Hi Kamil,

In comparison to DNS, any LES will have a coarser mesh (or a lower polynomial order in the same mesh), as the aim of an LES is to resolve the large and part of the intermediate flow scales. Note that as one keep refining an LES solution, one tends to the DNS limit where all the turbulent scales (down to the Kolmogorov ones) are resolved. The time step size should also be less stringent in an LES, but this will be naturally taken care of if a CFL-based step is used. In other words: when an LES solution is obtained from a DNS by coarsening the mesh (or reducing the polynomial order), a valid larger time step will automatically be set if the CFL number is mantained.

Dealiasing is a technique by which numerical integration is performed consistently. This is not is a must for well-resolved smooth flows, but is very important for under-resolved computations (such as implicit LES) as higher-order polynomial modes retain more energy.

In addition to that, as the LES mesh is coarsened from a DNS mesh, one will need extra dissipation added as the physical viscous term will be less effective in damping the turbulent scales (note that regular, second-order viscosity only acts significantly at the very small scales, which will be missing in an implicit LES). This is why you will need SVV, which is a type of high-order viscosity designed to affect the smallest resolved scales without interfering directly with the large ones. You can find information on how to set it up in the user guide (pages 92-93), but basically all you need to do is use: <I PROPERTY="SpectralVanishingViscosity" VALUE="True"/>. You can also tune it by setting the parameters SVVCutoffRatio and SVVDiffCoeff, but I would suggest starting with the default parameters.

Cheers,
Rodrigo
.

On Thu, Mar 17, 2016 at 12:53 PM, Kamil ÖZDEN <kamil.ozden.me@gmail.com> wrote:
Hi Rodrigo,

I'm running my analysis now by just adding the flag

<I PROPERTY="SpectralhpDealiasing" VALUE="True" />  inside section <SOLVERINFO> of my XML file and I wondered two things:

1) What is the difference between SpectralVanishingViscosity and SpectralhpDealising? Do I need to add anything to my xml file about SpectralVanishingViscosity to run my case with Implicit LES?

2) What about the grid size and structure? Do I need to change grid structure and/or grid size (number of elements, polynomial order etc.) to run Implicit LES when compared with DNS ?

3) What about the time step size? Do I need to increase or decrease the time step when compared with DNS?

I'll be glad if you answer my questions. Thank you very much.

Best Regards,
Kamil



02.03.2016 19:37 tarihinde Kamil ÖZDEN yazdı:

Hi Rodrigo,

Unfortunately I don't have any programming experience in any language except Matlab. I'll try to make a trial by just adding the flag needed for implicit Les to the xml file and share the results with you. Thank you very much for your help.

Best Regards,
Kamil

1 Mar 2016 14:34 tarihinde "Rodrigo Moura" <r.moura13@imperial.ac.uk> yazdı:
Dear Kamil,

For the incompressible solver you should definitely include the dealiasing flag, namely,
<I PROPERTY="SpectralhpDealiasing" VALUE="True" />
inside section <SOLVERINFO> of your XML file.

Please refer to Sec 2.4.1.2 of Spencer's book (Spectral/hp element methods for computational fluid dynamics) for an explanation about the dealiasing strategy used by the incompressible solver.

Note however the following. The success of our implicit LES approach via high-order CG methods (both in terms of discretization robustness and solution quality) is closely connect to a sensible use of SVV (spectral vanishing viscosity). This requires good knowledge of turbulence physics and of LES-like approaches and takes some programming expertise in Nektar++ since currently we do not have full control of the SVV operator only through the XML file. What I mean is that, for the purpose of implicit LES, you might need to modify the SVV operator inside Nektar and this is a somewhat advanced topic not covered by the user guide, as the CG+SVV-based implicit LES is a modern approach and not yet fully mature.

Please note the following study that discusses the behaviour of different SVV operators from a perspective of under-resolved turbulence computations:

Cheers,
Rodrigo
.

On Mon, Feb 29, 2016 at 11:22 AM, Kamil ÖZDEN <kamil.ozden.me@gmail.com> wrote:
Hi Rodrigo,

Thanks for your interest at first. I want to use the incompressible one. Do you know what settings should I do in the *.xml file to run an implicit LES?

Best Regards,
Kamil

29.02.2016 12:03 tarihinde Rodrigo Moura yazdı:

Hi Kamil,

Do you intend to use the incompressible solver or the compressible one? The incompressible is currently much more efficient in Nektar due to the possibility of using implicit time-stepping.

Cheers,
Rodrigo
.


---
Bu e-posta virüslere karşı Avast antivirüs yazılımı tarafından kontrol edilmiştir.
https://www.avast.com/antivirus




Bu e-posta Avast tarafından korunan virüssüz bir bilgisayardan gönderilmiştir.
www.avast.com