Just noticed that this had been buried in my outbox: On Thu, 21 Apr 2011, Vikram Garg wrote:
> Hey, > I am writing because this is either an interesting theoretical > requirement I am not understanding or a bug in my code/libMesh. For my > application, I want to set different penalties on different subsets of > the boundary. In particular, I want a penalty of epsilon on subset A > and -epsilon on subset B. A and B are share a single vertex, but are > otherwise disjoint. In my application, they also happen to be > orthogonal to each other, i.e. they form a corner. > > However, if I set penalties of epsilon and -epsilon on A and B, two > things happen > > 1) The first iteration of a uniform refinement loop gives a solution > that is unbounded. > > 2) The second iteration (a once uniformly refined mesh) gives a > jacobian verification error on an element, not the side. If I just use > the numerical jacobian, the solution is still unbounded. > > This happens even if I change my initial mesh size, i.e. the same > behaviour (1 and 2) repeats, no matter what my initial mesh size is. What's being penalized on each of A and B? It's definitely possible for two penalties to "cancel". But I wouldn't have expected it to happen in your case where you've been playing with different penalty terms for completely different parts of the physics. --- Roy ------------------------------------------------------------------------------ What Every C/C++ and Fortran developer Should Know! Read this article and learn how Intel has extended the reach of its next-generation tools to help Windows* and Linux* C/C++ and Fortran developers boost performance applications - including clusters. http://p.sf.net/sfu/intel-dev2devmay _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
