Bio: Josh Dolence is a scientist in the Computational Physics & Methods Group at Los Alamos National Laboratory and a founding member of the LANL Michigan SPARC, a permanent LANL presence at U-M in Ann Arbor. Before joining LANL, he received a PhD in Astronomy from UIUC in 2011 and spent three years in Astrophysical Sciences at Princeton University where he worked in computational astrophysics, studying topics like black hole accretion and supernovae. More recently, he leads the Methods for Multiscale, Multiphysics Accelerated Prediction project for LANL’s Advanced Simulation and Computing Program, focusing efforts on enabling unprecedented fidelity and scale in modeling complex systems like high energy density physics experiments.

Abstract: In many areas of computational science, developing new, state-of-the-art capabilities has become a high-cost, risky proposition. The complexity and diversity of models, methods, algorithms, and machines often lead to fundamental challenges in designing and building codes that enable advances in science and engineering. In fields like high energy density physics and astrophysics, multi-physics simulations leveraging adaptive meshes, particles, and a variety of numerical methods are foundational to progress but difficult to realize performantly on ever-evolving high-performance computing platforms. In this talk, I will present the Parthenon framework, an open-source code base that aims to facilitate the development of highly adaptive multi-physics codes that are fast, scalable, and capable of leveraging modern platforms with both CPUs and GPUs. I will describe the basic principles behind its design and some of its most enabling features and highlight the ~10 downstream codes it already supports.