Professional Documents
Culture Documents
Gaussian 16 Frequently Asked Questions
Gaussian 16 Frequently Asked Questions
com
News
Products
Support
Our People Search
Orders/Contact
I optimized a structure, then calculated the frequencies for it. The frequency calculation
showed the structure was not converged even though the optimization completed. Is my
structure reliable?
The structure you have is likely not reliable and can lead to incorrect results, including incorrect frequencies. If the frequency calculation does
not say “Stationary point found.”, then the structure is not a true stationary point: minimum, transition structure, or saddle point. This is true
even if the number of imagnary frequencies is correct (e.g., 0 for a minimum). In summary, the short answer is No. This question addresses
what this means in detail, and tells you what you can do about it.
You can run the frequency calculation automatically using Opt Freq in the route section. This allows you to verify that the final structure is a
true stationary point. If it is not, then it is not a true minimum (or transition structure); though, it is likely to be very close to one.
The following example illustrates the discrepancy. This first output section is from the optimization, performed using a DFT method:
This second part is from the frequency calculation at the “optimized” structure (performed using the same model chemistry):
As you can see, the frequency calculation indicates that the structure is not converged even though the optimization completed successfully.
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.
In this case, the displacements are different between the optimization and the frequency calculation because they rely on the Hessian: the
matrix of second derivatives of the energy with respect to displacement of the atoms. In a frequency calculation, this is calculated analytically
https://gaussian.com/faq3/ 1/3
9/10/21, 9:40 AM Gaussian 16 Frequently Asked Questions | Gaussian.com
for methods for which second derivatives are implemented (as is true for the DFT example here). In contrast, an estimated Hessian is used in a
geometry optimization unless you explicitly request a computed one using the CalcFC or CalcAll option to the Opt keyword.
In this example, the results of the convergence tests differ, leading to convergence in one case, but not the other. Clearly, the exact Hessian
from the frequency calculation is more accurate than the estimated Hessian from the optimization, so the frequency results are the ones which
indicate the true status of this molecular structure: very near a minimum but not a true minimum. As is well known, frequency and
thermochemistry results are based on a harmonic analysis that is only valid at true stationary points. Accordingly, some results will be
incorrect at non-stationary points.
Simply start a new calculation using the checkpoint file from the previous one, using a route section like the following:
%Chk=myfile
These changes tell Gaussian to read the geometry and initial guess orbitals from the checkpoint file, to read in the force constants (Hessian)
calculated in the frequency job, and then to do an optimization followed by a frequency calculation. You should check again at the end of the
second calculation to make sure that a stationary point is found.
In Gaussian 16, the default grid is the UltraFine grid (a 99,590 grid), which can help obtain smoother convergence to the stationary point in
cases where the potential energy surface is very flat. It is also useful for reducing numerical instability in general (e.g., DFT calculations in
solution), making such calculations generally more well behaved at a modest increase in computational cost. For these reasons, it was chosen
as the default grid for Gaussian 16.
Gaussian 09’s default grid is the smaller Fine grid; some Gaussian 16 users elect to use this grid as well. If this is the grid you are using, try
rerunning the calculation with Int=UltraFine.
Since modification of the numerical integration grid will bring changes in the predicted total energy, it must be considered an integral part of
the model chemistry. Thus, it is necessary to the same integration grid in all the calculations in the same study in order for the computed
energies and molecular properties to be comparable.
See the discussion of the Integral keyword for more information about numerical integration grids.
References
Keyword
[Keyword=]Option
https://gaussian.com/faq3/ 2/3
9/10/21, 9:40 AM Gaussian 16 Frequently Asked Questions | Gaussian.com
Follow
Follow @gaussian
https://gaussian.com/faq3/ 3/3