Re: [ff3d-users] Problem in Solving Laplace with Neumann BC's

[Stephane Del Pino]

Hello.

Default setting for ff3d to solve linear systems is Conjugate Gradient with and 
epsilon of convergence set to 1E-5.

Here, to get a better solution, you need to get a smaller epsilon. Taking 1E-8 
seems enough for your case:

Use the conjugate gradient option in your solver block:

...
solve(v) in O by M
  cg(epsilon=1E-8)
{
...

Beware that your are solving your problem in a box which corresponds exactly 
to your fictitious domain geometry. This can eventually lead to strange 
behaviors. If your mesh fits the geometry I recommend not to use the fictitious 
domain but directly the conformal FEM. It is very simple in your case, write
...
solve(v) in M
  cg(epsilon=1E-8)
{
...

Best regards,
Stéphane.