Some family of models take advantage of the MVT and combine it with NVT schemes to get what is usually reffered to implicit LES. But that's another story.

you could estimate the kolmogorov length scale

yes, you can. Plenty of resources in the wild. Pope is a good reference for that.

I just use the turbulent kinetic energy from the simulation results to calculate the Kolmogrov scale so you need ofc results for this

for a second I thought you're asking me to turn to God

that's from a precursor RANS which is recommended anyway

In this particular case, you can consider Stephen Pope an actual Pope of the 'Structural Turbulence' Church

i'll try all of the above. that was very enlightening! thanks

My practical understanding: With URANS your eddy viscosity represents the total viscosity effect of the eddies - ie refining the mesh more shouldn't have any effect on nut, you should reach a point where the mesh is refined enough and you get a converged solution. In this case, nut will be high and suppress the formation of any eddies For LES, the eddy viscosity only represents the sub grid contribution - the more you refine your mesh, the lower nut will go until you reach DNS What allows the hybrid methods to work is building in some sort of switch or blending factor to drop nut way down in regions where you meet the right conditions for LES, which allows appropriate turbulent eddies to form

The actual intuition for how URANS works seems a bit murky. I guess I conceptualise it like this: when we do Reynolds decomposition, we think of velocity as a mean component plus a fluctuating component like U=u+u'. The fluctuating component is parameterised to be represented by some turbulence model, eg k-epsilon. In regular RANS, that mean component is constant. When we move to URANS, we allow that mean component to change over time, while still modelling all of the fluctuating component

But that only makes sense if there's a big separation in frequency between fluctuations in the mean and the turbulent fluctuations

And things start to get real murky if that separation of scales starts to close up

Anyway, I don't really have any good sources to back that all up, so someone please let me know if the way I'm conceptualising that is incorrect

TL;DR - no amount of refining the mesh or time step of a URANS simulation will get you to an LES result, because the eddy viscosity won't decrease as a function of grid size and therefore nut will always be way too high and dissipate all your turbulent content

you mean DNS?

Yep, too many 3 letter acronyms

I think this is a fine summary 🙂

Thanks @User, I think I get it now. The 'smeared' wake (for example) is just a cloud of modeled fluctuations which won't change with finer meshes. I was confused because I did manage to gain more details from finer meshes but obviously the coarser versions were just too brutally coarse.

this change should stop with more refinement in case of a (U)RANS

at that point you have a grid independent result

To add to @User s comment. The limit of RANS and LES is different. A super fine resolution RANS will be just that, but a super fine LES will approach a DNS.

With refining RANS mesh you get a converged result, with refining a LES mesh, you run out of RAM and time.

these comments should be written somewhere for future n00bs to read

they are too good to be lost somewhere in this channel's history

but first @User needs to confirm 😄

@User you can pin messages, see the triple dot on the top right of a comment

that's cool

Food for thought: (1) turbulence is natural to the Navier-Stokes equations, not to models. Let's make an example: let's solve a turbulent channel flow for Re_tau = 171, and use a grid suitable for LES (196x128x96, 4pi x 2pi x 2). We will use PISO (assume, for now is energy-conserving) and 2order CDS, and the time scale is such that Co < 1. I run 3 cases: no Model, LES model, and a RAS model. Where will I get "turbulence"? More precisely, If I sample a point far from the wall, Will I get an Enegry spectrum (forget whether its slope is correct) in either case?

(2) If Spallart-Almaras is a RAS model, why a trivial change in one term makes it LES?

(3) If I use simpleFoam instead of PISO in (1), and run with Dynamic Smagorinsky LES, Will I get "turbulence"?