research

We accelerate materials design with computational and data-driven approaches at BAM and the University of Jena. In particular, we focus on the design of safe, sustainable, and reliable materials. To achieve this, we combine first-principles simulations, condensed matter physics, materials informatics, machine learning, and high-throughput computational methods.

Our research is organized around two complementary themes: (i) materials discovery with Materials Acceleration Platforms and (ii) machine-learned models for realistic materials simulations of advanced and chemically complex materials. In particular, we are interested in battery materials. These activities are supported by substantial contributions to open-source software and community research infrastructures. The research connects to the activity fields Materials Design and Electrical energy storage and conversion at BAM and includes collaborations with researchers across both areas.

Materials Acceleration Platforms and Data-Driven Materials Design

• Inverse design and generative models (e.g., evaluation of generative models)
• Synthesis prediction (e.g., via co-training and PU learning)
• Property forecasting (e.g., for thermal conductivity via chemical bonding concepts)
• Chemical heuristics and materials descriptors (e.g., for bond angle-based features for magnetic predictions or for bonding-based features for thermal conductivity prediction)
• Automated bonding analysis (e.g., as part of LobsterPy and atomate2)
• High-throughput calculations and data generation (e.g., for our bonding analysis database)
• Pre-trained and foundation models for materials discovery (e.g., as part of MACE-MP-0)
• Integration of experimental and computational data
• Autonomous optimization workflows

Machine-Learned Models for Realistic Materials Simulations

• Machine-learned interatomic potentials (e.g., benchmarking of MACE-MP-0)
• Automated training and benchmarking (e.g., development of autoplex)
• Vibrational, thermodynamic, and transport properties (e.g., for the computation of thermal conductivity)
• Chemically complex materials (e.g., see our recent review)
• Large-scale atomistic simulations
• Battery materials and interfaces
• Materials stability, degradation, and safety

Open-Source Software and Research Infrastructures

• Major contributors to the materials analysis software Pymatgen
• Maintenance and development of the workflow library Atomate2
• Maintenance and joint development of software autoplex for automated training of machine learning interatomic potentials (joint development with the group of Prof. Volker Deringer at the University of Oxford)
• Development of LobsterPy for automated bond analysis
• Dataset and software contributions to the Materials Project and contributions to the Materials Project Software Foundation
• Dataset and software contributions to NOMAD and involvement in FAIRmat 2.0

5 most important recent publications with contributions from our lab

• “Accelerated data-driven materials science with the Materials Project”
• “An automated framework for exploring and learning potential-energy surfaces”
• “Atomate2: Modular workflows for materials science”
• “A Quantum-Chemical Bonding Database for Solid-State Materials”
• “A foundation model for atomistic materials chemistry”