MICDE Seminar: Jaroslaw Knap, U.S. Army Research Laboratory, “Distributed Multi-scale Computing for Materials Modeling and Discovery” — March 15

Over the last few decades, multi-scale modeling has become a dominant paradigm in materials modeling. The practical impact of multi-scale modeling depends, to a great extent, on its ability to utilize modern computing platforms. The efforts focused on efficient utilization of the, often heterogeneous, distributed computing infrastructure has led to the emergence of distributed multi-scale computing. In distributed multi-scale computing, a multi-scale computer model is assembled from multiple interacting symponents representing individual scales in the model hierarchy. Each symponent may be endowed with a level of concurrency or be entirely serial. Also, symponents may be created and destroyed dynamically and communicate without predictable communication patterns. Today, most multi-scale computer models of materials are still developed on a case-by-case basis offering little to no reuse. We seek to formulate an adaptive computational framework for distributed multi-scale computing. Our focus is primarily on scalable algorithms applicable to a wide range of multi-scale modeling applications. We present a formulation of our distributed multi-scale computing framework. Subsequently, we describe a number of applications of the framework to modeling of materials, ranging from composites, bio-mechanics to energetic materials and electro-chemistry.