Re: [Libmesh-users] AMR in 3D

[John P. <pet...@cf...>]

Just in case those pngs didn't make it through....

http://www.cfdlab.ae.utexas.edu/~peterson/vectors.png
http://www.cfdlab.ae.utexas.edu/~peterson/u.png


> John Peterson writes:
> > Hi,
> > 
> > I looked again at the image you sent us.  I'm not sure there's
> > anything wrong ... it looks like you have "Use Log" enabled in
> > GMV, which is making some interior velocity vectors look like they
> > are coming outside the domain.  I attach a few AMR pictures from
> > ex18 of the x-velocity contours and the velocity vectors with
> > "Use Log" disabled.  The L2 imposition of the boundary conditions
> > sets u=0.5 along the top "edge".  There are some regions of localized
> > overshoots on the lid, seen in pink.
> > 
> > -John
> > 
> 
> Luca Antiga writes:
>  > Roy, John,
>  >   thanks for the prompt follow ups.
>  >   I reply on Lorenzo's behalf since we're working together and he's  
>  > out of the office today.
>  > 
>  > > What do you mean by "mesh refinement levels greater than one"?
>  > > Relaxing the default level-one hanging node restriction?  More than
>  > > one level of initial uniform refinement?  More than one level of AMR
>  > > on top of a uniform grid?
>  > 
>  > More than one level of AMR on top of a uniform grid, that was our test.
>  > 
>  > > I've never done Navier-Stokes with tets; is it possible that tet AMR
>  > > (with quadratic velocity, linear pressure, I assume) leads to LBB
>  > > violations?  You might see if you can reproduce the problem on a PDE
>  > > with a positive definite weak form.
>  > 
>  > Yes, P2P1 10-noded tets, LBB stable (without AMR).
>  > This is a good point, however we're seeing no spurious pressure modes
>  > in the refined meshes. Plus, it seems to work with the first level of  
>  > refinement.
>  > We'll test it with a positive definite weak form, that's a good idea.
>  > 
>  > Another test we want to perform is imposing Dirichlet boundary  
>  > conditions
>  > the old Lagrangian way, by setting the diagonal term to 1, the extra- 
>  > diagonal
>  > terms to 0 and the rhs to the bc value.
>  > 
>  > The fact is that no matter what we try, it seems we're not able to  
>  > impose boundary
>  > conditions exactly. We always see spurious velocities on side wall  
>  > elements in all
>  > children whose parents are adjacent to the top wall.
>  > 
>  > > If it is a libMesh bug, my first instinct would be to doublecheck that
>  > > the hanging node constraints are being properly created and satisfied.
>  > > Try the "DiscontinuityMeasure" object to make sure that your
>  > > interelement jumps on the failing meshes are on the order of machine
>  > > precision.
>  > 
>  > Another good suggestion. We still don't know the details of the AMR  
>  > implementation
>  > used in libmesh. We're learning as we go.
>  > 
>  > > There could be a problem with the tets+adaptivity.  Tim Kroger  
>  > > recently
>  > > introduced a new refinement pattern for tetrahedral elements which  
>  > > performs
>  > > an "edge swap" during refinement to avoid generating small angles.  I
>  > > think this was tested pretty extensively, but I could be wrong.
>  > 
>  > We are aware of this. We tried reverting to the old "unswapped"  
>  > refinement,
>  > and things apparently got worse. It doesn't look like a bug in the  
>  > swapping
>  > implementation.
>  > 
>  > Tomorrow we'll post a few images on our webpage for you to check out.
>  > 
>  > Thanks for the help you're providing
>  > 
>  > Luca
>  > 
>  > --
>  > Luca Antiga, PhD
>  > Head, Medical Imaging Unit,
>  > Bioengineering Department,
>  > Mario Negri Institute
>  > email: an...@ma...
>  > web: http://villacamozzi.marionegri.it/~luca
>  > mail: Villa Camozzi, 24020, Ranica (BG), Italy
>  > phone: +39 035 4535-381
>  > 
>  > 
>  > On Apr 3, 2007, at 6:13 PM, Roy Stogner wrote:
>  > 
>  > > On Tue, 3 Apr 2007, Lorenzo Botti wrote:
>  > >
>  > >> After your suggestions we've tested our navierstokes solver on the 3D
>  > >> lid driven cavity problem with both hex27 and tet10.
>  > >> Amr works great with hex and the problem we encountered with tet10 is
>  > >> limited to the selection of mesh refinemet levels greater than one.
>  > >
>  > > What do you mean by "mesh refinement levels greater than one"?
>  > > Relaxing the default level-one hanging node restriction?  More than
>  > > one level of initial uniform refinement?  More than one level of AMR
>  > > on top of a uniform grid?
>  > >
>  > >> Could this suggest a possible problem in libmesh? We're
>  > >> investigating this...
>  > >
>  > > I've never done Navier-Stokes with tets; is it possible that tet AMR
>  > > (with quadratic velocity, linear pressure, I assume) leads to LBB
>  > > violations?  You might see if you can reproduce the problem on a PDE
>  > > with a positive definite weak form.
>  > >
>  > > If it is a libMesh bug, my first instinct would be to doublecheck that
>  > > the hanging node constraints are being properly created and satisfied.
>  > > Try the "DiscontinuityMeasure" object to make sure that your
>  > > interelement jumps on the failing meshes are on the order of machine
>  > > precision.
>  > > ---
>  > > Roy
>  > >
>  > > ---------------------------------------------------------------------- 
>  > > ---
>  > > Take Surveys. Earn Cash. Influence the Future of IT
>  > > Join SourceForge.net's Techsay panel and you'll get the chance to  
>  > > share your
>  > > opinions on IT & business topics through brief surveys-and earn cash
>  > > http://www.techsay.com/default.php? 
>  > > page=join.php&p=sourceforge&CID=DEVDEV
>  > > _______________________________________________
>  > > Libmesh-users mailing list
>  > > Lib...@li...
>  > > https://lists.sourceforge.net/lists/listinfo/libmesh-users
>  >