I'm interested in studying interactions between n-hexane molecules and also interactions between chloroform molecules, what are the suggested dispersion-corrected density functionals to study these systems?
Unlike some people have stated above, B3LYP and Minnesota functionals do NOT have any dispersion built-in.
To properly model dispersion, you have to use either semiempirical corrections, like Grimme's -D3 corrections (which are available for B3LYP), or an ab initio van der Waals functional, like wB97X-V, B97M-V or wB97M-V by Mardirossian and Head-Gordon.
I don't think there is a definitive answer, but the empirical D3 parameters by Prof. Grimme are a pragmatic approach, especially if you are interested in van der Waals interactions for weakly bonded complexes. The Becke-Johnson damped version, D3(BJ) comes recommended. Recently a D4 parameterization was published by Grimme, involving DFTB-like parameters. We are working on incorporating that as well.
Other options: The Minnesota set of functionals include some dispersion by construction / parameterization, e.g. M05, M06, M11, M12, NM12, etc.
Also a pragmatic approach. Typically needs a bit higher integration accuracy.
I would say these two pragmatic options are also the most often used in practice.
Then there are other perhaps 'better' or 'more physically correct' options, such as the many-body dispersion (MBD) approaches (e.g. Tkatchenko & Scheffler, the XDM approach from Johnson, ...), the Langreth van der Waals functional, van Voorhis' functional, frozen-density embedding including van der Waals (FDE-vdW), and probably a few more I forgot.
if u can specify what atoms r u being used in ur complex n s d interaction between which molecules only then can i say which method is better. i agree with alexander ans to use b3lyp but for being more specific with d basis set which atoms r involved should b known as atoms also effect the basis set along withd choice of method.
I would avoid B3LYP for your problem. Goumans has already mentioned using the D3(BJ) correction and several other choices. A decent choice is omegaB97X-D or, even better omegaB97M-V. The former is available in Gaussian and both are likely to be available in QChem.
You may want to read section 10.4 of the 2nd edition of my book
Unlike some people have stated above, B3LYP and Minnesota functionals do NOT have any dispersion built-in.
To properly model dispersion, you have to use either semiempirical corrections, like Grimme's -D3 corrections (which are available for B3LYP), or an ab initio van der Waals functional, like wB97X-V, B97M-V or wB97M-V by Mardirossian and Head-Gordon.
B3LYP-D3 works normally ubnormally good for vdW-based systems.
Minnesota functionals DO include dispersion, while the enormous amount of parameters include results from datasets. In case of M15 there were even dispersion focused set, but even M06 would give you underestimated but at least attractive interaction, while majority of other functionals are much more repulsive. All that getting corrected by D3.
@Gess: Minnesota functionals, like B3LYP, are semi-local functionals. Their mathematical form does not physically allow modeling van der Waals dispersion interactions (i.e. the r^-6 tail).
This is brought in by the Grimme correction, or a proper non-local correlation functional, such as vdW-DF or VV10.
@Susi. Yes, it does not contain r^-6 or any other term responsible alone for description of vdW. However being strongly [over]parametrised functionals, Minnesota's describes the features that have place in systems included in sets used for parametrisations. Including vdW. For a system where vdW plays dominant role (noble gas dimers, benzene dimers, fullerene dimers, alkane dimers), without usage of Grimme correction (or NL or any other, also implemented directly into, like B97D), one will get repulsive interaction with majority of functionals, including B3LYP, may be slightly attractive with PBE or PBE0 or TPSS, and reasonably attractive with M06. Problem is that with Minnesota you never know will it overrate or underrate dispersion for particular system. M06-D3 will overestimate pretty surely.
Alternatively one can try to go to double hybrid (DH) functionals where unabilities of DFT are compensated by overshootings of MP2, or even to quite new DSD-DH and DOD-DH, where playing with spin components have place. I found quite funny that after adding to DFT MP2 and playing with spin components you still have to add rather sophisticated D3 to improve the results.
Finally, there are methods above DFT to get correct values, especially if you get reasonable geometries from DFT (that is normally the case). I would advise to have a look on SAPT-DFT and DLPNO-CCSD(T) approaches that can give you very accurate description of vdW interaction.
@Gess yeah, some Minnesotas get reasonable values for some systems by coincidence / because they were trained to do that, but other systems may result in nonphysical answers. E.g. the figures 1, 2, and 5 in
"How Accurate Are the Minnesota Density Functionals for Noncovalent Interactions, Isomerization Energies, Thermochemistry, and Barrier Heights Involving Molecules Composed of Main-Group Elements?"
in the article J. Chem. Theory Comput., 2016, 12 (9), pp 4303–4325
Funnily, "the main weakness is that none of them are state-of-the-art for the full spectrum of noncovalent interactions and isomerization energies" seems not to disturb the developers, who are creating new functionals that are reliable just for one-two properties. And yes, using seventy empirical parameters estimated from enormous datasets you can create functional that describe few selected properties pretty good. Of course you have to sacrifice smth, like reliable description of electron density (see Bushmarinov paper in Nature) but who cares on density?
Actually D3-corrected methods gives you geometry and density that corresponds to nonminimal point in case of noncorrected functional. (As long as D3 do nothing with density). Critics of D3 would name it nonphysical as well.
One can go to NL (in particular VV10, implemented in ORCA), that has stronger physical background (and no such a problem like "where from come 12 in second term?"), and normally provide result extremely close to D3. However there one would still need to specify purely empirial coefficient (plus cost, extremely noticable in SCNL case).
Prasanta, I would not add dispersion corrections to any of the Minnesota functionals (such as M06-2X) even though they do not contain explicit dispersion corrections. Minnesota functionals were parameterized to account for dispersion effects "implicitly". As Susi pointed out, this works well in some cases and poorly in other cases.
Sir, in case of non covalent interactions, especially in halogen bonding, which functional would you prefer that includes dispersion correction, apart from wb97xd?
None are granted to succeed. Do a model system with a high level method, like ccsd(t) or mp2, and try several functionals. Use the one that reproduces the interaction energy, geometry and dipole moment. Don't trust any recommendation that is not supported by facts: every system is different. PBE0 and B3LYP with Grimme's d3 correction on top might be good enough!
On a side note, if you want to use Truhlar's functionals, recall that the M06 family is 13 years old. Use the newer ones instead.
Jan M L Martin, Ajit J. Thakkar, Rubén Laplaza, Dmitry Sharapa, Susi Lehtola There are some technical difficulties regarding high level calculation, That's why I was opting for DFT.
I was studying some benchmark studies,
with GMTKN30 database, author of Phys. Chem. Chem. Phys., 2011, 13, 6670–6688 suggested wb97xd and m062x-d3. LC-wPBE-D3 was below average.
In MOLECULAR PHYSICS, 2017, 115, 19, 2315 article, wb97mv, wb97x, m062x, m05-d3(0) and PW6B95-D3 was opted better.
In J. Chem. Theory Comput. 2013, 9, 1918−1931authors have preferred m062x and wb97xd over b3lyp-D3 and PBE0-D3 with XB-51 and XB-18 dataset.
Phys. Chem. Chem. Phys., 2017, 19, 32184--32215 opts for wB97X-V, M052X-D3(0), and wB97X-D3 with GMTKN55 dataset.
J. Chem. Theory Comput. 2018, 14, 180−190 opts for LC-wPBE-D3(zero or BJ) is best for XB-51 and XB18 dataset.
As I said, your specific system may be different from those in benchmark tests. I am not telling you to use CCSD(T)/CBS, I suggest that building a *small, specific model* and benchmarking against it is probably the best way to choose a functional.
If your big system involves a particular halogen bond, your minimal system could be your halogen bond in a saturated closed shell environment, for example. That should be feasible with MP2 and a triple-z basis set.
I agree with Rubén Laplaza , your own benchmark on particular system that includes interaction of your interst is the most valuable. It is not even obligatory to perform full relaxation on benchmark-method, set of single points (rigid scan near DFT minimum) normally is sufficient. This way was used in S66x8 dataset and I applied same solution in work about interaction in fullerene dimer.
Also based on that work I would not believe in MP2 results - MP2 strongly overestimates noncovalent interaction. It was massively investigated by Hobza, true solution is MP2.5, MP2.C, MP2.X who includes MP3, or fudging parameters in SCS-MP2 approach. Nevertheless in my particular case, even SOS-MP2 is way worse than any D3-corrected DFT. So benchmark should be performed using DLPNO- or Canonical CCSD(T), DFT-SAPT or ACFD-ALDA.
I also agree with Ajit J. Thakkar that adding of D3 to any old Truhlars functionals is unrecommended. Parametrisation covers dispersion at least to some extend. Comparing uncorrected functionals on system with strong dispersion M06 will almost always among the best, simply because PBE or B3LYP don't care on dispersion at all. When D3 (or NL/VV10) is applied real competition starts.
Last. M06 as well as B3LYP and PBE is old functional, but I would not blindly switch to newer versions. Number of parameters were growing with years while applicability to all problems was decreasing. Minnesota is in my top-list for metal-containing systems, but for pure-organics i would not take it first.
In our latest revDSD benchmark paper, we both trained revised DSD double hybrids on the GMTKN55 set *and* evaluated a bunch of lower-level methods not already covered in Lars Goerigk ' GMTKN55 benchmark paper. It's possible to get down to 2.2 kcal/mol WTMAD using the most recent double hybrids, wB97M(2) of the Head-Gordon group and (with about 1/3 the number of empirical parameters) revDSD-PBEP86-D4 by ourselves. If you need to stay on rung four for reasons of computational cost, both Lars (in his follow-up paper to the original GMTKN55) and we found that Head-Gordon's wB97M-V (available in both ORCA and Q-CHEM) has a distinct edge over other functionals. http://doi.org/10.1021/acs.jpca.9b03157
Prasanta Bandyopadhyay wB97M-V and wB97X-V are available in e.g. PySCF and Psi4, which are both freely available. All functionals should also appear in a future release of ORCA.