ABSTRACT
The parameterization process will require much data, either from experiments or from ab initio calculations, as input for the simulation model. Generating billions of system configurations during a simulation requires evaluating an even more significant number of distances between pairs of atoms, angles between triplets or quadruplets of atoms, and, ultimately, interaction energies. It soon became apparent that neural networks could outperform empirical potentials for modeling contributions arising from many-body interactions. The development of neural network potentials started eight years later, with the first development of a complete potential predicted by an artificial neural network. The nanoporous material, or adsorbent, only impacts how the potential energy is calculated. The mathematical description for the reaction coordinate is expected to depend strongly on the phenomenon studied. During a conformational change, the molecule will change geometry.
