Simulations of the Experiments Ken Powell CRASH Review October, 2010
Dec 20, 2015
CRASH PreprocessorHyades is a Lagrangian rad-hydro code
that can model laser-plasma interactions
Used in the early stage (first 1.1 ns) of the simulations
Map Hyades Lagrangian result to CRASH Eulerian grid, via triangulation and interpolation
Ongoing work to build our own laser package (see Igor Sokolov’s talk)
Have also experimented with X-ray-driven initialization by CRASH or Hyades (See Eric Myra’s and Erica Rutter’s posters)
CRASH Radhydro Code: Hydro and Electron Physics
radiation/electron momentum exchange
radiation/electron energy exchange
electron heat conduction
Compression work collisional exchange
CRASH Radhydro Code: Multigroup diffusion
Radiation transport equation reduces to a system of equations for spectral energy density of groups.
Diffusion is flux-limited
For the gth group:advection compression work photon energy shift
Overview of Solver Approach
Self-similar block-based adaptive grid
Finite-volume scheme, approximate Riemann solver for flux function, limited linear interpolation
Level-set equations used to evolve material interfaces; each cell treated as single-material cell
Mixed Implicit/Explicit updateo Hydro and electron equations
Advection, compression and pressure force updated explicitly Exchange terms and electron heat conduction treated implicitly
o Radtran Advection of radiation energy, compression work and photon shift are evaluated
explicitly Diffusion and emission-absorption are evaluated implicitly
o Implicit scheme is a block-ILU-preconditioned Newton-Krylov-Schwarz scheme
CRASH Postprocessor
Synthetic radiographs generated by integrating absorption coefficients along lines of sight
Poisson noise is added to simulate finite photon count
Smoothing is done at scale associated with finite aperture in experiment
Tests included in verification suite – grid-convergence studies on problems with analytical solutions
Improvements to fidelity/efficiency finished this year
Electron/radiation physicso Flux limiting added - limit Spitzer-Harm flux by fraction of free-streaming heat fluxo Update based on total energy, but slope limiter applied on primitive variables
EOS and opacity calculationso Five material (Xe, Be, Au, acrylic, polyimide) EOS and opacity tables in placeo EOS tables made reversible (E→p→E or p→E→p puts you back where you
started)
Efficiency improvementso New block-adaptive-tree library (BATL); Efficient dynamic AMR in 1, 2 and 3Do Semi-implicit scheme, split by energy group
Requires less memory and CPU. Allows PCG.
Synthetic radiographs with blurringo Add Poisson noise due to finite photon count.o Smooth at the scale that corresponds to the pinhole size.
Pure Hydro Results3 geometries
o Straight tube (1200 μm diameter)o Step (1200 μm → 600 μm)o Nozzle (1200 μm → 600 μm)
250 μm Be disk, low laser energy
Shock speed ~ 20 km/s
Highest 3D resolution to dateo 2 μm spacingo 2400 x 480 x 480 uniform grido 550 million cells
Pure Hydro Results – Density Contours
Nozzle – Vertical cut
Nozzle – Horizontal cut
Step – Vertical cut
Step – Horizontal cut
Full Physics Results2 geometries
o 2D Straight tube (600 μm)o 3D Nozzle (1200 μm → 600 μm)
20 μm Be disk, nominal laser energy (3.8 kJ for 1 ns)
Shock speed ~ 160 km/s
Electron physics, five materials, 30 energy groups
Varying resolutionso 2D - 2 μm effective (1 AMR level)o 2D - 0.5 μm effective (3 AMR levels)o 3D - 4 μm effective (1 AMR level, 5 million cells)
The morphology conundrum persists independent of:
Mesh resolution (except on very coarse grids)
Flux function, limiter
Gray vs multigroup/number of groups
Treatment of electron physics
Number of materials used
Presence or absence of a symmetry axis
We CAN make a primary shock with realistic structure with different initial conditions (X-ray-driven) running CRASH alone
But it is hard to get the primary shock and the wall ablation to simultaneously match the experimental
result…
… and we get different results when initializing the same case using Hyades
Hyades-drivenX-Ray case
CRASH-drivenX-Ray case
The path aheadWe are further pursuing the X-ray-driven case,
comparing Hyades and CRASH to understand how the differences arise
We are developing a laser package, so we have an alternative preprocessor, one whose internal working we understand/have control over
We are working to improve the preconditioning of the implicit solve, to cut down the compute time (approximately 90% of compute time is spent here)