EUROPEAN CONGRESS AND EXHIBITION ON ADVANCED MATERIALS AND PROCESSES - EUROMAT 2021

Motivated by the limitations of conventional coarse-grained molecular dynamics for simulation of large systems of nanoparticles and the challenges in efficiently representing general pair potentials for rigid bodies, we present a method for interpolating general rigid body pair potentials, based on a specialized type of deep neural network, that maintains essential properties such as conservation of energy and invariance to the chosen origins of the particles. The network uses a specialized geometric abstraction layer to convert the relative coordinates of the rigid bodies to input more suitable to a more conventional artificial neural network, which is trained together with the specialized layer. This results in geometric representations of the particles optimized for the specific potential. The network can be trained directly on scalar values to fit a model without explicit gradient and then be used to efficiently evaluate the force and torque on the particles resulting from the potential. The network is then fitted to a number of interaction models, such as hard Gaussian overlap and Gay-Berne, to demonstrate its flexibility. The models sensitivity to noise in the training data is investigated, and the potential for directly fitting a network to data from molecular dynamics simulation is explored. Furthermore, generalization to soft bodies and potentials for polydisperse systems are discussed.