Uncertainty estimation and a stopping rule in nonlinear gyrokinetic simulations

Joint Varenna-Lausanne International Workshop on the Theory of Fusion Plasmas 2016

Oberparleiter, Michael^1 ; Nordman, Hans^1 ; Verdoolaege, Geert^2 ; Jenko, Frank^3

Department of Earth and Space Sciences, Chalmers University of Technology, Gothenburg

SE-412 96, Sweden^1

Research Unit Nuclear Fusion, Department of Applied Physics, Ghent University, Sint-Pietersnieuwstraat 41, Ghent

B-9000, Belgium^2

Department of Physics and Astronomy, University of California, Los Angeles

CA

90095, United States^3

关键词: Autocorrelation time;    Gyrokinetic simulations;    Nonlinear simulations;    Relative uncertainty;    Statistical treatment;    Temperature fluctuation;    Uncertainty estimates;    Uncertainty estimation;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/775/1/012009/pdf
DOI  :  10.1088/1742-6596/775/1/012009

来源: IOP

【 摘 要 】

We present a method to estimate the mean and uncertainty of fluctuating quantities, such as spatially averaged density and temperature fluctuations or radial fluxes, from initial value simulations of the Eulerian gyrokinetic code GENE[1, 2]. Since the time series are autocorrelated in time, the data is grouped into batches based on the autocorrelation time and their means form the sample for further statistical treatment, such as calculating the standard error of the mean.Based on this uncertainty estimate we develop a stopping rule for a nonlinear simulation: First, regression tests ensure that it has reached a stationary (quasisteady) state and data before this point is discarded. Then the previously described estimate is calculated. If the estimated relative error is below a prescribed threshold, the simulation is stopped. This scheme is applied to several previously performed GENE simulations ranging from simple benchmarks to modelling of JET and ASDEX discharges. It can be demonstrated that a number of simulations could be around 30% shorter if a maximal statistical relative uncertainty of 5% is desired for all monitored quantities.