive capabilities to small scales.
The novel MALA software stack developed by the researchers addresses these limitations. By harnessing the power of machine learning, MALA enables simulations at previously unattainable length scales. This advancement has the potential to propel research in drug design and energy storage forward, as it provides a simulation technique that combines high fidelity and scalability, overcoming the roadblocks that hindered progress in these technologies. The research findings have been published in the journal npj Computational Materials.
Hybrid approach based on deep learning
The researchers have introduced a groundbreaking simulation method called the Materials Learning Algorithms (MALA) software stack. In the field of computer science, a software stack refers to a collection of algorithms and software components combined to create a software application that addresses a specific problem.
Lenz Fiedler, a Ph.D. student and key developer of MALA at CASUS, explains that MALA integrates machine learning with physics-based approaches to predict the electronic structure of materials. The method employs a hybrid approach, combining deep learning, a well-established machine learning technique, for accurate prediction of local quantities, along with physics algorithms to compute global quantities of interest.
The MALA software stack takes the spatial arrangement of atoms as input and generates fingerprints called bispectrum components. These components encode the arrangement of atoms around each point on a Cartesian grid. The machine learning model in MALA is trained to predict the electronic structure based on these atomic neighborhoods. One notable advantage of MALA is that its machine learning model can be independent of the system size, enabling it to be trained on data from small systems and applied at any scale.
The researchers demonstrated the impressive effectiveness of this approach in their publication. They achieved a remarkable speedup of over 1,000 times for smaller system sizes, consisting of a few thousand atoms, compared to conventional algorithms. Moreover, the team showcased MALA’s ability to accurately perform electronic structure calculations at a large scale, involving more than 100,000 atoms. Notably, this accomplishment was achieved with modest computational effort, revealing the limitations of traditional density functional theory (DFT) codes.
Attila Cangi, the Acting Department Head of Matter under Extreme Conditions at CASUS, explains that as the system size increases and more atoms are involved, DFT calculations become impractical, while MALA’s speed advantage continues to grow. The key breakthrough of MALA lies in its capability to operate on local atomic environments, enabling accurate numerical predictions that are minimally affected by system size. This achievement opens up computational possibilities that were previously considered unattainable.
Boost for applied research expected
Cangi has ambitious goals of revolutionizing electronic structure calculations through the use of machine learning. He believes that MALA will bring about a significant transformation in this field by enabling simulations of much larger systems at unprecedented speeds. This breakthrough will have far-reaching implications, allowing researchers to tackle a wide range of societal challenges with an improved baseline. Examples include the development of new vaccines, the discovery of novel materials for energy storage, large-scale simulations of semiconductor devices, the study of material defects, and the exploration of chemical reactions for converting carbon dioxide into environmentally friendly minerals.
MALA’s approach is particularly well-suited for high-performance computing (HPC). As the system size increases, MALA takes advantage of the computational grid it utilizes by enabling independent processing on multiple grid points, effectively harnessing the power of HPC resources, especially graphical processing units.
Siva Rajamanickam, a staff scientist and parallel computing expert at the Sandia National Laboratories, explains that MALA’s algorithm for electronic structure calculations is highly compatible with modern HPC systems that have distributed accelerators. The ability to decompose work and execute parallel computations on different grid points across various accelerators makes MALA an ideal choice for scalable machine learning on HPC resources, resulting in unparalleled speed and efficiency in electronic structure calculations.