Hello!
In the messages that deal with the optimizer efficiency the bottom line seems to be to select Quick-Min when the forces are high or very small; and CG or LBFGS when the forces are moderate. Could someone elaborate what sort of forces are considered high or low in this context? Any examples of order of magintude? Does the number depend to the number of degrees of freedom?
Especially what forces are low enough that a switch to Quick-Min may be in order, as high forces usually present less of a problem for convergence.
What are the recommended switching thresholds for both "Force" and "Torque" in the case of Dimer method?
Thank you in advance-
Vlad
Choice of optimizer vs. force magnitude
Moderator: moderators
Re: Choice of optimizer vs. force magnitude
A good rule of thumb is that the forces should be ~ 1eV/Ang before switching to a second order optimizer such as CG or LBFGS. As long as your forces are accurate (small ediff) you can run until convergence with these. If you hit the noise in your forces, you can increase their accuracy or switch to a first-order optimizer such as quick-min, but the former option is preferred.
I would not say that this criterion is very sensitive to the number of degrees of freedom.
The same criterion would apply for the force on the dimer. There is an additional requirement to allow the dimer to rotate enough (DRotMax) or move slow enough (MAXMOVE) so that the dimer rotation is on the order of a few degrees each time the optimizer moves the center of the dimer.
I would not say that this criterion is very sensitive to the number of degrees of freedom.
The same criterion would apply for the force on the dimer. There is an additional requirement to allow the dimer to rotate enough (DRotMax) or move slow enough (MAXMOVE) so that the dimer rotation is on the order of a few degrees each time the optimizer moves the center of the dimer.