Madsen Group (TU Wien)

Theoretical Chemistry (TC)

Georg Madsen leads the research area of theoretical chemistry at the TU Wien. Research focuses on a broad range of aspects of theoretical materials chemistry. These include both method development within density functional theory, machine learning, potential energy exploration as well as transport theory. The methods are applied to the discovery of new materials, structure prediction, interface chemistry and thermochemistry. Specific recent research interests include:

We co-authored the highly-cited DFT codes WIEN2k and GPAW. Recently, we applied the optimization techniques developed for ML to construct 25 new functionals, which, for the first time, systematically document the compromises in accuracy inherent to mGGA DFT functionals.

We have authored several highly-cited codes, including almaBTE and BoltzTraP2, for the predictive calculation of transport properties such as the Seebeck coefficient or lattice thermal conductivity. These have been applied to a range of materials discoveries, including the high-throughput work on thermoelectric properties and the recent prediction of record thermal conductivity in the metal θ-TaN.

We have applied machine learning in atomistic studies, including the regression on invariant tensorial descriptors and a JAX-based machine-learned force field (MLFF) framework. The end-to-end differentiable MLFF framework made adversarial loss optimization possible, which together with the implementation of a learned optimizer and resulting in exponentially decreased learning speeds, enabled active learning.

We used the covariance matrix adaptation evolutionary strategy for the exploration of defect structures  and  surface reconstructions and combined our MLFF framework with thermodynamic integration to predict phase diagrams.

Link to Personal homepage: https://www.tuwien.at/tch/tc

ORCID: 0000-0001-9844-9145

Google Scholar: https://goo.gl/Fh4Qcz

---

Team in MECS

---

Open positions

–