Our goal

Fulfilling modern society’s demand for abundant, dependable and clean energy requires materials with very specific features. This is as true when it comes to generation as it is with respect to energy storage and conservation. Most often, the range of structures and compositions found in nature and the relatively few that have been synthesized in the lab are just a tiny fraction of chemical space, unlikely to comprise the most interesting candidate materials. In this context, the role of computational models has been completely redefined: while they used to be focused on reproducing experimental measurements for small sets of known materials, they are now expected to act as the map and the compass to navigate the space of all conceivable materials.

We develop predictive models of materials’ properties based on scale-bridging workflows. Each of those pipelines begins with first-principles calculations that capture the fundamental chemistry and electronic behavior of matter. At the other end we employ thermodynamics and transport theory to obtain parameter-free, quantitative results that can be directly compared with experiment.

The role of machine learning

Ab-initio-like accuracy, however, is normally tied to steep computational requirements. To actually guide searches through chemical space, we employ machine-learning methods to accelerate those calculations and therefore increase their throughput. Advanced regression, classification and unsupervised learning techniques also help us distill the significant quantities of data generated into physical insight and actionable materials design rules.

Our open-source software for transport calculations

We have been particularly active in developing predictive approaches to charge and thermal transport in semiconductors, based on a combination of first-principles calculations and the Boltzmann transport equation. Specifically, we have extended those techniques from single crystals to defect-laden structures, superlattices, interfaces and nanostructures (among other systems), always retaining excellent agreement with experiment. Those capabilities are essential for the design of new and improved technological solutions in areas ranging from microelectronics to thermoelectric energy harvesting. Thanks to those innovations, we have been able to point to previously overlooked candidate materials and uncover new physical phenomena. We have also released those workflows to the community in the form of open-source software: we maintain the BoltzTraP2, ShengBTE, opens an external URL in a new window and almaBTE, opens an external URL in a new window packages and manage their online user groups, comprising hundreds of researchers.

Logo von Boltz TraP2, blaue und graue schrift mit weißem Hintergrund und blauer Umrandung

© AG Madsen

ShengBTE Logo, blau-roter, flüssig aussehender Hintergrund, chinesisches Zeichen in rot und "BTE" in blau

© ShengBTE

almaBTE Logo, links eine Anordnung verschieden großer Rechtecke in gelb, blau, grün, violett und rot, die zusammen ein Quadrat ergeben, rechts das Wort almaBTE und darunter "Thermal Simulation" in Großbuchstaben

© almaBTE

Current research lines

  • Use of evolutionary algorithms to study surface reconstructions of complex oxides. See subproject P09 of the  SFB TACO.
Taco Logo
Logo of P09 Machine-learning methods for structure prediction of multi-component perovskites
  • Simulation of ionic matter through the use of automatically differentiable neural-network force fields.
  • Accelerated mapping of the the phase diagram of oxides.
  • Computational thermodynamics of defects in semiconductors.
  • Efficient calculation of thermal properties in quasi-1D structures.
  • Thermoelectric properties of two-dimensional materials, multilayers and heterostructures.
  • Generative machine-learning models for crystallographic structures.
  • Advanced solvers of the Boltzmann transport equation.

Selected Publications

  • A differentiable neural-network force field for ionic liquids, H. Montes-Campos, J. Carrete, S. Bichelmaier, L. M. Varela & G. K. H. Madsen, Journal of Chemical Information and Modeling 62 (2022) 88–101. D.O.I.: 10.1021/acs.jcim.1c01380
  • Ultrahigh thermal conductivity of θ-phase tantalum nitride, A. Kundu, X. Yang, J. Ma, T. Feng, J. Carrete, X. Ruan, G. K. H. Madsen & W. Li, Physical Review Letters 126 (2021) 115901. D.O.I.: 10.1103/PhysRevLett.126.115901
  • BoltzTraP2, a program for interpolating band structures and calculating semi-classical transport coefficients, G. K. H. Madsen, J. Carrete & M. J. Verstraete, Computer Physics Communications 231 (2018) 140-145. D.O.I.: 10.1016/j.cpc.2018.05.010
  • Exceptionally strong phonon scattering by B substitution in cubic SiC, A. Katre, J. Carrete, B. Dongre, G. K. H. Madsen & N. Mingo, Physical Review Letters, 119 (2017) 075902. D.O.I.: 10.1103/PhysRevLett.119.075902
  • almaBTE: A solver of the space–time dependent Boltzmann transport equation for phonons in structured materials, J. Carrete, B. Vermeersch, A. Katre, A. van Roekeghem, T. Wang, G. K. H. Madsen & N. Mingo, Computer Physics Communications 220 (2017) 351-362. D.O.I.: 10.1016/j.cpc.2017.06.023
  • Finding unprecedentedly low-thermal-conductivity half-Heusler semiconductors via high-throughput materials modeling, J. Carrete, W. Li, N. Mingo, S. Wang & S. Curtarolo, Physical Review X 4 (2014) 011019. D.O.I.: 10.1103/PhysRevX.4.011019