Search for: All records

Training machine learning interatomic potentials often requires optimizing a loss function composed of three variables: potential energies, forces, and stress. The contribution of each variable to the total loss is typically weighted using fixed coefficients. Identifying these coefficients usually relies on iterative or heuristic methods, which may yield sub-optimal results. To address this issue, we propose an adaptive loss weighting algorithm that automatically adjusts the loss weights of these variables during the training of potentials, dynamically adapting to the characteristics of the training dataset. The comparative analysis of models trained with fixed and adaptive loss weights demonstrates that the adaptive method not only achieves more balanced predictions across the three variables but also improves overall prediction accuracy. 

This study employs all-atomistic (AA) molecular dynamics (MD) simulations to investigate the crystallization and melting behavior of polar and nonpolar polymer chains on monolayers of graphene and graphene oxide (GO). Polyvinyl alcohol (PVA) and polyethylene (PE) are used as representative polar and nonpolar polymers, respectively. A modified order parameter is introduced to quantify the degree of two-dimensional (2D) crystallization of polymer chains. Our results show that PVA and PE chains exhibit significantly different crystallization behavior. PVA chains tend to form a more rounded, denser, and folded-stemmed lamellar structure, while PE chains tend to form an elongated straight pattern. The presence of oxidation groups on the GO substrate reduces the crystallinity of both PVA and PE chains, which is derived from the analysis of modified order parameter. Meanwhile, the crystallization patterns of polymer chains are influenced by the percentage, chemical components, and distribution of the oxidation groups. In addition, our study reveals that 2D crystalized polymer chains exhibit different melting behavior depending on their polarity. PVA chains exhibit a more molecular weight-dependent melting temperature than PE chains, which have a lower melting temperature and are relatively insensitive to molecular weight. These findings highlight the critical role of substrate and chain polarity in the crystallization and melting of polymer chains. Overall, our study provides valuable insights into the design of graphene-based polymer heterostructures and composites with tailored properties. 

We present a computational study on the solid–solid phase transition of a model two-dimensional system between hexagonal and square phases under pressure. The atomistic mechanism of phase nucleation and propagation are determined using solid-state Dimer and nudged elastic band (NEB) methods. The Dimer is applied to identify the saddle configurations and NEB is applied to generate the transition minimum energy path (MEP) using the outputs of Dimer. Both the atomic and cell degrees of freedom are used in saddle search, allowing us to capture the critical nuclei with relatively small supercells. It is found that the phase nucleation in the model material is triggered by the localized shear deformation that comes from the relative shift between two adjacent atomic layers. In addition to the conventional layer-by-layer phase propagation, an interesting defect-assisted low barrier propagation path is identified in the hexagonal to square phase transition. The study demonstrates the significance of using the Dimer method in exploring unknown transition paths without a priori assumption. The results of this study also shed light on phase transition mechanisms of other solid-state and colloidal systems.