DFT+U SCF CANNOT converge in CeO2 surface

[QuantumMisaka]

Details

We are trying to use DFT+U method in ABACUS 3.6.0 to do SCF in a CeO2 (110) surface

And the SCF cannot converge

Input and output files:

CeO2-dft-u.tar.gz

ABACUS: 3.6.0 Dependencies: Intel-OneAPI 2023.2 by Intel-toolchain Hardware: AMD EPYC 7b12 Pseudopotential and orbital : #3762

Task list for Issue attackers (only for developers)

• [X] Reproduce the performance issue on a similar system or environment.
• [ ] Identify the specific section of the code causing the performance issue.
• [ ] Investigate the issue and determine the root cause.
• [ ] Research best practices and potential solutions for the identified performance issue.
• [ ] Implement the chosen solution to address the performance issue.
• [ ] Test the implemented solution to ensure it improves performance without introducing new issues.
• [ ] Optimize the solution if necessary, considering trade-offs between performance and other factors (e.g., code complexity, readability, maintainability).
• [ ] Review and incorporate any relevant feedback from users or developers.
• [ ] Merge the improved solution into the main codebase and notify the issue reporter.

[kirk0830]

I have noted you used PD04.sp series pseudopotential, in which f-electron is explicitly treated. For other users previously have similar experience on ErO2 system, I would suggest you try 3+_f--core or 3+_f--core-icmod1 series counterpart, but convergence failure is indeed a problem. If you are interested, let's take a look at the test result posted on APNS: https://[kirk0830.github.io/ABACUS-Pseudopot-Nao-Square/pseudopotential/results/2024-03-27/Ce.html](https://kirk0830.github.io/ABACUS-Pseudopot-Nao-Square/pseudopotential/results/2024-03-27/Ce.html) What is first noted that GTH-LnPP1 can not get its converged value within range [0, 300] Ry of ecutwfc, but it might be okay because CP2K always uses a cutoff for planewave like 550 Ry or higher. In the log-plot, you can see convergence tendency of three properties, respectively: The pseudopotential you use corresponds to the yellow dashed line, the ecutwfc value for description of wavefunction (or charge density with 4*ecutwfc) getting consistent is higher than those of the other two PD04 series pseudopotentials, and the description seems to be unconvincing if ecutwfc lower than 50 Ry, since the tendency seems really unphysical. What cannot obtained from the test result is the correctness of results with pseudopotentials. Presently I would suggest you perform cell-relax test with the same O pseudopot-nao bundle but different Ce pseudopot-nao bundle, then compare with experimental results. Then both precision and efficiency you will have a rough idea about them, on pseudopot-nao bundles you use. Please note that even for lcao calculation, the ecutwfc still have impact on accurancy of estimation (resolution) of charge density, if I remember correctly.