Fig. 2: Iterative training process.
From: Graph neural network coarse-grain force field for the molecular crystal RDX
a Atomistic NVT simulations are performed for Gen 0. b Constrained atomistic simulations are performed in which the center of mass positions of the nitro and triazine groups are fixed in space. c CG model positions and forces from constrained atomistic simulations are used to train CG-GNNFF. d CG MD simulations with the trained CG-GNNFF are performed. e Decoder is used to reconstruct CG model trajectories into atomistic trajectories. Constrained atomistic simulations on these trajectories are performed to create the next generation of data.
