KNO3 optimization for polarization calcul.

[maxim]

Hi all,

I'm truing to perform constant-volume optimization (for berryopt -1) of ferroelectric gamma-KNO3 as it has already been done here http://www.ims.uconn.edu/~alpay/JPCM01.pdf. In this article, all authors report a 10% shortening of the c-axis, while I have it growing
The value of c has great importance in calcul. of polarization.

Maybe my k-points set is not enought (psps were downloaded here http://www.wfu.edu/~natalie/papers/pwpa ... index.html).

I will be glad for any advice

Thank you!

Code: Select all

# gamma-KNO3

getwfk -1

chkprim 0

#Parameters
nstep  1000
ecut 38
pawecutdg 95.
toldfe 1.0d-6

#Definition of occupation numbers
occopt 7
tsmear 0.2 eV

#Full optimization of the lattice
optcell 3
ionmov  3
ntime  500
tolmxf 2.0d-4
dilatmx 1.2
ecutsm  0.5

natom 15
natrd 3
ntypat 3
typat 1 2 3
znucl 19 7 8
xred 0.0000000000 0.0000000000 0.0000000000
0.0000000000 0.0000000000 0.4050000000
0.1309800000 0.8690200000 0.4340000000

spgroup 160
brvltt 0
acell 5.4870000000 5.4870000000 9.1560000000 angstrom
angdeg 90 90 120

# Kpoint Grid
kptopt 1 # option for automatic generation of k points
ngkpt 12 12 8
nshiftk 1
shiftk 0.0 0.0 0.0

# pawprtvol -3

# diemac 4.0

# Berry phase calculation of the polarization
# berryopt -1
# rfdir 1 1 1
# symmorphi 0

[maxim]

It should be noted when ABINIT starts optimization far away from the true structure it faces a lot of false local minima... In other words setting correct cell parameters and coordinates atoms as starting point for geometry optimization is critical. Now I'm trying to use more recent data from http://www.crystallography.net/informat ... if=2104622 instead of old data http://www.crystallography.net/informat ... if=1000051. But full optimization with these coordinates slips past the minimum and later I'll try to reduce ecutsm...

[mverstra]

What do you mean by slips past the minimum?

The local minimum problem can be real (true PES has lots of minima) - do you know if your PES is complex or whether the algorithm is having difficulty on its own?

your dilatmx is huge. ecutsm is less of a problem. Also your tolmxf is a bit high, and your ecut as well (really need 38 Ha?). Finally with toldfe 1.e-6 you are not guaranteed to have converged forces - have you checked this? If you converge the structure with unconverged forces it's kind of pointless...

M.

[maxim]

sorry for sluggishness
One more question... Do I need ecut corresponding 1mHa convergence (38Ha) when performing geom. optim?
thank you

[mverstra]

this has been hanging quite a while:

That's the whole point - you need to converge what you want to extract. Structures need converged forces. There is no absolute criterion on energy convergence which can guarantee a good structure. You should converge ecut and nkpt to get good forces on your initial structure, then relax.