FYI: difference in chemical shielding calculations between gpu4pyscf and pyscf

[steto123]

Hello,

for your information: i found some small differences between the isotropic shieldings calculated with gpu4pyscf and pyscf (i know the experimental character of this calculations). I have shared a pdf with detailed information because my former issue was not saved here.

Is there a simple reason for this differences. Is there a mistake or error in my code?

https://acrobat.adobe.com/id/urn:aaid:sc:EU:96d4d829-8366-4887-aef5-3e6c411dc82e

Many thanks for your great work.

Steffen

[wxj6000]

@steto123 Hi Steffen, can you share the xyz file? I didn't find the large discrepancy. The difference should be less than 1e-5 ppm. And I added some checks of consistency between PySCF and GPU4PySCF for a simple molecule.

[steto123]

Attached is the xyz file for acetonitrile. The structure comes from the test data of the CSESHIRE data collection.
http://cheshirenmr.info/CalculationFiles.htm
I have calculated all connections for the calibration used there with Orca5 and am in the process of doing the same for pyscf. The deviation in the calculated shielding between orca5 and pyscf lies in the 2nd to 4th decimal place, with deviations for C being larger than for H. At least that has been my observation so far. However, the differences between the values of pyscf and gpu4pyscf are larger, which I have not yet considered and tabulated systematically. I have therefore assumed that it may be due to the way I implemented the calculation based on examples from github. I can also provide you with the jupyter notebooks I used. However, these are commented in German by me.
Feel free to contact me directly as well. I can be reached at the anti-track email patient.kouprey-5619k@norton.me. I delete such emails from time to time. However, they are forwarded to my actual email, so publishing them here is not a problem.

Acetonitrile.xyz.txt

For security reasons i must rename this to .txt