From: Kevin Ellwood <kellwood@am...>  20040518 12:08:43

Hi All, I am getting to the point where I understand C++ well enough to use libmesh and I would like to do so, abandoning my own ancient monolithic "c" based code. The problem that I would like to first tackle is basically a nonlinear transient reaction diffusion problem. For this, the examples supplied with libmesh are very relevant. However, I have an additional unknown "u" that is assumed to be a lumped parameter, that is, it is constant throughout the spatial domain but is a function of time. It is couple to the rest of the unknowns through a surface integral along one edge. So I have, 1: dc/dt = div(D(c)*grad(c)) + f(c) 2: du/dt = int_s D(c)*grad(c).n ds 3: c = f(u) on s In my own "crappy" code, I introduce one extra local node in each element which are all linked to one extra global unknown, which represents "u". I enforce the Dirichlet BC using a penalty approach and I build eq. 2 as I go. Is there a proper way for doing something similar in libmesh and might anyone have a "libmesh" example that is close that they could share?  .*. Kevin R.J. Ellwood /V\ +++++++ ++++++ (/ \) gentoo linux ( ) +++++++ ++++++ ^^_^^ 