[QE-users] oscillating energy value during optimization
Nicola Marzari
nicola.marzari at epfl.ch
Mon Jul 20 16:29:32 CEST 2020
Hi Coralie,
it's not easy to converge "difficult" systems with iterative approaches,
so it's also a matter of expertise and experience, but here are a few
pointers:
1) decreasing degauss makes things worse, due to level crossing
instabilities. It should be around 0.02 Ry, or even larger (but not more
than 0.03 Ry).
2) I would suggest to use the SSSP pseudopotentials - we had to work on
N quite a bit. Use the suggested cutoffs for the "efficiency" version.
If in doubt, run everything on the materials cloud qe input generator.
3) is the number of bands correct? you want to have enough bands to
cover also states that are starting to be quite empty, but you do not
want to have a much larger number. how many electrons do you have in
your calculation? you want something like 10% or max 20% more nbnd than
the number of electrons, but I would just let the code decide this for
you, and then check that the highest states (print verbose) are almost
empty, but not a lot that are completely empty. If the highest one has
an occupation that is below 0.01 or 0.005 you are fine.
4) your convergence threshold seems also loose - what does the qe input
generator tells you? use that (with the same pseudos that come with that
- the convergence threshold depends on the number of electrons in your
system). if your calculations are not well converged, the forces are not
very accurate, so the atoms are moved in wrong directions during the
relaxation, and thus the total energy from one ionic step to the next
will not go down.
5) can you try a simpler system first?
6) the fact that you put cosab = 6.12323e-17 rather than zero gets me
worried - there is no reason to have a number different from zero. Of
course 10e-17 is identical to zero, but it belies some carelessness.
nicola
On 20/07/2020 16:05, Coralie Khabbaz wrote:
> Hello,
>
> My structure is a WN slab with molecules adsorbing on its surface. I
> always have problems with convergence: everytime I try to relax the
> structure, multiple scf steps converge, but then the energy value starts
> oscillating and it doesn't converge anymore. My structure gets stuck
> every time, even though I visualized it after every scf step and the
> structure seems fine. This is the input file I am using. I also tried
> decreasing the degauss to 1e-2, increasing ecutrho and ecutwfc by a
> little bit, decreasing mixing_beta to 6e-2 and also tried changing the
> mixing_mode to TF, but nothing worked! Does anyone have advice? Thank you!
>
> &CONTROL
> calculation = "relax"
> forc_conv_thr = 1.00000e-03
> max_seconds = 4.32000e+05
> nstep = 200
> pseudo_dir =
> "/home/coralee/projects/def-jkopysci/coralee/.pseudopot"
> /
>
> &SYSTEM
> a = 8.32716e+00
> angle1(1) = 0.00000e+00
> angle2(1) = 0.00000e+00
> b = 8.98689e+00
> c = 2.52767e+01
> cosab = 6.12323e-17
> cosac = 6.12323e-17
> cosbc = -1.85547e-01
> degauss = 2.00000e-02
> ecutrho = 4.75221e+02
> ecutwfc = 5.03902e+01
> ibrav = 12
> nat = 53
> nbnd = 480
> nspin = 2
> ntyp = 4
> occupations = "smearing"
> smearing = "gaussian"
> starting_magnetization(1) = 2.00000e-01
> starting_magnetization(2) = 2.00000e-01
> starting_magnetization(3) = 0.00000e+00
> starting_magnetization(4) = 0.00000e+00
> /
>
> &ELECTRONS
> conv_thr = 1.00000e-05
> diagonalization = "david"
> electron_maxstep = 200
> mixing_beta = 1.00000e-01
> mixing_mode = "local-TF"
> startingpot = "atomic"
> startingwfc = "atomic+random"
> /
>
> &IONS
> ion_dynamics = "bfgs"
> /
>
> &CELL
> /
>
> K_POINTS {automatic}
> 2 2 1 0 0 0
>
> ATOMIC_SPECIES
> N 14.00674 N.pbe-n-kjpaw_psl.1.0.0.UPF
> W 183.84000 W.pbe-spn-kjpaw_psl.1.0.0.UPF
> C 12.01070 C.pbe-n-kjpaw_psl.1.0.0.UPF
> H 1.00794 H.pbe-kjpaw_psl.1.0.0.UPF
>
> ATOMIC_POSITIONS {angstrom}
> N 3.210864 2.727236 14.728253
> N 7.377521 2.749437 14.765113
> N 3.206885 -1.748226 14.750672
> N 7.368425 -1.739825 14.763088
> W 3.182078 4.921673 14.584931
> W 7.371236 4.911599 14.572299
> W 3.210468 0.427880 14.572328
> W 7.375435 0.429218 14.573322
> N 1.114319 5.237998 14.590842
> N 5.297509 5.253239 14.643339
> N 1.127456 0.739579 14.633097
> N 5.291670 0.741857 14.637218
> W 1.130917 3.035176 14.411937
> W 5.301873 3.012738 14.402683
> W 1.122976 -1.443293 14.413944
> W 5.288433 -1.443790 14.404037
> N 3.219452 5.199606 12.448287
> N 7.366466 5.188650 12.437657
> N 3.207970 0.704887 12.465361
> N 7.371281 0.697722 12.461527
> W 3.205392 3.068813 12.535650
> W 7.371059 3.075963 12.535225
> W 3.208342 -1.417371 12.526978
> W 7.371423 -1.415702 12.526259
> N 1.126718 3.566876 12.355498
> N 5.287387 3.570864 12.349260
> N 1.127573 -0.926625 12.347835
> N 5.290289 -0.925589 12.354591
> W 1.124934 5.676097 12.168934
> W 5.300233 5.687505 12.124192
> W 1.125212 1.183514 12.166731
> W 5.291832 1.185520 12.166744
> N 3.207882 3.506729 10.422372 0 0 0
> N 7.371462 3.506729 10.422372 0 0 0
> N 3.207882 -0.986715 10.422372 0 0 0
> N 7.371462 -0.986715 10.422372 0 0 0
> W 3.207882 5.716584 10.333236 0 0 0
> W 7.371462 5.716584 10.333236 0 0 0
> W 3.207882 1.223140 10.333236 0 0 0
> W 7.371462 1.223140 10.333236 0 0 0
> N 1.126093 6.199382 10.052766 0 0 0
> N 5.289672 6.199382 10.052766 0 0 0
> N 1.126093 1.705938 10.052766 0 0 0
> N 5.289672 1.705938 10.052766 0 0 0
> W 1.126093 3.871912 10.000000 0 0 0
> W 5.289672 3.871912 10.000000 0 0 0
> W 1.126093 -0.621532 10.000000 0 0 0
> W 5.289672 -0.621532 10.000000 0 0 0
> C 4.904436 5.088998 16.341705
> H 4.817236 4.340738 17.152672
> H 4.810563 5.831388 17.155825
> H 3.209512 2.314472 17.745318
> H 3.207503 3.080782 17.746006
>
>
> _______________________________________________
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users
>
--
----------------------------------------------------------------------
Prof Nicola Marzari, Chair of Theory and Simulation of Materials, EPFL
Director, National Centre for Competence in Research NCCR MARVEL, EPFL
http://theossrv1.epfl.ch/Main/Contact http://nccr-marvel.ch/en/project
More information about the users
mailing list