Advances in stochastic optimization applied to interstitial segregation
Godehard Sutmann, Forschungszentrum Jülich, Jülich, Germany
Segregation of interstitial atoms in materials is a topic which is important for understanding materials properties and predicting its mechanical response. As a test system, single dislocation and dislocation networks were considered in a bcc Fe matrix and segregation was obtained as a result of a Monte Carlo energy minimization process. Applying a straight forward minimization procedure, which considers energy differences between subsequent configurations, which were obtained by homogeneous trial moves, provides a fast initial step of segregation, but extremely slows down in a later stage, which essentially avoids to get down to the minimal energy state. An analytical description has been obtained to analyse this behaviour and to make predictions about the number of steps, which are needed to get to the minimum. Thereby it is possible to distinguish between number of accepted steps in the MC procedure and rejected steps between two accepted steps. The formalism is developed for monotonous energy landscapes, but can be generalized for arbitrary rough energy surfaces. As a way out of the performance limitations, biased MC procedures were developed, which show a strong speedup w.r.t. the original stochastic approach, which is especially powerful for large systems.
Session M3: Moday, 25 June 2018
End: 06:15 p.m.