Top Page > Browsing
convegence problems with corehole calculations
Date: 2021/02/22 21:34
Name: Pavel Ondra&#269;ka   <pavel.ondracka@gmail.com>

Dear OpenMX users,

I'm having some troubles when calculating C1s core-level binding energies. My system is quite large metallic slab with an organic molecule on top. The initial state calculation proceed fine, but for the final state, I'm not able to get the desired convergence.

Initial state:
.....
SCF= 114 NormRD= 0.011270354777 Uele= -9089.475673747787
SCF= 115 NormRD= 0.004166053119 Uele= -9089.482412790698
SCF= 116 NormRD= 0.004221681470 Uele= -9089.480463992884
SCF= 117 NormRD= 0.002452479383 Uele= -9089.484680385594
SCF= 118 NormRD= 0.001494817522 Uele= -9089.482786454628
SCF= 119 NormRD= 0.012020815576 Uele= -9089.479252403156
SCF= 120 NormRD= 0.005807120780 Uele= -9089.483695256089
SCF= 121 NormRD= 0.007237372355 Uele= -9089.481885461102
SCF= 122 NormRD= 0.005478463835 Uele= -9089.483362516661
SCF= 123 NormRD= 0.007379230549 Uele= -9089.486446884825
SCF= 124 NormRD= 0.002591846672 Uele= -9089.487483838921
SCF= 125 NormRD= 0.002145402175 Uele= -9089.485290992994
SCF= 126 NormRD= 0.002689608405 Uele= -9089.483765546240
SCF= 127 NormRD= 0.001335981959 Uele= -9089.481873767179
SCF= 128 NormRD= 0.001212712930 Uele= -9089.480847465329
SCF= 129 NormRD= 0.003396245390 Uele= -9089.479695904614
SCF= 130 NormRD= 0.000822502227 Uele= -9089.481125661887
SCF= 131 NormRD= 0.001578554834 Uele= -9089.481151598757


Final state (atom 631, 1s 1)
...........
SCF= 282 NormRD= 0.158917825384 Uele= -9103.019708492246
SCF= 283 NormRD= 0.076997206802 Uele= -9103.055587345221
SCF= 284 NormRD= 0.049200580834 Uele= -9103.041121382084
SCF= 285 NormRD= 0.076797169753 Uele= -9103.062869821755
SCF= 286 NormRD= 0.192295867447 Uele= -9102.988948248669
SCF= 287 NormRD= 0.049221459031 Uele= -9103.047342859962
SCF= 288 NormRD= 0.312860170808 Uele= -9103.343974275827
SCF= 289 NormRD= 0.084819749175 Uele= -9103.048114274075
SCF= 290 NormRD= 0.409208742466 Uele= -9103.472523422823
SCF= 291 NormRD= 0.064801651458 Uele= -9102.994535544873
SCF= 292 NormRD= 0.121545166777 Uele= -9103.071732114371
SCF= 293 NormRD= 0.471939214286 Uele= -9103.082554061188
SCF= 294 NormRD= 0.165100890895 Uele= -9102.944222095535
SCF= 295 NormRD= 0.140732203933 Uele= -9102.944637562443
SCF= 296 NormRD= 0.094830259177 Uele= -9103.086367593169
SCF= 297 NormRD= 0.145883352565 Uele= -9103.002735412447
SCF= 298 NormRD= 0.050447216460 Uele= -9103.046188481858
SCF= 299 NormRD= 0.201910451143 Uele= -9103.221087996149
SCF= 300 NormRD= 0.103590824420 Uele= -9103.093293694754

Reading through the manual, I've already decreased the scf.Min.Mixing.Weight, increased both scf.Mixing.History and scf.Mixing.StartPulay and the scf.criterion is at 1e-4. I can make it somehow work if I increase the electronic temperature (like to at least 1000K) and also increase the scf.criterion. But it still doesn't really look like converged that well:

Final state (atom 631, 1s 1), electronic temperature 1000K, criterion 2e-4
SCF= 216 NormRD= 0.027479226643 Uele= -9095.812735925461
SCF= 217 NormRD= 0.019018535477 Uele= -9095.856211862683
SCF= 218 NormRD= 0.023605585288 Uele= -9095.869638786020
SCF= 219 NormRD= 0.019516041265 Uele= -9095.835435206580
SCF= 220 NormRD= 0.012707459808 Uele= -9095.826745068231
SCF= 221 NormRD= 0.011344694097 Uele= -9095.827195809532
SCF= 222 NormRD= 0.016335464431 Uele= -9095.823369007594
SCF= 223 NormRD= 0.016013689032 Uele= -9095.835625746116
SCF= 224 NormRD= 0.009938934179 Uele= -9095.834925893199
SCF= 225 NormRD= 0.015450779118 Uele= -9095.824127041793
SCF= 226 NormRD= 0.015648321523 Uele= -9095.826641568245
SCF= 227 NormRD= 0.012520307352 Uele= -9095.828724974021
SCF= 228 NormRD= 0.009105546672 Uele= -9095.830672967768
SCF= 229 NormRD= 0.011048213730 Uele= -9095.830549000135

I could use some help how to make it converge better and most importantly more reliably. I don't really care about the convergence speed, if it takes 500 or 1000 steps, its OK, but right now it feels like it starts oscilating at some point (I want to calculate BEs for all C atoms eventually, some C atoms from the organic molecule are a bit better with respect to the convergence, some worse).
メンテ
Page: [1]

Re: convegence problems with corehole calculations ( No.1 )
Date: 2021/02/22 21:37
Name: Pavel Ondra&#269;ka  <pavel.ondracka@gmail.com>

I was not able to include the whole structure as it is too big, so just the general settings

#---------------------------------------------------------------------------------------
#
# This was generated by OpenMX Viewer

System.CurrrentDirectory ./
System.Name abc
level.of.stdout 1
level.of.fileout 0
DATA.PATH /usr/local_rwth/sw/OpenMX/3.9/DFT_DATA

Species.Number 6
<Definition.of.Atomic.Species
H H6.0-s2p1 H_PBE19
C C7.0_1s-s4p3d2 C_PBE19_1s
C1 C7.0_1s_CH-s4p3d2 C_PBE19_1s
O O7.0_1s-s4p3d2 O_PBE19_1s
O1 O7.0_1s_CH-s4p3d2 O_PBE19_1s
Ti Ti7.0-s3p2d1 Ti_PBE19
Definition.of.Atomic.Species>

Atoms.Number 664
Atoms.SpeciesAndCoordinates.Unit Ang
<Atoms.SpeciesAndCoordinates
..........
Atoms.SpeciesAndCoordinates>

Atoms.UnitVectors.Unit Ang
<Atoms.UnitVectors
29.569000000000003 0.000000000000000 0.000000000000000
-14.784500000000001 25.607510000000001 0.000000000000000
0.000000000000000 0.000000000000000 38.000000000000000
Atoms.UnitVectors>

scf.XcType GGA-PBE
scf.SpinPolarization on
scf.ElectronicTemperature 300.0
scf.energycutoff 400.0
scf.maxIter 300
scf.EigenvalueSolver band
scf.Kgrid 1 1 1
scf.Mixing.Type rmm-diisk
scf.Init.Mixing.Weight 0.03
scf.Min.Mixing.Weight 0.005
scf.Max.Mixing.Weight 0.30
scf.Mixing.History 40
scf.Mixing.StartPulay 70
scf.criterion 1e-4

scf.dftD on
version.dftD 3

MD.Type nomd
#---------------------------------------------------------------

For final state add: (and rename the specific atom 631 from C to C1):
--------------------------------
scf.system.charge 1.0
scf.restart on
scf.coulomb.cutoff on
scf.core.hole on
<core.hole.state
631 s 1
core.hole.state>
-----------------------------

Also please note that the final state calculation needs to start from the initial state restart files, see http://www.openmx-square.org/forum/patio.cgi?mode=view&no=2654 for details.
メンテ
Re: convegence problems with corehole calculations ( No.2 )
Date: 2021/02/22 21:39
Name: Pavel Ondra&#269;ka  <pavel.ondracka@gmail.com>

Structure for download:
https://rwth-aachen.sciebo.de/s/OCbNcEWIZH9wt4b
メンテ
Re: convegence problems with corehole calculations ( No.3 )
Date: 2021/07/13 00:16
Name: Pavel Ondracka  <pavel.ondracka@email.cz>

I have some update here, so the issue is still present and it looks like the mixer has some issues with getting the spin-polarization properly in some cases. I have a non-magnetic system, so that when the core-hole is introduced, I would expect the total spin moment to be around 1 with the core-hole. This is the case for most of the calculations, however for some problematic atoms the total spin moment increases too much and the the system doesn't converge at all or only does so after several hundreds iterations.

Here is the initial state calculation:
https://drive.google.com/file/d/1fqAW6iWIsrvTluYgrtseC9iOjGulKhhi/view?usp=sharing

Here is the final state run for atom where it doesn't converge at all:
https://drive.google.com/file/d/1T6y5Ee1GQkznRfrr9cx2xwrOc_Bupban/view?usp=sharing (out file)
https://drive.google.com/file/d/1pTV7XpzsBBHb_ngQobt5t_bzTLgJgsxy/view?usp=sharing (standard output with the total spin moment)

How can I force the system to behave reasonably and not end with so large spin moment? Any help would be appreciated.
メンテ
Re: convegence problems with corehole calculations ( No.4 )
Date: 2021/07/13 00:18
Name: Pavel Ondracka  <pavel.ondracka@email.cz>

And here is another case, where the total spin moment went down to -6 before somehow stabilizing after several hundred iterations and the case converges finally.
https://drive.google.com/file/d/1i_2iG_g2oMfqzKAVoey7IYsZhvBTB591/view?usp=sharing (out file)
https://drive.google.com/file/d/1V74dGBsqJn-oo_DNuOorKVyAg861sKkg/view?usp=sharing (standard output with the total spin moment)
メンテ
Re: convegence problems with corehole calculations ( No.5 )
Date: 2021/07/19 15:49
Name: Pavel Ondracka  <pavel.ondracka@email.cz>

It seems the rmm-diisv does better than diisk but unfortunately isn't working 100% as well.
メンテ
Re: convegence problems with corehole calculations ( No.6 )
Date: 2021/07/20 10:11
Name: T. Ozaki

Hi,

Following the examples linked in the No.3 thread, I have performed the calculations with
the following basis functions:

<Definition.of.Atomic.Species
H H6.0-s2p1 H_PBE19
C C7.0_1s-s3p2d1 C_PBE19_1s
C1 C7.0_1s_CH-s3p2d1 C_PBE19_1s
O O7.0_1s-s3p2d1 O_PBE19_1s
O1 O7.0_1s_CH-s3p2d1 O_PBE19_1s
Ti Ti7.0-s3p2d1 Ti_PBE19
Definition.of.Atomic.Species>

and obtained the results shown below:
(the initial state: https://t-ozaki.issp.u-tokyo.ac.jp/calcs/init1.out )

For the insulating treatment:
Eb = -36978.596575750627+36989.222070037053-0.16945142009652 = 10.45604 Hartree = 284.52 eV
https://t-ozaki.issp.u-tokyo.ac.jp/calcs/final1.out

For the metallic treatment:
Eb = -36978.769900625644+36989.222070037053 = 10.45217 Hartree = 284.41 eV
https://t-ozaki.issp.u-tokyo.ac.jp/calcs/final2.out

I have noticed that the neighboring atoms of the core hole tend to be spin polarized
in the final state calculation, which makes the SCF convergence difficult.

I wonder that the induced spin polarization of the neighboring atoms is likely
to physically occur in some systems such as transition metal surface as discussed in
https://doi.org/10.1103/PhysRevLett.84.3970
https://doi.org/10.1103/PhysRevB.43.9558


Regards,

TO
メンテ
Re: convegence problems with corehole calculations ( No.7 )
Date: 2021/07/20 17:15
Name: Pavel Ondracka  <pavel.ondracka@email.cz>

Thanks for looking into it, I see you use explicit Kerker factor, I'll take a look at that and I'll try to experiment with the basis set a bit as well, so far I've just played with the mixing parameters. However as far as I can see even with your basis set the convergence was not obtained for the insulating treatment, it just ended after 1200 iterations. The metallic treatment seems to do better, but I have some reservations if it is suitable for the system (especially for C atoms not directly at the metallic surface), but the results seem to suggest it doesn't make a big difference so maybe it is OK after all.

BTW The problem is indeed looking like transition metal (in my case Ti surface) specific, I have a similar calculation with the same molecule on top of Al (it converges with no issues) and TiAl (some minor issues but nothing as pronounced as for the Ti). I did a brief look at the linked manuscripts, but I'm not sure if this is related to my issue. For me the total spin monent is zero for the initial state calculations so addition of the the molecule in itself does nothing, it is just when the core-hole is added, that everything goes wrong, while in the manuscripts the magnetic moment was there directly after the addition of the impurity.

Anyway thanks again for the pointers, I'll play with it a bit and report back.
メンテ
Re: convegence problems with corehole calculations ( No.8 )
Date: 2021/07/20 18:42
Name: T. Ozaki

Hi,

One may consider that the core hole behaves as a magnetic impurity, and the on-site Hund coupling
and intersite coupling lead to magnetization of neighboring atoms.

Regards,

TO
メンテ

Page: [1]

Thread Title (must) Move the thread to the top
Your Name (must)
E-Mail (must)
URL
Password (used in modification of the submitted text)
Comment (must)

   Save Cookie