National Center for Computational Sciences

Nuclear technology is a necessary part of any U.S. energy security strategy. Progress in nuclear technology, on the other hand, requires simulations that model many coupled physics systems, including Boltzmann transport equations, a powerful algorithm for analyzing transport phenomena with many-step gradients in both density and temperature.

For nuclear reactor simulation, the size of the equations to be modeled is tremendous—five orders of magnitude in space and 10 in neutron energy. Tom Evans of ORNL and his collaborators are extending an existing parallel transport solver called Denovo, to develop a first-of-a-kind, mathematically consistent, two-level approach to the multiscale challenge . With present algorithms, a solver that incorporates the separating out of processes for all scales would require 1017 to 1021 degrees of freedom (DOF) for a single time-step, which is beyond even exascale computational resources. Evans’ three-dimensional code scales to the petaflop level by uncoupling the multi-level, phase-space parameters of the equations. Denovo is a significant advance over current technology, because it allows fully consistent multi-step approaches to high-fidelity nuclear reactor simulations that cannot be performed with current technology. The team will use 10 million processor hours on ORNL’s petascale Jaguar supercomputer to advance this next generation reactor transport solver.