REPRESENTATION OF INTERMOLECULAR POTENTIAL FUNCTIONS BY NEURAL NETWORKS

We have investigated how a neural network representation of intermolec ular potential functions can be used to elevate some of the problems c ommonly encountered during fitting and application of analytical poten tial functions in computer simulations. For this purpose we applied fe ed-forward networks of various sizes to reproduce the three-body inter action energies in the system H2O-Al3+-H2O. In this highly polarizable system the three-body interaction terms are necessary for an accurate description of the system, and it proved difficult to fit an analytic al function to them. Subsequently we performed Monte Carlo simulations on an Al(3+ )ion dissolved in water and compared the results obtained using the neural network type potential function with those using a c onventional analytical potential. The performance and results of our c alculations lead to the conclusion that, for suitable systems, the adv antages of a neural network type representation of potential functions as a model-independent and ''semiautomatic'' potential function outwe igh the disadvantages in computing speed and lack of interpretability.