Felix Aertebjerg

About Me

PhD student at DTU Energy and supervised by Assistant Prof. Arghya Bhowmik, as well as co-supervised by Prof. Juan Maria García Lastra. My focus areas are geometric deep learning and probabilistic modeling applied to the sciences, with a particular focus on solids and battery materials. My approach is to embrace modern compute and algorithms and to adapt them to physically meaningful representations. Specifically, I believe that Machine Learning methods must be both explainable (trust-worthy) and efficient (scalable).

Outside of AI4X, my personal passions are time-series data and small language models. Weather forecasting is fx. a fascinating problem, it requires the model to both be able to capture complex physical systems and the dynamics of change therein. SMLs are cool too, they capture the pareto frontier of compute and intelligence so well. In the future, I hope to be able to run a Opus quality model on my laptop.

News

• Sep 2026

  "Complete Neural Electronic Initialization Accelerates Materials DFT" was accepted at the Sim2Science workshop at NeurIPS 2026! ☄️

• Sep 2026

  A new preprint is out: arXiv:2609.21759 ("Complete Neural Electronic Initialization Accelerates Materials DFT")! We initialize all the PAW DFT components and achieve end-to-end speed-ups in VASP for non-magnetic & magnetic crystals.

• May 2026

  ELECTRAFI was accepted to ICML! 🔥

• Jan 2026

  The preprint of ELECTRAFI is out! arXiv:2601.19966

Publications

• 

  Complete Neural Electronic Initialization Accelerates Materials DFT

  Felix Ærtebjerg, Jonas Elsborg, Arghya Bhowmik

  Preprint

  arXivBibTeX
• 

  Global Plane Waves From Local Gaussians: Periodic Charge Densities in a Blink

  Jonas Elsborg, Felix Ærtebjerg, Luca Thiede, Alán Aspuru-Guzik, Tejs Vegge, Arghya Bhowmik

  ICML

  arXivBibTeX

Research Interests

• MCMC, Gaussian Processes, Representation Learning
• Geometric GNNs, Efficient Attenion, Bayesian ML
• HPC, Scaling

Education

Work Experience