ABINIT Discussion Forums

Dear Abinit users,
I am trying to do a full structural relaxation, but it seems that there is a problem in my input file. the values of acell of the relaxed structure are far away from the experimental values and the same for the atomic possitions .
here is my input file:

#common declarations
natom 18
znucl 55 48 1
typat 1 1 1 1 1 1 2 2 3 3 3 3 3 3 3 3 3 3
ntypat 3
acell 8.43596 8.43596 12.6131 angstrom

xred 0.6825 0.1825 0
0.1825 0.3175 0
0.3175 0.8175 0
0.8175 0.6825 0
0 0 0.75
0 0 0.25
0 0.5 0.25
0 0.5 0.75
0.12318 0.62318 0.16510
0.37682 0.12318 0.16510
0.62318 0.87682 0.16510
0.87682 0.37682 0.16510
0.12318 0.62318 0.83490
0.37682 0.12318 0.83490
0.62318 0.87682 0.83490
0.87682 0.37682 0.83490
0 0 0
0 0 0.5

#angdeg 90 90 90
diemac 4 # le constant macroscopique dielectrique pour les
#isolants à large gap diemac varie de 2 à 4.
ecut 30
spgroup 140
brvltt -1
#prtcif 1
# structural optimization
#-------------------------------------------------------------
optcell 2
ionmov 3
dilatmx 1.0
ecutsm 0.5
tolmxf 1.0d-5
getwfk -1
ntime 100
natfix 6
natfixx 6
natfixy 2
natfixz 6
iatfix 5 6 7 8 17 18
iatfixx 5 6 7 8 17 18
iatfixy 5 6 7 8 17 18
iatfixz 5 6 7 8 17 18

ixc 11 # Integer for eXchange-Correlation choice,
#GGA, Perdew-Burke-Ernzerhof GGA functional
nband 23 #nombre de bandes,?

#Definition of the k-point grid
kptopt 1 # Option for the automatic generation of k
ngkpt 3 3 3
nshiftk 1
shiftk 0.5 0.5 0.5 # and shiftk 0.5 0.5 0.5 .

nstep 100
toldfe 1.0d-8
iscf 7

any help will be appreciate.
D.S

Hi!

There are several possible reasons.

1] Are your scf cycles converged?

2] Are your Broyden steps converged?

3] It's better to use tolvrs (e.x. 10^-18), than toldfe while relaxing geometry. http://www.abinit.org/documentation/helpfiles/for-v7.6/input_variables/varbas.html#toldfe

4] It's always advissed to relax the system in two steps: first optimize only the positions (xred) using optcell 0, then optimize cell size and positions together using optcell 2, starting with the relaxed positions from the first step. http://www.abinit.org/documentation/helpfiles/for-v7.6/input_variables/varrlx.html#optcell

5] Did you try another pseudopotential?

Best regards,

Igor

Hi Igor,
yes my scf cycles and Broyden steps are converged without error but the problem is: in the output files i find acell for a cubic structure (a=b=c) while in my case (a/= c) and the atomic positions are very different from the experimentel values.
thank you for the other suggestions, i will try them.
D. S

Regrading the acell, maybe you have to multiply a, b and c with respective rprim to get tetragonal acell.

http://www.abinit.org/documentation/helpfiles/for-v7.6/input_variables/varbas.html#acell
http://www.abinit.org/documentation/helpfiles/for-v7.6/input_variables/varbas.html#rprim
http://www.abinit.org/documentation/helpfiles/for-v7.6/input_variables/varbas.html#scalecart

Regrading xred, did you try tolvrs instead of toldfe as I told you in the previous post?

Igor

Hi Igor, sorry for this late
i tryed tolvrs 1.0d-18 with optcell 0 and ionmov 3 but the problem now that the Broyden steps are not converged even if nsteps 100, 1000, 10000
D.S

Hi!

You mention nstep while talking about Broyden steps. nstep variable regards SCF cycles, not Broyden algortihm. Did you want to say that your SCF cycles do not converge in 1000 steps?

1] If your scf doesn't converge, try to chagne iscf and/or add npulayit, nnsclo variables.

2] If your Broyden algorithm doesn't converge, try ionmov 2.



Igor

Hi Igor,
now i changed the pseudopotential to LDA, i set optcell 0 ionmov 2 and isf 2. the scf cycles and Broyden steps are converged, but when i put the relaxed xred in the input file with optcell 2 ionmov 2 here is the ERROR:
symatm.F90:221:ERROR
Largest error (above) is so large (0.01) that either input atomic coordinates (xred)
are wrong or space group symmetry data is wrong.
Action : correct your input file.

Hi!

Would you be so kind and try the optimization using a single input file which includes both relaxation options. See example below.

Igor


Single relaxation input example:

ndtset 2

# Set 1 : Internal coordinate optimization

ionmov1 2
ntime1 50
optcell1 0
tolmxf1 1.0e-8
natfix1 2
iatfix1 1 4
natfixz1 4
iatfixz1 2 3 5 6

# Set 2 : Lattice parameter relaxation (including re-optimization of
# internal coordinates)

dilatmx2 1.2
getxred2 -1
getwfk2 -1
ionmov2 2
ntime2 50
optcell2 2
tolmxf2 1.0e-8
strfact2 100
natfix2 2
iatfix2 1 4
natfixz2 4
iatfixz2 2 3 5 6

Hi Igor,
I try to optimize the structure using a single input file ( as in your exemple). it asked me each time to increase the value of dilatmx variable.

Sorry for the delay.

1] Well, increase it. I see that you put dilatmx 1.0. Try to increase it in steps of 0.05 until your algorithm works fine.

2] Do not go wild with this increasing. If it still gives you the same message even with 1.3, probably something else is amiss. Then maybe your initial xreds are too far away from the equilibrium positions.

Igor

Hi Igor,
I already tried to increase dilatmx up to 1.3 and it still shows the same error.
I took The xred values from an experimental work, do you mean that there is a problem in the experimentals values ​​of xred?
if the initial xreds are far away from the quilibrium positions, what can I do?
D.S

firstime wrote:I took The xred values from an experimental work, do you mean that there is a problem in the experimentals values ​​of xred?

No. I think that theoretical method we all use gives the position too far away from those experimental ones. It depends on the approximations used (pseudos, variable convergence, etc.). Theoretical values converge toward the exact theoretical results, and cannot be better than those. And exact theoretical values are different from the experimental ones.

Try to move a bit the initial xreds. Of course, only those not fixed by crystal symmetry. If you have time, look into your log file and try to figure out which positions are maybe blown away during the optimization. Or try to figure out which ones change most from the initial ones. Probably, these are the ones that need to be initially modified.

Igor