The tol parameter in multilevel_solver.solve() should be relative to the norm of the right-hand-side, as it is for the other iterative methods.

http://groups.google.com/group/pyamg-user/browse_thread/thread/b1096fd0ff9d0e49

What I do in the Krylov routines is, # Scale tol by normb normb = norm(b) if normb != 0: tol = tol*normb

The larger that b is, then the more tol should be relaxed. Is this what you mean? If so, then you or I can just pop this code into multilevel.py at the start of the solve(...) routine.

Yep, that's all that is needed. Go ahead and make the changes.

Done

moved to github: https://github.com/pyamg/pyamg/issues