Reliable quantification of uncertainty in time averages of turbulence simulations
Time: Thu 2022-12-15 10.30 - 11.30
Location: Faxén, Teknikringen 8
Participating: Donnatella Xavier
Abstract: Time averages computed from simulations of statistically stationary turbulence can contain temporal uncertainties due to lack of ergodicity, initialization bias and sampling error. The statistical tools cited in literature typically used to quantify these uncertainties contain hyperparameters that have not been analyzed for turbulent flows. In this talk, I will discuss my work on these uncertainty quantification tools and analysis of their hyperparameters, so that they can be used to accurately estimate the variance in the sample means and higher order moments of any given time series of a turbulent flow variable. I will focus on the autoregressive model and batch means estimators for estimating the uncertainty in the time averages and how their hyperparameters are easily understood from the autocorrelation of the time series. I will also present a method for the automatic detection of the initial transient and a computationally inexpensive way to use the autoregressive model that applies to statistically stationary turbulent flow.