ABINIT Discussion Forums

Hi,
I am investigating a real material according to entries in the ICSD database: hexagonal P63mmc BaTiS3.
However, my phonon calculation "explodes" --- the ETOT during several iterations starts off looking for a minimum, and then grows to >1.0x10E18.

I believe I have followed best practices in line with what I have read on this forum and in the tutorials, so I am completely at a loss as to what I am doing wrong. Can you help?

1. My procedure. First, using occopt 7 tsmear 0.005 nkpt 16 16 16 and shiftk 0 0 0.5, (these are common inputs in my lab group), I relax a 10 atom BaTiS3 to <1.0d-6 and get force output:
###########
cartesian forces (eV/Angstrom) at end:
1 0.00000000000000 0.00000000000000 0.00000000000000
2 0.00000000000000 0.00000000000000 0.00000000000000
3 0.00000000000000 0.00000000000000 0.00000000000000
4 0.00000000000000 0.00000000000000 0.00000000000000
5 -0.00004212512204 -0.00002432095055 0.00000000000000
6 0.00000000000000 -0.00004864190110 0.00000000000000
7 0.00004212512204 -0.00002432095055 0.00000000000000
8 0.00004212512204 0.00002432095055 0.00000000000000
9 0.00000000000000 0.00004864190110 0.00000000000000
10 -0.00004212512204 0.00002432095055 0.00000000000000
###########

I take the xred coordinates and acell from this relaxation and use it as inputs to my 2 step phonon calc, which I amended following the Abinit tutorial. here is my input file:

###############3
ndtset 2

#self-consistent run (do a relaxation prior to this to get the lattice params and xred you need)
iscf1 3
kptopt1 1
tolvrs1 10d-18

#Set 2 : response-function calculations for all needed perturbations
getwfk2 -1
iscf2 3
kptopt2 2
nqpt2 1
qpt2 0 0 0
rfphon2 1
rfatpol2 1 10
rfstrs2 3
rfdir2 1 1 1
tolvrs2 1.0d-10

#common input data
occopt 7
tsmear 0.005
angdeg 90 90 120

#Definition of the atom types and atoms
ntypat 3
znucl 56 22 16
natom 10
typat 2*1 2*2 6*3
nband 56
acell 1.24619849627768E+01 1.24619849627768E+01 1.07370991566719E+01
xred
3.33333333333333E-01 6.66666666666666E-01 7.50000000000000E-01
6.66666666666666E-01 3.33333333333333E-01 2.50000000000000E-01
1.22447328210655E-40 0.00000000000000E+00 0.00000000000000E+00
1.22447328210655E-40 7.34683969263930E-40 5.00000000000000E-01
3.39768817052261E-01 1.69884408526131E-01 7.50000000000000E-01
1.69884408526131E-01 3.39768817052261E-01 2.50000000000000E-01
8.30115591473869E-01 1.69884408526131E-01 7.50000000000000E-01
6.60231182947739E-01 8.30115591473869E-01 2.50000000000000E-01
8.30115591473869E-01 6.60231182947739E-01 7.50000000000000E-01
1.69884408526131E-01 8.30115591473869E-01 2.50000000000000E-01

nstep 120

ecut 25.0
ecutsm 0.5
ixc 2

ngkpt 16 16 16
nshiftk 1
shiftk 0 0 0.5
#############################
Now, the choice of occopt 7 and not doing the elec response or piezo response should be OK as the BaTiS3 evaluates to a metal, ie. a grep eV *.log leads to no gap being shown.

As well, I have used enough bands, as there are several that remain empty (as seen in the initial relaxation and in the phonon calc).

My forces after the SCF run for the overall phonon script shows:
################cartesian forces (eV/Angstrom) at end:
1 0.00000000000000 0.00000000000000 0.00000000000000
2 0.00000000000000 0.00000000000000 0.00000000000000
3 0.00000000000000 0.00000000000000 0.00000000000000
4 0.00000000000000 0.00000000000000 0.00000000000000
5 -0.00007294376740 -0.00004211410374 0.00000000000000
6 0.00000000000000 -0.00008422820748 0.00000000000000
7 0.00007294376740 -0.00004211410374 0.00000000000000
8 0.00007294376740 0.00004211410374 0.00000000000000
9 0.00000000000000 0.00008422820748 0.00000000000000
10 -0.00007294376740 0.00004211410374 0.00000000000000
#####
So, these are still "good" and < 1meV/A.

Yet my ETOT "runs away" after the 2nd iteration following SCF convergence. Here is an excerpt of grep ETOT *.log:
#### (begin): (SCF)
ETOT 1 -225.90670490190 -2.259E+02 3.430E-01 1.715E+04 5.629E-01 5.629E-01
ETOT 2 -206.39500359010 1.951E+01 1.922E-01 1.198E+05 1.740E+00 1.177E+00
.
.
.
ETOT 27 -229.23466236540 -6.253E-13 7.511E-04 1.433E-18 1.207E-11 1.638E-06
#########(end SCF --- completed and converged) (Begin 1st ph. iter.
-ETOT 1 1818.9711124103 -6.132E+02 1.873E+01 9.163E+06
ETOT 2 61.167930658041 -1.758E+03 2.778E+00 1.093E+05
ETOT 3 34.129221637761 -2.704E+01 9.608E-01 1.414E+05
ETOT 4 5.9236058427327 -2.821E+01 3.370E-01 8.533E+03
ETOT 5 4.8440606112617 -1.080E+00 1.429E-01 2.072E+03
ETOT 6 4.5561568039211 -2.879E-01 6.096E-02 2.930E+02
ETOT 7 4.4636071256909 -9.255E-02 7.224E-02 9.180E+01
.
.
.
ETOT 38 4.4472053190284 -1.819E-11 2.647E-03 4.157E-11
#####(end 1st ph iter. -- completed and converged)
#######(Next -- 2nd ph iter:)
-ETOT 1 1290.1103474881 -5.053E+02 1.122E+01 5.996E+06
ETOT 2 1326.8274580917 3.672E+01 1.472E+01 3.222E+06
ETOT 3 1415.2531194128 8.843E+01 2.364E+01 3.600E+06
ETOT 4 4.7527898395319 -1.411E+03 3.653E+00 1.280E+04
ETOT 5 2.0968711495229 -2.656E+00 6.336E-01 1.770E+03
ETOT 6 3.5041825349390 1.407E+00 2.218E-01 4.079E+03
ETOT 7 1.7620269946558 -1.742E+00 3.961E-02 3.242E+02
ETOT 8 1.6898365527311 -7.219E-02 7.670E-02 1.642E+01
ETOT 9 1.6918354621810 1.999E-03 2.245E-02 1.378E+01
ETOT 10 1.7005789700797 8.744E-03 3.245E-02 3.049E+01
ETOT 11 1.6873802847776 -1.320E-02 1.834E-02 2.192E-01
ETOT 12 1.6874065023553 2.622E-05 1.537E-02 2.003E-01
.
.
.
ETOT 22 1.6894999623280 -1.365E-03 2.954E-03 5.699E+00
ETOT 23 1.6946198310934 5.120E-03 6.582E-03 1.793E+01
ETOT 24 1.6940648305585 -5.550E-04 2.457E-03 1.855E+01
ETOT 25 1.7125899772252 1.853E-02 5.160E-03 6.756E+01
ETOT 26 1.7024428741986 -1.015E-02 2.649E-03 3.784E+01
ETOT 27 1.7643608210922 6.192E-02 3.466E-03 2.108E+02
ETOT 28 1.7303419903008 -3.402E-02 1.586E-03 1.112E+02
ETOT 29 1.8414102403406 1.111E-01 5.993E-03 3.813E+02
ETOT 30 1.8270411359167 -1.437E-02 1.704E-03 3.870E+02
ETOT 31 2.1817082137736 3.547E-01 5.181E-03 1.300E+03
ETOT 32 2.0076583212669 -1.740E-01 6.202E-03 8.152E+02
ETOT 33 3.3247089729214 1.317E+00 8.949E-03 4.519E+03
ETOT 34 2.5379309409338
.
.
.
ETOT 61 1338206.2896201 9.448E+05 1.259E+04 3.297E+09
ETOT 62 1152023.7156858 -1.862E+05 8.048E+03 3.159E+09
ETOT 63 4736588.7198249 3.585E+06 4.079E+04 1.282E+10
ETOT 64 2514249.3483283 -2.222E+06 2.062E+04 6.244E+09
ETOT 65 13070246.569097 1.056E+07 8.947E+04 3.541E+10
ETOT 66 7312298.1510583 -5.758E+06 5.835E+04 1.918E+10
ETOT 67 25945974.485643 1.863E+07 2.472E+05 6.389E+10
ETOT 68 22360490.540047 -3.585E+06 1.389E+05 6.173E+10
ETOT 69 87120036.501026 6.476E+07 7.887E+05 2.328E+11
ETOT 70 49031869.552340 -3.809E+07 4.737E+05 1.231E+11
ETOT 71 260046222.10705 2.110E+08 1.808E+06 7.119E+11
.
.
.
ETOT116 2.23578776634417E+17 -1.950E+17 1.675E+15 5.910E+20
ETOT117 8.16473058031813E+17 5.929E+17 9.988E+15 2.010E+21
ETOT118 6.41600462211735E+17 -1.749E+17 5.113E+15 1.765E+21
ETOT119 2.75484568285127E+18 2.113E+18 2.301E+16 7.434E+21
ETOT120 1.34755566995016E+18 -1.407E+18 1.306E+16 3.357E+21
######### (and that's it.....it has "run away"...... and even though I stop this after nstep 120, I don't
think that going further would help?)


Any ideas on what I am doing wrong and how I can correct?

Thanks,
John B.

I can think of one idea, but not sure why it would/would not help: changing iscf from anderson mixing to CG method.

What pseudopotentials are you using?

I am using LDA based PSPs created using the Opium software and they are termed norm-conserving, designed non-local with 25 Ha cutoff in energy for the plane waves. Thus, my ecut in ABINIT is limited to a maximum 25 Ha as well.

So, my change from iscf 3 (Anderson mixing) to iscf 5 (Congruent Gradient) worked for BaTiS3 in P63mmc.

But, before declaring victory, I tried a derivative compound, PbTiS3 also arranged in P63mmc.
The iterations on ETOT were going along smoothly, but then hit a rough patch near the 3rd to the end and 2nd to the end iterations: I am using a converge criterion of tolvrs 1.0d-10 in the phonon convergence area. Here are a few excerpted line from a grep on ETOT on the 3rd to last iteration:
ETOT 34 1.6209813372991 -2.489E-10 1.885E-03 3.698E-10
ETOT 35 1.6209813371501 -1.490E-10 2.964E-03 3.817E-10
ETOT 36 1.6209813370590 -9.101E-11 1.782E-03 2.352E-10
ETOT 37 1.6209813370045 -5.451E-11 2.793E-03 1.725E-10
ETOT 38 -29.964148297439 -3.159E+01 2.633E-03 1.632E-10
ETOT 39 -29.964151805387 -3.508E-06 2.483E-03 1.559E-10
ETOT 40 -29.964153844183 -2.039E-06 2.343E-03 1.489E-10
ETOT 41 -29.964180462241 -2.662E-05 2.211E-03 1.386E-06
ETOT 42 -29.964156178580 2.428E-05 2.088E-03 1.060E-10
ETOT 43 -29.964155342722 8.359E-07 1.972E-03 8.250E-11

So, it almost gets there, but has problems and restarts
I am given this message in the log file:

Potential-based CG line minimization has trouble to converge.
The algorithm is restarted with more secure parameters.

Now, I am not sure if I should worry about this. But the change from a positive value to a negative value has me concerned. The 2nd to last iteration has the same behavior but occurs at tolvrs = 1.82E-07:
ETOT 32 1.6209778892280 -7.612E-10 5.027E-03 1.820E-07
ETOT 33 -29.967261605789 -3.159E+01 4.724E-03 7.068E-08

So, since the criteion of tolvrs 1.0d-10 is arbitrary, I re-ran this structure using tolvrs 1.9d-10 to eliminate one of these "oddities". The affect on the obtained phonons was negligible.

So, my question is: I this algorithm re-start a problem?

Thanks,
JB

I should add that it is not really ETOT that is being converged, but the 2nd derivative of the energy

Hi, first I would say that the calculation was not "going along smoothly" because it was trying to converge to a positive ETOT, which would indicate an unbound system. Finally it jumped over into a true minimum and then converged. This kind of restart in my experience happens much more with metals than anything else. Is this system metallic? If so you might try 1) adding more conduction bands; 2) investigating dependence on occopt and tsmear.

I think there has been a misunderstanding here: Though abinit labels iterations with ETOT, for structure convergence I thought ETOT meant etotal and the log file labels it as such.

The ETOT iterations I describe earlier seem to have a column heading of 2nd derivative of energy, as I'd expect from the tie-in to phonons. So, they could be positive. Indeed, wouldn't we want them to be positive to indicat a local minimum has been reached?