gxtb point charge gradient (pcgrad)

[c36247954-max]

Hi,

I recently tried running QM/MM optimizations with orca as an external tool, and tried to use g-xTB for that purpose (via ExtOpt). However, I recognized, that the pcgrad file is empty and that in gxtb there is a TODO related to POINT CHARGES. I assume that this includes the point charge gradients. However, the ptchrg.xyz file is created. I am not sure, but to me it appears, that I could in principle parse the ptchrg file in a second call to the gradient subroutine and create potentially the point charge gradients? Or am I missing something more deeply rooted in the gxTB code? Any suggestions, hints, or a time-frame at what time these gradients would be available would be highly appreciated.

Best Regards.

[stgeo]

Hmm, there are several points here. I assume you mean g-xTB/MM, rather than QM/g-xTB (which I'm not sure would work currently).

• To get the correct gradients of the QM atoms, the electrostatic interactions with the point charges need to be included. This is something xtb can do: https://xtb-docs.readthedocs.io/en/latest/pcem.html so we can include it there (although ORCA already has an internal interface to xtb), but it's not a (documented) feature of the current gxtb version.
• The gradients of the point charges themselves (or rather the MM atoms) is something ORCA will do if any of them are active in the optimization. Passing them to g-xtb as if they were real atoms would certainly not be correct.
• The TODO refers to the last line in the input file ORCA writes, which includes the file name of the point charge file. We don't read that, because we can't use it yet.

[c36247954-max]

Hi, thank you for your quick reply!
to be more specific:
I am trying to run a AMBER QM/MM geometry optimization with ORCA as a driver for the QM part.
In ORCA I try to use g-xtb for the minimization. The minimization step works fine, and I get a converged structure in ORCA, however, AMBER in the second step fails, as it tries to read the .pcgrad file, which is empty.

Running a similar optimization with XTB1/XTB2 provides the required .pcgrad file. I was wondering why it so far is missing in the ExtOpt with g-xtb. But I assume your last point concludes my question, as if I understand your last point correctly, the point charges that are provided by AMBER (ptchrg.xyz) are not read and processed in g-xtb?

Thanks, and best regards.

[stgeo]

So does AMBER drives the optimization and ORCA needs to write the .engrad and .pcgrad files, or are you running a geometry optimization with ORCA's native QM/MM using an AMBER forcefield?

Either way, yes, since gxtb doesn't support point charges yet, the QM atom gradients will be wrong (since they are missing the point charge term), and the gradients on the point charges also cannot be computed.

[c36247954-max]

Ok, so it does not make sense (for now) to run this approach.
AMBER (the program) drives the optimization and calls ORCA in each step. I did not use ORCA's native QM/MM implementation.

Anyway, thank you very much for your quick reply and your help!