What is it about?

Computer simulations of large molecular systems have become a vital tool to aid our understand of biological and chemical processes. One of the main challenges in this field is to rapidly analyse huge amounts of data stemming from simulations. In particular, it is often important to automatically determine large-scale molecular motions, which take place on much longer timescales than elementary atomic vibrations. Our paper presents a new approach to this problem by combining three ideas: a variational principle, which is the mathematical description of slow molecular motions, kernel methods, a well-known learning technique in data science, and random Fourier features, a data-efficient approach to solve the resulting matrix equations.

Featured Image

Why is it important?

While the mathematical theory underlying slow molecular motion has been in place for several years, many previous analyses required at least some degree of prior knowledge about how to best describe the system to be analysed. The class of kernel methods we considered here require very little prior knowledge, while the random approximation technique we employed make our method computationally efficient. Therefore, we believe this work can pave the way towards a more straightforward analysis pipeline for molecular simulation data, which can integrate more easily into the computational modeling process.

Read the Original

This page is a summary of: Efficient approximation of molecular kinetics using random Fourier features, The Journal of Chemical Physics, August 2023, American Institute of Physics, DOI: 10.1063/5.0162619.
You can read the full text:




The following have contributed to this page