The fix ave/spatial method will give the correct total surface
tension. Local stress is more tricky but the fix ave/spatial method should
be fairly close. If you want a more exact answer you can try Hardy's
approach (or the Irving-Kirkwood method). Depending on which
component of the pressure tensor you need, you could also try the
method of planes (MOP)
http://link.aps.org/doi/10.1103/PhysRevE.52.1627 which wouldn't be too
hard to implement in LAMMPS. It can't give you all the components of
the pressure though.
Stan