SOC convergence problem for BCC Fe

[hongriTianqi]

Describe the bug

scf_nmax is set to 200. Finally drho 5.496971e-06, energy difference = 1.226634e-03 eV. The energy difference is very large.

Expected behavior

Get converged results.

To Reproduce

soc_convergence.tar.gz

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).

[WHUweiqingzhou]

I suggest you can use so-called Angle Mixing method (PR #3356), which is expected to give a better convergence than DIIS method for non-colinear calculations. Since drho is defined by summing over all spinor, drho is quite hard to converge into 1e-6 and you can release it a little bit, f.e. 2e-6 or 4e-6. You can set

mixing_angle   1.0
mixing_beta     0.2
mixing_beta_mag   1.0
scf_thr      2e-6

[hongriTianqi]

Thanks for your suggestion. The results come very well: https://app.bohrium.dp.tech/abacustest/?request=GET%3A%2Fapplications%2Fabacustest%2Fjobs%2Fjob-abacustest-v0.3.108-538675

[hongriTianqi]

@WHUweiqingzhou Thanks for your suggestion. Sorry to inform that some cases still met convergence difficulty with the above setting:

GE181  -6.819687e+03  -3.368666e-07  2.619e-06  2.029e+01
 GE182  -6.819687e+03  5.968269e-07   3.183e-06  2.027e+01
 GE183  -6.819687e+03  -6.700650e-07  3.033e-06  2.030e+01
 GE184  -6.819687e+03  4.082850e-07   3.415e-06  2.029e+01
 GE185  -6.819687e+03  -4.099856e-07  2.604e-06  2.025e+01
 GE186  -6.819687e+03  6.698856e-07   3.364e-06  2.028e+01
 GE187  -6.819687e+03  -5.939329e-07  3.104e-06  2.035e+01
 GE188  -6.819687e+03  3.345604e-07   3.137e-06  2.030e+01
 GE189  -6.819687e+03  -4.127807e-07  2.616e-06  2.032e+01
 GE190  -6.819687e+03  6.707603e-07   3.300e-06  2.032e+01
 GE191  -6.819687e+03  -5.941463e-07  3.141e-06  2.027e+01
 GE192  -6.819687e+03  3.321969e-07   3.075e-06  2.028e+01
 GE193  -6.819687e+03  -4.087041e-07  2.631e-06  2.030e+01
 GE194  -6.819687e+03  6.696188e-07   3.322e-06  2.032e+01
 GE195  -6.819687e+03  -5.947372e-07  3.046e-06  2.033e+01
 GE196  -6.819687e+03  3.326717e-07   3.132e-06  2.035e+01
 GE197  -6.819687e+03  -4.088140e-07  2.639e-06  2.027e+01
 GE198  -6.819687e+03  6.673551e-07   3.317e-06  2.029e+01
 GE199  -6.819687e+03  -5.930790e-07  3.090e-06  2.046e+01
 GE200  -6.819687e+03  3.324552e-07   3.138e-06  2.028e+01
 !! CONVERGENCE HAS NOT BEEN ACHIEVED !!

See the attached case:

soc_converge.tar.gz

[WHUweiqingzhou]

@hongriTianqi,
I make some tests, and find old DIIS mixing with mixing_beta=0.1 can reach a quite low EDIFF~1e-10, which is better than mixing_angle=1.0 method.