dolfin team mailing list archive

[dolfin@xxxxxxxxxx]

Commit from logg (2005-02-11 04:16 CET)
----------------

Fixed Poisson solver:

- Solver now gives the same results as the old solver
- Changed mapping of boundary ids from i+1 to i (why was that so?)
- Corrected creation of function u for saving (removed extra argument
  dim = 1)
- Added the old code in solveOld() for comparison

Remains to do:

1. Cleanup Poisson solver and move specification of boundary conditions
somewhere else. I suggest that the assembler does this automatically
if given an extra argument bc. The assembler can also provide a function
setBC(). We can overload and have a version assemble(..., bc) that calls
both assemble(...) and setBC(bc). We should probably create the proper
wrapper for PETSc MatZeroRows() if we can find a good name for it,
perhaps NewMatrix::ident(uint rows[], uint n)?

2. Keep function solveOld() temporarily so we can do a benchmark for
different meshes.

3. Implement virtual real NewFunction::operator(const Point& p) and
move specification of right-hand side to main.cpp

  dolfin  src/demo/solvers/poisson/main.cpp      1.21
  dolfin  src/modules/poisson/PoissonSolver.cpp  1.22