Multiwfn forum / Potential issues with CHELPG calculations

Re: Potential issues with CHELPG calculations

dummy@example.com (sturlaugson) — Tue, 14 Jul 2020 16:59:53 +0000

Thank you for the suggestion to have ORCA generate the natural orbital output. I will try that and compare. I will also try the comparing DFT outputs as well. Also--thanks for the note about the faster CHELPG procedure in version 3.7!

Re: Potential issues with CHELPG calculations

dummy@example.com (sobereva) — Sat, 11 Jul 2020 01:17:58 +0000

This is the correct map of showing fitting points of CHELPG method. The points are exported by CHELPG module of Multiwfn and the map is plotted by VMD.

The old map in Multiwfn manual is misleading.

Re: Potential issues with CHELPG calculations

dummy@example.com (sobereva) — Fri, 10 Jul 2020 21:42:43 +0000

I am not familiar with the CHELPG calculation function of ORCA. I strongly suggest you first compare the CHELPG charge computed by ORCA and Multiwfn under DFT level, because in the case of MP2, the thing is more complicated: The orbitals written into wfn/molden file may depend on your specific keywords. In the MP2 calculation, though the finally reported energy is MP2, if you do not explicitly request ORCA to generate natural orbitals, then the orbitals in the exported wfn/molden file should still be the HF ones, while the charges at HF and MP2 levels always differ significantly.

The implementation of CHELPG charge calculation in Multiwfn is completely correct, and the result is exactly identical to Gaussian. The illustration figure in the CHELPG section of Multiwfn manual is indeed wrong, I will revise it in the next release.

PS: If you are using Multiwfn 3.6, please update to the latest version, because very recently the ESP calculation speed in Multiwfn has been significantly improved, the cost of calculating CHELPG charge has reduced by more than one order of magnitude.

Appendix: The correct way of generating .molden file containing MP2 orbitals
Run a calculation like this

! mp2 def2-TZVP
%mp2
Density relaxed
NatOrbs true
end
* xyz 0 1
 C                  0.00000000    0.00000000    0.56221070
 H                  0.00000000   -0.92444774    1.10110546
 H                 -0.00000000    0.92444774    1.10110546
 O                  0.00000000    0.00000000   -0.69618936
*

Then change the resulting .mp2nat file to .gbw file, and then use orca_2mkl to convert it to .molden file.

Potential issues with CHELPG calculations

dummy@example.com (sturlaugson) — Fri, 10 Jul 2020 15:40:37 +0000

I have recently run ORCA SPE calculations, MP2/cc-pVTZ(-f), on the ion dimethylimidazolium that has been geometry optimized using RHF/6-31G(d). I wanted to use the output to calculate the CHELPG charges on the atoms. ORCA has a built-in CHELPG routine which I initially used, but it gave suspect results (not consistent with some literature values, and some charges too large). I then ran the ORCA wavefunction output through Multiwfn, and got significantly different CHELPG charges than what ORCA produced, (but closer to the literature and more physically reasonable). Both Multiwfn and ORCA CHELPG calculations were done under identical parameters (vdw radii, cutoff length, grid density). See data below. Clearly, there is a discrepancy in how ORCA and Multiwfn determine the CHELPG charges for the same ORCA output. Given the greater difference between the ORCA CHELPG calculations and the literature values, I assume there is an problem with the way ORCA is calculating the CHELPG charges, but I don't know that for sure. I do not have Gaussian or GAMESS or another electronic structure package to compare additional CHELPG routines.

I am posting this to see if anyone else has had experience with discrepancies like this between different CHELPG routines?

Atom ORCA CHELPG Multiwfn CHELPG    DOI: 10.1021/jp0362133    DOI: 10.1016/j.molliq.2015.03.040 (in Supplementary Info)
NA    0.278    0.147    0.152    0.146
CR    -0.141    -0.047    -0.106    -0.046
NA    0.282    0.136    0.153    0.146
CW    -0.244    -0.152    -0.127    -0.135
CW    -0.242    -0.173    -0.127    -0.135
C1    -0.422    -0.191    -0.174    -0.177
HCR 0.243    0.222    0.206    0.202
C1    -0.426    -0.177    -0.173    -0.177
HWR 0.263    0.222    0.205    0.202
HWR 0.263    0.228    0.205    0.192
H1    0.192    0.128    0.131    0.125
H1    0.188    0.135    0.131    0.125
H1    0.192    0.135    0.131    0.142
H1    0.193    0.126    0.131    0.142
H1    0.188    0.131    0.131    0.125
H1    0.193    0.131    0.131    0.125

Finally, there are some errors in documentation out there for the CHELPG procedure. According to the original paper (DOI: 10.1002/jcc.540110311), the grid spacing is 0.3 Angstroms and the cut-off length is 2.8 Angstroms. Occasionally, these get converted, INCORRECTLY, to 3 pm and 28 pm. (See https://www.cup.uni-muenchen.de/ch/comp … lpg.html.) This unit conversion error is also present on page 72 of the Multiwfn 3.6 manual (and presumably other versions). Perhaps some of the issues I've seen in inconsistent results between different CHELPG routines are due to some conversion errors?

Any insight here from anyone would be appreciated. Thanks.