Ways to deal with non convergence in optimizing the structure and get the maximum optimization cycle

Questions
#1

ways to deal with non convergence in optimizing the structure and get the maximum optimization cycle

#2

The keyword GEOM_OPT_MAX_CYLCES allows you to change the number of maximum optimization cycles. But in order to improve convergence while optimizing a structure there are several approaches you can take.

  1. Try to optimize the structure with a very simple basis set (like 6-31G), that way the computational cost is low, then once that structure has been optimize increase the basis set to the desired basis with the converged structure.
  2. Just start the optimization over from the last known structure, increasing optimization cycles.
  3. Start with a frequency calculation then read in the Hessian for optimization, rough input below. *Warning this can be computationally costly depending on size of system and basis set.
$molecule
...
$end
$rem
jobtype = freq
...
$end

@@@

$molecule
read
$end

$rem
jobtype = opt
geom_opt_hessian = read
...
$end
#3

In addition to Peter’s suggestions, one last thing to check is what coordinates are being used. By default the optimization is performed in “delocalized internal coordinates”, as that is MUCH more efficient than other systems. However, this requires an iterative back-transformation at each step that can fail, in which case Q-Chem switches automatically to Cartesian coordinates and continues the optimization. Cartesian coordinates are strongly coupled and therefore optimization is slow, and I’ve found that your best bet in such cases is to halt the optimization (i.e., kill the job) and restart, because this sometimes restores internal coordinates. (This is a known issue with the optimizer that will be fixed some day, I hope.)

The check for this is to search for “GEOMETRY OPTIMIZATION IN CARTESIAN COORDINATES”.