Parts | Days | Selection | Search | Updates | Downloads | Help
MM: Fachverband Metall- und Materialphysik
MM 12: Postersitzung I
MM 12.28: Poster
Monday, March 14, 2011, 17:30–19:00, P5
Construction of High-Dimensional Neural Network Potentials Based on Atomic Pairs — •Jovan Jose Kochumannil Varghese and Jörg Behler — Ruhr-Universität Bochum, D-44780 Bochum
An accurate description of the interatomic potential is the crucial step in theoretical simulations. Consequently, a large number of potentials of varying form and complexity has been reported in the literature. Still, for some systems the accuracy that can be achieved is not satisfying. Artificial Neural Networks (NN) have become a promising new tool for the construction of efficient and accurate potentials due to their flexible functional form. We present a new high-dimensional NN approach based on an expansion of the total energy in terms of environment-dependent atom pairs. The advantages and drawbacks of this approach are discussed and compared to the alternative approach employing a summation of atomic energy contributions.