Hello there:

I have started working on LES implementation into an existing incompressible NS code. I wanted to know some basics about LES> So I read basic fundamentals abt LES. However, when it came to coding ....I got few problems.

I wasnt sure about the filter function implementation. The filter function G(x) etc..where is it used ?

since the sub-grid scale viscosity is computed using

nuT = 2.C*Sij ...etc. or some other models..with a filter width etc...but I dont find the presence of filter function..

where is the filter function used? Why is the filter function important?

Can someone guide me to basic implementation of LES to a normal incompressible N-S equations?

Thanks,

CFDtoy

I'm not sure about your doubt (and I should confess that I'm not a LES expert...) but if you're working with LES the simplest filter function, that is implicitly employed in mesh-based computations, is your mesh itself. Note that since you're not solving a DNS problem, your mesh is not able to model the smallest scale phenomena related to the flow. In this sense, your mesh "filters out" these small effects implicitly and models only the effects that are "visible" in its size scale.

I naive explanation: Imagine that in a turbulent flow you'll have big, medium and small swirls all mixed and exchanging energy in some sense. Your mesh will only be able to represent the swirls that are in the same order of magnitude of its elements (or cells). The small swirls will be "hidden" (filtered out) within the element volume and you'll need to model these small effects in some way. In LES computations it's done by some auxiliary model like Smagorinsky, k-e, etc..., etc..., etc... . These models will try to represent these small effects that were losen the by lack of resolution of your mesh. Of course, some models are richest than others, but they're only trying to represent something that was lost...

Regards

Renato.

p.s.: Since I'm not a LES guy, I'm waiting someone correcting me. Note that I haven't employed any kind of formalist in my description and I should emphasize that it's only my point of view from the LES texts that I already read.

You apply your filter to the fields at each iteration.

Very few people explicitly apply a filter to LES. Explicit filtering (through the use of a Gaussian filter for example) is very rare. In general, it is as Renato says. You simply model your flow field as unsteady and include the effect of the sub-grid through the use of a sub-grid model. e.g. The Smagorinsky model. So, if you have a RANS code, you simply include a sub-grid model and rather than iterating until you reach a converged solution, you simply advance in time. Make sure you Reynolds number is high enough that you have a turbulent flow, otherwise you won't see the effects of turbulence! Furthermore, the Smagorinsky model will act to drain energy (this is the basic idea of what the isotropic sub-grid scales do) even if the flow is Laminar. Dynamic variations of the Smagorinsky model (Calculating the Cs value dynamicially rather than setting it to a value (i.e. 0.1) will result in zero values of Cs in Laminar regions) eliminate this problem. Conceptually LES is simple and quiet beautiful. All the mumbo-jumbo about filtering can in effect be ignored since it (the mumbo-jumbo!) is more a mathematical construct than anything else. For general LES calculations of engineering interest you need not concern yourself with it.

I hope this is of some help.