The development of advanced materials relies on a continuous feedback loop between computational prediction, experimental synthesis, and characterization. Computational methods support this process in many ways. In this contribution, we focus on two complementary aspects: screening vast chemical spaces to identify promising candidates for experimental investigation and simulating materials behavior at scales beyond the reach of conventional first-principles methods. To address both challenges, we combine quantum-chemical calculations, workflow automation, and machine learning (ML).

For materials discovery, we develop automated high-throughput quantum-chemical workflows for large-scale materials screening, as implemented in the open-source framework atomate2.[1, 2] These workflows enable the systematic generation and analysis of high-quality electronic-structure data for known and hypothetical compounds. A central theme of our work is integrating chemical bonding concepts into machine learning models. Using descriptors derived from quantum-mechanical bonding analyses, we develop chemically informed ML approaches for predicting vibrational and thermal transport properties, combining interpretability with efficiency.[3, 4] Beyond property prediction, we address the challenge of predicting new stable compounds and identifying synthesizable compounds among large numbers of hypothetical compounds. To this end, we develop metrics to assess generative models and employ positive-unlabeled learning strategies that enable reliable classification even without true negative examples.[5–7]

A second focus of our research is developing machine-learned interatomic potentials for advanced atomistic simulations. Through our software autoplex,[8] we automate training-data generation, model construction, and validation, enabling efficient development of both general-purpose and system-specific potentials. These models extend simulations to time and length scales inaccessible to conventional first-principles calculations while maintaining near-first-principles accuracy. More recently, we have investigated data-efficient fine-tuning strategies for pretrained interatomic potentials, showing that accurate phonon and thermal-property predictions can be achieved with only limited additional first-principles data.[9, 10] Applications of autoplex include investigating structurally complex and disordered materials such as amorphous arsenic.[11]

Together, these developments show how electronic-structure theory, chemical bonding analysis, workflow automation, and machine learning can be combined to accelerate both new-material discovery and the simulation of complex material behavior while maintaining a close connection to the underlying physics and chemistry.

References:

1. M. K. Horton, P. Huck, R. X. Yang, et al., “Accelerated data-driven materials science with the Materials Project,” Nature Materials 24 (2025): 1522–1532, https://doi.org/10.1038/s41563-025-02272-0.

2. A. M. Ganose, H. Sahasrabuddhe, M. Asta, et al., “Atomate2: Modular workflows for materials science,” Digital Discovery 4 (2025): 1944–1973, https://doi.org/10.1039/D5DD00019J.

3. A. A. Naik, C. Ertural, N. Dhamrait, P. Benner, J. George, “A Quantum-Chemical Bonding Database for Solid-State Materials,” Scientific Data 10 (2023): 610, https://doi.org/10.1038/s41597-023-02477-5.

4. A. A. Naik, N. Dhamrait, K. Ueltzen, et al., “A Critical Assessment of Bonding Descriptors for Predicting Materials Properties,” Advanced Intelligent Discovery (2026): e70129, https://doi.org/10.1002/aidi.70129.

5. P. Hagemann, S. Müller, J. George, P. Benner, “Transport Novelty Distance: A Distributional Metric for Evaluating Material Generative Models,” Journal of Physics: Materials (2026), https://doi.org/10.1088/2515-7639/ae869a.

6. P. Hagemann, K. Ueltzen, S. Müller, J. George, P. Benner, 2026, arXiv preprint, DOI: 10.48550/arXiv.2607.28776.

7. S. Amariamir, J. George, P. Benner, “SynCoTrain: a dual classifier PU-learning framework for synthesizability prediction,” Digital Discovery 4 (2025): 1437–1448, https://doi.org/10.1039/D4DD00394B.

8. Y. Liu, J. D. Morrow, C. Ertural, et al., “An automated framework for exploring and learning potential-energy surfaces,” Nature Communications 16 (2025): 7666, https://doi.org/10.1038/s41467-025-62510-6.

9. J. Grandel, P. Benner, J. George, “Targeted fine-tuning of machine-learning interatomic potentials for phonons and phase transitions,” npj Computational Materials (2026), https://doi.org/10.1038/s41524-026-02338-w.

10. I. Batatia, P. Benner, Y. Chiang, et al., “A foundation model for atomistic materials chemistry,” The Journal of Chemical Physics 163 (2025): 184110, https://doi.org/10.1063/5.0297006.

11. Y. Liu, Y. Zhou, R. Ademuwagun, et al., “Medium-Range Structural Order in Amorphous Arsenic,” Journal of the American Chemical Society 148 (2026): 9400–9412, https://doi.org/10.1021/jacs.5c18688.