Optimization completed on the basis of negligible forces

Question

I am performing relaxed scan in Gaussian (B3LYP/6-31G(d,p)). There are two small organic molecules, one of which is a radical, the other has a benzene ring.

I'm trying to understand the optimization output. Specifically, the line Optimization completed on the basis of negligible forces. I understand that these forces are the maximum and RMS forces above.

Some examples of "negligible forces" from my output file:

        Item               Value
Maximum Force            0.000004
RMS     Force            0.000001
Maximum Displacement     0.019917
RMS     Displacement     0.004747

        Item               Value
Maximum Force            0.000002
RMS     Force            0.000000
Maximum Displacement     0.002231
RMS     Displacement     0.000490

        Item               Value
Maximum Force            0.000002
RMS     Force            0.000001
Maximum Displacement     0.003132
RMS     Displacement     0.000805

However, some small values like

        Item               Value
Maximum Force            0.000002
RMS     Force            0.000001
Maximum Displacement     0.000666
RMS     Displacement     0.000149

are followed only by Optimization completed. So what does "negligible" mean? How small do they need to be to say they are negligible? Or is this connected to at least one NO in convergence criterion section?

Next question, how reliable is optimization when such message pops up? If it's unreliable, what do I do about this? Would altering the initial geometry fix it when in only happens in a couple of scan steps?

Answer 1

According to the Gaussian Frequently Asked Questions:

In general, a stationary point is found when one of two criteria is met for a point in an optimization calculation:

• All four values listed in the output are smaller than the indicated thresholds.
• The Maximum Force and RMS Force are two orders of magnitude smaller than the thresholds shown, regardless of the values of the displacements.

You indicate three molecules where the second criteria activate - the forces are really, really small. So the optimization gives up.

In your final case, you can see that the displacements are also small, so the first criteria activate.

You then want to know how "reliable" the geometries are when the forces are small, but the displacements are not. That depends a lot on your needs. The optimization quits because such small forces imply a very flat potential energy surface and it may take a long time before "complete" convergence. My guess is that you still have some imaginary frequencies, but that might not matter depending on your needs.

It's worth pointing out that many quantum chemical programs have different criteria for declaring an optimization "finished." (In other words, your calculation in a different program might be declared done because of negligible energy and force/gradient change.)

If you want to push for a true minima, you should take that optimized geometry, run a force / frequency calculation to get an accurate Hessian, and save the result to a checkpoint, then re-start the optimization using the saved Hessian:

%Chk=myfile
# B3LYP/6-31G(d,p) OPT=ReadFC Freq Geom=AllCheck Guess=Read

Of course, that means probably breaking your relaxed scan calculation into a set of separate steps. (Honestly, this is how I usually do relaxed scans anyway - a set of separate jobs with each stepped geometry.)

One other very important point. If you're doing a scan of intermolecular interactions with B3LYP, you really should be using a dispersion correction such as EmpiricalDispersion=GD3BJ. I often see incorrect repulsive potential energy curves from B3LYP without dispersion corrections.