b'Automated DependencyNewly developed techniques improve the speed and efficiency of high-fidelity Graph Partitioning innuclear reactor simulations.Multiphysics Object- H igh-fidelity nuclear reactor simulation is critical to ensuring the continued operation of the existing nuclear reactor fleet and the development of Oriented Simulationnext-generation reactors. These simulations are incredibly complex, EnvironmentFramework combining neutronics, heat conduction, solid mechanics, fluid flow, and material science. The MOOSE framework developed by INL has made solving these complex problems possible. This research investigated and developed techniques that can significantly improve the efficiency/speed of existing MOOSE simulations, enable MOOSE to solve problems with greater complexity, and reduce the cognitive overhead and maintenance burden on MOOSE developers and users.TOTAL APPROVED AMOUNT:One significant efficiency issue relates to how FEM calculations are performed. These $125,000 over 1 year calculations use a mesh of geometric-shaped elements to represent the physical PROJECT NUMBER:geometry. Performing simulations requires calculating an intricate sequence of 20A1054-012 quantities associated with mesh locations that all depend on each other. Because many of these calculations can depend on each other, they are currently computed PRINCIPAL INVESTIGATOR:using a set of hard-coded loops over the mesh. For many MOOSE-based applications, Robert Carlsen this results in unnecessary looping over the mesh with each loop requiring CO-INVESTIGATORS: many redundant calculations to be performed. To overcome this inefficiency and Andrew Slaughter, INL other related limitations, a system of annotated dependencies for each of these Logan Harbour, INL calculations and an associated partitioning algorithm for automatically executing these calculations were developed and studied. No other FEM or multiphysics codes currently provide similar high-level automated mesh loop generation capability.Minimizing the total number of mesh loops while honoring all constraints imposed by the needs of each calculation is challenging. Exhaustively checking all possible ways to combine the calculations into loops scales with exponential complexity and is infeasible. While correctly grouping the calculations into loops is relatively easy, doing this in an optimal way is extremely difficult. This problem is effectively a graph partitioning and manipulation problem. Discovering an effective heuristic to maximally combine an arbitrary graph partitioning turned out to be the crux of this research. A heuristic was developed that generates an irreducible set of graph partitionsno further partitions can be combined without first splitting one or more partitions.Conflicting consolidation: When assigning calculations to mesh loops, some combinations are suboptimal because they result in a larger total number of groups. Consider the case of four aggregation calculations: C depends on A, and D depends on B. Combining A and D prevents any further consolidation (a). It would prevent combining C and B (shown) as well as other combinations, such as A and B. However, combining A and B still allows C and D to be combined (b). Much more complicated versions of this problem must be handled by the algorithm. Understanding and solving this phenomenon in all its complex forms was the crux of the research.47'