I am trying to use AdaptiveLinearVariationalSolver to solve Poisson equation in 3D, starting with a quite coarse mesh. As a result I either get solution with the unrefined mesh, or “not enough memory error”, depending on the tolerance set.
The problem is I am not even sure what I am doing.
I have based my code on a documented example: https://fenicsproject.org/olddocs/dolfin/1.6.0/python/demo/documented/auto-adaptive-poisson/python/documentation.html
Unfortunately, I do not know, what norm is used to estimate the a posteriori error. If I understood the example correctly, the estimate is not L1 norm (integral of abs(u) dx), but an absolute value of the integral of u dx. I have tried to use either sqrt(u * u * dx()) or (u ** 2 * dx())**0.5 to use L2 norm, but neither approach works.