strange fluctuation in PBE0 convergence

[jasius]

Hi all, I am trying to calculate DOS of a large supercell with PBE0. I use PBE and the performed non SC calculation, reading both charges and wavecar. During the iteration 19 the SCF which seemded to be converging all of a sudden experience in a jump. While it recovers afterwards it is still struggling to converge afterwards. I was wondering on why this happens

thank you

vasp.5.4.4.18Apr17-6-g9f103f2a35 (build May 01 2020 14:29:13) complex

POSCAR found type information on POSCAR P Mg O N H Fe
POSCAR found : 6 types and 459 ions
scaLAPACK will be used
LDA part: xc-table for Pade appr. of Perdew
found WAVECAR, reading the header
WARNING: stress and forces are not correct
POSCAR, INCAR and KPOINTS ok, starting setup
FFT: planning ...
reading WAVECAR
the WAVECAR file was read successfully
charge-density read from file: Chabazite_01Ns0
magnetization density read from file 1
entering main loop
N E dE d eps ncg rms rms(c)
DAV: 1 0.126662339203E+04 0.12666E+04 -0.94101E+02 16816 0.276E+01
DAV: 2 0.126305172425E+04 -0.35717E+01 -0.35707E+01 18240 0.603E+00
DAV: 3 0.126299480254E+04 -0.56922E-01 -0.56916E-01 19664 0.982E-01
DAV: 4 0.126299336572E+04 -0.14368E-02 -0.14368E-02 19488 0.127E-01
DAV: 5 -0.310805978586E+04 -0.43711E+04 -0.70579E+02 18432 0.249E+01
DAV: 6 -0.314310852629E+04 -0.35049E+02 -0.75327E+01 17616 0.935E+00
DAV: 7 -0.314554139101E+04 -0.24329E+01 -0.35985E+00 22256 0.218E+00
DAV: 8 -0.314569168046E+04 -0.15029E+00 -0.21215E-01 20160 0.571E-01
DAV: 9 -0.314570489048E+04 -0.13210E-01 -0.12067E-02 22720 0.178E-01
DAV: 10 -0.314570706766E+04 -0.21772E-02 -0.12024E-03 21160 0.696E-02
DAV: 11 -0.314570765406E+04 -0.58640E-03 -0.33528E-04 22960 0.338E-02
DAV: 12 -0.314570786161E+04 -0.20755E-03 -0.20898E-04 17624 0.199E-02
DAV: 13 -0.314570795941E+04 -0.97793E-04 -0.17665E-04 16248 0.135E-02
DAV: 14 -0.314570802211E+04 -0.62697E-04 -0.16361E-04 13880 0.102E-02
DAV: 15 -0.314570807178E+04 -0.49674E-04 -0.15584E-04 11928 0.845E-03
DAV: 16 -0.314570811536E+04 -0.43578E-04 -0.15019E-04 11328 0.737E-03
DAV: 17 -0.314570815523E+04 -0.39874E-04 -0.14514E-04 11080 0.664E-03
DAV: 18 -0.314570819236E+04 -0.37130E-04 -0.14032E-04 10840 0.608E-03
DAV: 19 -0.314702171630E+04 -0.13135E+01 -0.13111E+01 10608 0.559E-03
DAV: 20 -0.314554350030E+04 0.14782E+01 -0.58320E+01 16856 0.237E-01
DAV: 21 -0.314570760791E+04 -0.16411E+00 -0.34123E-02 17216 0.930E-02
DAV: 22 -0.314570827373E+04 -0.66582E-03 -0.30131E-03 14344 0.260E-02
DAV: 23 -0.314570834754E+04 -0.73807E-04 -0.22274E-04 14408 0.873E-03
DAV: 24 -0.314570837770E+04 -0.30165E-04 -0.12097E-04 13664 0.486E-03
DAV: 25 -0.314570840408E+04 -0.26378E-04 -0.11259E-04 11128 0.409E-03
DAV: 26 -0.314570842931E+04 -0.25230E-04 -0.10892E-04 10056 0.377E-03
DAV: 27 -0.314570845361E+04 -0.24302E-04 -0.10560E-04 9848 0.354E-03
DAV: 28 -0.314570847707E+04 -0.23453E-04 -0.10245E-04 9528 0.334E-03
DAV: 29 -0.314570849971E+04 -0.22648E-04 -0.99296E-05 9384 0.317E-03
DAV: 30 -0.314570852162E+04 -0.21904E-04 -0.96402E-05 9264 0.299E-03
DAV: 31 -0.314570854281E+04 -0.21192E-04 -0.93507E-05 9088 0.284E-03
DAV: 32 -0.314570856332E+04 -0.20511E-04 -0.90688E-05 9240 0.272E-03
DAV: 33 -0.314570858318E+04 -0.19863E-04 -0.87999E-05 8984 0.261E-03

ISTART = 1
#INIWAV = 1
ICHARG = 11
NWRITE = 2
LPETIM = F ! timer
LWAVE = T ! write "WAVECAR"
LCHARG = T ! write "CHGCAR"
LVTOT = F ! write "LOCPOT"
LELF = F ! write "ELFCAR"
LAECHG = F ! the core charge is written to AECCAR0 and the valance charge to AECCAR2
LADDGRID = F
# ------------------------------------------------------------------------------------
GGA = PE ! Perdew-Burke-Ernzerhof
PREC = ACCURATE
ENCUT = 520
NELM = 55
NELMIN = 4
NELMDL = -5
EDIFF = 0.000001
EDIFFG = -0.005
ISIF = 2 ! The cell shape and the volume is relaxed
ISYM = 1 ! Symmetry is ON
LCORR = T
ISMEAR = -5
##SIGMA = 0.05
LREAL = AUTO

##IVDW = 12

ISPIN = 2
##MAGMOM = 16*0 15*0 160*0 15*0 252*0 1*5
LORBIT = 11

LHFCALC = .TRUE.
ALGO = Normal
IMIX = 1
AMIX = 0.1

#-------------------
#parallel stuff
NCORE = 24

[alexey.tal]

Dear jasius,

Thank you for your question.

I see that you are running a spin-polarized calculation. I think that this jump in the total energy is very likely related to the change in the magnetic state, but it is hard to tell without seeing the OUTCAR file from this calculation. Please see our forum guidelines for what should be attached with the post in order for us to be able to help you with your question.

Best wishes,
Alexey

[jasius]

thank you and please see attached

OUTCAR.zip

[alexey.tal]

Unfortunately, this OUTCAR file appears to be damaged and I'm not able to open it. Also, it would be helpful if you could provide all the input files.
If you would like to achieve a more smooth convergence you can try to follow suggestions from this wiki page. In particular, you should try reducing the mixing.

[jasius]

My apologies Alexei, I attached all the files and made sure they are all readable. This is a large system and takes long time for me to run but not only it flips the spin (I see in PBE calculation negative spin when it converges) but struggles to converge at the end. I would appreciate the suggestions (and I will try making linear mixing).

The link you referred to suggest restarting and reading it from the WAVECAR. I would like that, e.g., converging with PREC=NORMAL and CUToff 400 eV and then perhaps converge with higher precision. But when I try that using the input below it still goes through 4 steps of positive energy and it starts what it seems like with initial magnetization in MAGMOM. Could you recommend the most efficient way of restarting the lower level calculation?

thank you

[alexey.tal]

I cannot read the content of the OUTCAR after line 2502. See the screenshot below:

pic_resized.png

I see that you are using the PBE0 functional. I would suggest that you first converge your calculation with PBE with the hard cutoff (should be way cheaper) and then restart from the WAVECAR with PBE0. Also, with the hybrid functional calculation I would suggest that you use ALGO=Damped or ALGO=All.

In the PBE calculation with ALGO=Normal, I think that the linear mixing should help in this situation. Another thing would be to try setting MAGMOM, if you know what magnetic state should be the lowest. That should speed up the convergence a lot.