ICAMS / Interdisciplinary Centre for Advanced Materials Simulation


High-performance computing in materials science: current applications and challenges

Date: 14.02.2013
Place: Festcolloquium zur JUQUEEN-Einweihung, Foschungszentrum Jülich, Jülich, Germany

Alexander Hartmaier

To understand whatever binds the world’s innermost core together” has always been a dream of mankind. While the theme of Dr. Faustus aims at an understanding of the metaphysical world, current research activities in computational materials science have the objective of deriving physical laws that connect the atomic structure of materials with their observable, i.e. macroscopic behavior. From quantum mechanics we know that the electronic structure of matter is responsible for its mechanical properties, like stiffness, strength, hardness, toughness. However, materials science teaches us that macroscopic material behavior – taking place on much larger length and time scales – is usually dominated by the interaction and competition of many mechanisms rather than by the occurrence of a singular one. Hence, one primary focus of high-performance computing in materials science is to develop methods and algorithms that allow us to upscale atomistic methods such that they can unravel the complex interplay and competition of different fundamental mechanisms occurring during deformation of materials. In different examples it will be demonstrated that such upscaling requires a change of paradigm in the traditional workflow of simulations, because postprocessing of ultra-large atomistic samples of hundreds of millions of atoms becomes virtually impossible. Instead, on-the-fly-analysis tools need to be developed that allow the analysis of the physical properties of atomistic samples during the simulation. Apart from these atomistic methods, physical descriptions of material behavior inevitably lead to the mathematical problem of solving partial differential equations for large systems. Examples will be given the show how current research in this field points into the direction of developing new efficient solvers for those mathematical problems such that full use can be made of the novel architectures of high-performance computing systems like JUQUEEN.

« back