Generative machine learning in the grand canonical ensemble

• Applicant: Professor Dr. Tristan Bereau
• Subject Area: Statistical Physics, Nonlinear Dynamics, Complex Systems, Soft and Fluid Matter, Biological Physics ; Methods in Artificial Intelligence and Machine Learning
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 554514035

We use physics-informed generative machine learning (ML) to describe particle configurations in the grand-canonical ensemble. We map the problem of coupling the interactions of the Hamiltonian to a denoising diffusion probabilistic model (DDPM). Our DDPM is capable of accurately coupling hundreds of degrees of freedom, which offers new perspectives on the calculation of the chemical potential. We propose to expand on these ideas by supporting more complex force fields, intramolecular degrees of freedom, and a body-order decomposition of the energy function. Finally, we propose to learn the equation of state of complex molecular systems: molecular fluids and phospholipid membranes.

• DFG Programme: Research Grants