It's probably a bit hard to tell if you want to go into the characteristics of the specific methods you're using. You might perhaps find some rough explanations in their basic behavior if you dig deep enough, but even then, it's questionable how useful such interpretations would be. Contrary to HF, DFT methods try to capture electron correlation effects by empirical parametrization, so putting up one against the other might be a bit of an unfair comparison. But the way that DFT models are set up in my view also discourages any deeper analysis of their results, because they were basically "made to work" for some specific problems by clever choice of parameters.

Some general thoughts on your models however... if you want to benchmark your interactions on model systems, you should of course strive for the most accurate methods you can afford. As mentioned before, HF is definitely not the end of the flagpole here - rather just the starting point for more accurate things because electron correlation effects are almost fully absent in it. Also, 6-31+G(d) as a basis set is really comically undersized by today's standards, especially if you're just working with small model systems. So that would be one point you could try to improve apart from the method. With your methods, you're basically looking at the results that two fundamentally deficient approximations in your Schrödinger equation (both for the Hamiltonian and the wavefunction) give you. Whether that serendipitously holds any merits is hard to tell from the results alone. Quantum chemistry tends to be intransparent like that, I'm sure you've made the experience before.

Again, ab initio methods really gain a lot if you move away from that quantum chemistry package to others which provide much more efficient algorithms and approximations. I could look into some more advanced things if you want.

Dear Anthony,

To your calculations, I would like to add one with MP2  (result attached).

In my opinion, it is only the effect of introduction of correlation functional (or correlation interactions for MP2).

A thing that would be interesting if you can add here the energies of each method and the distances of N-H and O-H.

Hi Joaquim, 

Many thanks for your answer. I'll see if I can get hold of the energies and distances. 

What system size do you have in mind for your calculations? That will obviously decide much of what you can or can't try.

Completely independent of size, you'll always want to use dispersion correction for B3LYP. The keyword would be "empiricaldispersion=gd3bj" (unless you're in G09 RevD, where it was buggy). Otherwise, the results will be fundamentally deficient because dispersion is not included in B3LYP at all. - The same is true for HF actually, together with a more fundamental lack of correlation effects; so as Joaquim said, you should look at post-HF methods. But they can become too demanding if you're not doing some better implementations, none of which are available in the software package with the big "G".

Hi Matthias. It was just two amino acid backbones (side chains capped with -H; essentially glycine then). Each backbone was charged, so COO- and NH3+. I ran three quick basic calculations, independent of one another:

HF/6-31+G(d)

B3LYP/6-31+G(d)

WB97XD/6-31+G(d)

I noticed that in both B3LYP and WB97XD, the backbones do not hold the charge i.e.COO- and NH3+ quickly exchange to become COOH and NH2.

I was just wondering, what in the theory of hybrid DFT, enabled that behaviour, that HF would not.  

Just for context, this is a study where I would be taking "rough-estimations" of a post-translational amino acid into a molecular dynamics simulation. 

Thanks

It's probably a bit hard to tell if you want to go into the characteristics of the specific methods you're using. You might perhaps find some rough explanations in their basic behavior if you dig deep enough, but even then, it's questionable how useful such interpretations would be. Contrary to HF, DFT methods try to capture electron correlation effects by empirical parametrization, so putting up one against the other might be a bit of an unfair comparison. But the way that DFT models are set up in my view also discourages any deeper analysis of their results, because they were basically "made to work" for some specific problems by clever choice of parameters.

Some general thoughts on your models however... if you want to benchmark your interactions on model systems, you should of course strive for the most accurate methods you can afford. As mentioned before, HF is definitely not the end of the flagpole here - rather just the starting point for more accurate things because electron correlation effects are almost fully absent in it. Also, 6-31+G(d) as a basis set is really comically undersized by today's standards, especially if you're just working with small model systems. So that would be one point you could try to improve apart from the method. With your methods, you're basically looking at the results that two fundamentally deficient approximations in your Schrödinger equation (both for the Hamiltonian and the wavefunction) give you. Whether that serendipitously holds any merits is hard to tell from the results alone. Quantum chemistry tends to be intransparent like that, I'm sure you've made the experience before.

Again, ab initio methods really gain a lot if you move away from that quantum chemistry package to others which provide much more efficient algorithms and approximations. I could look into some more advanced things if you want.

Some very interesting advice. So, my contract comes to an end at which time I will no longer have access to that software package. However, over the last 4 years, I've picked up enough spare parts to build my own small cluster (just 3 nodes of 16 cores on each) running Linux. any recommendations for a free Quantum Chemistry package, ideal for biomolecules and with a good user interface?

Well I have to admit, I've never been strong on the "biomolecule" front since my work so far has mostly been concerned with accurate methods for small-ish systems. In terms of "free", Orca might have some things for you to look at. It's not really strong on the user interface front, nothing will beat the "Let's ban our Nobel laureate company founder from using our software" guys in that regard I guess. But it would be the first thing that comes to my mind. Especially since Orca also includes some rather cheap semiempirical methods designed by Grimme, maybe that would work for you.

Thanks, Matthias. I'll give Orca a try when I get the chance. 

Hi Anthony,

as you have noticed in your studies,

in both B3LYP and WB97XD, the backbones do not hold the charge i.e.COO- and NH3+ quickly exchange to become COOH and NH2.

The same was observed by our group while performing DFT studies involving biomolecules. So, we took uncharged groups, on the backbone and side chain groups were left charged depending on its physio-chemical property. 

like side chain of Lys residue was protonated so that it can acquire charge. Also we observed that this charged side-chains plays crucial role in the catalysis of reaction.

We have gaussian 09 in our lab. not sure of any freeware DFT calculating program.

But surely, G09, gave accurate result in predicting TS, intermediates as well as final product which were found during one-pot synthesis method using biomolecules.

Hi Tanuj,

I too have been having some big problems with holding the charge in the correct place of a biomolecule. You may have seen a number of posts where I am trying to capture the corresponding transition state for the protonation/deprotonation of a biomolecule (a glycated lysine-side chain). I've been using WB97XD functional and I've yet to find the corresponding TS. 

Dear Anthony,

As you are a gaussian user, I recommend you gamess software (in my opinion is the free reverse of gaussian). With the new version, it is more user-friendly than older version (highlighting in windows). Gamess can perform all the calculations, corrections, etc, as gaussian do. The only thing that you must check are the corrections (in the gaussian algorithm) because some of them, it is required to include the command line. This is the unique problem. If you decide to use it, I can help you to do the first steps in gamess if you have some trouble.

Many thanks for that recommendation and the offering of your help, Joaquim. It is much appreciated and I may get in touch depending on what my next job contract looks like. Many thanks again. 

Hallo, Anthony!

  I suppose that you use standard Mulliken population analysis. In such case the non-balanced basis sets with relatively many polarisation and diffuse basis functions (which are unphysical) cannot produce reliable results on atomic charges and bonding (my exprience is that 6-31+G(d) is the absolutely worst case).

 I propose either use of Mulliken population analysis with basis sets without polarisation and diffuse functions (e.g. pure 6-31G)  or use of other type population analysis (such as NBO),  preferably with larger basis sets.

   Best wishes

                     mb