SI2-SSE: A Software Element for Neutrino Radiation Hydrodynamics in GenASiS

Multi-physics computational modeling is an integral part of the scientific study of many complex natural phenomena. These phenomena often involve the physics of radiation transport. For example, neutrino radiation hydrodynamics is a key element of the physics governing environments with hot and dense nuclear matter. Such extreme environments include the Early Universe, during primordial nucleosynthesis of light nuclei such as hydrogen through lithium just after the Big Bang; the merger through inspiraling of neutron star-neutron star or neutron star-black hole binaries; and the death throes of massive stars, more than ten times the mass of the Sun, in stellar explosions known as core-collapse supernovae, which are responsible for elements such as oxygen and calcium without which life as we know it would not exist. Radiation transport and kinetic theory of particles besides neutrinos---photons, electrons, or neutrons---are also relevant to many areas of astrophysics, as well as a broad range of other science applications, including materials science, plasma physics, neutron transport, multiphase flows, and high-energy-density physics. As such, the availability of a software element to solve radiation transport problems is highly valuable to researchers.

This project will create and deploy a software element to solve radiation hydrodynamics problems on modern supercomputers featuring 'hybrid' architectures that include traditional CPUs plus 'accelerators' or 'coprocessors,' such as GPUs or Intel Many Integrated Core processors, respectively. This radiation hydrodynamics functionality will be developed within GenASiS (General Astrophysical Simulation System), a new framework being developed to facilitate the simulation of astrophysical phenomena on the world's leading capability supercomputers. In particular, the radiation transport solver will utilize the extant capabilities of GenASiS for adaptive computational 'mesh refinement,' whereby the representation of the natural continuum is captured adaptively on a mesh of points foundational to any computational model in order to maximize the fidelity of the computational model for a given computational cost. We will use the so-called M1 approach, solving directly for the zeroth and first angular moments (energy density and momentum) of the radiation field, with higher-order moments given by 'closure relations,' expressing them in terms of the zeroth and first moments. The energy dependence of the radiation field will be retained, with the zeroth and first angular moments discretized into 'energy bins.' Our computational approach to neutrino radiation transport will be an 'implicit-explicit' (IMEX) scheme. Interactions between radiation and matter will be handled with a time-implicit subsolver, which will involve the inversion of dense matrices local to each node of the machine to exploit all available hardware in the node, including accelerators and coprocessors when available. Algorithms and software resulting from this project will be made available to the community. GenASiS, as an extensible, object-oriented simulation framework, will be valuable to researchers seeking to experiment with and implement different kinds of solvers for multi-physics problems. In particular, the neutrino hydrodynamics solver developed in this project is of high interest to astrophysics modelers.