This presentation begins with an overview of computational methods that we have developed for analysis and post-processing of fully resolved atomistic simulations of materials. Our goal? To transform the results of large-scale MD simulations into higher-scale descriptions that are more meaningful and scientifically useful.
We'll explore some of the longstanding and more recent examples of our work, including the Dislocation Extraction Algorithm (DXA), grain segmentation, and identification and tracking of surfaces and pores in materials. These complex algorithms are easily usable for everyone thanks to the integration in our software OVITO. But they suffer from practical limitations when it comes to ultra-large simulations and parallel scalability. A gap has emerged between the growing size of simulation models and the researchers’ ability to adequately process the computed trajectories.
We'll delve into the reasons behind these limitations and present new ideas for developing massively parallel methods that can upscale atomistic models into coarser, high-level data representations.
In short, we'll address an important question of exascale materials modeling: Why is it so simple to simulate microstructural features using molecular dynamics, but so difficult to analyze these simulations and extract meaningful outputs?
Copyright. All Rights Reserved.