R2SCAN convergence problem

[QuantumMisaka]

Details

For SCF of same structure （CO and H2 adsorption in Fe5C2(510) surface, the latest v2.1(dev) NAO will not converge in 300 steps, while the official v2.0 NAO will converge:

INPUT：

INPUT_PARAMETERS RUNNING ABACUS-DFT

#Parameters (1.General)
suffix                  clean  # suffix of OUTPUT DIR
#ntype                   4   #  number of element
nspin                   2   #  1/2/4 4 for SOC
symmetry                0   #  0/1  1 for open, default
esolver_type            ksdft  
dft_functional          MGGA_X_R2SCAN+MGGA_C_R2SCAN  # same as upf file, can be lda/pbe/scan/hf/pbe0/hse
ks_solver             genelpa  # default for ksdft-lcao

#Parameters (2.Iteration)
calculation             relax 
ecutwfc                 200
scf_thr                 1e-7
scf_nmax                300
relax_nmax              400
relax_method            bfgs  
force_thr_ev            0.05  # ev
# stress_thr            5

#Parameters (3.Basis)
basis_type              lcao  # lcao or pw
kspacing                0.14 0.50 0.14 # replace KPT
# gamma_only            1  # 0/1, replace KPT

#Parameters (4.Smearing)
smearing_method         mp    # mp/gaussian/fixed
smearing_sigma          0.002  # Rydberg

#Parameters (5.Mixing)
mixing_type             broyden  # pulay/broyden
mixing_ndim               20

#Parameters (6.Calculation)
cal_force          1
cal_stress         1
out_stru           1  # print STRU in OUT
out_chg            0  # print CHG or not
out_bandgap        0
out_mul            1  

#Parameters (7. Dipole Correction)
efield_flag        1   # open added potential, if 0, all below useless
dip_cor_flag       1   # open dipole correction
efield_dir         1   # direction of dipole correction, 0,1,2 for x,y,z

the 1st SCF in relax process will not converge by using the v2.1(dev) NAO:

 GE297    1.81e+02   2.12e+02  -2.63782113e+05  -7.03217024e-03   1.6101e-04   1.0818e-03  60.76
 GE298    1.81e+02   2.12e+02  -2.63782115e+05  -1.96780108e-03   1.5566e-04   1.1212e-03  60.71
 GE299    1.81e+02   2.12e+02  -2.63782118e+05  -3.54125372e-03   1.0910e-04   1.8743e-03  60.62
 GE300    1.80e+02   2.12e+02  -2.63782137e+05  -1.90117122e-02   1.1790e-04   2.9315e-03  60.62
 >> Leave SCF iteration.
 * * * * * *
 !! CONVERGENCE HAS NOT BEEN ACHIEVED !!

While in v2.0 official NAO, this SCF can converge

 GE184    1.75e+02   2.01e+02  -2.63754454e+05  -2.48189066e-06   1.5504e-07   4.6288e-07  69.56
 GE185    1.75e+02   2.01e+02  -2.63754454e+05   5.37807322e-06   1.4567e-07   5.2853e-07  69.81
 GE186    1.75e+02   2.01e+02  -2.63754454e+05  -1.73654140e-05   1.3080e-07   4.4374e-07  69.84
 GE187    1.75e+02   2.01e+02  -2.63754454e+05   7.34717246e-06   1.2054e-07   5.5272e-07  69.71
 GE188    1.75e+02   2.01e+02  -2.63754454e+05   9.64171606e-06   1.0511e-07   5.2776e-07  69.58
 GE189    1.75e+02   2.01e+02  -2.63754454e+05  -5.70485401e-06   9.4869e-08   3.5031e-07  69.49
 >> Leave SCF iteration.
 * * * * * *

Attachment:
Fe5C2-CO-3fold-SCAN.tar.gz

Task list for Issue attackers (only for developers)

• Reproduce the performance issue on a similar system or environment.
• Identify the specific section of the code causing the performance issue.
• Investigate the issue and determine the root cause.
• Research best practices and potential solutions for the identified performance issue.
• Implement the chosen solution to address the performance issue.
• Test the implemented solution to ensure it improves performance without introducing new issues.
• Optimize the solution if necessary, considering trade-offs between performance and other factors (e.g., code complexity, readability, maintainability).
• Review and incorporate any relevant feedback from users or developers.
• Merge the improved solution into the main codebase and notify the issue reporter.

[QuantumMisaka]

NAO v2.1(dev) used in this issue

v2.1-std.tar.gz

and NAO v2.0 used is the official standard version

[QuantumMisaka]

It seems related to #4802 and #4058 , waiting for mixing part refinement

[QuantumMisaka]

In 3.7.3 version ,this convergence problem seems not solved, but even worse.

I'm doing and checking more tests

[WHUweiqingzhou]

@QuantumMisaka
could you try PR #4859, and make some tests with different mixing_eps, like 1e-10, 1e-12, 1e-14, 1e-15?

[QuantumMisaka]

@WHUweiqingzhou
I've tried them , all of them cannot lead SCF converge

(base) [2201110432@wm2-login01 test-mixeps]$ ls
abacus.err  abacus.slurm  eps14  INPUT                KPT         slurm.hosts  time.json
abacus.out  eps12         eps15  JobProcessing.state  OUT.ABACUS  STRU
(base) [2201110432@wm2-login01 test-mixeps]$ grep GE abacus.out | tail -n 1
 !! CONVERGENCE HAS NOT BEEN ACHIEVED !!
(base) [2201110432@wm2-login01 test-mixeps]$ grep GE abacus.out | tail -n 2
 GE300    1.83e+02   2.13e+02  -2.63782924e+05   1.91234615e-03   7.2299e-05   2.1935e-04  38.53
 !! CONVERGENCE HAS NOT BEEN ACHIEVED !!
(base) [2201110432@wm2-login01 test-mixeps]$ grep GE eps12/abacus.out | tail -n 2
 GE300    1.83e+02   2.13e+02  -2.63782926e+05   3.92081852e-04   5.9905e-05   3.6945e-04  19.12
 !! CONVERGENCE HAS NOT BEEN ACHIEVED !!
(base) [2201110432@wm2-login01 test-mixeps]$ grep GE eps14/abacus.out | tail -n 2
 GE300    1.83e+02   2.13e+02  -2.63782931e+05   8.10737831e-04   5.3708e-05   6.9621e-05  19.33
 !! CONVERGENCE HAS NOT BEEN ACHIEVED !!
(base) [2201110432@wm2-login01 test-mixeps]$ grep GE eps15/abacus.out | tail -n 2
 GE300    1.82e+02   2.13e+02  -2.63782936e+05   1.67324681e-04   5.2416e-05   5.3002e-04  19.18
 !! CONVERGENCE HAS NOT BEEN ACHIEVED !!

[mohanchen]

It seems related to charge mixing or smearing, it is not a diagonalization problem.

[mohanchen]

Convergence issues always exist in many examples, so let's change the issue to discussion panel.