I'm pretty sure that the geometry approximation is responsible of the convergence degradation. I've experienced this problem with the ringleb flow (2D Euler) using inviscid wall bcs. If you do p refinement instead of h refinement you should see that the entropy error actually stagnates and you get almost zero convergence. This is due to the error in the geometry representation. Imposing the bcs on the approximated geometry according to the exact solution of the ringleb flow allows to recover optimal convergence rates. The geometry approximation error is neglected. Unfortunately I don't think that the bump has an exact solution, does it? Otherwise you could try this way.
Lorenzo On May 15, 2013 7:58 PM, "Manav Bhatia" <[email protected]> wrote: > On Wed, May 15, 2013 at 1:17 PM, John Peterson <[email protected]> > wrote: > > > On Wed, May 15, 2013 at 6:36 AM, Manav Bhatia <[email protected]> > > wrote: > > >> > > >> > > >> One other thought: do you have some O(h) stabilization terms present? > > >> Then if you aren't refining the grid they don't get smaller... > > >> > > > > > > > > > There is a GLS stabilization term with the typical tau matrix. I have > > > experimented with different tau definitions, but get the same behavior. > > > > > > Essentially, almost all definitions in 1-D have the form > > > > > > (sum_{i=1,n_shape_funcs} | a dNi/dx | ) ^-1 > > > > > > where N is the shape function and 'a' is the velocity. The numeric > value > > of > > > this expression reduces as the polynomial order is increased. > > > > Really? The lowest order hierarchics are the Lagrange basis > > functions, and dNi/dx ~ 1/h is certainly true for those. > > > > Hi John, > > I have attached a pdf which plots the expression for a 2-noded and a > 3-noded element using Lagrange interpolation functions. The element has > unit length. The 2-noded value is constant at 2 for the entire domain of > the element, while the 3-noded element shows a variation in the element > domain. It equals 2 at xi=0, but rises to about 8 on either end. Each > quadrature point uses the inverse of this function value. > > This is my basis for stating that the the influence of higher > polynomial order is included, although I can't say if this is constraining > the order of convergence. > > Manav > > > ------------------------------------------------------------------------------ > AlienVault Unified Security Management (USM) platform delivers complete > security visibility with the essential security capabilities. Easily and > efficiently configure, manage, and operate all of your security controls > from a single console and one unified framework. Download a free trial. > http://p.sf.net/sfu/alienvault_d2d > _______________________________________________ > Libmesh-users mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/libmesh-users > > ------------------------------------------------------------------------------ AlienVault Unified Security Management (USM) platform delivers complete security visibility with the essential security capabilities. Easily and efficiently configure, manage, and operate all of your security controls from a single console and one unified framework. Download a free trial. http://p.sf.net/sfu/alienvault_d2d _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
