[overture] Re: cgins parameters

  • From: Bill Henshaw <henshaw@xxxxxxxx>
  • To: "overture@xxxxxxxxxxxxx" <overture@xxxxxxxxxxxxx>
  • Date: Fri, 20 Jan 2012 12:34:43 -0800

Alessandro:
Correct -- you can look at the vorticity which is equivalent to Du -- normally 
boundary layers or shear layers are the hardest regions to resolve.
In the end the key question is what information ("quantity of interest") do you want from the flow? 
It is always recommended to run with more than one grid resolution in order
to see if the quantity of interest has converged.

...Bill

Alessandro Orchini wrote:

Thanks Bill,
so I have to look at the local gradient of the velocity Du when I run a simulation, and consequently estabilish if the h I chose was small enough, is that right?
About the second question I'm not really an expert in the "grep" command, but it seems it's pretty useful, thanks for the suggestion. 

Regards

Alessandro




------------------------------------------------------------------------
Date: Fri, 20 Jan 2012 10:07:54 -0800
From: henshaw@xxxxxxxx
To: overture@xxxxxxxxxxxxx
Subject: [overture] Re: cgins parameters

Hi Alessandro,

Alessandro Orchini wrote:


    Hi all,

    I have a few questions about cgins.

    1) I read the Reference Manual "Cgins: A Solver for the
    Incompressible Navier–Stokes Equations" and I also looked at the
    reference "On the smallest scale for the incompressible
    Navier-Stokes equations", but I can't figure out if and how much
    the artificial diffusion (second order) influence the simulations.

    For example, if I run a simulation with \nu=1.0e-4, is it ok to
    choose values for artificial diffusion ad21 = ad22 = 1? How can I
    know /a priori/ if they're too big?
Suppose you want to solve the INS equations with a given value of nu. If you want to have a fully resolved computation then you should choose the grid spacing according 
to the formula on page 13 of the Cgins ref. manual:
                    h =  C * sqrt( nu/ Du )
The local grid spacing h is related to the local gradient of the velocity, Du. The constant C seems to normally be about 1. Thus you really can't tell the correct value for h 
without knowing what the solution looks like (approximately).
If you choose h in this way then you should need NO artificial dissipation.
If you can't afford to use a very fine grid then you can use artificial dissipation. You can think of this as an LES turbulence model. To have as small effect as possible you should choose the dissipation to be as small as possible BUT large enough so the solution remains stable. These leads to the recommendations in the ref. manual.
If you want to know if the dissipation is affecting your result then you should run the problem with a finer grid and see if the things you are interested in change. If they 
don't change very much then you are probably OK.



    2) I'd like to know precisely how the "project initial condition"
    option actually work, does anyone know if there's some reference
    about it, or where the source code is located in Overture?
Unfortunately, I don't recall ever writing up the precise details of the projection.
However, if you want to know where the projection is computed in the code a good guess would be provided by using "grep" to find out who is printing the messages that are written to 
the screen when the projection is performed:
projectVelocity>>> iteration=0, (new div)/(old div)= 0.065, divergence after projection= 1.61e+00 projectVelocity>>> iteration=1, (new div)/(old div)= 0.127, divergence after projection= 2.05e-01 projectVelocity>>> iteration=2, (new div)/(old div)= 0.902, divergence after projection= 1.85e-01 projectVelocity>>> iteration=3, (new div)/(old div)= 0.998, divergence after projection= 1.84e-01 

You would then find that this is done in the file cg/ins/src/project.C

Regards,
  Bill


    Thanks for the help

    Alessandro

Other related posts:

  1. Home
  2. overture
  3. Archive
  4. 01-2012