Flash Center for Computational Science University of Chicago of 16 ANL February 2011 Milad Fatenejad DOE NNSA ASCR Flash Center University of Chicago Progress on FLASH Target Application Simulations 1
Flash Center for Computational Science University of Chicago
of 16
ANL February 2011
Milad Fatenejad DOE NNSA ASCR Flash Center
University of Chicago
Progress on FLASH Target Application Simulations
1
Flash Center for Computational Science The University of Chicago
of 16
Table of Contents
Reasons for choosing the CRASH experiment as the target application
Progress on FLASH HEDP development Progress in understanding physics/data
requirements through a new collaboration
2
Flash Center for Computational Science The University of Chicago
of 16
The CRASH experiment was chosen as the first target application for several compelling reasons
Experiments have already been performed and data is available
The key validation features (shock position, structure) are robust and reproducible
It is possible to perform NIF experiments using this setup for further validation
Experiments exercise most of the basic capabilities needed for HEDP simulations
3
Flash Center for Computational Science The University of Chicago
of 16
Data is available from CRASH experiments which have features that have proven to be extremely robust
Radiographs from many CRASH experiments show a planar shock wave
Images from different times show that the shock persists for a long time
26 ns 14 ns
F. Doss, et al., HEDP 6, 157 (2010) Drake, et al., Conf. Proc. RPDHM2010, HEDP, submitted
4
Flash Center for Computational Science The University of Chicago
of 16
CRASH experiment simulations require modeling several types of important HEDP physics
The following physics is required to model the target application: 2T Ion/electron hydrodynamics with radiation transport
Thermal conduction
Laser energy deposition
Multiple materials (Be, Xe, Au, polyimide, acrylic)
EOS and opacities for these materials
Once these capabilities are implemented and the target application is satisfactorily simulated, sophisticated, inline, non-LTE opacity and EOS models will be implemented and tested
These capabilities combined with Exascale computing power will enable transformative HEDP simulations and science
5
Flash Center for Computational Science The University of Chicago
of 16
New physics and numerical methods are needed in FLASH to simulate the target application
1-T Hydrodynamics must be extended to support separate ion/electron temperatures with radiation transport Properly controlling the differential heating of ions/electrons near a
shock is an unsolved problem for Godunov style Eulerian fluids codes
Several hydrodynamic models are being incorporated into FLASH to determine the best approach
An implicit diffusion solver is necessary for modeling thermal conduction and radiative transfer with flux-limited multigroup diffusion (FLMGD) We plan to support several external libraries, including HYPRE and
PETSc
We are using HYPRE in implicit diffusion solvers for uniform grids and plan to use it for AMR
6
Flash Center for Computational Science The University of Chicago
of 16
New physics and numerical methods are needed in FLASH to simulate the target application
Laser ray-tracing is needed to drive the simulation Rapid progress has been made in this area because of the block-
structured mesh in FLASH and by leveraging the pre-existing particles capability
Opacities are needed for the multigroup diffusion calculation Preliminary work has focused on supporting tabulated opacities
Future work will focus on implementing sophisticated in-line non-LTE models; this approach will greatly benefit from Exascale capabilities
Sophisticated multi-material equation of state (EOS) models are needed in many HEDP simulations The multi-material infrastructure is undergoing development with
preliminary work focusing on tabulated equations of state
Future work targeting in-line EOS calculations will also benefit from Exascale capabilities.
7
Flash Center for Computational Science The University of Chicago
of 16
Implementation of HEDP capabilities is progressing rapidly
FLASH Capability Future Development
Ongoing Development
Completed and Under Testing
RAGE Hydro X
CRASH Hydro X
Entropy Advection X
HYPRE Diffusion X
Multigroup Radiation Diffusion X
Electron Conduction X
Laser Ray-Tracing X
Tabulated Multimaterial EOS X
Opacity X
Inline non-LTE Opacity X
Inline EOS X
8
Flash Center for Computational Science The University of Chicago
of 16
Verification with analytic solutions and code-to-code comparisons are continuously performed
Comparison of 2T shock structure between analytic Shafranov1 solution, a Lagrangian code solution, and FLASH – entropy advection show excellent agreement.
Comparison of 2T shock structure between FLASH using RAGE-Like hydro and the RAGE code itself show reasonable agreement. Further investigation is ongoing.
1V.D. Shafranov, Soviet Physics JETP 5, 1183 (1957)
9
Flash Center for Computational Science The University of Chicago
of 16
A new collaboration with labs/universities has led to further understanding the simulation requirements
Collaboration with national laboratories and universities is necessary for successfully simulating the target application
National laboratories: LLNL and LANL The national laboratories have extensive knowledge and experience in
running HEDP simulations They have provided useful data which has helped identify where better
treatments of the physics are needed (e.g. EOS and opacities)
FLASH/RAGE code-to-code comparisons are helping to improve both codes
Universities: FLASH Center, CRASH, University of Wisconsin-Madison The CRASH center has provided essential experimental/simulation data Code-to-code comparisons with the CRASH code are ongoing UW is helping develop the FLASH Center’s EOS/opacity capability
10
Flash Center for Computational Science The University of Chicago
of 16
The collaboration has already led to a deeper understanding of the physics/data needed to model the experiment
Previous HYDRA simulations1 performed by CRASH demonstrated that flux-limited multigroup diffusion (FLMGD) is sufficient for obtaining agreement with the experiment
Recent RAGE simulations have shown little qualitative difference between FLMGD and IMC
Consultation with LLNL has provided strong evidence that the EOS/opacity can be accurately described using an LTE model
1F. Doss, et al., HEDP 6, 157 (2010) 11
Flash Center for Computational Science The University of Chicago
of 16
Comparison with sophisticated LANL polyimide opacity has shown reasonable agreement
12
Flash Center for Computational Science The University of Chicago
of 16
Analysis of LANL data has shown that bound-bound transitions dominate in the critical frequency range
13
Flash Center for Computational Science The University of Chicago
of 16
Analysis of LANL data has shown that bound-bound transitions dominate in the critical frequency range
14
Flash Center for Computational Science The University of Chicago
of 16
Current university opacity models are unable to accurately model the bound-bound behavior
15
Flash Center for Computational Science The University of Chicago
of 16
The collaboration will continue to answer critical questions and improve the fidelity of the simulations
How sensitive is the solution to the Xenon and Polyimide opacity? How does the underlying hydrodynamic model (ALE vs. Eulerian)
affect the solution? Is laser ray-tracing necessary to model the experiment or can X-
ray drive suffice? How sensitive is the solution to zoning, refinement, and spatial
resolution Code-to-code comparisons with FLASH / RAGE / CRASH will
continue to improve all three codes
16