MC/MD Coupling for Scale Bridging Simulations of Solute Segregation in Solids: An Application Study

Parallelization comparison and optimization of a scale-bridging framework to model Cottrell atmospheres

“…In this sense, the present work provides more insight on the performance of the parallel MC approach for modeling the interstitial solute segregation, and also discussing the design decisions to reduce the execution time and exploit computational resources efficiently without systematically affecting the simulation. As it is difficult to hand pick any existing parallel approach for this particular application of Cottrell atmo-spheres modeling, because of several challenges with the different system requirements and fluctuating acceptance rate, a new scalable parallel scheme called 'distributed manager-worker' is presented, and it is compared in performance with an earlier developed scheme, referred to as 'general manager-worker' and implemented by the same authors [7].…”

“…Such introduced C atom creates a tetragonal lattice distortion in the ferritic host matrix and therefore exerts tension on the two first nearest neighbors and compression on the second nearest neighbors. However, the lattice distortion in the host matrix created by this impurity atom has shown to be short ranged [7], and this fact opens different possibilities for the parallelization of the MC algorithm. Each trial move involves two spherical regions in the system, which are in principle allowed to overlap, with one of the chosen atoms in the center of each region.…”

“…After that, the local regions around the target atoms are relaxed within a radius R = 2 nm and a new possible value for the local energy minimum is computed. Following the principle of localized energy computations in spherical subsystems, non-overlapping spheres in the system may be executed as independent MC trial moves in parallel [7], using the cut-off radius for the definition of spheres as the reference.…”

“…Therefore, Classical MD and Metropolis [6] Monte Carlo (MC) are here coupled using virtual atoms to overcome the timescale limitation. Each virtual atom is defined as a placeholder, that is, a site in the system which may potentially be suitable to place a C atom: consequently, the number of virtual atoms in a system depends strongly on the crystal structure [7]. Considering this scope, two different parallelization approaches applied for the MC simulation in the coupling scheme are compared for different scenarios according to its performance and statistical efficiency, so that the prediction capabilities of the underlying physics and the computational performance of these algorithms is evaluated.…”

“…Considering this scope, two different parallelization approaches applied for the MC simulation in the coupling scheme are compared for different scenarios according to its performance and statistical efficiency, so that the prediction capabilities of the underlying physics and the computational performance of these algorithms is evaluated. Both parallel approaches are based on the manager-worker scheme: the first approach is a general manager-worker based on a speculative scheme developed by the authors [7] and the second approach is newly proposed scheme with a distributed manager-worker that uses domain decomposition combined with a load balancing routine. The main contributions of the paper are (1)…”