Computational Fluid Dynamics (CFD) provides an approximation to the governing equations of fluid motion for a given system. CFD is termed an approximation due to errors in the numerical solution procedure; some of which can be controlled, and others, which are a function of the solution approach used are less controllable.

There are three main approaches that can be taken to solve the Navier-Stokes equations, varying in complexity and computational expense, DNS, LES and RANS.

Direct Numerical Simulation (DNS) of a given flow field resolves all of the events in the domain. This approach to turbulence modelling yields the highest resolution with very little uncontrollable error. However, to sufficiently resolve the flow field for all turbulent events the computational grid and time step must be extremely fine. This means that simple problems consume very large amounts of memory and computation time.

The Large Eddy Simulation does not resolve the flow to the same accuracy as DNS. Instead, the model uses filters with a length scale to differentiate between higher energy (larger) eddies and lower energy (smaller) eddies. The argument is that the majority of the flow activity is provided by large eddies, and it is not necessary to resolve the smaller scale eddies due to their relative impact on the flow. The LES approach is "cheaper" than DNS, as extremely fine grid resolution is not required to capture large eddies in the flow. Follow this link to learn more about LES CFD.