Investigation of water desalination/purification with molecular dynamics and machine learning techniques

Abstract

This paper incorporates a number of parameters, such as nanopore size, wall wettability, and electric field strength, to assess their effect on ion removal from nanochannels filled with water. Molecular dynamics simulations are incorporated to monitor the process and a numerical database is created with the results. We show that the movement of ions in water nanochannels under the effect of an electric field is multifactorial. Potential energy regions of various strength are formed inside the nanochannel, and ions are either drifted to the walls and rejected from the solution or form clusters that are trapped inside low potential energy regions. Further computational investigation is made with the incorporation of machine learning techniques that suggest an alternative path to predict the water/ion solution properties. Our test procedure here involves the calculation of diffusion coefficient values and the incorporation of four ML algorithms, for comparison reasons, which exploit MD calculated results and are trained to predict the diffusion coefficient values in cases where no simulation data exist. This two-fold computational approach constitutes a fast and accurate solution that could be adjusted to similar ion separation models for property extraction.