I was optimizing U value for W in WO3 system. I found without U it is showing 1.3 eV and with U=6 eV (varying 2, 3, 4 etc.) it is showing 1.54 eV and for 10 eV it is showing 1.75 eV. It is going 2.5 eV with U =16 eV. It was approaching experimental bandgap at higher U value 16 eV. I have not seen such high value for U in literature. It seems nonphysical.
On increasing U value lattice constants are increasing and ground stat energy is also increasing as U=8 energy =-226.23 eV for U=10 energy = -212.56 eV etc.
My other query is that I have seen many literature on WO3 but in most cases people have not done DFT+U calculation. I am not clear why they have not done DFT+U while W is d 4 state element. So how to decide which d element we have to do DFT+U ?
Other than I want to study Eu doped WO3 system. Eu is 4f 7 element for which I have to do DFT+U. I tried to optimize U value for Eu in WO3 without U value for W element.
When I measured bandgap of WO3 then it was 1.3 eV and with Eu doping without DFT+U for Eu it was 1.93 eV. When I am optimizing U value for Eu up to U=8eV, it was showing bandgap 1.93, 1.86, 2.03, 1.96 etc (varies U value in steps of 1 eV). It was showing fluctuating bandgap maximum 2.03 eV.
So I want to ask which U value I should take into account ?
Thank You
Shilendra
Using DFT+U is always a risky business in my opinion.If you do not calculate U from first principles there are usually a lot of reasonable values to choose. Also you should not forget that you also can choose J (unless you use a formulation that only explicitly considers U-J and you can set J to 0) You could try to match the bandgap or the equilibrium lattice parameter of experiments. Or match the shape of the valence band to a higher level calculation (hybrid, GW,...). It really depends what you are interested in!
Matching the experimental bandgap is usually the way to go if your results require an "accurate" bandgap. However, 16eV is definitely very large and you are probably right to deem it unphyiscal. You could try to do a benchmark hybrid calculation in a small (undoped) cell. How does the DOS vary in comparison to normal GGA and the GGA+U? Is the bandgap close to experimental values?
If you are able to describe all relevant aspects of your system except the bandgap with a reasonable U, you might look into using a scissor operator like people did in the paper I attached (I have not read that, that was just a quick google search to give you an example of a scissor operator) However, I do not think that this procedure is considered state of the art.
I hope that helps.
cheers, Michael
Article First-principles study of electronic structures and optical ...
Thank You Dear Dr. Michael Wolloch for your help. Can we calculate U value from first principle and how ?
My other query about hybrid calculation as hybrid calculation takes long time compare to GGA. What is the right way to do hybrid calculation:
(1) We should optimize structure with hybrid functional or
(2) We should optimize structure with GGA and after that take optimize geometry and we have to do single point calculation with hybrid functional.
Please guide me.
Shilendra
You can also evaluate the U parameter from first principles using constrained RPA method (cRPA). However I think that not all softwares are able to do this types of calculations. F. Aryasetiawan has some really nice bibliography about this topic. Maybe its worth trying.
http://arxiv.org/abs/cond-mat/0603138
http://www.cond-mat.de/events/correl11/manuscripts/aryasetiawan.pdf
Thank you dear Kuntal Kabra and Oier Arcelus for suggestions and discussion.
Shilendra
- Badges
- Science method
- Similar topics
- Methodology
- Crystallography
- X-ray Diffraction