UT theoretical chemistry code forum

Dear Administrator,

I have few dimer runs, which has really zero energy gradiance. I took this and did a single point geometry optimization using VASP and found higher energy gradiances of about 0.04 eV/angstrom.

Why is this happening? Or am I doing some thing wrong?

This could be our error. Our optimizers come before the convergence check. In earlier versions of the code, the optimizers take an extra step just before vasp decides that the forces low enough for convergence. For quick-min, this is not a big problem, but for conjugate gradients and lbfgs, they take a finite difference step which move the configuration a significant distance, and could increase the force by 0.04 eV/Ang.

You can check if this is the problem by comparing the last position listed in the OUTCAR with the coordinates in the CONTCAR/CENTCAR. If they are different, the last coordinates in the OUTCAR will be the saddle.

This problem has been fixed in the current version, v2.01a, (28/03/07).

One last possibility, if the dimer run was killed before convergence, is that the CONTCAR is displaced from the center of the dimer, by the finite difference step. If this is the case, use the CENTCAR, which always contains the best estimate of the saddle.

Dear Administrator,

Still I face the similar problem of increased energy gradiances when I do a single point calculation to check the forces.

Do you have any new version of the dimer method where this has been taken care? Please do let me know, so that I can use it.

Thanks for all your valuable support

The center of the dimer is now being properly written to the CENTCAR. This has been fixed in v2.03.

Dear Administrator,

I have downloaded and used the new version 2.03. Still the problem of having different energy gradiances for dimer output and for single point output, exists. Please check again. The problem still exists

I've tested v2.03 more carefully, and I don't see this error. It would be very helpful to have a (modest) calculation which demonstrates that this error is still present.

I've updated the vtstcode to 2.03a in which an additional line is printed in the DIMCAR. This does not change the OUTCAR or the final CONTCAR/CENTCAR. The only difference is an extra line in the DIMCAR which prints the final force and energy on the converged geometry. If you were comparing the force and energy in the DIMCAR with a subsequent single point energy calculation, this may be resolved, although nothing is really different. If the problem is still in your runs, however, we need to fix it.