MECI energy convergence

Questions
#1

I am currently using Q-Chem 5.4 in order to calculate MECI using SP-TDDFT. I have checked that the criterion that is used to converge a calculation is based on the gradient, energy change and atomic displacement. I have been doing it for many small molecules and in many cases what has happened to me is that the calculation is converged in terms of displacement and gradient but not in energy. Even in that case, it is considered by the program to be converged, as shown below:

                         Maximum     Tolerance    Cnvgd?
     Gradient           0.000204      0.000300     YES
     Displacement       0.000277      0.001200     YES
     Energy change     -0.000184      0.000001      NO

Final energy is -499.68276174754624

** OPTIMIZATION CONVERGED **

These are the keywords and instructions of my input:

$rem
JOBTYPE = opt
MECP_OPT = true
SPIN_FLIP = true
MECP_METHODS = mecp_direct
GEOM_OPT_MAX_CYCLES = 1200
CIS_N_ROOTS = 5
MECP_STATE1 = [0,1] ! Looking for state of charge 0, 1st singlet ie/ the GS !
MECP_STATE2 = [0,2] ! Second singlet
METHOD = PBE0
BASIS = 6-31+G*
SCF_CONVERGENCE = 9
SYM_IGNORE = true
$end

Is there anyway of making the calculation converge also in terms of energy?

Thank you

#2

There is no way to include energy in the converge criteria in a simple way. The way that the convergence criteria is checked is that the absolute value of the maximum gradient component is lower than the tolerance [ < 0.000300] and either the absolute value of the maximum displacement [ < 0.001200] OR the energy change is lower than the tolerance [ < 0.000001].

So in this case that you have provided, the code is working as intended saying the the optimization is converged as the gradient and displacement have both converged. You can enforce tighter convergence criteria by changing the rem variables found here Optimization Controls

specifically, the follow variables:

GEOM_OPT_TOL_GRADIENT 
GEOM_OPT_TOL_DISPLACEMENT  
GEOM_OPT_TOL_ENERGY

which change the tolerances for convergence. So if you want the energy to converge over the displacement, then you should tighten the displacement tolerance so that way the energy is the only thing that can trigger the convergence:

GEOM_OPT_TOL_DISPLACEMENT = 100

BUT, this could cause your optimization calculation to take far longer to converge or may never converge. Since your displacement is already an order of magnitude smaller than the convergence criteria [ 0.000277 << 0.001200 ]. In addition, since the energy change convergence criteria is on the difference of the total energy, therefore with the energy of your system on the order of 5e2, the resulting energy change on the order of 1e-4 is reasonable. With 1e-6 may never be obtainable in a computationally reasonable time, particularly if the gradient and displacement are already significantly small/converged.

1 Like
#3

https://manual.q-chem.com/5.4/sect_MECP.htmlI agree with everything Peter said, for both geometry optimizations and for MECP optimizations. For the latter, I would just like to add that the energy gap is part of the objective function that is being optimized.

Discussion of MECP-optimization algorithms in the manual: 9.8.3 Minimum-Energy Crossing Points‣ 9.8 Nonadiabatic Couplings and Optimization of Minimum-Energy Crossing Points ‣ Chapter 9 Exploring Potential Energy Surfaces: Searches for Critical Points and Molecular Dynamics ‣ Q-Chem 5.4 User’s Manual

1 Like