Absolutely agreed

Oh, lovely

I am so happy when I came across with that kind of things

If I can find hardcopy and it is affordable, I do not hesitate to buy it

Then, I can scan it (I assume that it is ~ 300 pages, if more, than I could have doubt on it)

How does it make sense I can run interFoam as “laminar” I thought the turbulence model closes the equation after Reynolds averaging. Does it just ignore the reynold stresses all together?

It jsut solves the navier stokes equations without any turbulence closures

Well, it is true that turbulence modelling is needed to solve the closure problem when applying Reynolds averaging (space-time filtering in general) to the NSE, but Reynolds Averaging is not a condition for the numerical solution of the NSE. These two are separate concepts. You can run any simulation using PISO/SIMPLE without a turbulence model. In FOAM, that means the turbulence fluxes and viscosity are identically zero.

Thank you to both

I am trying to understand how adequate my LES simulation is, using Pope's prescription: finding a metric M, $$M = \frac{X_{resolved}}{X_{resolved}+X_{modeled}}$$ where typical discussions suggest that >80% should be the resolved component, or $$M > 0.8$$. The choice of X for my case is the kinetic energy, so : $$M = \frac{k_{resolved}}{k_{resolved}+k_{modeled}}$$ However, I am confused by two formulations for choice of the resolved component.

choice 1: $$ k_{res} = \frac{1}{2}U^2$$ , which is basically the kinetic energy contained in the resolved velocity field, and choice 2: $$ k_{res} = \frac{1}{2} ({U-\bar{U}})^2$$, which is basically the kinetic energy contained in fluctuations about the average of the resolved field. In other words, the resolved "turbulence". The first is straightforward to implement, while for the second I basically use trace(Uprime2Mean). I haven't been able to decide which is the more appropriate metric here, so would appreciate help on this.

Regarding this, I do recall that we had a discussion about this around here: https://discord.com/channels/397772779870486529/726377664213942323/927818681470304336 But, from the point of view of approaching a DNS, why do we not want to use the total kinetic energy directly (mean included)?

It's the second choice

What subgrid model are you using?

Any definite source on that? I am not doubting it in any way, just the reasoning behind excluding the mean flow

Smagorinsky for now, but I also have WALE and kEqn in mind.

I mean I've run those three for initial cases...Now I am looking at fine tuning the process.

It's not really excluding the mean flow, although I guess you could interpret it that way

You're interested in the energy associated with the eddies, not the bulk flow

Does this question bear importance for the above ? (In case I'm missing something important)

Interesting. I think I did not catch this argument in the way Pope frames this (in his paper, 10 questions about LES) He talks about any parameter of interest, and how it is split between resolved and modelled. So I figured I may very well apply it to all of KE. But what you said brings to mind an argument I read about spectral solvers, where we can eliminate pressure from the equations... It might be analogous to filtering out the bulk/mean flow effects from analysis. Sad for me as my resolution index just dropped to around 0.6 🥲

Nah just changes how you calculate it, easiest to get subgrid from kEqn

But yeah that's the problem, I wouldn't expect the mean flow to change that much with increasing grid resolution

I've always seen it done with the TKE, and never with the overall KE

True. Actually smagorinsky has an intermediate k (and also epsilon) variable in its formulation, solved from a quadratic equation. I wonder if there's a switch in dictionary to write the generic turbulence->k() variable.. for now I simply had to recompile my solver with a variable to store k(), but would be much easier to just use a switch in turbulence model, since right now I am blocking memory unnecessarily, just to maintain compatibility with RAS and LES.

Ah yeah usually you can output that sort of field, there's a special post proc function

One sec I'll find it

Okay, I think it's the writeObjects function object