Dresden 2026 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 14: Machine Learning in Dynamics and Statistical Physics II

DY 14.6: Talk

Monday, March 9, 2026, 16:15–16:30, HÜL/S186

Learning microstructure in active matter — •Writu Dasgupta, Suvendu Mandal, Aritra Mukhopadhyay, and Benno Liebchen — Technische Universität Darmstadt, Darmstadt, Germany

Understanding the full parameter dependence of microscopic structure in active matter remains a central challenge, particularly for strong activity and high density, where simulations become increasingly expensive. Here, we present a data-driven approach that learns radial and angular correlations in terms of the pair-correlation function g(r,θ) of passive and active Brownian particles. Our predictions are in close quantitative agreement with Brownian dynamics simulations, even for parameter values that the neural networks had not previously encountered during training. Our predictions are subsequently distilled into compact, closed-form expressions using symbolic regression, providing an interpretable description of the underlying structure. Our approach offers a unified and computationally efficient route to understanding non-equilibrium correlations.

Keywords: Machine learning; Symbolic regression; Pair-correlation function; Active Brownian particles