SCF results depend on the number of MPI processes (007_PW_UPF201_USPP_Fe, nspin=2, kpar>1)

[chengleizheng]

Describe the bug

For nspin=2 plane-wave SCF calculations with kpar>1, the converged total energy and stress depend on the number of MPI processes, even though the input is identical and pw_seed is set (which is supposed to make the random initial wavefunctions process-count independent).

Reproduced on tests/01_PW/007_PW_UPF201_USPP_Fe (Fe, nspin=2, kpar=3, 8 k-points, pulay mixing, pw_seed=1):

np	FINAL_ETOT (eV)
3	-673.83493470
4	-673.83493069
6	-673.83493470
8	differs by ~1e-6
12	-673.83493470

• np=3/6/12 (uniform pool sizes) agree to ~2e-9
• np=4/5/8 (non-uniform pools: 2,1,1) deviate by 1e-6 ~ 7e-6 eV; stress diagonal differs by up to ~4e-3 kbar

The difference appears already in the first-iteration drho (e.g. 6.8291 vs 6.8287), so it is not a convergence-tolerance artifact: the SCF reaches different fixed points.

Expected behavior

Identical results for any number of MPI processes (np >= kpar)

To Reproduce

cd tests/01_PW/007_PW_UPF201_USPP_Fe
mpirun -np 3 /path/to/abacus_basic_para   # INPUT: nspin 2, kpar 3, pw_seed 1, mixing_type pulay
mpirun -np 4 /path/to/abacus_basic_para
# compare FINAL_ETOT_IS and TOTAL-STRESS in OUT.autotest/running_scf.log

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).