Nowadays, humans can examine and design materials in the computer down to the quantum-mechanical structure of the electrons. This is what the JuDFT codes, developed in our group, enable, at the highest level of precision that is essential for studying quantum materials, and with high performance. Moreover, their deep integration with the AiiDA workflow engine enables high-throughput studies of thousands of materials in a simple, reproducible way.

Such capabilities are essential to discover and design materials for the next technological revolutions, such as quantum technology. However, these calculations are so expensive they can only handle materials systems with about a thousand atoms in them. This is a lot in the quantum world, but almost nothing on the scale of technological devices. Imagine if there was a way to model materials systems with millions of atoms, or more, down to the electronic structure. This could revolutionize materials design.

This is what the application of AI, more specifically of machine learning methods, to electronic structure calculation offers. Instead of by equations, these methods get informed by data, and they can improve with time. However, it is paramount to ensure that the results remain accurate, explainable and generalizable. This is the goal of my research.

My research is supported by the Helmholtz School for Data Science in Life, Earth, and Energy (fellow), as well as by the French-German strategic research partnership AIDAS - AI, Data Analytics and Scalable Simulation. I am a member of the European network Daemon-COST Action - Data-driven applications towards the engineering of functional materials.

Have a look at my PhD project. I share all of my related notes, codes and presentations there.

You can find my publications on JuSER, ORCiD, Google Scholar and Zenodo.

I maintain the Best of Atomistic Machine Learning List - the largest list of AML tools on the web.

I am also active on GitHub, BlueSky, LinkedIn, Mastodon, Twitter and YouTube.

PhD Students