Validation LAUR 11 04905 Bruce Fryxell Center for Radiative Shock Hydrodynamics Fall 2011 Review Code comparison collaboration includes researchers from three institutions CRASH University of Michigan ID: 310516
Download Presentation The PPT/PDF document "Code Comparison and" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Code Comparison and ValidationLA-UR 11-04905Bruce Fryxell
Center for Radiative Shock Hydrodynamics
Fall 2011 ReviewSlide2
Code comparison collaboration includes researchers from three institutions
CRASH – University of Michigan
Bruce Fryxell, Eric Myra
Flash
Center – University of Chicago
Milad
Fatenejad
, Don Lamb, Carlo
Grazianni
Los
Alamos National Laboratory
Chris Fryer, John
WohlbierSlide3
The CRASH problem has inspired this collaboration
When output
from
H2D
at 1.1 ns is used as the initial conditions for
CRASH, the primary shock is not planar, but shows a large protruding feature at the center of the tubeWall shock appears similar to that seen in experimentsSlide4
We are comparing several HEDP codes
Codes
currently in the test suite
CRASH (University of Michigan)
FLASH (University of Chicago)
RAGE, CASSIO
(LANL)
HYDRA
(LLNL)
Our goal
is to understand
differences between results of the CRASH experiment and simulations
This will be accomplished by comparing the codes on a wide range of problems, from simple tests to full HEDP experimentsSlide5
The codes in the test suite cover a range of numerical algorithms and physics models
Grid
CRASH –
Eulerian
AMR, block structured
FLASH – Eulerian AMR, block structuredRAGE/CASSIO – Eulerian AMR, cell-by-cell refinementHYDRA – ALE (Arbitrary Lagrangian-Eulerian
)
Hydrodynamics
CRASH – Second-order Godunov, dimensionally
unsplit
FLASH – Piecewise-Parabolic Method,
Strang
splitting
RAGE/CASSIO – Second-order Godunov
HYDRA – Lagrangian with remapSlide6
Treatment of material interfaces differs significantly between the codes
CRASH
Level set method
– no
mixed cells
FLASH Separate advection equation for each speciesInterface
steepener
- consistent mass advection algorithm
Opacities in mixed cells weighted by number density
Common
T
i
in each cell used to compute other quantities
RAGE/CASSIO
Interface preserver or volume of fluid
Opacities in mixed cells weighted by number density
EOS in mixed cells assume temperature and pressure equilibration
HYDRA
No mixed cells in Lagrangian modeSlide7
Both radiative diffusion and transport are represented in the test suite
Radiative Transfer
CRASH / FLASH / RAGE
Multigroup
flux-limited
diffusionEmission term treated explicitly (implicitly in CRASH)Equations for electron energy and each radiation group advanced separately
CRASH includes frequency advection
RAGE uses implicit gray calculation for radiation/plasma energy exchange
CASSIO
Implicit Monte Carlo
HYDRA
Multigroup
flux-limited diffusion
Emission term treated implicitly
Equations for electron energy and each radiation group advanced simultaneously
Implicit Monte Carlo (not yet exercised for this study
)Slide8
A variety of three-temperature methods and drive sources are included
Three-temperature
approach
CRASH
/ FLASH / RAGE / CASSIO
Compression/shock heating divided among ions, electron, and radiation in proportion to pressure ratios
FLASH has option to solve separate electron entropy equation to apply shock heating only to ions
HYDRA
Only ions are shock heated by adding an artificial viscous pressure to the ion pressure
Drive source
CRASH – Laser drive from Hyades, X-ray drive, laser package
FLASH – X-ray drive, laser package under development
RAGE – X-ray drive, laser package under development
CASSIO – Mono-energetic photons
HYDRA – Single-beam laserSlide9
First code comparison attempt was the “1d shifted
problem”
One
-dimensional version of the CRASH problem shifted into a frame of reference in which the Be
disk is stationarySlide10
The first attempt showed significant differences in shock structure between RAGE and FLASHSlide11
Results on 1D shifted problem have led us to consider a suite of simpler testsTemperature relaxation tests
Diffusion tests
Conduction
Radiative diffusion
Hydrodynamic tests
These tests are still in progress – some tests have been completed with only a subset of the code suite, while others have not yet been attempted with any of the codesSlide12
Temperature relaxation testsInitial conditionsInfinite Medium – no spatial gradients
Ion, electron, and radiation temperatures initialized to different values
Fully ionized helium plasma with density 0.0065
gm
/cm
3Gamma-law EOSIndividual testsIon/Electron equilibration
Ion/Electron equilibration + radiation
Constant opacity
Electron-temperature-dependent opacity
Energy-group-dependent opacity
4 groups or 8 groups
Constant (but different) opacity in each groupSlide13
CRASH, FLASH and RAGE give identical results for the simplest relaxation problems
Ion-electron
equilibration
Ion-electron-radiation equilibrationSlide14
RAGE and FLASH show differences in multigroup
tests
8 energy
groups – constant but different opacity in each group
Significant
differences in energy density in each group S
maller differences
in
temperatures
Differences
not yet
understood
Comparison with future CRASH results may help track down differencesSlide15
Diffusion testsElectron conduction
Electron conduction + ion/electron equilibration
Gray radiation diffusion
Electron conduction + ion/electron equilibration + gray radiation diffusion
Electron conduction + ion/electron equilibration +
multigroup
radiation diffusion
Tests run with and without flux limitersSlide16
Electron conduction test led to discovery of bug in FLASH
Initial temperature profile
Before bug fix in FLASH
After bug
fix in FLASH
t = 1.5 ns
t = 1.5 nsSlide17
Codes agree on diffusion tests 2) and 3)
Conduction + ion/electron coupling
Gray radiation diffusion
All three codes give identical results
t = 1.5 ns
t = 2.e-5 nsSlide18
Codes still agree with “full physics”
Gray diffusion, emission/absorption, electron conduction, electron/ion coupling
t = 0.2 nsSlide19
Hydrodynamics tests – not yet completedHydrodynamics (shifted 1d simulations)
Hydro + ion/electron equilibration
Hydro + electron conduction
Hydro + radiation diffusion + electron conductionSlide20
We have learned a great deal from these simple test problemsAs
a result of these tests we were able to
Understand some of the differences in the codes more clearly
Find bugs in codes
Improve the physics models within the codes
Test physics that is difficult to verify using analytic solutionsUnderstand time step size requirements for each type of physicsSlide21
Xe opacity comparisons
Data plotted for a single matter temperature and density relevant to the CRASH experiment
Relevant photon energies are those below ~300
eV
.
T = 50
eV
,
r=0.011
gm
/cm
3Slide22
Magnified view of relevant region
T = 50
eV
,
r=0.011
gm
/cm
3Slide23
Shock morphology is sensitive to Xe opacity
Simulations used SESAME gray opacities
Xe
opacities multiplied by constant scale factor of 1, 10, and
100
For future studies, different scale factors
may
be
used for each energy groupSlide24
More complex comparisonsTwo-dimensional shifted simulations with X-ray drive
Two-dimensional simulations of full CRASH experiment with X-ray drive
Two-dimensional simulations of full CRASH experiment with input from
H2D
with laser drive
Two-dimensional simulations of full CRASH experiment with self-contained laser driveSlide25
Tuning CRASH with X-ray drive caneliminate axis feature
These two simulations are identical except for the temperature of the X-ray driveSlide26
Initial untuned FLASH simulation with X-ray drive produces the anomalous axis feature
Initiated with mono-energetic X-ray drive
Time = 6
nsSlide27
Low grid resolution can producemisleading results
CASSIO initiated
with X-ray drive (mono-energetic photons)
No
protruding axis
feature at low resolution
CASSIOSlide28
High-resolution untuned CASSIO simulation
with IMC transport produces axis feature
Initiated with X-ray drive (mono-energetic photons)
time = 15 ns
High resolution – 1.5 micron
Protruding feature on axis is
presentSlide29
Low resolution HYDRA simulation with laser drive produces a small axis feature
30 ns
Higher
resolution simulation is
needed
before definitive conclusion can be reached about the axis featureSlide30
CRASH hydrodynamic validation studyJacobs’ Richtmyer-Meshkov instability experiment
Instability generated by shock impulsively accelerating an interface between two materials
Sinusoidal perturbation of interface – amplitude grows in time
Performed in vertical shock tube
Materials used were air and SF
6 (density ratio ~ 1:5)Shock Mach number = 1.21Shock reflects from end of tube and re-shocks the interfaceSlide31
Results at 6 ms (before re-shock)
128 grid points per wavelength
256 grid points per wavelength
Experiment
Experiment shows more roll up than simulationsSlide32
Growth rate agrees well with experiment
Re-shockSlide33
SummaryDetailed comparisons of five HEDP codes have begun
Good agreement on many test problems
Discrepancies still exist for some simple test problems
Comparisons have already led to the discovery of a number of bugs and code improvements
Non-planar primary shock has been seen in simulations of the CRASH experiment at high resolution using four of the codes in the test
suiteValidation simulations of Richtmyer-Meshkov
instabilities produced good agreement with Jacobs’ experiments – especially before re-shock