P. F. Fischer and C. P. Tzanos, "Filtered Simulations of Turbulence in a Reactor Rod Bundle Flow," Preprint ANL/MCS-P1213-0105, January 2005. [pdf]
Although RANS (Reynolds averaging of the Navier-Stokes equations) turbulence models provide adequately accurate solutions for many engineering problems at a reasonable computational cost, there is no single RANS model of universal applicability. Moreover, in many important applications the RANS model predictions for flow distributions, heat transfer coefficients, and thermal mixing can be significantly off. Examples of such applications are flows in pressurized water reactor rod bundles with mixing vanes and flows in gas cooled reactors where buoyancy effects are significant. Because direct numerical simulation (DNS) of turbulence for large Reynolds numbers and for system-size scales of interest are not still computationally practical, LES (large eddy simulation) and coarse DNS models of turbulence, which are computationally less demanding than DNS, hold the promise to remedy the shortcomings of the RANS models in many important applications. LES is predicated on a scale separation mechanism, usually in the form of a filter, to isolate the resolved (simulated) scales from the subgrid scales (SGS). An essential part of LES is to account for the effects of the missing SGS terms; these typically appear as enhanced diffusion in the Navier-Stokes equations.