In Gaussian, the totally energy of my system has been reported using "Sum of electronic and zero-point energy" (output from the FREQ). This value though, hasn't been scaled to take into account fluctuations at 0 K. I say this because a) I can't find a scale factor for the combination of wb97xd/6-311++g(2df,2p), and b) I am calculating relative interaction energies so my reported values are compared between different conformations rather than against experimental data.
However, I am currently running MP2/ag-cc-pvtz single point energy calculation to get a wave function perspective on the interaction energies (plus MP2 has been known to produce reliable energies for non-bonded interactions). But, without performing FREQ (which would a) require too much resources and b) a whole new OPT step) I can't take in to account the ZPE.
Therefore, would it be appropriate to report three relative interaction values:
1) wb97xd/6-311++g(2df,2p) non-scaled zero-point corrected relative interaction energy
2) wb97xd/6-311++g(2df,2p) single point interaction energy total energy (no ZPE)
3) MP2/ag-cc-pvtz single point interaction energy total energy (no ZPE)
I say this, because unless I am wrong, I don't think it's appropriate to compare single-point energy (which does not take into account zero-point energy correction) to an energy value which has taken account of the zero-point energy.
Thanks
Anthony,
if You use semi-empirical methods, then zero-point energy is included in final result automatically through empirical parameters.
Eugene,
Thanks for your answer. Given that MP2 is a post-Hartree Fock ab initio method, are you suggesting I consider a semi-empirical method like INDO, NNDO, MOPAC, AMPAC etc.?
Hi Smruti - I'm interested in ab initio calculation - not Semi-Empirical.
For me PM3 in MOPAC is the best initial approximation, Anthony, because I consider carbon nanostructures.
Hi Eugene - I am studying proteins. I will investigate which semi-empirical method would be best for accurate energy calculations for proteins. Thanks.
Proteins are too large for me, Anthony. Such problems must be solved by the parts. You can not see them as a whole. You can only choose and study interesting for You their properties.
In general zero point energy and electron-phonon coupling is the great problem, because Born-Oppenheimer approximation is not enough in this case.
Good luck!
Hi Eugene, my apologies, by protein I meant 'part' of a protein i.e., two sides chains and a sugar compound. Thanks for your suggestions.
Hello Anthony,
I think that some studies of Truhlar could interest you.
I hope it helps you
Well, in my opinion it is justified in many cases to combine the ZPE from DFT with ths single point energies from good ab initio calculations. DFT is known to give quite good frequencies. Besides, while computing relative energies of the conformers, you are left with only a fraction of an error (a difference of two rather similar values, both of which having more ore less the same error).
The wB97xD functional is not the best one for frequencies as it contains up to 100% of non-local exchange, so the scaling factor can be expected to approach that oh HF (about 0.92 - this is the value obtained by my students for a couple of small molecules :). You can use either the non-scaled values or use the estimeted scaling factor of 0.92 with the hand-wave argument concerning the HF-exchange contribution in the functional.
Good luck!
Marcin
Hello Anthony,
One possibility is make two suppositions: 1) DFT used gives reliable frequencies but not correct "electronic energy" which is not rare. 2) MP2 gives reliable electronic energy but not the frequency.
From these very simple suppositions you could do a DFT opt+freq. Then subtract their electronic energy and add the MP2 electronic energy. For example the scheme could be: E(system+ZPE)= E(DFT+ZPE)-E(DFT)+E(MP2) or the same with enthalpy or Gibbs energy. The calculation method could be: MP2/aug-cc-pVTZ // wb97xd/6-311++g(2df,2p).
Truhlar has some studies where different levels of calculus are used, extrapolation methods with different levels in same molecule, etc... (actually I do not have access to cite properly).
Also you can think in ONIOM calculations for example or make your own scheme of calculus or use one of Truhlar.
Other think that you could do to perform the MP2 energy is an extrapolation. In this case the best option should be TZV and QZV, but if QZV is too expensive, an extrapolation with DZV and TZV should be better than only TZV. Exist a lot of extrapolation methods, for example, for two variables exist the Parthiban and Martin Exponential formula (2001) and other....
I hope it helps you
Hi Joaquim, that was an excellent answer! Thank you for your help. I've currently got some SPE calculations of MP2/aug-cc-pVTZ running at the moment over 48 cores. My earlier attempts at 12 cores over 48 hours didn't complete in time!
Thanks again.
Hello,
Depending on the time with 48 cores you could think in QZV and make the extrapolation. Sincerely, I think that extrapolation results are very good compared with only one result and one calculation. The time is more expensive but, in my opinion, it is a good inversion. For example, in a CCSD(T) calculation the difference is remarkable but in the case of MP2 (which is much less accurate) could be very important and more, with MP2 you can consider variational theorem so the energies extrapolated always will be lower than your single point calculations. If you have some reference value and you can compared, first you could try with CBS (D-T)ZV and if it is not enough then try with (T-Q)ZV extrapolation.
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Energy