Transport Physics and UQ Marvin L. Adams Texas A&M University CRASH Annual Review Ann Arbor, MI October 28-29, 2010
The integrated team has produced significant results this year. Collaboration has been fruitful and essential. o UQ is a tightly integrated UM/TAMU/SFU effort. o Theoretical understanding has advanced via collaboration (UM/TAMU). o Radiation has been an integrated UM/TAMU effort; this continues on more fronts as PDT & CRASH-MG mature. This talk describes recent TAMU contributions. o Includes UQ, Radiation, and theory. o Much involves collaboration with UM and/or SFU; the remainder is part of integrated CRASH plan. We are an integral part of the team.
Radiation effort is challenged by prohibition on coupling. Key task is assessment of diffusion model error. o diffusion model error ≈ [hi-res transport] – [hi-res diffusion] o Must translate no-hydro results to rad-hydro problem o Must address diffusion discretization error High-res transport tool is PDT (TAMU code). We employ a no-hydro “CRASH-like” test problem. We have developed a technique for using a transport code (e.g., PDT) to help assess diffusion model error.
CRASH-like test problem helps us assess model & discretization errors Constant energy deposition to electrons at “shock” Can assess effects of o discretization in energy, direction, space, and time o transport vs. diffusion Current focus is on ablation layer in plastic o See Morel’s talk 4 mm.3125 mm Be g/cc Au 19.3 g/cc Xe g/cc Xe 0.1 g/cc Xe g/cc plastic 1.43 g/cc electron energy source
PDT now solves CRASH-relevant problems. Continually adding verification tests (McClarren poster) Performance improvements have enabled solution of relevant problems o 40x serial speedup o 67% efficiency on 12k cores o Team effort (NE+CPSE at A&M) o see poster (“Massively Parallel...) There have been many other improvements o electron-energy sources, flexible initial and boundary conditions, CRASH opacities, better parallel I/O, improved visualization, diffusion preconditioner (debug phase), improved spatial discretizations, improved quadrature sets, etc.
PDT can produce high-resolution transport results for this problem. Example: o 50 energy groups, S18 quadrature (360 directions), 128 cells in first 0.5 micron of plastic (!), fully implicit solution o Weekend run, 1024 cores We are confident that we can assess discretization error and diffusion-model error for this problem o See Morel’s talk
We’ve developed and applied advanced BMARS to CRASH UQ Recent BMARS progress o Improved BMARS code, comparison with GP (see posters, papers, Stripling thesis) o Applied to H2D shock breakout (calibrated flux limiter, wall opacity, and Be EOS) o Built H1D emulator using BMARS and GP (paper accepted) o Contributors included Mallick, McClarren, Stripling, Ryu, Bingham, Holloway, and others from UM See poster on calibration of H2D parameters for shock breakout (Stripling, et al.)
We have developed and disseminated new theoretical results Theory of thin/thick radiating shocks o Physics of Plasmas, McClarren/Drake/Morel/Holloway Verification solutions o JQSRT, McClarren/Wohlbier o Also see McClarren poster Diffusion model error in radiating shock o JQSRT, Drake/McClarren o Also see Morel talk
We are improving discretization, iteration, parallel, and UQ methods Assessment of diffusion model and numerical errors: underway; high priority in coming year New STAPL: PDT transition has begun Positive spatial discretization: Maginot, et al. poster Long characteristics spatial discretization: Pandya, et al. poster Provably optimal sweep schedules: Adams, et al. poster Diffusion preconditioners for DFEM transport: in progress Uncertainties from uncertain opacities: dimension-reduction effort in progress
Next year should see further significant advances PDT will become a more capable CRASH tool o more efficient temperature iterations, including use of diffusion preconditioner o RZ geometry; space-time characteristics; DG diffusion o more flexible source and boundary conditions (for verification tests) o Must scale well on BG/L We will continue to advance UQ methods o include uncertainties in “x” inputs o improve emulator o assess model and discretization error o compensate for model error? We will continue theoretical developments o more verification problems, including analytic 3T solutions o track-based sweeps o analyses of iteration and time-differencing methods
Questions?