Machine Learning Assisted Screening of Two-Dimensional Materials for Water Desalination.

There exists a vast expanse of data in the literature which can be harnessed for accelerated design and discovery of advanced materials for various applications of importance ─ for example, desalination of seawater. Here, we develop a machine learning (ML) model, training it with ∼260 molecular dynamics (MD) computation results, to predict the desalination performance of 2D membranes that exist in the literature. The desalination performance variables of water flux and salt rejection rates are correlated to 49 material features related to the chemistry of the pores and the membranes along with applied pressure, salt concentration, partial charges on the atoms, geometry of the pore, the mechanical properties of the membranes, and the properties of water for the water model used. We used the ML model to screen 3814 structurally optimized 2D materials for maximum water flux and salt rejection rates from the literature. We found some candidates that perform ∼4 times better than the more popularly known 2D materials such as graphene and MoS. This result is verified using data obtained from MD simulations performed on several representative 2D membranes for different classes. Such validated statistical frameworks using literature data can be very useful in guiding experiments in the field of functional materials for varied applications.