Hi Pierre, Sorry for taking so long to get back to you on this.
On Fri, Mar 1, 2013 at 9:14 AM, Pierre de Buyl <[email protected]> wrote: > On Fri, Feb 22, 2013 at 12:10:51PM -0500, Daniel Wheeler wrote: > > On Wed, Feb 20, 2013 at 1:54 PM, Pierre de Buyl <[email protected]> > wrote: > > Following my 2D experience, I went on to define a 3D mesh in gmsh that is > made > of two concentric spheres, one at r=sigma and one at r=20*sigma ("far > away"). > The mesh is only defined on a slice of an eighth of a sphere. This gives > very > good results, at the expense of more memory and more CPU time. As the > precision > needs to be higher close to r=sigma, I defined the typical cell size in > gmsh. > This sounds good at least. > > I am just unsure of the precision of the solution. What I need, as a final > result, is > \int_0^\pi n(R,theta) sin(theta) cos(theta) dtheta > at some fixed value R. In the absence of flow, the solution should be > perfeclty > symmetrical and the above integral should equal 0. I get, depending on the > size > of the sphere, a significantly non-zero result (from 0.1 to 0.3). > Does it improve as the mesh is refined? > Is there any typical way of checking wether the solution is precise enough > for > this kind of computation? > In general, one refines the grid and tries to determine the order of accuracy. One way would be to plot dx versus the integrral error and see how that looks on a loglog plot. You many also want to use a uniform gird to determine accuracy and then switch to a non-uniform grid once that has been determined. -- Daniel Wheeler
_______________________________________________ fipy mailing list [email protected] http://www.ctcms.nist.gov/fipy [ NIST internal ONLY: https://email.nist.gov/mailman/listinfo/fipy ]
