How the 'adaptmesh' is supposed to work when there is solution only by refining the mesh?

Hi,

The problem I'm struggling with is a nonlinear elliptic pde. working with 'pdenonlin', it converges to the solution only when the mesh size is small enough in some particular parts of domain. So there is a need to a smart refining approach which seems to be 'adaptmesh'. but as I experienced, it doesn't do the refinement when it can not converge to the solution at the first attempt (I mean by the initial mesh structure passed to it).

Is that true? If yes how can I do the smart refinement till I get the solution?

thanks for any hint and suggestion,

mohammad,