Why the AtomicForce is bigger than the opt setting ?

[fangyongxinxi]

Dear Sr,
 
       We want to get the AtomicForce of a single atom wire like below. Frist, opt this model ( constraint all the atoms except the center atom line ) with the setting: 0.05eV/A, success, and get the opt.vnl and opt.nc. Then , we calculate the AtomicForce using the configuration in opt.vnl, but get the AtomicForce is bigger (some atoms in the line is 1.223*e-2Ry/B=0.3eV/A  ) than the setting value. we restore the opt.nc file to calculate the AtomicForce, get the similar result. But testing the lih2li, get the right answer.
        Why this happens ?
        

[Anders Blom]

There could be at least two things at play. First, the relaxation is carried out with periodic boundary conditions (equivalent bulk), but the forces computed with open (two-probe). On the one hand, since the methods are a bit different, different forces can be expected. Also, you mentioned that you have constrained some atoms in the optimization (which is natural), so the system is not completely relaxed. Also, I would generally be cautious to trust the absolute forces from the two-probe calculation to a high degree, but rather look at how they might change as a bias is applied, or some other change made to the system. Some improvements of the force calculations for two-probe systems can be expected in upcoming releases of ATK.