Ti2B2 monolayer the result doesn't converge

Hello, I am currently working on Ti2B2 monolayer that has a hexagonal structure. The structure is the same as Al2B2 from this paper https://arxiv.org/pdf/2110.09136.pdf and I was able to converge successfully (same structure as code below). However, for Ti2B2 (code below) the The ETOT values are jumping and as reached the max number of iterations for one cycle.

I have tried toldfe to 1.0d-6, toldff, and changing diemac and checking other input variables but nothing seems to be working. The only improvement I have been able to make with toldff is each step converging but the overall TIM_DEN doesn’t converge. Does anyone have any suggestions on how to converge this or does the structure not compatible with the material?

I have also read another paper https://arxiv.org/pdf/1802.08473;Hexagonal stating that the global minimum phase of Ti2B2 is a 2 by 2 cell. Can some explain the different between the papers and if this could be the reason why the file isn’t converging?


acell 5.7324837864 5.7324837864 20 # used a = b = 3.0335 A value from paper converted to bohr

rprim sqrt(0.75) -0.5 0.0 # FCC primitive vectors (to be scaled by acell)
sqrt(0.75) 0.5 0.0 # not fully sure
0.0 0.0 1.0

ntypat 2
znucl 22 5
pp_dirpath “$ABI_PSPDIR”
pseudos “PAW/Ti.GGA_PBE-JTH.xml, PAW/B.GGA_PBE-JTH.xml”

Definition of the atoms

natom 4 #
typat 1 1 2 2 # 1 - Al 2 - B
0 0 0.15242057 # calculated from the vertical height given divided by 2 and vacuum
0 0 -0.15242057
1/3 1/3 0
2/3 2/3 0

#Numerical parameters of the calculation
ecut 20

pawecutdg 40

Definition of the k point grid

kptopt 1
ngkpt 10 10 1
nshiftk 1
shiftk 0 0 0.5

Parameters for the SCF procedure

nstep 1000
toldfe 1.0d-10

diemac 12.0

tolmxf 5.0d-5

#Optimization of the lattice parameters
optcell 1
ionmov 2
ntime 1000
dilatmx 1.05
ecutsm 0.5

I’m in the middle of my research and got stuck here. If you can shed any light, it would be greatly appreciated. Thank you.

Hi e3uydl

Ti is not Al - perhaps the compound is not as stable, but the scf cycle should work.

  1. try norm conserving psp, with semi core states for the Ti
  2. change the PAW dataset for Ti, as a test. It’s a delicate atom to pseudize
  3. is it metallic? If so diemac 12 is not good - leave the default, and if it’s still problematic try iprcel 45 and related options to precondition the SCF cycle.
  4. check your lattice vectors in the log file. The input looks ok but I’m not sure the sqrt will work as expected

Thank you so much for your response. I have a few questions regarding your suggestions.

  1. How do use norm conserving PSP? Do I just replace pseudos “PAW/Ti.GGA_PBE-JTH.xml, PAW/B.GGA_PBE-JTH.xml” with something else? if so, where do I get the pseudopotentials from?

  2. https://arxiv.org/pdf/1802.08473;Hexagonal in the article I tried running the 2 by 2 cell relaxation and it also didn’t converge, does it make a difference where I try the new psps on?

  3. I have tried the default diemac and doesn’t make a difference.

  4. In my log file (for the two by two cell) I got this warning but I’m not sure what it means, do you mind explaining.

src_file: m_drivexc.F90
src_line: 711
message: |
Density went too small (lower than xc_denpos) at 48 points
and was set to xc_denpos = 1.00E-14. Lowest was -0.30E-04.
This might be due to (1) too low boxcut or (2) too low ecut for
pseudopotential core charge, or (3) too low ecut for estimated initial density.
Possible workarounds : increase ecut, or define the input variable densty,
with a value larger than the guess for the decay length, or initialize your,
density with a preliminary LDA or GGA-PBE if you are using a more exotic xc functional.

Thanks for your help.