I am doing tddft calculations of spin dynamics. There are quite different results when I use elk-8.5.10 and elk-8.8.26 with the same input file. I am not sure which ontcome is better. I am wandering are there any important changes between these two versons on the part of time evolution of the electronic state on femtosecond time-scales with TDDFT? Or is there any critical parameter I ignored in the input file? The following is my inpit file, as well as the results. Could anyone help me understand this problem?
Thanks in advance.
All the best,
Yilyu
tasks
0
1
450
460
xctype
3
mixtype
1
beta0
0.01
lradstp
2
! no symmetry operations used for the ground-state run
nosym
.true.
autolinengy
.true.
rgkmax
5
tshift
.false.
spinpol
.true.
fsmtype
-2
spinorb
.true.
wrtdsk
.false.
! fairly large number of empty states required for magnetic cases
nempty
5
! DFT+U block
! here AMF double counting is used (dftu=2)
! inpdftu=1 corresponds to provide U and J as input
dft+u
1 1 : dftu, inpdftu
1 2 0.06614878 0.002939946 : is, l, U, J
avec
19.4499999999999990 0.0000000000000036 0.0000000000000000
-1.9450000000000001 3.3688388207214701 0.0000000000000000
0.0000000000000000 0.0000000000000000 20.0000000000000000
TDDFT on top of a FSM ground-state will be fairly unstable, so it's important to converge the calculation well.
The later version of Elk uses more single-precision arithmetic in the TDDFT time step which can enhance this instability.
I ran your input file with
epspot
1.e-8
nempty
10
and obtained much more stable results with version elk-8.8.26. This indicates that your ground-state calculation was not converged enough for a stable TDDFT run.
You may also like to look at convergence with respect to the the k-point set and the plane-wave cut-off rgkmax.
Regards,
Kay.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Dear Elk community,
I am doing tddft calculations of spin dynamics. There are quite different results when I use elk-8.5.10 and elk-8.8.26 with the same input file. I am not sure which ontcome is better. I am wandering are there any important changes between these two versons on the part of time evolution of the electronic state on femtosecond time-scales with TDDFT? Or is there any critical parameter I ignored in the input file? The following is my inpit file, as well as the results. Could anyone help me understand this problem?
Thanks in advance.
All the best,
Yilyu
tasks
0
1
450
460
xctype
3
mixtype
1
beta0
0.01
lradstp
2
! no symmetry operations used for the ground-state run
nosym
.true.
autolinengy
.true.
rgkmax
5
tshift
.false.
spinpol
.true.
fsmtype
-2
spinorb
.true.
wrtdsk
.false.
! fairly large number of empty states required for magnetic cases
nempty
5
mommtfix
1 1 0 0.000 1.222
1 2 0 -1.162 0.377
1 3 0 -0.718 -0.988
1 4 0 0.718 -0.989
1 5 0 1.161 0.378
tshift
.false.
tddos
.true.
tdmag3d
.true.
plot3d
0 0 0
0 0 1
0 1 0
1 0 0
80 80 80
ntswrite
400
ntsbackup
100
sppath
'~/workspace/22-11/elk/species/'
! DFT+U block
! here AMF double counting is used (dftu=2)
! inpdftu=1 corresponds to provide U and J as input
dft+u
1 1 : dftu, inpdftu
1 2 0.06614878 0.002939946 : is, l, U, J
avec
19.4499999999999990 0.0000000000000036 0.0000000000000000
-1.9450000000000001 3.3688388207214701 0.0000000000000000
0.0000000000000000 0.0000000000000000 20.0000000000000000
scale
1.8897261339212518764104149343425
atoms
2 : nspecies
'Ni.in' : spfname
5 : natoms; atpos, bfcmt below
0.0000000000000000 0.0000000000000000 0.4999999999999999 0 0.0000000 -0.0001222
0.2000000000000000 -0.0000000000000001 0.4999999999999999 0 0.0001162 -0.0000377
0.4000000000000000 -0.0000000000000003 0.4999999999999999 0 0.0000718 0.0000988
0.6000000000000000 -0.0000000000000004 0.4999999999999999 0 -0.0000718 0.0000989
0.8000000000000000 -0.0000000000000006 0.4999999999999999 0 -0.0001161 -0.0000378
'I.in' : spfname
10 : natoms; atpos, bqfcmt below
0.0666666666666673 0.6666666666666651 0.5764889302304050 0.00000000 0.00000000 0.00000000
0.1333333333333330 0.3333333333333360 0.4235110697695960 0.00000000 0.00000000 0.00000000
0.2666666666666669 0.6666666666666650 0.5764889302304050 0.00000000 0.00000000 0.00000000
0.3333333333333329 0.3333333333333360 0.4235110697695960 0.00000000 0.00000000 0.00000000
0.4666666666666670 0.6666666666666651 0.5764889302304050 0.00000000 0.00000000 0.00000000
0.5333333333333329 0.3333333333333361 0.4235110697695960 0.00000000 0.00000000 0.00000000
0.6666666666666670 0.6666666666666641 0.5764889302304050 0.00000000 0.00000000 0.00000000
0.7333333333333328 0.3333333333333360 0.4235110697695960 0.00000000 0.00000000 0.00000000
0.8666666666666669 0.6666666666666641 0.5764889302304050 0.00000000 0.00000000 0.00000000
0.9333333333333330 0.3333333333333360 0.4235110697695960 0.00000000 0.00000000 0.00000000
!this k-point set is too small for calculation of accurate moments
ngridk
4 4 1
! total simulation time
tstime
3307
dtimes
0.2
! laser pulse parameters
! 1 - 3 : polarisation vector (including amplitude)
! 4 : frequency
! 5 : phase in degrees
! 6 : chirp rate
! 7 : peak time
! 8 : full-width at half-maximum
pulse
1 : number of laser pulses
30 0.0 0.0 0.01433223562 0.0 0.0 1747.06 1033.53
wplot
3000 100 1 : nwplot, ngrkf, nswplot
-1 1 : wplot
Hi Yilyu,
TDDFT on top of a FSM ground-state will be fairly unstable, so it's important to converge the calculation well.
The later version of Elk uses more single-precision arithmetic in the TDDFT time step which can enhance this instability.
I ran your input file with
and obtained much more stable results with version elk-8.8.26. This indicates that your ground-state calculation was not converged enough for a stable TDDFT run.
You may also like to look at convergence with respect to the the k-point set and the plane-wave cut-off rgkmax.
Regards,
Kay.
Thank you Kay. It's very helpful for me. I tried to increase the rgkmax, and it also works.
Have a nice day!
Kind regards,
Yilyu