Quantifying Uncertainties in Radiative Shock Experiments
Carolyn C. Kuranz
CRASH Annual Review
Fall 2010
Why is it important to understand experimental uncertainty for this project?
Creates realistic input parameter space for predictive studies
Understanding dominant sources of uncertainty can help us to focus on those areas to reduce the uncertainty
Helps us to understand and improve the predictive capability of the model Important for future experiments
Partial list of experimental inputs that have uncertainty associated with them
Laser energy
Laser pulsewidth
Laser spot size
Observation time
Be disk thickness
Be surface roughness
Xe gas pressure
Diagnostic x-ray signalBackground signalSource broadening
Target geometryAngle between Be disk and tubeAngle between tube and diagnostic
Pre and post-shot probability distributions functions (PDFs) for these
uncertainties often differ!
Summary of the CRASH calculation
X - Experiment parametersθ - Physical ConstantsN - Numerical ParametersYS - Results to be analyzed with data by statistical methods
CRASHPre-Processor
XH
CalibrationData (D)
CRASHRadiation-Hydrodynamics
Simulation Code
XCθC
NC
YHP YC YS
CRASHPost-Processor
XRθR
I will be discussing the uncertainties in some of the experimental inputs
Types of PDFs for experimental inputs
Tails of PDFs are often complex (details and examples to follow)
Laser Energy is an example of quasi-Gaussian distribution
Mean values of experimental days are within 3% of nominal but standard deviation is ~1% or less on individual day
Be disk thickness is an example of a quasi-uniform distribution
Several parameters have a “uniform” distribution with low-amplitude, long tails
In this case, the tails of the distribution correspond to cases in which there is a malfunction of a simple measuring instrument or disregard of measuring procedures
Understanding experimental uncertainties is very complex: observation time in Y2 experiment
Recent experiments measured the amount of time it takes for the shock to move through the Be diskEach experiment used 3 instruments for the
measurement The most sensitive instrument had 10 ps resolution
Time
Spa
ce
VelocityInterferometer
}548 ps
Shock breakout
But these instrumental uncertainties were not the dominant uncertainty
-0.5 0.5 1.50
200000000000000
400000000000000
600000000000000
800000000000000
Time (ns)
Las
er I
rrad
ian
ce
(W/c
m2)
t0
Time
Spa
ceDiagnostic fiducial
Total uncertainty was ± 50 ps even though instrumental uncertainty was smaller
Largest uncertainty came from measuring time interval between the drive laser and diagnostic fiducial laser
}548 ps
Always look behind the curtain…
Often the analysis of experimental data focuses on the detail of these small error bars
The uncertainty in this measurement is dominated by a larger systematic error
ConclusionsUnderstanding and quantifying the uncertainties in our
experiments is complex and sometimes surprisingThere are 2 types of PDFs for these uncertainties: quasi-
Gaussian and quasi-uniformThe tails of these PDFs are often complex
The PDF for a given parameter can be different pre-shot and post-shot
We are continuing to work towards identifying and quantifying uncertainty in our experiments