Gaussian calculation problem - Maxcycle values for Opt and SCF

Question

I've been trying to calculate vertical ionization potentials (IP) and electron affinities (EA) for a bunch of prostaglandin derivatives (23-25 heavy atoms). My calculation setup in Gaussian09 looks mostly like that:

# opt=(tight,maxcycle=1000) 
freq=noraman cphf=noread 
b3lyp/6-31g(d) geom=connectivity 
integral=grid=ultrafine scf=maxcycle=1000

As you can guess, I have been struggling with converging the geometry and finding the stationary point for my molecules. Some were kind of stubborn and wouldn't converge, so in desperation I increased the Maxcycle values for Opt and SCF from a 1000 up to 5000, and I succeeded with some of them. Now, my question is if I need to recalculate all other prostaglandins that have already nicely converged with lower Opt(Maxcycle) or SCF(Maxcycle) options (I used them in different combinations like 1000, 3000 and 5000)? The thing is that I want to be able to objectivelly compare the IP and EA values between different derivatives.

I am concerned because I remember that someone told me that if I went for SCF=QC option I would need to recalculate the whole set of molecules with the same keyword option. However, as far as I know SCF=QC is like a substantial change in the setup as it uses another algorithm for calculations. Hopefully, changing the SCF(Maxcycle) value is not that dramatic, is it? How about SCF=Verytight - would it demand recalculating everything or not?

Another thing that bothers me is Opt=Tight. What if I decided to go for Opt=Verytight. Would I need then to recalculate all of the molecules with the same setting?

And lastly, what if I changed Integral(grid=ultrafine) to Integral(grid=ultrafine, acc2e=12) which is supposedly the most precise approach?

Answer 1

You are mixing two independent, but slightly related topics.

First is the SCF convergence, i.e. obtaining the wavefunction. This step is necessary for any calculation, without it the Gaussian crashes. For well behaved molecules, as the prostaglandines (organics) most probably are, the solution should be found fast (<30 SCF cycles). It also does not matter which converger you choose, i.e. QC or DIIS. How "tight" you decide for the SCF affects the final energy, but only the number of digits (precision). The SCF does not have any "verytight" option, according to manual, only the Conver=N.

On the other hand, the integral and grid settings do have effect on the energies, so all calculations have to be done with the same grid. I won't recommend changing the acc2e settings, the error is negligible.

Regarding the geometry convergence, you should obtain an optimized geometry in less than 100 steps for reasonable organic systems. If this does not happen, something is rotten in your setup and you should look deeper into the problem. In principle, you cannot converge the geometry to "tight" without sufficient accuracy of the SCF, namely the ultrafine grid, which you now do use.

How good the optimization was can be guessed by the frequencies which should be zero (translations and rotations). If they are significantly off, optimize the geometry more tightly. But as said above, you cannot optimize on noisy surface, where the noise is produced by the grid and SCF convergence.