I'm currently doing some partime-DFT calculation with Gaussian. I have some questions regarding the pre-calculation step. Please help me make it clear from a practical point of view.
1. I understand that the result of optimization is a local minimum. And from what I observed, the output configuration is not much different from the input one, it means the local minimum is very close to starting point.
So what could be done to approach to the global minimum (for complicated molecules, with 100's of atoms)?
Should we just draw the molecule with Gausview, let the Gausview auto-locate the positions, distances of atoms then set up calculation with DFT and basis sets (like LANL2DZ, 6-31G...)?
Or should we run a optimization at Hartree-Fock level first, then use that optimized configuration as input for DFT calculation?
Or should we draw molecule in other software, like Jmol, Avoga, try to optimize using these software then use that output configuration as input for DFT?
2. If we start with the configutation in the cif. file, do we need to run optimization before start single point calculation?
3. From your experience, how good is the agreement between theoretical and calculation, in term of bond distances, bond angles. How close is considered accepted? Should we pay attention to this comparison?
Hello.
The problem with large molecules isn't the size, but the computational effort to search the local minima.
I recommend you to draw the molecule in a software (e.g. MOE), and minimize it. As such, you will have the minimum for that particular program. The problem is: that minimum can be very far from the local minimum calculated in Gaussian. The calculation time will be proportional to that difference.
In Gaussian, i've done minimmizations with almost 120 atoms in a 12-core computer, and took 2 days processor time (12 hours).
Best regards.
Thank you, Ricardo for your useful comment.
And how about using .cif file as input? should it be considered optimized configuration already?
And how is your experience regarding the agreement between expt data and theoretical calculation data.
Could you please share!
Thank you!
Hello.
Again, the problem isn't the input. Is the conformation of the molecule within.
You must start with a minimized molecule, and that will depend of the program used to minimize.
You can do a semi-empirical first (with MOPAC), and then go to Gaussian to further refinement. Or, you can simply minimize it in a program (MOE, ChemDraw, etc), and hope that it will be good for Gaussian.
Best Regards.
Keep in mind that there is a *huge* difference between a local minimum and a global minimum. A global minimum means that structure is the minimum of all conformers, at the very least. Most optimization algorithms will find the nearest stationary point which should be confirmed as a local minimun or saddle-point with a frequencies run.
It is common to "pre-optimize" to reduce computational cost (in time), and as such this step is done with a lower level of theory than the actual calculation. This is expected to provide a better starting guess to the calculation at the desired level of theory, but you could also have the opposite (a worst guess) happen, although less common.
Thank you, Ricardo and Daniel for your interesting comments.
Please let me ask the last question regarding the agrement between theoretical calculation and experiment data (X-ray diffraction), how is your case?
in my case, some time the different in the distance might be up to 1Ao, how about your experience? What is the difference "limit", beyond that limit the calculation is considered "something wrong".
Thank you!
It depends on how much you are willing to spend computationally. But even DFT results can be quite good, with deviations of less than 3 pm.
You might want to have a look on this article:
Bühl, Michael, e Hendrik Kabrede. “Geometries of Transition-Metal Complexes from Density-Functional Theory”. J. Chem. Theory Comput. 2, nº 5 (2006): 1282–1290. http://dx.doi.org/10.1021/ct6001187
- Badges
- Science topic