The Polymer Society of Korea

Abstract

Machine learning (ML) offers a powerful and efficient way to predict the physical and chemical properties of materials using large databases. While methods like density functional theory (DFT) provide accurate predictions, they are computationally expensive and limited in scope. To overcome this, we developed ML-based force fields (ML-FFs) that enable fast and accurate predictions of structural changes at the atomic scale. Our approach combines high-dimensional neural network potentials (HDNNPs), Gaussian process models, and a universal potential, trained on DFT and ab-initio molecular dynamics (AIMD) data. These ML-FFs are applied to both nanomaterials and soft materials, the latter requiring special consideration due to their flexible and dynamic structures. This framework supports efficient multiscale simulations, offering valuable insights for designing high-performance and functional materials.