Karlsruhe Institute of Technology: Machine learning speeds up material simulations

Research, development and production of new materials depend crucially on fast and precise simulation methods. Machine learning, in which artificial intelligence (AI) independently acquires and uses new knowledge, will make it possible in the future to develop complex material systems in a purely virtual manner. How this works and which applications benefit from it is explained by a researcher from the Karlsruhe Institute of Technology (KIT) together with colleagues from Göttingen and Toronto in an article in the journal Nature Materials. (DOI: 10.1038 / s41563-020-0777-6)


Digitization and virtualization are becoming more and more important in a wide variety of scientific disciplines. This also applies to materials science: Research, development and production of new materials depend crucially on fast and precise simulation methods. This, in turn, benefits a wide variety of applications – from efficient energy storage devices, which are indispensable when using renewable energies, to new drugs whose development requires an understanding of complex biological processes. Methods of AI and machine learning can significantly advance material simulations. “Compared to conventional simulation methods based on classical or quantum mechanical calculations, a clear speed advantage can be achieved with neural networks specially tailored to material simulations, ”explains physicist and AI expert Professor Pascal Friederich, head of the research group AiMat – Artificial Intelligence for Materials Sciences at the Institute for Theoretical Computer Science (ITI) at KIT. “Faster simulation systems will enable scientists in the coming years to develop larger and more complex material systems in a purely virtual manner, to understand them down to the atomic level and to optimize them.” Head of the research group AiMat – Artificial Intelligence for Materials Sciences at the Institute for Theoretical Computer Science (ITI) of KIT. “Faster simulation systems will enable scientists in the coming years to develop larger and more complex material systems in a purely virtual manner, to understand them down to the atomic level and to optimize them.” Head of the research group AiMat – Artificial Intelligence for Materials Sciences at the Institute for Theoretical Computer Science (ITI) of KIT. “Faster simulation systems will enable scientists in the coming years to develop larger and more complex material systems in a purely virtual manner, to understand them down to the atomic level and to optimize them.”


High precision from the atom to the material

In an article published in the journal Nature Materials, Pascal Friederich, who also works as an associated group leader in the field of Nanomaterials by Information-Guided Design at the Institute for Nanotechnology (INT) at KIT, gives one together with researchers from the University of Göttingen and the University of Toronto Overview of the basic principles of machine learning used for material simulations, the data acquisition process and active learning processes. Algorithms for machine learning enable artificial intelligence not only to process the data entered, but also to find patterns and correlations in large data sets, learn from them and make predictions and decisions independently. In material simulations, it is important to achieve high precision over various time and size scales – from the atom to the material – and at the same time limit the computing costs. In their article, the scientists also deal with various current applications, such as small organic molecules and large biomolecules, structurally disordered solid, liquid and gaseous materials as well as complex crystalline systems – for example organometallic framework compounds that are used for gas storage or material separation, for sensors or for Let the catalysts use.


Even more speed with hybrid methods

In order to expand the possibilities of material simulations in the future, the researchers from Karlsruhe, Göttingen and Toronto propose to develop hybrid methods: These combine machine learning (ML) and molecular mechanics (MM) processes. MM simulations use so-called force fields in order to calculate the forces acting on each individual particle and thus to predict motion sequences. The similarity of the ML and MM potentials allows close integration with variable transition areas. Such hybrid methods could, for example, significantly accelerate the simulation of large biomolecules or enzymatic reactions in the future.

Comments are closed.