I see, that makes sense since the subgrid is modelled. I'll set up some simulation today and confirm your findings. Thanks for the input, very interesting stuff!

We did two point correlations and we came to the result that with that kind of geometrie the computational domain should be longer, but still some researcher getting some very good result for pressure loss using shorter geometries

And also we still got very good result for the pressure loss, with shorter geometries than what we got out of the two point corellation.

Of course if you want to be sure the two point correlation is the best way, but I am wondering why it does still work

Did you use the same resolution and sub grid models?

yes

Was the numerical model the same? I suspect they used a higher order method. If everything is the same?

Hard to tell

ah not I got that wrong, the researcher used other grid resolutions and other numerical models

they use third or fourth order, but does this make such a big difference for the length of the computational domain?

No the order not that much but the resolution yes. As long as you use 2nd order or greater and time stepping is non dissipating (like RK4 used by @slopezcastano ) it should be doable.

With a decent resolution ofcourse. However I saw a large difference between a spectral element and fvm for capturing structures and other features.

> And also we still got very good result for the pressure loss, with shorter geometries than what we got out of the two point corellation. @User Pressure fluctuations on the wall are typically modulated by the outer region (the log region) of the boundary layer, not so much by the inner (viscous) sub-layer, indicating that the scales related with pressure are INERTIAL. The autocorrelations (which measure stochastic scales) are not representative. If you resolve the log-region accurately (regardless of the physics close to the wall), then you should obtain good mean values for pressure

> No the order not that much but the resolution yes. As long as you use 2nd order or greater and time stepping is non dissipating (like RK4 used by @User ) it should be doable. @User You can also use dissipative first (or higher) order time schemes, as long as the (artificial)viscous time scale of is smaller wrt the Kolmogorov's time scale

> With a decent resolution ofcourse. However I saw a large difference between a spectral element and fvm for capturing structures and other features. @User turbulence is local, then either method (FVM/SEM/FR/DG), may be able to resolve turbulence. The problem resides on HOW the non linear term in NS is "discretized", and in FV whether you are respecting the Peclet limit for the linear schemes.

Thanks for that very nice informations, I have to think about it 🙂

@slopezcastano thanks for the details, that's correct. I have been using QUICK for advection in OpenFOAM and seems to work well. All the other terms are second order. What's your setup?

> @User thanks for the details, that's correct. I have been using QUICK for advection in OpenFOAM and seems to work well. All the other terms are second order. What's your setup? @User For resolved LES? Gauss linear, leastSquares, and CrankNicolson/RK4, but I dont use PISO, and I use a mixed lagrangian smagorinsky model. For the rest of cases, I go with IDDES SA. BTW, QUICK is also second-order scheme when used for the advection term.

I meant for the DNS case you showed. Yes Quick indeed is 2nd order for the advection term.

why CrankNicolson over backward?

> why CrankNicolson over backward? @User I use semi-implicit schemes for describing the non-linear term. Besides, there is no point in going fully implicit if you have Co<1

the choice of backward over euler in FOAM is more due to historical reasons

the Euler time scheme used to be stabilized with an artificial flux in the old days. This is not good for LES

Hence, the reason why they suggested to use backward

> I meant for the DNS case you showed. Yes Quick indeed is 2nd order for the advection term. @User Gauss linear, leastSquares, CrankNicolson where corresponds

Ohh yes, and full hexahedral grids

> @User Pressure fluctuations on the wall are typically modulated by the outer region (the log region) of the boundary layer, not so much by the inner (viscous) sub-layer, indicating that the scales related with pressure are INERTIAL. The autocorrelations (which measure stochastic scales) are not representative. If you resolve the log-region accurately (regardless of the physics close to the wall), then you should obtain good mean values for pressure @User But I am not sure about this when there are ribs inside the pipe/channel, or am I wrong?

In a pipe with ribs is not the case because the driving force plus the pressure acting on the ribs MUST balance the shear stress

then the pressure is modulated by viscous scales as well

yeah and thats why I am wondering why some people, like mentioned before, get so got results even with ripped pipes 😄