[syavash]

Dear All,


I have come across numerous threads discussing about stability and accuracy of different divergence schemes. However, I would like to know about the diffusivity of the divergence schemes for scalar transport, particularly those convenient for Large-eddy Simulation (LES). I have read that SFCD scheme is far too diffusive for LES. On the other hand, I am not sure if linearUpwind or limitedLinear are suitable for scalar transport in LES.


Then, what would you suggest as a minimum-diffusion scheme for passive scalar transport when doing LES? A scheme which also shows good stability at the same time?


Regards,
Syavash

[pete20r2]

Hi Ehsan,
I am by no means an expert on this but typically diffusivity and stability are two sides of the same coin. That is, low diffusivity usually implies low stability. The source guide explicitly recommends Gauss linear for LES simulations, citing low diffusivity (dissipation) as the reason.
https://www.openfoam.com/documentati...ce-linear.html
If you are looking for stability, you might want to look into the blended scheme which I believe was implemented in OF foundation V6 and should be in v1806 also.
With this scheme you can specify two schemes (for you I'd say linear and upwind) and OF will use a different scheme per cell depending on a number of factors.
See:
https://www.openfoam.com/releases/op...6/numerics.php
I'll let you read through that.
Make sure to read the log at the start of the run to make sure the function is working properly.
If you need a hand getting it working, post in here again.

[syavash]

Originally Posted by pete20r2

Hi Ehsan,
I am by no means an expert on this but typically diffusivity and stability are two sides of the same coin. That is, low diffusivity usually implies low stability. The source guide explicitly recommends Gauss linear for LES simulations, citing low diffusivity (dissipation) as the reason.
https://www.openfoam.com/documentati...ce-linear.html
If you are looking for stability, you might want to look into the blended scheme which I believe was implemented in OF foundation V6 and should be in v1806 also.
With this scheme you can specify two schemes (for you I'd say linear and upwind) and OF will use a different scheme per cell depending on a number of factors.
See:
https://www.openfoam.com/releases/op...6/numerics.php
I'll let you read through that.
Make sure to read the log at the start of the run to make sure the function is working properly.
If you need a hand getting it working, post in here again.

Hi Peter,


Thanks for your reply. I forgot to mention that I am currently using OF 2.3 for my simulations. I know that having both stability and accuracy at the same time can be difficult. However, I would like people to share their experience regarding schemes for scalar transport.

As you may already know, Gauss linear which is second-order central scheme can also be used for scalar, however it may cause fluctuations in the scalar value, as well as unrealistic values out of the range (unbounded).
Also, I came up with the idea of using blended scheme which is a combination of upwind/linear schemes but I was not sure how this could affect the solution of the scalar field.


I think it would be great if people with experience on using divergence schemes for passive scalar, particularly in LES, can suggest a couple of options for other people like myself so it will not be necessary to go through the various schemes to examine each of them one by one.


Regards,
Syavash


P.S. At the moment I am using vanLeer01 for the passive scalar, however I am not sure that is a good choice for LES.

[Santiago]

Originally Posted by syavash

Hi Peter,


Thanks for your reply. I forgot to mention that I am currently using OF 2.3 for my simulations. I know that having both stability and accuracy at the same time can be difficult. However, I would like people to share their experience regarding schemes for scalar transport.

As you may already know, Gauss linear which is second-order central scheme can also be used for scalar, however it may cause fluctuations in the scalar value, as well as unrealistic values out of the range (unbounded).
Also, I came up with the idea of using blended scheme which is a combination of upwind/linear schemes but I was not sure how this could affect the solution of the scalar field.


I think it would be great if people with experience on using divergence schemes for passive scalar, particularly in LES, can suggest a couple of options for other people like myself so it will not be necessary to go through the various schemes to examine each of them one by one.


Regards,
Syavash


P.S. At the moment I am using vanLeer01 for the passive scalar, however I am not sure that is a good choice for LES.

Some comments:

The only good choice for resolved LES in foam is linear (leastSquares, or skewCorrected for interpolation) when using models based on boussinesq hypothesis (e.g. Smagorinsky Family) and inverse deconvolution. This forces the user to have cell peclet numbers < 2.

The fact that some scheme is unbounded doesnt make it unsuitable to work with! Anyway your system might be unbounded unless you go full upwind on time and space.

[pete20r2]

I wasn't clear enough when I said blended, it's actually localBlended. Read the link.

[syavash]

Originally Posted by Santiago

Some comments:

The only good choice for resolved LES in foam is linear (leastSquares, or skewCorrected for interpolation) when using models based on boussinesq hypothesis (e.g. Smagorinsky Family) and inverse deconvolution. This forces the user to have cell peclet numbers < 2.

The fact that some scheme is unbounded doesnt make it unsuitable to work with! Anyway your system might be unbounded unless you go full upwind on time and space.

Dear Santiago,


Thanks for your comment and I should say that I see your point. I have a passive scalar which has value 1.0 at the inlet and 0 elsewhere. When using Gauss linear I get fairly stable solution but I also get higher than 1 values in the downstream which I am not sure are physical or merely are there due to numerical scheme I have chosen.


Nevertheless, I wonder what the purpose of strictly bounded scheme such as vanLeer01 or Gamma01 had been in the first place if we suppose to have higher than 1 value in the domain?!


Regards,
Syavash

[syavash]

Originally Posted by pete20r2

I wasn't clear enough when I said blended, it's actually localBlended. Read the link.

Oh I see. I checked out the link you provided and I can see that the blending factor is calculated automatically in the field. However I am restricted to use OF 2.3.1 and currently cannot use that scheme.


Regards,
Syavash

[Santiago]

Originally Posted by syavash

Dear Santiago,


Thanks for your comment and I should say that I see your point. I have a passive scalar which has value 1.0 at the inlet and 0 elsewhere. When using Gauss linear I get fairly stable solution but I also get higher than 1 values in the downstream which I am not sure are physical or merely are there due to numerical scheme I have chosen.


Nevertheless, I wonder what the purpose of strictly bounded scheme such as vanLeer01 or Gamma01 had been in the first place if we suppose to have higher than 1 value in the domain?!


Regards,
Syavash

Again, the transport diffusion equation in algebraic form loses boundedness, and the degree of unboundedness is dictated by the linear solver you use, and certain properties of the coefficient/residual matrix...

Best way to go is to clip the scalar field after solving for it

[syavash]

Originally Posted by Santiago

Again, the transport diffusion equation in algebraic form loses boundedness, and the degree of unboundedness is dictated by the linear solver you use, and certain properties of the coefficient/residual matrix...

Best way to go is to clip the scalar field after solving for it

Thanks for the elaboration. I see that vanLeer01 which provides boundedness between 0 and 1 is doing the clipping. I wonder how another bounded scheme such as Gamma01 would perform.

Maybe I should compare the solution between the linear scheme and a bounded scheme such as vanLeer01 in term of time-averaged scalar quantity.
Further comments are appreciated.

Regards,
Syavash