Neural networks for fitting PES data distributions of asphaltene interaction

Neural networks methodology is a tool to get potential energy surface (PES) in cases where there is too much dispersion of data; hence, a binding energy fitting can be found with this methodology on asphaltene-asphaltene molecular interaction. A data distribution of intermolecular pair potential (U_AA) interaction in a vacuum between two molecular asphaltenes systems using compass classical force field has been previously reported (Energy Fuels 2006, 20, 195). In the latter, all possible interactions between the species were taken into account. Focusing in their data distribution, we have applied neural networks on the following molecule-molecule geometry orientations with the purpose of obtaining energy vs. contact distance, which is the minimum distance where the interacting species is not equal to zero: face-to-face distribution of asphaltene-asphaltene interactions, all geometry asphaltene-asphaltene discrete distributions, and the random distribution of asphaltene-asphaltene interactions. Neural networks fit provide a potential energy surface through a function approximation for a data distribution of high dispersion; hence a binding energy is found with this methodology.