UT theoretical chemistry code forum

Dear VTSTT-Developers,

at which points are the energies evaluated in the current version of VTSTTools? Is the energie at the midpoint still interpolated as described in the original paper of Prof. Henkelman, or is the VTSTTools implementation following the approach by Heyden et al. where energies and forces are evaluated at the midpoint?

(What about the dimer-implentation in the EON code?)

I'm also interested in information about the maximum distances of the two dimer images. From the posts in this forum, it seems to me that separations large than 1e-2 A are strongly discouraged from being used. Why that? Simply beacuse the finite differences become too bad, or is it, beacuse the energy and forces at the dimer-midpoint (which finally converges to the transition state (TS)) are interpolated and therefore, for too large separations, the midpoint cannot converge very exactly to the TS?

Best,
Markus

Both use forward difference, with the energy and force evaluated at the midpoint and a displaced point.

Since the energy and force is evaluated at the midpoint, the finite difference length does not effect the final energy and force. As you say, it leads to a poor estimation of the curvatures which decreases the efficiency of the method.

Dear Prof. Henkelman,

thank you very much for the answer.

Are there any important differences in the implementations of the dimer-method between the EON and VaspTSTTools; or differ both implementationts in the same details from the original description (JCP 111, 7010 ) of your method? Cold you give a short list of differences between the current implementations and the original paper?

Thank you very much.

Best,
Markus

The most important changes are

(i) forward difference [J. Chem. Phys. 121, 9776 (2004)]

(ii) reduction of a force call in the improved dimer of Andreas Heyden [J. Chem. Phys. 123, 014106 (2005)]

(iii) LBFGS for rotation and translation as described by Johannes Kaestner [J. Chem. Phys. 128, 014106 (2008)] as well as a small trick to efficiently check for rotational convergence

(iv) solid state dimer to allow for variable cell [J. Chem. Phys. 140, 174104 (2014)]

The eon and vtstcode implementation should be equivalent, but it has been on my to-do list to verify this (for a long time now).

thank you very much