Questions Regarding Using HamGNN Potential Trained with ABACUS Data for HamEPC Calculations

[BraunSakai]

Dear Dr. Zhong,

I hope this message finds you well.

I am currently planning to use the HamGNN potential trained with ABACUS datasets to perform HamEPC calculations, and I have a couple of questions regarding this process:

Are there any specific details or considerations I should keep in mind when using the HamGNN potential for HamEPC calculations? Should I adjust any parameters or configurations?

Is it possible to perform the entire HamEPC calculation process solely within ABACUS, without involving OpenMX? If so, are there any additional settings or workflows required to achieve this?

Thank you very much for your time and guidance. I look forward to your response.

Best regards

[QuantumLab-ZY]

Dear @BraunSakai,

Thank you for reaching out. Unfortunately, the integration of HamEPC with ABACUS has not been fully implemented yet. The main challenge lies in automating the preprocessing step for the Hamiltonian in ABACUS. Additionally, exporting the HSE Hamiltonian from ABACUS is quite complex because the Hamiltonian matrix elements for HSE include values beyond the cutoff radius of two atoms.

Currently, the script we provide for packaging the HSE Hamiltonian from ABACUS contains a known bug. Specifically, the script only exports Hamiltonian matrix elements within the cutoff radius of two atoms. While this works fine for PBE calculations, it causes issues for HSE due to the extended range of matrix elements. Regrettably, we do not have time to address this issue in the short term.

We appreciate your understanding and patience as we work toward improving this functionality.

Best regards,
Yang Zhong

[BraunSakai]

Dear Dr. Zhong,

Thank you so much for this clarification. My understanding is that the postprocess script provided based on ABACUS 3.5 has issues when extracting the HSE Hamiltonian. To use HamEPC, adjustments would need to be made to the data extraction step from HamGNN. Could you kindly confirm if this is correct?

Sincerely

[Youhaojen]

Dear Yang Zhong,

I noticed that utils_abacus from HamGNN has been updated. Since I'm not using the HSE Hamiltonian and only working with the PBE functional, does that mean a suitable grad_mat_prep for ABACUS is now available?

Best regards,
Hao-Jen You