Spectral denoising for accelerated analysis of correlated ionic transport

We propose a novel method for analyzing and calculating mass/charge transport in media with non-negligible correlations from atomistic simulations [1]. While widely adopted thanks to its rapid convergence, the dilute uncorrelated approximation is inaccurate. On the other hand, the exact Green-Kubo method is prohibitively expensive for complex and large systems. The approach we present automatically calculates and utilizes the collective diffusion eigenmodes of the displacement correlation matrix to denoise the calculation of the transport properties. It can also be adopted to discover collective diffusion modes in an unsupervised fashion. The approach is universally applicable and provably superior to previously available methods, exhibiting speed ups of several orders of magnitude.