[lammps-users] minimization


I have a question about conjugate gradient minimization procedure: the forces on
atoms, espicially free surface atoms, are not close to zero. For example, the
forces on interior atoms are 10^(-4)ev/A(metal units), but the force normal to
the free surface can reach 0.1ev/A. The maximum iterations and force
evaluations are set to be very large and are not excedeed in simulations. Since
the force is closely associated with the gradient of the energy, do you think
the conjugate gradient method works well for my case?

Besides, there are other two criteria to stop the minimization
1. the change in energy between outer iterations is less than the tolerance
2. the 3N dimensional force vector goes to zero
I decrease the tolerance but cannot improve the result.

What kind of modification I should make if I want the second criterion (the
force vector goes to zero) to be the only one can stop the minimization?

If you have a large system the stopping criteria are averaged
over the whole system and can be a problem. You might
try a 2nd min where you freeze the bulk and just relax
the surface and see if that helps.

You can also run dynamics with fix viscous and damp out
all the velocity which should relax all forces.