RT-TDDFT calculations with hybrid functional PBE0 were very different when setting md_nsteps and estep_per_md

[xuan112358]

Describe the bug

When running RT-TDDFT calculations with hybrid functional PBE0 and NVE emsemble, results of setting md_nsteps and estep_per_md were very different in energy revolution, current revolution and absorption spectrum.
I think this two settings are supposed to be the same.
This phenomenon just happend in this 1D hydrogen chain case, but seem to not happen in 2D hBN and 0D molecule system.
Results of setting estep_per_md seemed strange and totally wrong, no matter with out_current as 1 or 2. Below are energy revolution, current revolution and absorption spectrum, respectively.

Results of setting md_nsteps seemed a bit more normal, but somewhat strange. Below are energy revolution, current revolution and absorption spectrum, respectively.

Expected behavior

No response

To Reproduce

No response

Environment

No response

Additional Context

No response

Task list for Issue attackers (only for developers)

• Verify the issue is not a duplicate.
• Describe the bug.
• Steps to reproduce.
• Expected behavior.
• Error message.
• Environment details.
• Additional context.
• Assign a priority level (low, medium, high, urgent).
• Assign the issue to a team member.
• Label the issue with relevant tags.
• Identify possible related issues.
• Create a unit test or automated test to reproduce the bug (if applicable).
• Fix the bug.
• Test the fix.
• Update documentation (if necessary).
• Close the issue and inform the reporter (if applicable).

[ESROAMER]

Since the PBE results are fine, I suspect the issue might be with the initialization at each estep. Could you provide your input files and the PBE current results for checking?

[ESROAMER]

Also, when you replaced estep with md_nstep, have you checked if all the atoms were fixed?

[xuan112358]

Since the PBE results are fine, I suspect the issue might be with the initialization at each estep. Could you provide your input files and the PBE current results for checking?

@ESROAMER
Sure, this is the PBE current results.

And these are all the input&output files and pictures of HSE with md_nstep/estep_per_md and PBE with estep_per_md, just some large files removed.

github_files (2).zip

I'm sure that all the atoms were fixed wirh md_nstep

[ESROAMER]

So we now have two problems:
1.The results from estep and md_nstep are inconsistent.
2.The result obtained from md_nstep is incorrect.
For the first peoblem, I will check whether there is any missing part in the initialization of estep. For the second problem, I consulted Dr. Jiyuyang. For PBE0 using the Spencer–Alavi scheme, having too few k-points can significantly affect the convergence of the results. You can try increasing the number of k-points, switching to HSE, or changing the exx_singularity_correction option to revised_spencer for testing.

[xuan112358]

@ESROAMER
I have already tried several methods you suggested, but unfortunately, their results are the same as before.
I switched to HSE, but the energy changes and absorption spectrum results were almost the same as above.
I found the number of k-points were very unsufficient for PBE0. If I changed the exx_singularity_correction option to revised_spencer, k-points setting were almost convered. I calculated absorption spectrum with exx_singularity_correction as revised_spencer, k-points as before or setting the number of kpoints of y&z directions from 1 to 2 (achieved convergence for k-points in SCF), but the results were the same, nothing better.
I'm worried there are still some bugs.

[ESROAMER]

@ESROAMER I have already tried several methods you suggested, but unfortunately, their results are the same as before. I switched to HSE, but the energy changes and absorption spectrum results were almost the same as above. I found the number of k-points were very unsufficient for PBE0. If I changed the exx_singularity_correction option to revised_spencer, k-points setting were almost convered. I calculated absorption spectrum with exx_singularity_correction as revised_spencer, k-points as before or setting the number of kpoints of y&z directions from 1 to 2 (achieved convergence for k-points in SCF), but the results were the same, nothing better. I'm worried there are still some bugs.

Sorry I was busy with other things and didn't get to test this issue thoroughly until today. I found some bugs related to the hybrid gauge and will fix them in two days. The velocity gauge is not affected according to my tests, so you may test the velocity gauge example first and wait for my fixes.

[ESROAMER]

I have added the fix in #7198 . Please take a look.