[Wien] Continuing confusion in dos...

[Peter Blaha]

Just a few guidelines how to select a proper k-mesh and the nband parameter:


a) Convergence parameters from "other" calculations are only a 
guideline, but your system could be very different.

b) How to select the k-mesh: Obviously, since -hf is very expensive, you 
should test this with a normal GGA. Do various calculations with 
different k-meshes for the property you are interested in (eg. DOS) 
until you know how it changes.

Once this is done, you could go to -hf, BUT: this might be too 
expensive. Then the option of a "reduced k-mesh" comes into play, since 
this can very much decrease computer time. Eg: You found out that you 
need a 4x4x4 kmesh for a reasonable DOS, use this, but also a reduced 
mesh of 2x2x2,

c) selection of "nband": also here, you have to "think", not just do 
what others have done for a very different system.

Suppose you do Si: It has 4 electrons/atom, i.e. 2 bands/atom = 4 
bands/cell are occupied. Now consider the number of bands if the 
electronic shell would be fully occupied, i.e. all s+p states, i.e. 8 
e/atom, 4 bands/atom = 8 bands/cell. So nband should be at least twice 
as large as the number of occupied bands.

Consider a Cu atom in some cell: it has 11 electrons/atom, so 
contributes 5.5 bands/atom. A full shell would be minimum 12 e (3d+4s), 
better 18e (3d+4sp). Nband does not need to be too large as compared to 
the occupied bands.

If you, however, have a Ti atom, it has just 4 e/atoms (2 bands), but 
still probably all unoccupied 3d states should be included in a HF 
calculation, so probably you have to go from 4 to 12e/Ti atom.

Now add this up for all your atoms. And in addition, if it is a 
spin-polarized case, you cannot divide by 2.

d) Don't ask others if your calculation is converged. How should we know 
without doing a calculation. Unless you are very experienced (or have 
anyway unlimited computer ressources), start with low/medium converged 
parameters, do the calculation. Increase the parameters and compare your 
results. WIEN2k is pretty efficient, because it can use your previous 
scf density and may converge in few cycles with the better parameters 
(unless you do the better calculation in a new directory with a new 
initialization. NEVER DO THIS.)

e)  The next wien2k release will help you a lot in this respect, as it 
has a new parameter -prec 0-3 in init_lapw. It will select k-mesh and 
Rkmax automatically based on your -prec level AND the number/type/RMT of 
your atoms in your cell.

Am 29.11.2021 um 09:09 schrieb Gavin Abo:
> There is the following, but maybe it will not help since it is just a 
> list of links to previous posts already given in the mailing list.
>
> At [1], it advises to use "at least one iteration".
>
> In your post at [2] if I have interpreted it correctly, you have a 3D 
> structure (cubic spinel supercell with 56 atoms).
>
> With regards to your 2x2x1 k-mesh, I have seen k-meshes, such as 3x3x1 
> [3], with 1 k-point in the z-direction like that used for 2D 
> structures [modeled in 3D as a supercell with vacuum (which might also 
> go by the names surface [4], slab, or monolayer)].  Though, maybe more 
> than 1 k-point in the z-direction is need for some 3D structures.
>
> Post at [5] mentions a 2x2x2 is a fine start.  Then, from there you 
> could increase the k-mesh (e.g., 3x3x3, ..., 5x5x5, ...) for k-point 
> convergence testing.
>
> In post [6], it looks like 6x6x6 was a k-mesh were k-point convergence 
> was being reached in many cases.
>
> It also looked like in [7] for DOS with -hf that it could be important 
> that convergence testing had been done with nband in case.inhf.
>
> [1] 
> https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg20756.html
> [2] 
> https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg21511.html
> [3] 
> https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg21083.html
> [4] Slide 29 titled "Surfaces": 
> http://susi.theochem.tuwien.ac.at/events/ws2006/WIEN2k-getting_started.pdf
> [5] 
> https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg20861.html
> [6] 
> https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg20384.html
> [7] 
> https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg21069.html
>
> On 11/28/2021 10:47 PM, shamik chakrabarti wrote:
>> Dear Dr. Tran,
>>
>>                  I have applied 4k points using the procedure as you 
>> have suggested. However, for one iteration with 4k points, it is 
>> coming as "the energy in scf is not converged". Also, the DOS looks 
>> different from 1 k points. I am attaching two plots here. Is it that, 
>> to get the correct DOS we have to run more than one iteration 
>> until the energy is converged?
>>
>> Looking forward to your suggestions in this regard.
>>
>> with regards,
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> SEARCH the MAILING-LIST at: 
> http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html

-- 
-----------------------------------------------------------------------
Peter Blaha,  Inst. f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-158801165300
Email: peter.blaha at tuwien.ac.at
WWW:   http://www.imc.tuwien.ac.at      WIEN2k: http://www.wien2k.at
-------------------------------------------------------------------------