Linearization energies in Lu2O3

Dear Users,

I am running calculations on doped cubic Lu2O3. I use default specie for Lu with RMT=2 and a modified O specie with RMT=1.45 and without the local orbital of order 3 present in the default specie. I use RPP potential-only meta-GGA.

With these species, everything runs smooth most of the times. In case that matters - linearization energy of f-type local orbital of Lu turns out to be slightly positive (like, 0.1-0.2 Ha), while its initial value is -0.3291. Fermi energy is something like 0.3-06 Ha.

However, sometimes I get severe problems with linearization enegries, which manifest in the respective warnings, but, more importantly, in wrong MT charges on all atoms. Normally, converged calculations result in MT charge on Lu of ~67.5, while in problematic ones it's 65+-10. The problematic calculations sometimes converge fine, sometimes do not converge, and sometimes converge with Lu MT charge if 65 - in the latter case 1 or 2 linearization energies are not found.

I wonder if removing some of the LOs from Lu can help. How can I know if I can safely remove a LO? If I make a test calcualtion to check the effect, what do I look at?

Thank you in advance.
Best regards.
Andrew

EDIT: I remebered another hack I use sometimes. I have a script which reads LINENGY.OUT from a given calculation, calculates average values of linearization energies per specie and orbital, and writes those values to species files. With this, I can transfer the linengies from a converged calculation to a problematic one. Sometimes it helps. Sometimes it does not.

Also, I did not mention that I tried playing around with rgkmax, gmaxvr, trimvg and msmoth - again, sometimes it helps, but mostly it does not.

Last edit: Andrew Shyichuk 2019-03-14

Andrew Shyichuk - 2019-03-14

I've run some test calculations using LDA. Lu2O3 with default Lu specie and with Lu without the f LO converge smoothly (in 22 and 23 loops, respectively), differ negligibly in density of states plots, and their total energies are -469908.936416 and -469908.933963 Ha, respectively. Can I conclude that the f LO on Lu was not necessary (in this case)?

However, calculations without either s or p LO on Lu fail to converge. Noteworthy, the one without p LO exhibits strange behavior of MT charges I mentioned above.

Last edit: Andrew Shyichuk 2019-03-14

Lars Nordström - 2019-03-15

Hi!

You say you use default species file for Lu.in, which assumes looks like
'Lu' : spsymb
'lutetium' : spname
-71.0000 : spzn
318945.3295 : spmass
0.237356E-06 2.8000 57.8887 700 : rminsp, rmt, rmaxsp, nrmt
21 : nstsp
1 0 1 2.00000 T : nsp, lsp, ksp, occsp, spcore
2 0 1 2.00000 T
2 1 1 2.00000 T
2 1 2 4.00000 T
3 0 1 2.00000 T
3 1 1 2.00000 T
3 1 2 4.00000 T
3 2 2 4.00000 T
3 2 3 6.00000 T
4 0 1 2.00000 T
4 1 1 2.00000 T
4 1 2 4.00000 T
4 2 2 4.00000 T
4 2 3 6.00000 T
4 3 3 6.00000 F
4 3 4 8.00000 F
5 0 1 2.00000 F
5 1 1 2.00000 F
5 1 2 4.00000 F
5 2 2 1.000000 F
6 0 1 2.00000 F
1 : apword
0.1500 0 F : apwe0, apwdm, apwve
0 : nlx
7 : nlorb
0 2 : lorbl, lorbord
0.1500 0 F : lorbe0, lorbdm, lorbve
0.1500 1 F
1 2 : lorbl, lorbord
0.1500 0 F : lorbe0, lorbdm, lorbve
0.1500 1 F
2 2 : lorbl, lorbord
0.1500 0 F : lorbe0, lorbdm, lorbve
0.1500 1 F
3 2 : lorbl, lorbord
0.1500 0 F : lorbe0, lorbdm, lorbve
0.1500 1 F
3 3 : lorbl, lorbord
0.1500 0 F : lorbe0, lorbdm, lorbve
0.1500 1 F
-0.3291 0 T
0 3 : lorbl, lorbord
0.1500 0 F : lorbe0, lorbdm, lorbve
0.1500 1 F
-2.2383 0 T
1 3 : lorbl, lorbord
0.1500 0 F : lorbe0, lorbdm, lorbve
0.1500 1 F
-1.0908 0 T

where the lines
4 3 3 6.00000 F
4 3 4 8.00000 F
means you have the 4f as valence states. Then you ought to take care of these low lying states with a proper LO with a linearisation energy to adjust to the position of the semi-core states. But maybe something goes wrong as you end up with linearization energy in valence region ...
You can check: How does your partial 4f DOS look like? At what energies are the 14 bands corresponding to Lu-4f situated in EIGENVAL.OUT? Do these shifts among the problematic calculations.

You can try to either lower the starting guess of the semi-core state from -0.3291 Ha or to put the 4f in core. The latter should be much less problematic. Then what are the 4f eigen-values in the core, deep enough?

Good luck,
Lars

Andrew Shyichuk - 2019-03-15

Dear Lars,

Yes, this is the specie, other then RMT, which was 2 in my case.

I've taken another look at my calculations - it looks like the f linengy is not the only one making trubles, the Lu s and p ones are also often not found.

Please consider the DOS plot (RPP09 functional) with colored PDOS, zero is at eigenvalue zero (i.e. I had Fermi energy added to the x axis of the as-obtained DOS plot).
While f linengy matches the peak of f PDOS, it is not so for s linengy (-1.7957) and p linengy (-0.6236). Minimal eigenvalue was -1.82, so it looks like the two linengies stick to 1s and 2p states, Given 5s and 5p treated as valence orbitals, should the respective linengies match their PDOS peaks (i.e. ~0.022 and 0.047), just like the f linengy does?

Thank you.
Andrew

Last edit: Andrew Shyichuk 2019-03-15

TDOS-Lu2O3-RPP.png

Lars Nordström - 2019-03-15

Hi!!
I would guess you have semi-core s and p states below 0 energy at the corresponding linesrisation energies too.
Isuggest you try f in core to compare
Best eishes,
Lars

Andrew Shyichuk - 2019-03-15

Dear Lars,

Indeed, 5s and 5p must be lower. The s- and p-PDOS peaks of lutetium in the image are very weak (were enlarged 50 times) and are a side effect of Lu-O overlap. The area in the image is mostly populated by O s and p (and Lu f) states.
The way I see it, it is not ok to treat higher-lying Lu 4f states as core, while there are lower-lying semi-core / valence Lu 5s and 5p states (although I might be wrong and I will try it).

That, however, brings me back to square 1: if the 4f states are not the only problem, how can I improve linearization energy search for Lu 5s and 5p states? A test calculation with demaxbnd=5 (also, it is not on the manual) is running. Anything else?

Thank you.
Andrew

UPD: I've made a different DOS plot, in a broader range. Interestingly. Lu p linengy does not match the Lu p PDOS peak, while at Lu s linengy there are no Lu s states whatsoever.

- Last edit: Andrew Shyichuk 2019-03-15
  
  TDOS-Lu2O3-RPP-2.png

Lars Nordström - 2019-03-15

OK, to be clear.
That the 4f states are at positive energies I think is a LDA error, they are lying to high ...
What happens when you put the f in core?
Remember, you do not expect any f-bonding in rare earths as LDA/GGA always predict.
Therefore it is common to treat them as core or to use LDA+U to localize them.
Best,
Lars

Lars Nordström - 2019-03-15

Sorry I was probably too brief ...
When you have very narrow bands close to Fermi energy, as your f-bands, details in bonding and hybridization will change occupations drastically leading to oscillations and unstable numerics.
The cause of this problem is that these bands are unphysically (unchemically) high in energy due to LDA/GGA errors.
Have a nice weekend,
Lars

Andrew Shyichuk - 2019-03-15

Dear Lars,

In case of Lu2O3 and RPP09 functional, the localization of f orbitals is quite legit. I am not saying it's exact, but, for instance, Dorenbos model predicts that they must be something like 2 eV below the top of valence band, and that is where I get them.
My calculation with 4f in core is running, as of 2nd iteration the Lu charges look fine, which is a good sign.

Thank you, and have a nive weekend too.
Andrew

Andrew Shyichuk - 2019-03-18

Dear Lars,

I've tried to do the f-in core calculation, both with the respective LO and without it. Both were really bad, not even approaching convergence after 160 loops. Thus, the problem must lay somewhere else.

Best regards.
Andrii