Self-Adaptive, Unstructured Mesh, NURBS Enhanced, Polyhedral Schemes, with Hybrid Multicore CPU and Manycore GPU Solution Algorithms, for Nuclear Reac

点击翻译按钮获取中文摘要

(a) Development of computationally efficient, numerically accurate, spatially self-adaptive NURBS-enhanced VE methods in 2D and 3D. This will build upon prior R&D work performed by the nuclear engineering group at Imperial College London.(b) Development of mathematically rigorous, dual-weighted-residual, goal-based or goal-oriented error measures using the primal (forward) and adjoint solutions of the neutron transport equation with an associated prescribed functional. This will be applied to both eigenvalue (Keff of nuclear criticality) problems as well as fixed (or prescribed) neutron source problems. In this case typical "goals" will be global quantities of interest, such as the degree of criticality (or Keff), the local fission/capture reaction rate within prescribed regions of the computational domain, or local quantities such as scalar/angular neutron flux, neutron/gamma-ray photon fluence, radiation dose or radiation damage (displacement per atom or DPA) at prescribed points within the computational domain.(c) Development of computationally efficient, massively parallel, hybrid multicore CPU/manycore GPU discrete ordinate (SN) sweep based neutron transport solvers on NURBS-enhanced unstructured polygonal/polyhedral meshes. This will take advantage of hybrid OpenMP (shared), MPI (distributed) and CUDA/OpenCL (GPU) parallel software libraries. A key aspect will be investigating what specific aspect of the discrete ordinate (SN) sweep based neutron transport solver algorithm will be more optimal in terms of the use of the multicore CPU versus use of the manycore GPU and how data can be effectively passed between these two different types of compute units as efficiently as possible for sweep algorithms.(d) A final aspect of the project will be applying the methods developed to standard multidimensional and multiscale nuclear reactor physics and reactor shielding benchmark verification test cases. These benchmark verification test cases will be used to investigate the computational efficiency, scalability and numerical accuracy of the self-adaptive, massively parallel, NURBS-enhanced virtual element methods using hybrid multicore CPU and manycore GPU workstation and HPC systems. These will include the OECD/NEA 3D C5G7 UOX/MOX colour-set nuclear reactor physics benchmark verification test case and the extended quarter-core OECD/NEA KAIST Light Water Reactor (LWR) nuclear reactor physics benchmark test case. For reactor shielding the OECD/NEA SINBAD coupled neutron and gamma-ray photon transport ASPIS and Crank-duct benchmark verification test cases will be investigated.