# les-ras
- z
Zino
09/21/2025, 9:08 PMYep, you look at the ratio - z
Zino
09/21/2025, 9:09 PMDepending on which rule of thumb you use, you should have 80% or more or the energy resolved, the rest modelled - s
silentspirit
09/22/2025, 7:09 AMSo my refinement region should be size of boundary layer, especially near body refinement? - s
silentspirit
09/22/2025, 7:10 AMAnd how to ensure that ratio during meshing? - z
Zino
09/22/2025, 7:10 AMYou should have Y+ around 1, but your cells near the wall should be cubes, not thin inflation layers - z
Zino
09/22/2025, 7:10 AMIncrease refinement, don't use inflation layers - s
silentspirit
09/22/2025, 7:13 AMAnd if i do unstructured meshing then there will be quads etc instead of cubes, will that work? Or do i need to do structured meshing - s
silentspirit
09/22/2025, 7:14 AMAnd after the mesh is made is there any way to check if the mesh is capable of running les or something like that? - s
silentspirit
09/22/2025, 7:21 AMThanks!! - l
littleB123
10/02/2025, 6:51 AMQuick question - wolf dynamics recommend using grad(omega) for div(phi,k) Gauss linearUpwind grad(omega); - why is that ? Why not grad(k)? https://cdn.discordapp.com/attachments/598785402391887872/1423200626359140412/image.png?ex=68df725f&is=68de20df&hm=1672dc7046101739642317858665322f0bbc873d50202162d3750f4fe5fe2144& - p
Pflegeverantwortungsöl
10/02/2025, 11:25 AMIt is not wrong to use grad(k) but it is more likely to activate the limiter, increase diffusion and if your mesh is bad you're likely to crash - p
Pflegeverantwortungsöl
10/02/2025, 11:25 AMYou can try grad(k) for good meshes but if you want to be sure with almost the same accuracy then just go for grad(w) - l
littleB123
10/02/2025, 11:25 AMThanks! - p
Pflegeverantwortungsöl
10/02/2025, 11:27 AMI've seen NASA work on that I can try to find it gimme a sec - q
qr
10/02/2025, 11:27 AMWhere can we learn more about these choices? - k
kandelabr
10/02/2025, 11:38 AMinteresting... I simply assumed that was a typo, "corrected" it and moved on - s
slopezcastano
10/03/2025, 11:15 AMIt should be grad(k) and not grad(omega). - s
slopezcastano
10/03/2025, 11:17 AM - q
qr
10/03/2025, 11:18 AMAa yes thanks but I meant to ask about what OP mentioned about grad(k) being poor compared to grad(omega), that kind of details - s
slopezcastano
10/03/2025, 11:20 AMwell, grad(k) doesnt activate any limiter, it applies a skewness correction on non-orthogonal cells - s
slopezcastano
10/03/2025, 11:23 AMit is a mistake to use grad(omega) as a corrector for kEqn because these two variables tend to be inversely "proportional" related. Regions of large omega correspond to low k, and regions of high k gradients transport omega away - s
slopezcastano
10/03/2025, 11:23 AMsaid more scientifically, omega "destroys" k - s
slopezcastano
10/03/2025, 11:24 AMand grad(k) only "transports" omega - s
slopezcastano
10/03/2025, 11:24 AMsee the equations - s
slopezcastano
10/03/2025, 11:25 AMif anything grad(k) should be the "limiter" in this case - s
slopezcastano
10/03/2025, 11:29 AMI want to see the link where they justify this choice - s
slopezcastano
10/03/2025, 11:35 AMMost likely is a typo. I dont find a reason why it should be that way, except when you have an explicit treatment of the divergence of one variable as a term in the equation of another variable - z
Zino
10/03/2025, 11:38 AMYeah that's my impression too - I wonder why it forces you to specify the gradient field though, if in the overwhelming majority of cases you're just going to be feeding it the gradient of that field... - s
slopezcastano
10/03/2025, 11:39 AMThere are valid reasons to use different gradients