Convergence Failure in Constrained Optimization

Questions
#1

Hello, We are trying to optimize the geometry of an organic molecule by the constrained method. The distance between two unconstrained atoms in equilibrium geometry is 12.05 angstrom and we want the constrained optimized structures by increasing the distance between two terminal anchor atoms in increment of 0.05 angstrom. The structures are fairly optimized up to the distance between the terminal anchor atoms of 14.85 angstroms. But beyond that, the optimization gives convergence failure. We want it up to 20.30 angstrom. Is there any solution for that? We have also increased the MAX_SCF_CYCLES and also tries with different SCF_ALGORITHM, but it gives the same problem. Error is always “Maximum optimization cycles reached”. Input is as follows:

$rem
JOBTYPE Opt
EXCHANGE B3LYP
BASIS 6-31G(d)
GUI=2
MAX_SCF_CYCLES 500
SCF_ALGORITHM = diis
THRESH 9
symmetry false
sym_ignore true
$end

$comment
constr-opt-100
$end

$molecule
0 1
C -0.76710 2.73128 0.25250
C -0.70554 3.40362 1.65187
C 0.68435 3.98698 1.28792
C 0.79473 2.94944 0.14414
H -1.27306 3.35601 -0.48642
H -0.63066 2.64483 2.43503
H -1.51185 4.10193 1.89264
H 1.46173 3.92718 2.05179
H 0.62528 5.00836 0.90075
H 1.12366 3.30257 -0.83791
C -1.29788 1.32086 0.17566
O -1.06080 0.44794 0.98432
O -2.07265 1.14147 -0.92232
C 1.61785 1.73141 0.51532
O 2.18135 1.55218 1.57124
O 1.64419 0.85189 -0.51768
C -2.60414 -0.18258 -1.12653
H -2.77893 -0.26301 -2.20141
H -1.87196 -0.92895 -0.81122
C 2.26687 -0.42213 -0.25802
H 1.82584 -1.11117 -0.98151
H 2.02683 -0.74901 0.75566
C 3.77276 -0.32789 -0.46044
H 4.23617 0.31632 0.29040
H 4.01545 0.06169 -1.45387
C -3.90638 -0.34376 -0.35446
H -3.73555 -0.29697 0.72421
H -4.62752 0.43262 -0.62742
O 4.26974 -1.67321 -0.32621
C 5.60869 -1.82278 -0.48593
O 6.35471 -0.90523 -0.74195
O -4.41856 -1.63983 -0.71402
C -5.61010 -1.97409 -0.15508
O -6.22266 -1.24250 0.58852
C -6.03759 -3.35581 -0.58930
H -7.00540 -3.59020 -0.14539
H -6.10469 -3.40606 -1.68097
H -5.29460 -4.09545 -0.27290
C 6.01181 -3.26659 -0.30113
H 5.47846 -3.90254 -1.01528
H 7.08779 -3.36341 -0.44825
H 5.74165 -3.60654 0.70420
$end

$opt
CONSTRAINT
stre 35 39 17.103
ENDCONSTRAINT
$end

#2

Is there a reason you are using thresh = 9? This is an extremely tight SCF convergence criteria, especially for a large floppy molecule such as the one you have provided. If you use a thresh = 8 which is default for geometry optimization, the geometry optimization job completes in 82 optimization cycles for the job you provided.

@yzhang is looking into the SCF convergence of this job, to try help provide possible solutions for converging the SCF to thresh = 9.

#3

Thank you for your reply. We have also tried without thresh keyword, it gives the same problem.

#4

So I ran the job without thresh = 9 adding in geom_opt_max_cycles = 100 (see $rem below) and the optimization job converged with no issues. Do you not experience the same? If that is the case, can you copy the failure from the output file? What version of Q-Chem are you using?

$rem
JOBTYPE Opt
EXCHANGE B3LYP
BASIS 6-31G(d)
GUI=2
SCF_ALGORITHM = diis
geom_opt_max_cycles = 100
symmetry false
sym_ignore true
$end
#5

I tried to set thresh = 14 and find no SCF convergence issue. When you set thresh = 9 for such a optimization job, you are actually lowering the integral threshold than the default (= 12 for geom optimizations). Why do you want to lower the integral threshold?

#6

Thank you both for the solution. Herewith I have copied the failure from outputs
(without geom_opt_max_cycles = 100):

  1. constr-opt-without_thresh.out
  2. constr-opt-thresh14.out

Nuclear Repulsion Energy = 1722.3688911456 hartrees
There are 84 alpha and 84 beta electrons

mpirun noticed that process rank 2 with PID 9449 on node node34 exited on signal 6 (Aborted).

remove work dirs /scratch/qchem9433.0 – /scratch/qchem9433.3
rm -rf /scratch/qchem9433.0
rm -rf /scratch/qchem9433.1
rm -rf /scratch/qchem9433.2
rm -rf /scratch/qchem9433.3
rm -rf /scratch/qchem9433

And there is no particular reason for using lower integral thresold. First, we have not found the minima without thresh and after that we try it by decreasing the thresh. Our error was not to add ‘geom_opt_max_cycles = 100’ in $rem section. And we are using Q-Chem version: 5.1.2

#7

Those files did not attach, if you can send them to support@q-chem.com we can provide additional assistance.

Both @yzhang and myself were able to perform this job by simply removing the thresh = 9 and adding geom_opt_max_cycles = 100. If that did not work for you send us that output to our support as well.

#8

Peter, why do you say that THRESH=9 is extremely tight? For geometry optimization the default value is THRESH=11 (corresponding to SCF_CONVERGENCE=8), but in my experience for large molecules sometimes even that is not tight enough. We have better luck with THRESH=12 (or THRESH=14 as the user says, but at somewhat increased cost).