Energy explosion and "Can't find even an electron!" error during relaxation (ION MOVE 2)

[DanielZhao0432]

Describe the bug

I am performing a structure relaxation (relax) for a system consisting of a phosphate group (PO₄³⁻) adsorbed on a Mg-Ba alloy surface using ABACUS.
The first ionic step (ION MOVE 1) converges successfully with a Kohn-Sham energy around -17691 Ry.

However, at the beginning of the second ionic step (ION MOVE 2), the energy fluctuates wildly:
ION MOVE 2, ELEC ITER 1: KS energy drops to ~ -34077 Ry.
ION MOVE 2, ELEC ITER 2: KS energy explodes to ~ 280984 Ry.
Eventually, the calculation crashes with the error: Can't find even an electron!.

Expected behavior

The relaxation should proceed with reasonable energy changes between ionic steps, or the SCF should be robust enough to handle the new geometry.

To Reproduce

1. System: Mg-Ba alloy surface (10-10) + Phosphate group.
2. Calculation type: relax.
3. ABACUS version: v3.9.0.23
4. Pseudopotentials: ONCV
5. Command: mpirun -np 32 abacus

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).

[DanielZhao0432]

MgBa_PO4.zip

This compressed file contains the following input files: INPUT, STRU and KPT, as well as a folder called 'PP_ORB', which contains all the pseudopotentials and orbitals used in this calculation.

P.S. To reproduce these calculation results, please modify the addresses of the pseudopotentials and orbitals in the 'INPUT' file.

[DanielZhao0432]

This test case was attempted on LTSv3.10.0, v3.9.0.14, and v3.9.0.23 with the following results:

1. Relaxation runs normally on LTSv3.10.0, with no energy jumps appearing in the log file.
2. Version 3.9.0.14 exhibited issues similar to those in v3.9.0.23.

Therefore, this bug can be considered introduced between v3.10.0 and v3.9.0.14.

Furthermore, when performing calculations using these 3 versions of ABACUS, I ran 32 MPI processes on a single compute node, with each process executing a single thread, utilizing a total of 32 cores.