(11d) Self-Assembly of Non-Spherical Colloids Using Coarse-Grained Simulations and Machine Learning Approaches

Modern synthesis methods have enabled the fabrication of colloids with various different shapes. In addition, plastic waste degrades to nano- and microplastics particles in the environment, also leading to irregular shaped colloidal particles. These particles, polluting our habitats, have unknown implications for our health and environment. Therefore, understanding self-assembly of these non-spherical particles is crucial not only for many applications, like bottom-up fabrication of large ordered systems, but also for understanding the fate of microplastic particles in the aquatic environment.

To simulate the non-spherical colloids, we utilized the commonly used composite bead approach to represent the different shapes. We find that this approach is not only expensive as the complexity of the colloidal shape increases, due to inter-body distance calculations, but also cumbersome as it is not always obvious how the composite beads should interact with each other. Here, we will highlight the use of data-driven methods to accelerate our simulations. We trained neural networks to directly predict energy, or. forces and torques between non-spherical rigid particles based on their configuration, bypassing the need for inter-bead distance calculations. We show that MD simulations performed with neural net predicted forces and torques can accurately reproduce the structure and kinetics of the traditional simulations with explicit distance calculations. This approach can yield significant computational speedup and can be applied to irregular shapes with any pair interaction, given sufficient training data.