The University of Luxembourg, a leading international research university, is seeking a postdoctoral researcher for the UMLFF project within the Faculty of Science, Technology and Medicine (FSTM). This project focuses on developing uncertainty-aware machine learning force fields (MLFFs) by integrating state-of-the-art equivariant neural network architectures with robust uncertainty estimates. The goal is to enable fully automated active learning in configurational and chemical space, advancing general-purpose MLFFs that are accurate and reliable across diverse chemical systems.

The successful candidate will join the Theoretical Chemical Physics group, working closely with Dr. Igor Poltavskyi and Prof. Alexandre Tkatchenko, both recognized experts in machine-learning force-field development. The project benefits from access to high-performance computing resources, including the MeluXina supercomputer, and strong international collaborations with institutions such as the Max Planck Institute, Google DeepMind, and TU Berlin.

Key scientific themes include intrinsic uncertainty estimation in MLFFs (epistemic and aleatoric), negative log-likelihood and calibration methods for force-field training, feature-space and orbit-based analysis of atomic environments, detection of extrapolation and low-reference data regimes, active learning in configurational and chemical space, and training and benchmarking of large-scale foundational MLFF models. The postdoctoral researcher will develop and implement uncertainty-aware MLFF architectures, design and test loss functions and calibration methods, analyze feature-space structure and model applicability domains, train and benchmark MLFF models on diverse molecular and materials datasets, integrate uncertainty estimates into active learning pipelines, and publish results in leading journals and conferences.

Applicants must hold a PhD in physics, chemistry, materials science, computational science, or a related field, with a strong background in machine learning and/or computational chemistry. Experience with deep neural networks, scientific programming in Python, molecular simulations, force fields, or atomistic modeling is essential. Highly desirable qualifications include experience with equivariant or graph neural networks (e.g., NequIP, MACE, SO3LR, Allegro), MLFF training and validation, knowledge of uncertainty estimation (Bayesian or ensemble methods), experience with HPC environments and large-scale model training, and familiarity with DFT codes (e.g., FHI-aims) and molecular dynamics.

The University of Luxembourg offers a modern, dynamic, and multilingual environment, with high-quality equipment and close ties to the business world and Luxembourg labour market. The position is a fixed-term contract for 12 months, full-time, with a yearly gross salary of EUR 85176. Applications should be submitted online through the HR system, including a CV with publication list, cover letter, PhD diploma or expected defense date, transcript of university-level courses, and contact details of 2-3 referees. Early applications are encouraged as review begins upon receipt. Applications by email will not be considered. The university promotes an inclusive culture and encourages applications from individuals of all backgrounds.