Lattice Boltzmann Method, Large Language Models, and Me

• Date: 29. 10. 2026
• Time: 14:00 - 15:30
• Address:
  Sokolovská 83, Praha
• Room: K3
• Speaker: Jakub Klinkovský

The lattice Boltzmann method (LBM) is a mesoscopic approach to computational fluid dynamics: instead of discretizing the Navier–Stokes equations directly, it evolves discrete distribution functions on a lattice through a simple collision-and-streaming procedure. This makes it conceptually elegant, easy to implement, and very well suited to GPU acceleration. I will give an accessible introduction to LBM for listeners with no prior experience: where it comes from, how one iteration works, and how it compares with classical CFD methods. I will then summarize what we implemented in TNL-LBM, our open-source framework published earlier this year, which scales from single-GPU workstations to GPU supercomputers. In the final part I will describe my own recent experiment: using AI agents for a substantial development task, namely adding static mesh refinement to our LBM solver. I will discuss what this way of working looks like in practice, where it accelerates development, and where it fails.