Neural Density Functional Theory in Higher Dimensions with Convolutional Layers

Abstract

Based on recent advancements in using machine learning for classical density functional theory for systems with one-dimensional, planar inhomogeneities, we propose a machine learning model for application in two dimensions (2D) akin to density functionals in weighted density forms, as e. g. in fundamental measure theory (FMT). We implement the model with fast convolutional layers only and apply it to a system of hard disks in fully 2D inhomogeneous situations. The model is trained on a combination of smooth and steplike external potentials in the fluid phase. Pair correlation functions from test particle geometry show very satisfactory agreement with simulations although these types of external potentials have not been included in the training. The method should be fully applicable to 3D problems, where the bottleneck at the moment appears to be in obtaining smooth enough 3D histograms as training data from simulations.