problem with Iron antiferromagnetic structure E vs V calculation

[Praveen]

Hello,
I am new to DFT and elk.I am trying to plot energy vs volume curves for Iron-FCC-antiferromagnetic structures.I have 7 combinations of antiferromagnetic fcc.But the nature of plot is not parabolic for few combinations.The input script for one such combination for which the plot is not parabolic,is as shown below.

! FCC_AntiFerro-magnetic_1: Fe

tasks
0
10
21
! use Broyden mixing of the potential
mixtype
3
xctype
20
sppath
'/home/praveen/elk-2.2.5/species/'
avec
6.481761000 0.000000000 0.000000000
0.000000000 6.481761000 0.000000000
0.000000000 0.000000000 6.481761000

scale
1.04036802621

autolinengy
.true.

autormt
.true.

beta0
0.001

betamax
0.3

atoms
1 : nspecies
'Fe.in' : spfname
4 : natoms; atposl, bfcmt below
0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 1.00000000
0.00000000 0.50000000 0.50000000 0.00000000 0.00000000 1.00000000
0.50000000 0.00000000 0.50000000 0.00000000 0.00000000 1.00000000
0.50000000 0.50000000 0.00000000 0.00000000 0.00000000 -1.00000000
ngridk
45 45 45

autokpt
.false.
gmaxvr
14

vkloff
0.5 0.5 0.5
spinpol
.true.

! empty states
nempty
10
epspot
0.1e-7

!force calculation
tforce
.true.

The energy vs volume plot is smoth and quadratic till it reaches minimum.But after that,it becomes linear.Can someone please please help me know where I am going wrong? Thanks in advance.

Praveen

[Praveen]

Hello,
I am new to DFT and elk.I am trying to plot energy vs volume curves for Iron-FCC-antiferromagnetic structures.I have 7 combinations of antiferromagnetic fcc.But the nature of plot is not parabolic for few combinations.The input script for one such combination for which the plot is not parabolic,is as shown below.

! FCC_AntiFerro-magnetic_1: Fe

tasks
0
10
21
! use Broyden mixing of the potential
mixtype
3
xctype
20
sppath
'/home/praveen/elk-2.2.5/species/'
avec
6.481761000 0.000000000 0.000000000
0.000000000 6.481761000 0.000000000
0.000000000 0.000000000 6.481761000

scale
1.04036802621

autolinengy
.true.

autormt
.true.

beta0
0.001

betamax
0.3

atoms
1 : nspecies
'Fe.in' : spfname
4 : natoms; atposl, bfcmt below
0.00000000 0.00000000 0.00000000 0.00000000 0.00000000 1.00000000
0.00000000 0.50000000 0.50000000 0.00000000 0.00000000 1.00000000
0.50000000 0.00000000 0.50000000 0.00000000 0.00000000 1.00000000
0.50000000 0.50000000 0.00000000 0.00000000 0.00000000 -1.00000000
ngridk
45 45 45

autokpt
.false.
gmaxvr
14

vkloff
0.5 0.5 0.5
spinpol
.true.

! empty states
nempty
10
epspot
0.1e-7

!force calculation
tforce
.true.

The energy vs volume plot is smoth and quadratic till it reaches minimum.But after that,it becomes linear.Can someone please please help me know where I am going wrong? Thanks in advance.

Praveen

[Markus]

Dear Praveen,

I'm not sure this will help you, but I see some problems in your input file:

a) avoid Broyden mixing in magnetic systems if you don't have the time to carefully adjust the parameters. It may flip the magnetic state wildly around, giving you arbitrary E-V-curves. (By the way, the beta-parameters only work for mixtype 1, adaptive linear mixing.)

b) You set magnetic fields of 1 and -1 on the Fe atoms. This is huge and should be reduced in every SCF-cycle by setting reducebf to, eg,, 0.7.

Good luck with that!

Markus

PS: I recommend you use the latest elk version, which is a lot faster than 2.2.5, typically by almost 50% for me.

[Praveen]

Dear Markus,
Thank you for your feedback.I will take your suggestions and proceed with my calculations. It may sound very silly,but do I have to again perform k-point convergence and then plot E vs V curve or I can use previous k-point convergence data and proceed with it? Sorry for asking such silly questions

Praveen

[Lars Nordström]

Dear Praveen,

I agree with Markus two points.

I just want to add that your curve looks like two crossing curves and that the Broyden mixing might have let you to jump from one solution to another. How do the atomic moments look like as function of volume?

Lars

[Praveen]

Dear Lars,
Thank you for your suggestion.No I have not checked how atomic moments vary as function of volume.I calculated other magnetic structures with broyden mixing.They turned out fine(of Fe) except this one structure. I will check how atomic moments look.Also it would be very helpful if you could answer my silly question "do I have to again perform k-point convergence and then plot E vs V curve or I can use previous k-point convergence data and proceed with it?".Sorry for such silly question.

Praveen

[J. K. Dewhurst]

Hi Praveen, Markus and Lars,

The elk.in file does not contain 'reducebf' which is required if you start with a large (~1) magnetic field. Remember that at field of 1.0 in elk.in corresponds to 1715 Tesla.

Also, 45x45x45 is far too large a k-point grid for something simple like the moment.

Here is my modified elk.in file:

tasks
 0

spinpol
 .true.

xctype
 20

sppath
 '../../species/'

avec
6.481761000 0.000000000 0.000000000
0.000000000 6.481761000 0.000000000
0.000000000 0.000000000 6.481761000

scale
 0.95

nempty
 10

atoms
 1 : nspecies
'Fe.in' : spfname
 4 : natoms; atposl, bfcmt below
 0.00000000 0.00000000 0.00000000 0.00000000 0.00000000  0.001
 0.00000000 0.50000000 0.50000000 0.00000000 0.00000000  0.001
 0.50000000 0.00000000 0.50000000 0.00000000 0.00000000  0.001
 0.50000000 0.50000000 0.00000000 0.00000000 0.00000000 -0.001

ngridk
 8 8 8

... which gives:

Finally, it's good practice to use 'highq=.true.', this should give nicely converged results in most cases.

Regards,
Kay.

[Praveen]

Thank you John Kay Dewhurst . Now I am getting proper curves.