Dear Dr. Timrov,
Thank you very much for your detailed and helpful response.
Your explanation has clarified the role of Hubbard_alpha and the need to avoid
treating DFT+U as a black-box method. I will go through the tutorials and
papers you recommended, and I will also check the projected density of states
to see the contribution of Cu 3d states near the band edges. I also appreciate
your comments on the occupations, nbnd, and the use of the QE input generator.
Thank you once again for your time and valuable advice.
Kind regards,
Stephen Mwenda
On Friday, 15 May 2026 at 18:24:50 GMT+2, Timrov, Iurii
<[email protected]> wrote:
Dear Stephen,
I recommend going through the tutorial that includes YouTube videos for
DFT+U:https://sites.google.com/view/hubbard-koopmans/program
Regarding your input files, try to use the QE input
generator:https://qeinputgenerator.materialscloud.io/You can upload your
existing input file, and this tool will suggest how to improve it.
DFT+U and band gaps: If you apply +U to states that form the edges (top of the
valence or bottom of the conduction), then the band gap will be changing.
Otherwise, if you apply +U to states that are far from the edges, the band gap
will not change or change little (since those deep states that you move with +U
will change the hybridization with those states that form the edges). I
recommend reading this paper on this
topic:https://www.mdpi.com/2076-3417/11/5/2395
- Is it reasonable to apply Hubbard U only to Cu 3d states in Cu2O using
this old syntax?
Check the projected density of states and see where the Cu-3d states are. Are
they close to the band edges?
- Is the use of Hubbard_alpha(1) = -0.08 appropriate in a production DFT+U
calculation, or should it only be used for linear-response/perturbative
calculations?
Hubbard_alpha is the strength of the perturbation that is used to compute U
using linear-response theory based on supercells and finite differences,
described here:https://journals.aps.org/prb/abstract/10.1103/PhysRevB.71.035105
You need to vary Hubbard_alpha around zero and then compute the response
occupations. You need to find the range of Hubbard_alpha values where the
response is linear. More is explained in the paper above.
There is a more efficient way to compute U. Namely, using the HP code, which is
based on density functional perturbation theory. It gives the same result as
above, but it is computationally faster. Check here if
interested:https://journals.aps.org/prb/abstract/10.1103/PhysRevB.98.085127https://www.sciencedirect.com/science/article/pii/S0010465522001746
However, unfortunately, for the system that you want to study (Cu2O) there is a
problem with this method. The U will explode if you compute it
self-consistently. Read more about this
here:https://pubs.aip.org/aip/jcp/article/140/12/121105/211929/Communication-Comparing-ab-initio-methods-of
You can also consider other methods to compute U for such systems (i.e. with
closed shells):https://arxiv.org/abs/2512.16803
- Are there any concerns with using occupations = 'fixed' for Cu2O in
vc-relax/scf and occupations = 'tetrahedra' for nscf/DOS?
It is ok since Cu2O is diamagnetic.
- Is nbnd = 40 reasonable for a 6-atom Cu2O cell with 56 electrons for
band-structure and DOS analysis?
You should check how many occupied states there are in this system (it also
depends on the pseudopotentials, i.e. whether the semicore states are included
or not). Then add some number of empty states. The higher in energy you go for
the empty states, the higher nbnd must be. There is no universal number. It is
system-dependent and it depends in which energy interval for PDOS you are
interested.
- Are there any syntax issues or methodological concerns that I should
correct before finalizing the calculations?
Regarding the syntax, use QE input generator. Regarding the methodological
concerns, you should not use DFT+U as a black box. There are many intricacies
that one should be aware of before using this method. It looks simple, but at
the same time it is very tricky.
HTH
Greetings,Iurii
----------------------------------------------------------Dr. Iurii
TIMROVTenure-track scientistLaboratory for Materials Simulations (LMS)Paul
Scherrer Institute (PSI)CH-5232 Villigen,
SwitzerlandProfile:www.psi.ch/en/lms/people/iurii-timrovGroup
website:www.timrovresearch.com/From: users
<[email protected]> on behalf of Stephen Mwenda via
users <[email protected]>
Sent: Friday, May 15, 2026 17:46
To: [email protected] <[email protected]>
Subject: [QE-users] Advice on Cu2O DFT+U input setup in Quantum ESPRESSO
Dear Quantum ESPRESSO users,
I hope you are well.
I am carrying out DFT calculations on cuprous oxide, Cu2O, using Quantum
ESPRESSO. My aim is to study how the electronic band gap changes with Hubbard U
applied to the Cu 3d states.
I would like to kindly ask for advice on whether my input setup is reasonable,
especially regarding the old-style DFT+U syntax, pseudopotentials, and use of
the Hubbard alpha parameter.
I have attached one representative input file for the vc-relax calculation at U
= 1 eV, since this file includes the Hubbard U setup. The same workflow is then
repeated for U = 2 to 6 eV by changing the value of Hubbard_U(1). For U = 0, no
Hubbard correction is applied.
My calculation setup is as follows:
- Material: Cu2O, cubic cuprite structure
- Exchange-correlation functional: PBE
- Pseudopotentials:
- Cu.pbe-dn-rrkjus_psl.1.0.0.UPF
- O.pbe-n-rrkjus_psl.1.0.0.UPF
- ecutwfc = 60 Ry
- ecutrho = 480 Ry
- k-points for vc-relax and scf: 6 6 6 0 0 0
- k-points for nscf/DOS: 12 12 12 0 0 0
- Number of atoms: 6
- Number of bands in nscf: 40
For U = 1 eV, the relevant part of my &SYSTEM block is:
&SYSTEM
ibrav = 1
a = 4.25220
nat = 6
ntyp = 2
ecutwfc = 60
ecutrho = 480
occupations = 'fixed'
lda_plus_u = .true.
Hubbard_U(1) = 1.0
Hubbard_alpha(1) = -0.08
/
My ATOMIC_SPECIES block is:
ATOMIC_SPECIES
Cu 63.54600 Cu.pbe-dn-rrkjus_psl.1.0.0.UPF
O 15.99940 O.pbe-n-rrkjus_psl.1.0.0.UPF
Therefore, Hubbard_U(1) and Hubbard_alpha(1) are intended to apply to Cu.
The relaxed structure remains cubic, and the calculated direct band gap at Γ
increases from about 0.46 eV at U = 0 to about 0.64 eV at U = 6 eV. I
understand that PBE/PBE+U may still underestimate the experimental Cu2O band
gap, but I would like to confirm whether the input syntax and general setup are
technically correct.
My specific questions are:
- Is it reasonable to apply Hubbard U only to Cu 3d states in Cu2O using
this old syntax?
- Is the use of Hubbard_alpha(1) = -0.08 appropriate in a production DFT+U
calculation, or should it only be used for linear-response/perturbative
calculations?
- Are there any concerns with using occupations = 'fixed' for Cu2O in
vc-relax/scf and occupations = 'tetrahedra' for nscf/DOS?
- Is nbnd = 40 reasonable for a 6-atom Cu2O cell with 56 electrons for
band-structure and DOS analysis?
- Are there any syntax issues or methodological concerns that I should
correct before finalizing the calculations?
I would be grateful for any advice or corrections from experienced users.
Kind regards,
Stephen
_______________________________________________________________________________
The Quantum ESPRESSO Foundation stands in solidarity with all civilians
worldwide who are victims of terrorism, military aggression, and indiscriminate
warfare.
--------------------------------------------------------------------------------
Quantum ESPRESSO is supported by MaX (www.max-centre.eu)
users mailing list [email protected]
https://lists.quantum-espresso.org/mailman/listinfo/users
