Center for Turbulence Research Annual Research Briefs 2016 259 A framework for estimating epistemic uncertainty in LES closures By L. Jofre, S. P. Domino† AND G. Iaccarino 1. Motivation and objectives Over the past decade, large-eddy simulation (LES) has gained significant importance as a high-fidelity reference technique for the numerical resolution of turbulent flow. One of the main reasons is the tremendous growth in available computational power, which has made its superior accuracy attractive with respect to other cost-effective methods like the Reynolds-averaged Navier-Stokes (RANS) equations. Moreover, despite the presum- ably further increase in computing resources through the deployment of upcoming exas- cale supercomputers — 1-10k times augmented floating-point capacity is foreseen (DoE 2012) —, the expectation in the computational community is that LES will continue its consolidation as a workhorse methodology for engineering applications and multiscale problems, whereas direct numerical simulation (DNS) will remain as the gold standard technique, affordable only in very expensive scientific studies. In comparison to DNS, LES approaches reduce the computational cost of solving turbulent flow by removing small-scale information from the governing equations via low-pass filtering. However, the effects of the small scales on the resolved flow field are not negligible, and therefore their contribution in the form of subfilter stresses needs to be modeled. As a consequence, the assumptions introduced in the closure formulations result in potential sources of structural uncertainty that can affect the quantities of interest (QoI). Hence, it is of remarkable utility the development of a framework capable to effectively estimate these effects on complex scenarios. Even with the widespread utilization of LES in many scientific and technological areas, there have been few studies in which model-form incertitude has been analyzed from an uncertainty quantification (UQ) viewpoint. In general, most are based on non-intrusive methodologies applied to simple flow configurations, and are concerned mainly with sen- sitivities to LES closure parameters (Lucor et al. 2007), such as model coefficients (Meldi et al. 2011), filter characteristics (Meyers & Sagaut 2007a ) or mesh resolution (Meyers & Sagaut 2007b ). This type of analyses, although useful from the practitioner’s perspective, present important impediments to generalization due to their dependency on the under- lying structure of the models utilized. In order to overcome this limitation, this work aims to develop a framework for the estimation of structural uncertainty in LES closures that is independent of the initial model form. The strategy feeds from the methodology pre- viously introduced in the context of RANS approaches (Gorl´ e & Iaccarino 2013; Emory et al. 2013), although there are important differences due to the inherent distinction between the two turbulence-resolution techniques. In short, the framework is based on introducing perturbations to the modeled turbulent stress tensor. These correspond to discrepancy in the magnitude (trace), shape (eigenvalues) and orientation (eigenvectors) of the normalized subfilter stresses with respect to a given tensor state. † Sandia National Laboratories