CCL: Strange geometry optimization with breaking of H-bonds



Dear Grigoriy,

I have a couple of questions:

1) Why didn't you include D3 or D4 corrections in your first calculation? I might be wrong, but dispersion corrections will improve geometries in most cases, if not all.
2) Why use ORCA 5.0.3 when version 5.0.4 has been out for quite some time now, and many bugs were corrected or detected?

Looking at the release notes for version 5.0.3:

Regarding the use of D4 in ORCA, in the release notes for version 5.0.3, the first post in the ORCA forum says:

"Unfortunately, the C++ implementation of the DFT-D4 code was affected by a bug in the gradient and some further minor bugs.

This affects all DFT gradient (and thus also hessian) calculations using D4 in ORCA mainly manifesting in sometimes erroneous geometry optimizations.

In most cases this breakdown is indicated by geometry optimizations running up in energy/gradient. The magnitude of the error is not generally assessable and should be checked for each individual case (it can be negligibly small for many systems). In any case, it is recommended to check the convergence behavior/final geometry in comparison to a D3 calculation.

Further, some elements (Li, Be, Na, Mg, K, Ca, Sc, Ti, V, Rb, Sr, Y, Zr, Nb, Cs, Ba, Hf, Ta) did not make use of the revised reference polarizabilities given in DOI: 10.1039/D0CP00502A resulting in minor energy differences compared to the current DFT-D4 standalone code provided by Grimme and co-workers.

These bugs also affect the r2SCAN-3c composite method. D3, GFN-xTB and GFN-FF methods are NOT affected!

In any case, all energies computed with the ORCA implementation of D4 should be checked with the stand alone code (current stable release version 3.5.0, https://github.com/dftd4/dftd4).

The bug will be fixed in the upcoming ORCA 5.0.4 release.

The Grimme Group DevTeam & ORCA Team"

It may be advisable to re-run these calculations with version 5.0.4 using D3 or D4 corrections, checking if you get the same problems.

Best regards,

Marcos
---

Prof. Dr. Marcos Verissimo Alves
Prof. Adjunto II, Curso de Física Computacional
Instituto de Ciências Exatas
Universidade Federal Fluminense
Volta Redonda - RJ, Brasil


Em ter., 23 de abr. de 2024 às 19:47, Grigoriy Zhurko reg_zhurko%chemcraftprog.com <owner-chemistry.^-^.ccl.net> escreveu:
I have an Orca 5.0.3 job:

!B3LYP 6-31++G(D,P) Opt PAL3 CPCM(Water) ANFREQ

* xyz 0 1
...

The geometry optimization produced a structure at attach 1, with O16-H23 distance 2.63944, but with negative frequencies. To get rid of them, I repeated the job with these keywords:

!B3LYP 6-31++G(D,P) Opt PAL3 CPCM(Water) AnFreq
!TightOpt DefGrid3 VeryTightSCF

* xyz 0 1
...

Finally I got an optimized geometry without negative frequencies, but the structure is different - see attach 2. The distance O16-H23 is now 3.89712.
Isn't that strange that the optimized structure does not have this O:H hydrogen bond?
At attach 3 you can see one more optimized (at the same level ) structure, which seems to be ok - the O15..H21 distance is 2.27375. My intuition tells me that this is a correct calculation.
The structure from attach 1 was also computed at B3LYP-D4 level and everything seems ok - see attach 4 (O16..H23 distance is 2.21156):

!B3LYP D4 6-31++G(D,P) Opt PAL3 CPCM(Water) ANFREQ

* xyz 0 1
...

Maybe this is a bug in Orca?
Grigoriy Zhurko