Discrete modeling via function approximation methods towards bridging atomic- and micro-scales

Discrete modeling of processes at the atomic-scale affords practical approaches to complex materials of interest commercially and to the US Air Force. Reductions in computation times can be large, suggesting the possibility of real-time modeling of thin film growth and the consequent development of processing routes to achieve specific physical and chemical properties. Formulation of the model to be used is critical in achieving such computational gains. Frameworks for these models such as Monte Carlo and molecular dynamics can be used conceptually, but they cannot be applied in practice because of the high number of required computations per time step. The simplest discrete model involves the Potts model to simulate energies, then to create a partition function of probabilities for various states and configurations, followed by a decision algorithm that determines the state of surface atoms. Although the inclusion of defects, dopants, atom complexes, surface reconstruction and crystal orientations can be included directly in this modeling approach, the resulting collection of behaviors is very entangled with logical and mathematical functions. Hence, the time to exercise the model increases noticeably. It is shown that this problem can be reduced dramatically by employing neuro-computing methods.