Atomic or metallic?

Total energy, geometry optimization, DFT+U, spin....

Moderator: bguster

Locked
fizixer
Posts: 2
Joined: Sun Mar 11, 2018 10:03 pm

Atomic or metallic?

Post by fizixer » Sun Mar 11, 2018 10:13 pm

I'm trying to do calculation of a large metallic element (Bismuth but later other large metals too) and I'm wondering about doing it like tutorial 1 atomic (one atom only) vs tutorial 4 metallic (using a known crystal structure config).

I tried to do it similar to tutorial 1, but for atomic Bi not molecular Bi2, and I'm getting the warnings "density went too small". Also I'm a bit confused as to the right values of the following:

- occopt (I'm using 3?)
- tsmear (should I use it? not necessary?)
- nkpt (1 nkpt is not enough? I've heard metals need high nkpt but is that for metallic calc or atomic calc?)
- diemac (I commented out, since for metals it has to be very high and default value works)
- ecut (again the warning says I should use a high ecut but I'm already using 30 or 50 Ha)
- nsppol (I read in another thread that turning this on fixed convergence issues)

A related question is that I'm probably using the word 'metal' with two different meanings: metal as in electropositive element or one with high electrical conductivity, vs metal as in in a crystal structure state, like tutorial 4, and not "gaseous" atomic state, like tutorial 1.

Could you help me be more systematic with my calculations, and any resources to read up on beyond the tutorials, and yet not like papers (that have tons of formalism and math are a bit too complicated for what I'm trying to learn).

fizixer
Posts: 2
Joined: Sun Mar 11, 2018 10:03 pm

Re: Atomic or metallic?

Post by fizixer » Sun Mar 11, 2018 11:20 pm

I'm coming across this term "ensemble DFT" originally described by Marzari, Vanderbilt, Payne in 1997 (PRL 79, 7) and I see no mention of it in abinit website, documentation, or forums.

The technique helps eliminate convergence issues associated with metallic calculations.

How can I do ensemble DFT, or some equally effective method in eliminating convergence issues, in ABINIT? Quantum ESPRESSO has it through the CP package (another plus for QE since CP is car parrinello, a much more efficient form of MD).

I've been a long time user of ABINIT, when most of my friends used QE, and I'm a little disappointed to see something in QE that ABINIT doesn't seem to have.

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Atomic or metallic?

Post by mverstra » Tue Mar 13, 2018 2:03 pm

Dear friend of abinit!

Don't worry, we are here. There are several confusing terms (metal and ensemble DFT, at least) and we need to be clear about what you actually want to do:

To get a real metal you need a system which is not finite (at least a wire, plane, 3D solid). Your atom will not be a real metal, even if it's last orbitals are open shell. If this is the case, you need to either include spin and occupy them explicitly (example for Bismuth with 5 valence electrons):

Code: Select all

nsppol 2
nkpt 1
nband 6
occ
# spin up
1.0  1.0  1.0  0.0  0.0  0.0
# spin down
1.0  1.0  0.0  0.0  0.0  0.0
occopt 2
wtk 1.0


or use a "metallic" occupation

Code: Select all

occopt 3
tsmear 10 K

which gives a statistical weight to each orbital and should automatically enforce something close to:

Code: Select all

occ
# spin up s and p orbitals filled partially
1.0  0.5  0.5  0.5  0.0  0.0
# spin down idem
1.0  0.5  0.5  0.5  0.0  0.0

This is what you did, and it's fine, but it simulates a thermodynamical ensemble of atoms or molecules, in the sense of Marzari's PRL and his phd thesis (very pedagogical if you want the details of occupation functions). ABINIT also has occopt 4 & 5 which are two flavors of Nicola's optimal occupations - these get the 0K total energy of solids well, even if the smearing is large (which is nice because it accelerates the kpt convergence). I would not recommend this for atoms, or for systems with delicate Fermi surfaces - the occupations are not monotonic and could give you something weird for Bi IMHO (to be tested).

Nowadays this is usually called thermal occupation or modified Fermi functions, etc.

Other meanings of ensemble DFT: there are more recent techniques which combine explicitly the ground and excited states of a KS system, to access the ionization potential and other quantities not linked just to a single KS Slater determinant. These are experimental and not implemented in any distributed code AFAIK. Here is one recent publication on the topic https://journals.aps.org/prl/abstract/1 ... 119.033003

Check out the new input variable documentation:
https://www.abinit.org/sites/default/fi ... vars.html#

Now back to your real problems:

[*] "density went too small" this is not a problem, just numerical noise where the density fluctuates slightly below 0 and is reset.

[*] The default tsmear is very large (0.04 I think). You should set it, as above.

[*] diemac is useful to improve convergence for isolated systems with vacuum. For you atom it may be necessary, try 4 or 10 if the SCF cycle is having a hard time.

[*] ecut is probably fine, Bi is a soft element for the psp, unless you are including the semicore d electrons, but even then 50Ha should be ok (use the pseudo-dojo potentials online, or the HGH ones if you need SOC).
Matthieu Verstraete
University of Liege, Belgium

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Atomic or metallic?

Post by mverstra » Tue Mar 13, 2018 2:06 pm

PS: what is the problem, actually? Did your calculation converge? Did you obtain something unexpected for the atom? There are already tons of publications using abinit for Bi, but it is not clear from your initial email what the issue is.
Matthieu Verstraete
University of Liege, Belgium

Locked