Advancing Hybrid Fluid-Kinetic Methods for Plasma Simulation

The complex, multiscale nature of plasmas necessitated the early adoption of computers by plasma physicists to model their behavior more than half a century ago.Their limited computational resources demanded fast, efficient algorithms for simulating plasmas.In general, these algorithms fall into two categories: kinetic and fluid.Kinetic algorithms retain much detailed information about how velocities are distributed in particles of the plasma, and as a consequence, they are very widely applicable but computationally expensive.In contrast, fluid algorithms discard much of this information to reduce computational cost, but must make assumptions about the plasma to compensate.Such algorithms have kept pace with developments in computing technology and are still in wide use today, owing to their inherent parallelizability.However, as high-performance computing moves into the exascale era, plasma physicists have become inspired to attempt increasingly large simulations, as evidenced by the trend toward whole-device modeling in the study of magnetic confinement fusion.The widening of the computational bottleneck has exposed flaws in the traditional paradigm of using either purely kinetic or purely fluid algorithms.It is now clear that a hybrid approach, combining the strengths of both kinetic and fluid methods, is necessary to continue efficiently modeling plasmas at the largest desired scales.Although the study of hybrid techniques is not new, the problems associated with kinetic or fluid simulations of large systems all but necessitates their accelerated development.To that end, this work is broadly concerned with the construction and implementation of hybrid methods for plasma simulation on modern high-performance computing systems.A framework for the hybridization of kinetic and fluid methods is presented.This framework was then developed into the hPIC2 plasma simulation code, which overcomes the difficulty of targeting many computing architectures through the use of the Kokkos performance portability library.hPIC2 incorporates a number of innovations deemed indispensable for large-scale plasma simulation, including a novel fault tolerance scheme and advanced finite element methods furnished by the MFEM library.This work also explores conversions between fluid and kinetic descriptions.A method was developed that identifies when a kinetically modeled portion of a plasma satisfies the fluid assumptions and dynamically converts it into a fluid, which is of particular utility in highly collisional hybrid plasma simulations.The algorithms and code presented here permit the simultaneous use of fluid and kinetic descriptions for the plasma in a single simulation and adaptive conversion between them.They have been designed for and tested on architectures composing the largest supercomputers currently available.The result of this work is a significant step toward the realization of plasma simulations spanning many scales.This is particularly important for pulsed-power physics and magnetic confinement fusion, whose devices are large and complex and stress the limits of the traditional, non-hybrid approach.This work would not have been possible without the love and support of my family.My parents, Jenn and Doug, always encouraged me to pursue a career in science, and I will be forever thankful.You fostered curiosity and a love of learning that will never die in me.My breaks back at home in Colorado, long hikes in the Rockies, longer soaks in the hot tub, and the occasional brewery crawl all worked to restore me after long stints in Illinois.My sisters, Shay, Rachel, and Sarah, helped me to become the person I am today.Trips to see a car show and Rip and Bo in Laporte, or to work out and terrify Beef across the street, were welcome respites from the lab.Just kick it!I was lucky to have a close-knit group of friends during my time in Illinois.The qual-fueled bonfires, sea shanties on Lake Shelbyville, poker nights, and basement shows set a memorable and energizing backdrop to my years of study.My partner, Meirah, was a supporting and calming presence in the most stressful times of my graduate studies.I am not sure why you waited six months for me while I was in New Mexico, but I am supremely thankful that you did.I am extremely grateful to my advisor, Davide Curreli, for his guidance and mentorship during my graduate studies.Our impromptu brainstorming sessions on the blackboard inspired me to take risks in science, most of which paid off in the end.Your leadership was a vital component in the success of this work, and I hope to emulate it during my career.The Laboratory of Computational Plasma Physics has also been an extremely helpful resource for collaboration, discussion, and friendship.Special thanks go to Huq, for enthusiastically adopting hPIC2 as your primary code, even when it was beset by bugs; Mohammad, for helping with early verification and bringing hPIC2