(2eq) Artificial Neural Network Model for Capturing the Effect of Local Atomic Environment on Surface Diffusion in Metals and Theirs Alloys

The rate constants of thermally activated atomic scale processes are influenced by their local chemical environment. The surface diffusion in metals and their alloys is much slower kinetic process, specially at low temperature. These rare events are difficult to capure by regular molecular dynamics. In situations where rate constants vary over several orders of magnitude, dynamical simulations of material systems require rate constant models that can accurately capture the effect of the chemical environment on the rates. An artificial neural network (ANN) model is developed that can be used to calculate the activation barrier of adatom hopping on (001) surfaces of different metals and their alloys. Different composition of the alloy is allowed. The hopping adatom can be either species in the alloy composition. The ANN model is trained using activation barriers calculated for different atomic environments with binary alloys. First, local environment with many (say 96 in one of the case) different sites including various adatom and substrate sites are explored for the creation of a database of activation barriers. The database is next analysed with help of decision trees to identify the top few (say 17 important sites) most important sites. These sites form the input layer of the ANN model. Good predictions are obtained with the ANN model for the different composition of alloys. The ANN model can be useful for modeling morphological changes in nanomaterials involving metal alloys.

Keywords: Decision trees, Cluster expansion models, Neural networks, Activated process, Molecular simulations, Machine learning

Research Interests: Molecular Simulations, Surface Science, Modeling and Simulation, Optimization, Machine Learning