SCF convergence detoriates and oscillates after certain steps during structure relaxation with vacancies

Running Abinit Ground state
1

Hai,
I am doing positron lifetime calculation in Au 3x3x3 supercell with 4 vacancies

SCF convergence suddenly deteriorates and starts oscillating

TC-DFT STEP 1 - ELECTRONIC GROUND-STATE CALCULATION
ETOT 1 -14370.949818419 -1.437E+04 3.327E-04 1.252E+02 2.056E-02 2.056E-02
ETOT 2 -14369.885027813 1.065E+00 2.879E-05 4.583E-01 3.268E-03 1.729E-02
ETOT 3 -14369.788157398 9.687E-02 3.758E-05 7.578E-01 1.903E-03 1.919E-02
ETOT 4 -14369.780950526 7.207E-03 2.757E-05 7.040E-03 3.757E-04 1.919E-02
ETOT 5 -14369.782041279 -1.091E-03 8.689E-06 2.215E-03 2.671E-04 1.945E-02
ETOT 6 -14369.782148266 -1.070E-04 9.593E-06 5.818E-04 4.515E-05 1.948E-02
ETOT 7 -14369.782142363 5.903E-06 8.083E-06 5.763E-05 4.448E-05 1.952E-02
ETOT 8 -14369.782144917 -2.554E-06 2.338E-06 1.829E-05 1.343E-05 1.951E-02
ETOT 9 -14378.181294990 -8.399E+00 1.224E+01 2.552E-01 1.413E-02 1.930E-02
ETOT 10 -14369.724128201 8.457E+00 6.569E-03 2.497E-03 1.406E-02 1.950E-02
ETOT 11 -14369.773722613 -4.959E-02 1.882E-03 2.826E-04 7.493E-05 1.950E-02

looking at the log following warning occurs when a sudden jump occurs

!COMMENT
src_file: m_lobpcg2.F90
src_line: 665
message: |
An old style abi_xorthonormalize happened but now I’ll try to continue :wink:
…

au333_4_corner_vac_sft3_middle_lt_relax.abi (9.4 KB)
au333_4_corner_vac_sft3_middle_lt_relax.abo (118.3 KB)
log.txt (257.6 KB)

ps: I am using custom potential generated with 5s5p5d6s (added txt extension to upload)
this potential is verified to reproduce lattice parameter and bulk modulus

Au.LDA-PW-paw_5s5p5d6s.abinit.txt (694.5 KB)

Looking forward to suggestions for this convergence issue
Thanks
-rajaraman

2

Hello Rajaraman,

Your approach looks perfectly fine, but I would just increase the number of bands. I tried a quick test with 100 more bands and it worked fine.
Good luck!
Bogdan

3

Thank you Bogdan.

I never thought of increasing nband (tried removing manual definition and allowed code to calculate optimum nband, which I used as a guideline)

I just started tinkering nline and nnsclo.

I will try your suggestion.

Is it the reason for the
" An old style abi_xorthonormalize happened but now I’ll try to continue" warning when it start oscillating?

Thanks a lot
-rajaraman