From my point of view, dispersion correction is required for any geometric calculations. I use the D3 for a long time and now D3BJ. Of course this is only a palliative of ab Initio calculations, but the geometry improves significantly.

Yes it will, especially when calculating weak intermolecular interactions, for example, I used uncorrected B3LYP and corrected B3LYP-D3 when I was investigating aromatic pi contacts and halogen bonding in crystals, and it shows that interaction energies observed with B3LYP-D3 are much better and some intermolecular interactions cannot be treated without dispersion corrections. Although Dn corrections are just geometric parameters added to conventional DFT methods, and I think it has no point to use them if investigating molecular geometry in gas phase, but when it comes to calculating geometry of molecular interactions it must be used.

The consideration of dispersion interactions is crucial to obtain accurate NMR shifts for systems presenting weak intra- or intermolecular contacts. If such contacts are expected to be less important, you may obtain excellent results without consideration of dispersion effects (see for instance http://dx.doi.org/10.1021/jp302128r ). 

Note that some dispersion corrected methods can lead you to poorer results than uncorrected ones because some methods significantly overestimate the dispersion. The problem is that such overestimation is dependent on the system. Maybe this review article http://dx.doi.org/10.1039/C3RA40713F can be useful to guide you in the selection of the dispersion corrected method. 

http://dx.doi.org/10.1021/jp302128r

http://dx.doi.org/10.1039/C3RA40713F

It depends on the functional you choose.  I was recently at a talk by Grimme (who developed the D3 dispersion correction), and he showed that for functionals like Truhlar's M062X, and M11, which were parameterized for disperion, addition of D3 did not have any significant effect.  I do agree that dispersion corrections make a difference for functionals like B3LYP, and PBE, However, I would be cautious about using B3LYP at all for this system, as it has the potential of a large self correlation error, and B3LYP has only about 20% exchange built in.  If your computational program has M11 or M062X, I'd just use that.

I have a feeling that with organosilicons MP2 should work well. So why DFT?

What about RI-MP2? It is not much computationally expensive.  Does it produce good geometry optimization?

Even though MP2 is generally robust there are systems for which it fails.  If the organosilicon compounds contain aromatic structures, and the crystallization involves interaction between the aromatics, then MP2 is not a good choice.  It has been shown to fail dramatically for stacked pi systems.

Are you only considering structure at zero pressure, or are you interested in multiple pressures? If you are only considering zero pressure, TS, DF, or D3 might work just fine; but testing is very important, especially given that some systems experience more thermal expansion than others; hence it is important to be mindful of how a correction has been parametrized since you'll be comparing to structures at some finite temperature. If you are interested in a range of pressures (and temperatures), then taking thermal expansion into account is a good idea (see below).

It was mentioned that some corrections can greatly overestimate the dispersion; however, care must be taken in deciding whether or not this is actually the case. Keep in mind that DFT fails to include expansion caused be the free energy of the system, hence structural comparison with systems at low temperature, let alone 300 K, should make the minimum energy structure calculated by DFT+D (whatever dispersion correction) seem over-bound. Just by how much depends on the system and correction. For us, Grimme D2 works best, providing the most binding out of the corrections we've tested. Typically, we see about 1-2% expansion in volume due to zero-point energy alone for the molecular solids we study. Add about another 2% due to temperature at 300 K. However, we take the extra step to calculate phonons, and then correct for thermal expansion using the quasi-harmonic approximation, which of course also requires multiple structure calculations at different pressures. As a result, we see pretty good agreement with experimental structures across a range of temperatures and pressures.

@Aaron Landerville

Are you using D2 instead of D3 specifically?

Yes, for our purposes (nitramine based molecular solids) D2 gives better results when combined with the quasi-harmonic approximation. D3 under-predicted dispersion, but gave results closer to experimental values of volume at 300 K.  Hence, D3 gave us volumes about 3-4% larger than D2, which might be fine if we didn't care about thermal properties, and were simply aiming for direct prediction of the experimental structure.

Dear José, Jean-Pierre, Alexey, Aaron, Marko, Samuel,

thank you all for your answers and helpful contributions!

Let me summarize the topic as far as I understood it up to this point:

Dispersion correction is important for systems with weak intermolecular interactions.

The consideration of dispersion interactions is crucial to obtain accurate NMR shifts for systems presenting weak intra- or intermolecular contacts.

Several methods were proposed during the discussion:

• B3LYP-D3 was used for investigating aromatic pi contacts and halogen bonding in crystals.

• TPSS-D3(BJ) and PBE-D3(BJ) are used as default methods for geometry optimizations for moderately large molecules.

• For VERY large molecules BLYP-D3(BJ) was proposed.

• Truhlar's M06-2X, and M11

• MP2 was proposed as alternative

There is also another important point: DFT+D adds a "D" correction to the energy. This does not affect electron density, which remains a pure DFT density. Therefore, if you want to calculate properties from the density, they will be of the same quality, as from the original DFT functional.

Hello,

I agree with the above but one should keep in mind that performing geometry optimization with a dispersion correction may lead to a different geometry (than without this correction). Hence, with a compound different geometry its properties may be different. 

Hi Jean,

I just wanted to point out that the electron density obtained from a DFT-D method may be different from that based on ordinary DFT calculations. It may happen so when the geometry optimization is ivolved. Thus, even using a simple DFT-D technique the electron density may be affected (because of the resulting compound geometry).

I'm not referring to a crystal packing at all :)

I think like others already mentioned, it will depend on the functional you chose. For functionals like wB97XD, M06-2x, they already included dispersion corrections to some degree. But for other functionals like BLYP, PBE or B3LYP, you want to include D2 or D3 when you are dealing with weak interactions.

However, i think there is another point in your problem. To repeat NMR results, you would need a rather large basis set. So in addition to the question whether to add D2/D3, you also need to consider whether the basis set is large enough (with diffusion components added) and the calculation with the chosen functional and basis set is feasible to carry out.

follow this question:

What is the best dispersion-corrected DFT method?

by

Ahmad Alrawashdeh

https://www.researchgate.net/post/What_is_the_best_dispersion-corrected_DFT_method

It is all very true, but what would be the effect ot including dispersion corrections to,say, B3LYP harmonic frequencies of benzene or anthracene? Has someone such imformation first hand?

I think to add dispersion factor to your calculation depend on whether your compounds have dominant dispersion energy or not , you can know this from studying the literature of them or you can simply add the dispersion factor , if you obtain better results it may suggest the presence of dispersion .For example I add in the input file : ( empiricaldispersion = gd3) for compounds with halogen bond and obtained reasonable results confirmed by EDA.

Prasanta Bandyopadhyay

Some DFT functionals contain different portions of dispersion correction. For example, the hybrid dispersion corrected ωB97X-D functionals are often employed for transition state search, TD-DFT calculation.. The functional ωB97X-D utilities 100% exact long-range exchange, a minor fraction of exact short-range exchange, the B97 correlation density functional, and empirical dispersion correction. Other functional like M06-2X is also used for a similar purpose. These functionals are prevalent for transition state search, TD-DFT calculation.

See my article-

• Article Synthesis of 5, 6-diaroylisoindoline-1, 3-dione and computat...

Does dispersion correction of b3lyp influence dipole moments of gas phase large molecules?

Dispersion correction is just an empirical correction to the energy of a system. It does't alter the electron density. Though, it may affect the optimization process as in the optimization this energy is taken into account. Hence, using a dispersion correction you may result in a different geometry than without it. As such, dispersion correction is only an empirical correction to the energy. Nothing more.

Thanks I am interested in possible structure changes with added dispersion. Can you orient me to a list of dispersion modified functionals? Info on that is equaly dispersed in the literature.

Best regards

Laszlo Nemes

Basically, any functional can be modified with dispersion correction. Some functional do not even need it. You will get more idea by reading this paper.

Article Thirty years of density functional theory in computational c...

Which is the best functional with respect to charge distribution (el ectric dipole moment.