Geometry Optimisation Problem

Questions
#1

Hi Q-Chem developers!

I’m currently running multiple harmonic frequency calculations but I have found a problem with one particular molecule not converging properly (COClF - smiles code: C(=O)(F)Cl). Both the displacement and energy change converge to the desired value. However, the gradient gets close to converge but it doesn’t after all optimisation cycles; it basically stays in the same value pretty much through all the calculation (hope the data below helps visualise this):

     Gradient               0.000064	  0.000015	NO
     Displacement       0.000006	  0.000060     YES
     Energy change     -0.000000	  0.000001     YES

I’ve tried multiple things including using different basis sets, turning the symmetry off, and using alternative initial geometries (including a Z-matrix representation), but it just doesn’t work. I’m wondering if you guys could help with it? (it might not be a problem related with the program, but maybe there’s a keyword that I’m missing). Here’s a copy of the input file:

$comment
QCHEM - COClF
Geometry optimisation and vibrational frequencies
$end

$molecule
0 1
C 0.4911656443 0.1216536157 -0.0000048240
O 1.1314045225 1.1042136062 0.0000015411
F 1.0009499982 -1.1013888005 0.0000010333
Cl -1.2356929429 0.0205216859 0.0000004304

$end

$rem
JOBTYPE OPT
SYMMETRY FALSE
METHOD TPSS0
BASIS AUG-cc-pVTZ
SCF_CONVERGENCE 8
GEOM_OPT_MAX_CYCLES 100
GEOM_OPT_TOL_GRADIENT 15
GEOM_OPT_TOL_DISPLACEMENT 60
XC_GRID 3
$end

@@@

$molecule
READ
$end

$rem
JOBTYPE FREQ
SYMMETRY FALSE
METHOD TPSS0
BASIS AUG-cc-pVTZ
SCF_CONVERGENCE 8
XC_GRID 3
$end

Cheers,
Juan

#2

With the default optimization convergence criteria, I find that this optimization converges in 5 cycles. Therefore I think the problem is the very tight criteria that you have requested (GEOM_OPT_TOL…). In general there is no guarantee that optimizations can be converged arbitrarily tightly, is there a reason to tighten the convergence in this case? The defaults are fairly sensible in my experience.

1 Like
#3

Thanks jherbert! I was trying to get the best geometry possible for the harmonic calculations but you’re right, accepting the geometry is not totally converged but converged enough makes sense.

Cheers,
Juan

#4

I only mess with the GEOM_OPT_TOL… criteria if my frequency calculation returns a very small imaginary frequency. Even then, maybe reduce the defaults by factor of 2. The ones you tried are really tight.