Convergence difficulties met in LCAO scf calculation with Broyden and Kerker charge mixing method

[hongriTianqi]

Describe the Bug

Convergence difficulties are met in LCAO scf calculation with Broyden and Kerker charge mixing methods, but not in the corresponding PW calculation.

To Reproduce

Steps to reproduce the behavior:

1. cat 206_NO_BD_chargemixing/OUT.autotest/running_scf.log
2. cat 206_NO_KK_chargemixing/OUT.autotest/running_scf.log
3. cat 106_PW_BD_chargemixing/OUT.autotest/running_scf.log
4. cat 106_PW_KK
   chargemixing.tar.gz
   _chargemixing/OUT.autotest/running_scf.log

[mohanchen]

@YuLiu98 Could you take a look at these examples and propose your opinions?

[hongriTianqi]

This issue has not been addressed, so it should be kept open. @Unboundwill @dyzheng @YuLiu98 @mohanchen

[caic99]

@Qianruipku @mohanchen Would you look into this issue?

[Qianruipku]

Maybe it is associated with issue #484. scf aims to find the minimum of "dr2" when it is thought to be converged. However, "dr2" is differently defined for lcao and pw. Thus, we must use different mixing methods for lcao and pw. In fact, "pulay" and "broyden" are both the pulay method. The former is for lcao while the latter is for pw.