Re: [QE-users] Different implementations of the Hubbard 'U' between QE and VASP
On 06/05/2020 22:02, Shivesh Sivakumar wrote: Nicola, Sorry to bring up the subject again but I still can't reconcile something. I understand that 'U' is not universal and not completely portable between different PPs. But what I find odd is the effect of 'U'. On QE (with NCPP), 'U' seems to suppress the band gap whereas in VASP (with PAW), 'U' seems to widen the band gap. This is what looked very strange. Best, Shivesh Not sure what to say - i.e. one would really need to delve deep into the case study. I think the VASP PAW projectors that are used for the U are strictly localized inside the core, and so might be quite a bit different from those of say a NCPP pseudo. But this is the kind of question that can be answered only by a detailed study - do you get with no U the same results? Where did you get the Cr norm-conserving pseudo? Is it any good? Our recommendation is the USSP from the GBRV work of David Vanderbilt: https://www.materialscloud.org/discover/sssp/plot/precision/Cr nicola -- Prof Nicola Marzari, Chair of Theory and Simulation of Materials, EPFL Director, National Centre for Competence in Research NCCR MARVEL, EPFL http://theossrv1.epfl.ch/Main/Contact http://nccr-marvel.ch/en/project ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
Re: [QE-users] Different implementations of the Hubbard 'U' between QE and VASP
Nicola, Sorry for the barrage of messages but it seems like I found the origin of this odd behaviour. Even though the band structures show very good agreement (barring the band gap), the contribution of orbitals is different near VB and CB edges - I did a fatbands calculation on QE and then on VASP and it turned out that there are differences in hybridization between the two band structures atleast near the VB and CB edges. I believe this is why inclusion of 'U' shows odd behaviour ( Since it shifts only d orbitals and not s or p orbitals). Best, Shivesh On Wed, May 6, 2020 at 1:02 PM Shivesh Sivakumar wrote: > Nicola, > > Sorry to bring up the subject again but I still can't reconcile something. > I understand that 'U' is not universal and not completely portable between > different PPs. But what I find odd is the effect of 'U'. On QE (with NCPP), > 'U' seems to suppress the band gap whereas in VASP (with PAW), 'U' seems to > widen the band gap. This is what looked very strange. > > Best, > Shivesh > > On Wed, May 6, 2020 at 10:22 AM Shivesh Sivakumar < > shiveshsivaku...@gmail.com> wrote: > >> Nicola, >> >> Thank you very much for your valuable input. It definitely makes >> everything more clear to me. >> >> Best, >> Shivesh >> >> On Wed, May 6, 2020 at 9:50 AM Nicola Marzari >> wrote: >> >>> >>> >>> >>> Dear Shivesh, >>> >>> >>> this is really important: as Iurii and Paolo already mentioned, U is >>> *not* a universal parameter. For the same element and for the same >>> material, it can be easily 3eV or 7eV (e.g. if the Hubbard manifold it >>> acts upon has been taken say from a pseudopotential (NCPP/USPP/PAW) >>> generated in a neutral configuration or in an oxidized one). >>> >>> So, that means that the same identical material - say, MnO - should be >>> studied using a U for Mn of 3.13 if you use pseudopotential "A", and >>> 7.56eV when you use pseudopotential "B" for Mn. For a slightly >>> compressed unit cell, maybe the U for "A" would be 3.10eV, and for "B" >>> 7.74eV. >>> >>> Looking at the literature wouldn't help - especially (IMNSHO) because >>> almost noone has understood what the U is. >>> >>> nicola >>> >>> >>> >>> On 06/05/2020 18:25, Shivesh Sivakumar wrote: >>> > Timrov and Paolo, >>> > >>> > Thank you very much for your wonderful inputs. I just wanted to >>> clarify >>> > a few things that I might have left out in my original message. >>> > When I said I used the same parameters for both calculations, this is >>> > what I meant: >>> > a) Relaxed it in QE ( and then moved to scf etc..). For VASP, I >>> used the >>> > relaxed structure to directly perform scf calculation, therefore the >>> > structure shouldn't cause any problems between the two. >>> > b) Used a well tested (for convergence) k point grid on QE. Used the >>> > same k mesh for VASP. >>> > c) The same smearing parameters and same hubbard U value ( of 3 eV - >>> for >>> > the Chromium atom in my system) for both calculations. >>> > >>> > Regarding your final question, yes, I did not compute 'U' from >>> > first-principles. It was assumed after referring to various >>> > literature about the same material. >>> > >>> > Best, >>> > Shivesh >>> -- >>> -- >>> Prof Nicola Marzari, Chair of Theory and Simulation of Materials, EPFL >>> Director, National Centre for Competence in Research NCCR MARVEL, EPFL >>> http://theossrv1.epfl.ch/Main/Contact http://nccr-marvel.ch/en/project >>> >> ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
Re: [QE-users] Different implementations of the Hubbard 'U' between QE and VASP
Nicola, Sorry to bring up the subject again but I still can't reconcile something. I understand that 'U' is not universal and not completely portable between different PPs. But what I find odd is the effect of 'U'. On QE (with NCPP), 'U' seems to suppress the band gap whereas in VASP (with PAW), 'U' seems to widen the band gap. This is what looked very strange. Best, Shivesh On Wed, May 6, 2020 at 10:22 AM Shivesh Sivakumar < shiveshsivaku...@gmail.com> wrote: > Nicola, > > Thank you very much for your valuable input. It definitely makes > everything more clear to me. > > Best, > Shivesh > > On Wed, May 6, 2020 at 9:50 AM Nicola Marzari > wrote: > >> >> >> >> Dear Shivesh, >> >> >> this is really important: as Iurii and Paolo already mentioned, U is >> *not* a universal parameter. For the same element and for the same >> material, it can be easily 3eV or 7eV (e.g. if the Hubbard manifold it >> acts upon has been taken say from a pseudopotential (NCPP/USPP/PAW) >> generated in a neutral configuration or in an oxidized one). >> >> So, that means that the same identical material - say, MnO - should be >> studied using a U for Mn of 3.13 if you use pseudopotential "A", and >> 7.56eV when you use pseudopotential "B" for Mn. For a slightly >> compressed unit cell, maybe the U for "A" would be 3.10eV, and for "B" >> 7.74eV. >> >> Looking at the literature wouldn't help - especially (IMNSHO) because >> almost noone has understood what the U is. >> >> nicola >> >> >> >> On 06/05/2020 18:25, Shivesh Sivakumar wrote: >> > Timrov and Paolo, >> > >> > Thank you very much for your wonderful inputs. I just wanted to clarify >> > a few things that I might have left out in my original message. >> > When I said I used the same parameters for both calculations, this is >> > what I meant: >> > a) Relaxed it in QE ( and then moved to scf etc..). For VASP, I >> used the >> > relaxed structure to directly perform scf calculation, therefore the >> > structure shouldn't cause any problems between the two. >> > b) Used a well tested (for convergence) k point grid on QE. Used the >> > same k mesh for VASP. >> > c) The same smearing parameters and same hubbard U value ( of 3 eV - >> for >> > the Chromium atom in my system) for both calculations. >> > >> > Regarding your final question, yes, I did not compute 'U' from >> > first-principles. It was assumed after referring to various >> > literature about the same material. >> > >> > Best, >> > Shivesh >> -- >> -- >> Prof Nicola Marzari, Chair of Theory and Simulation of Materials, EPFL >> Director, National Centre for Competence in Research NCCR MARVEL, EPFL >> http://theossrv1.epfl.ch/Main/Contact http://nccr-marvel.ch/en/project >> > ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
Re: [QE-users] Different implementations of the Hubbard 'U' between QE and VASP
Nicola, Thank you very much for your valuable input. It definitely makes everything more clear to me. Best, Shivesh On Wed, May 6, 2020 at 9:50 AM Nicola Marzari wrote: > > > > Dear Shivesh, > > > this is really important: as Iurii and Paolo already mentioned, U is > *not* a universal parameter. For the same element and for the same > material, it can be easily 3eV or 7eV (e.g. if the Hubbard manifold it > acts upon has been taken say from a pseudopotential (NCPP/USPP/PAW) > generated in a neutral configuration or in an oxidized one). > > So, that means that the same identical material - say, MnO - should be > studied using a U for Mn of 3.13 if you use pseudopotential "A", and > 7.56eV when you use pseudopotential "B" for Mn. For a slightly > compressed unit cell, maybe the U for "A" would be 3.10eV, and for "B" > 7.74eV. > > Looking at the literature wouldn't help - especially (IMNSHO) because > almost noone has understood what the U is. > > nicola > > > > On 06/05/2020 18:25, Shivesh Sivakumar wrote: > > Timrov and Paolo, > > > > Thank you very much for your wonderful inputs. I just wanted to clarify > > a few things that I might have left out in my original message. > > When I said I used the same parameters for both calculations, this is > > what I meant: > > a) Relaxed it in QE ( and then moved to scf etc..). For VASP, I used the > > relaxed structure to directly perform scf calculation, therefore the > > structure shouldn't cause any problems between the two. > > b) Used a well tested (for convergence) k point grid on QE. Used the > > same k mesh for VASP. > > c) The same smearing parameters and same hubbard U value ( of 3 eV - for > > the Chromium atom in my system) for both calculations. > > > > Regarding your final question, yes, I did not compute 'U' from > > first-principles. It was assumed after referring to various > > literature about the same material. > > > > Best, > > Shivesh > -- > -- > Prof Nicola Marzari, Chair of Theory and Simulation of Materials, EPFL > Director, National Centre for Competence in Research NCCR MARVEL, EPFL > http://theossrv1.epfl.ch/Main/Contact http://nccr-marvel.ch/en/project > ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
Re: [QE-users] Different implementations of the Hubbard 'U' between QE and VASP
Dear Shivesh, this is really important: as Iurii and Paolo already mentioned, U is *not* a universal parameter. For the same element and for the same material, it can be easily 3eV or 7eV (e.g. if the Hubbard manifold it acts upon has been taken say from a pseudopotential (NCPP/USPP/PAW) generated in a neutral configuration or in an oxidized one). So, that means that the same identical material - say, MnO - should be studied using a U for Mn of 3.13 if you use pseudopotential "A", and 7.56eV when you use pseudopotential "B" for Mn. For a slightly compressed unit cell, maybe the U for "A" would be 3.10eV, and for "B" 7.74eV. Looking at the literature wouldn't help - especially (IMNSHO) because almost noone has understood what the U is. nicola On 06/05/2020 18:25, Shivesh Sivakumar wrote: Timrov and Paolo, Thank you very much for your wonderful inputs. I just wanted to clarify a few things that I might have left out in my original message. When I said I used the same parameters for both calculations, this is what I meant: a) Relaxed it in QE ( and then moved to scf etc..). For VASP, I used the relaxed structure to directly perform scf calculation, therefore the structure shouldn't cause any problems between the two. b) Used a well tested (for convergence) k point grid on QE. Used the same k mesh for VASP. c) The same smearing parameters and same hubbard U value ( of 3 eV - for the Chromium atom in my system) for both calculations. Regarding your final question, yes, I did not compute 'U' from first-principles. It was assumed after referring to various literature about the same material. Best, Shivesh -- -- Prof Nicola Marzari, Chair of Theory and Simulation of Materials, EPFL Director, National Centre for Competence in Research NCCR MARVEL, EPFL http://theossrv1.epfl.ch/Main/Contact http://nccr-marvel.ch/en/project ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
Re: [QE-users] Different implementations of the Hubbard 'U' between QE and VASP
Timrov and Paolo, Thank you very much for your wonderful inputs. I just wanted to clarify a few things that I might have left out in my original message. When I said I used the same parameters for both calculations, this is what I meant: a) Relaxed it in QE ( and then moved to scf etc..). For VASP, I used the relaxed structure to directly perform scf calculation, therefore the structure shouldn't cause any problems between the two. b) Used a well tested (for convergence) k point grid on QE. Used the same k mesh for VASP. c) The same smearing parameters and same hubbard U value ( of 3 eV - for the Chromium atom in my system) for both calculations. Regarding your final question, yes, I did not compute 'U' from first-principles. It was assumed after referring to various literature about the same material. Best, Shivesh On Wed, May 6, 2020 at 1:59 AM Timrov Iurii wrote: > Dear Shivesh, > > > > While performing PBE+U calculations on a 2-D material, I wanted to try > using NCPPs and then PAW PPs. For the former, I obviously tried QE and for > the latter, VASP. > > > In Quantum ESPRESSO, DFT+U can be used with any type of pseudopotentials > (NC PPs, US PPs, PAW PPs). So your motivation to switch to VASP because of > PAW PPs looks strange to me. > > > However, it is important to note that DFT+U for PAW PPs in QE and VASP is > not implemented in the same way. There are some technical differences (in > QE the projections are done on atomic/ortho-atomic orbitals while in VASP > on \beta projector functions). This aspect will be discussed briefly in the > paper which we are working on (Timrov, Marzari, Cococcioni). > > > > The electronic bandstructures look very similar (the parameters are > completely the same, except for plane wave cutoffs, as you might imagine) > but the band gap is different. > > > Which parameters are completely the same? Do you refer to a k points > sampling, smearing, lattice parameter? Or Hubbard U? > > > > With NCPPs, the Hubbard 'U' seems to suppress the band gap (compared to > U=0) whereas for VASP calculation, it seems to increase the band gap > (compared to U=0). Is there a qualitative explanation for this? > > > Which values of U do you use in the two codes? How did you choose them? > > > Please note that *the Hubbard U parameter is not a universal parameter*. > If you compute it from first principles (e.g. using linear response theory, > cRPA, or other approach), its value depends on many factors (type of the > Hubbard manifold, pseudopotentials, functional, oxidation state, etc.). So > if you use the same value of U in two different codes (i.e. different > implementations of DFT+U), with different PPs, with different Hubbard > manifolds, etc., then the effect of the Hubbard correction with same U will > be different. > > > The correct procedure is the following: to compute Hubbard parameters from > first-principles for a given set of computational parameters (Hubbard > manifold, PPs, functional, etc.), and *use it consistently with exactly > the same set of parameters*. In QE, Hubbard parameters can be computed > using the HP code. Hubbard U is not global, and hence it is not portable > (i.e. you cannot compute it in one code with one set of parameters, and > then use it in another code with another set of parameters; even in the > same code you cannot compute with one set of parameters and then use it in > the same code with another set of parameters). > > > Greetings, > > Iurii > > > -- > Dr. Iurii Timrov > Postdoctoral Researcher > STI - IMX - THEOS and NCCR - MARVEL > Swiss Federal Institute of Technology Lausanne (EPFL) > CH-1015 Lausanne, Switzerland > +41 21 69 34 881 > http://people.epfl.ch/265334 > -- > *From:* users on behalf of > Shivesh Sivakumar > *Sent:* Wednesday, May 6, 2020 12:35:30 AM > *To:* Quantum ESPRESSO users Forum > *Subject:* [QE-users] Different implementations of the Hubbard 'U' > between QE and VASP > > Hello all, > > While performing PBE+U calculations on a 2-D material, I wanted to try > using NCPPs and then PAW PPs. For the former, I obviously tried QE and for > the latter, VASP. The electronic bandstructures look very similar (the > parameters are completely the same, except for plane wave cutoffs, as you > might imagine) but the band gap is different. > With NCPPs, the Hubbard 'U' seems to suppress the band gap (compared to > U=0) whereas for VASP calculation, it seems to increase the band gap > (compared to U=0). Is there a qualitative explanation for this? > > I am very sorry if this question is not totally appropriate for the forum > but I just want to know if conceptually, there are some differences in > implementation that I am missing out. > > Best, > Shivesh Sivakumar > University of Washington > Seattle, WA-98105 > ___ > Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso) > users mailing list users@lists.quantum-espresso.org >
Re: [QE-users] Different implementations of the Hubbard 'U' between QE and VASP
Iurii is perfectly right: U is not a portable parameter. This being said, it seems to me strange that two calculations performed with different approaches but with the same value of U may yield opposite results. It seems to me more likely that one of the two calculations is not correctly set up. Paolo On Wed, May 6, 2020 at 10:59 AM Timrov Iurii wrote: > Dear Shivesh, > > > > While performing PBE+U calculations on a 2-D material, I wanted to try > using NCPPs and then PAW PPs. For the former, I obviously tried QE and for > the latter, VASP. > > > In Quantum ESPRESSO, DFT+U can be used with any type of pseudopotentials > (NC PPs, US PPs, PAW PPs). So your motivation to switch to VASP because of > PAW PPs looks strange to me. > > > However, it is important to note that DFT+U for PAW PPs in QE and VASP is > not implemented in the same way. There are some technical differences (in > QE the projections are done on atomic/ortho-atomic orbitals while in VASP > on \beta projector functions). This aspect will be discussed briefly in the > paper which we are working on (Timrov, Marzari, Cococcioni). > > > > The electronic bandstructures look very similar (the parameters are > completely the same, except for plane wave cutoffs, as you might imagine) > but the band gap is different. > > > Which parameters are completely the same? Do you refer to a k points > sampling, smearing, lattice parameter? Or Hubbard U? > > > > With NCPPs, the Hubbard 'U' seems to suppress the band gap (compared to > U=0) whereas for VASP calculation, it seems to increase the band gap > (compared to U=0). Is there a qualitative explanation for this? > > > Which values of U do you use in the two codes? How did you choose them? > > > Please note that *the Hubbard U parameter is not a universal parameter*. > If you compute it from first principles (e.g. using linear response theory, > cRPA, or other approach), its value depends on many factors (type of the > Hubbard manifold, pseudopotentials, functional, oxidation state, etc.). So > if you use the same value of U in two different codes (i.e. different > implementations of DFT+U), with different PPs, with different Hubbard > manifolds, etc., then the effect of the Hubbard correction with same U will > be different. > > > The correct procedure is the following: to compute Hubbard parameters from > first-principles for a given set of computational parameters (Hubbard > manifold, PPs, functional, etc.), and *use it consistently with exactly > the same set of parameters*. In QE, Hubbard parameters can be computed > using the HP code. Hubbard U is not global, and hence it is not portable > (i.e. you cannot compute it in one code with one set of parameters, and > then use it in another code with another set of parameters; even in the > same code you cannot compute with one set of parameters and then use it in > the same code with another set of parameters). > > > Greetings, > > Iurii > > > -- > Dr. Iurii Timrov > Postdoctoral Researcher > STI - IMX - THEOS and NCCR - MARVEL > Swiss Federal Institute of Technology Lausanne (EPFL) > CH-1015 Lausanne, Switzerland > +41 21 69 34 881 > http://people.epfl.ch/265334 > -- > *From:* users on behalf of > Shivesh Sivakumar > *Sent:* Wednesday, May 6, 2020 12:35:30 AM > *To:* Quantum ESPRESSO users Forum > *Subject:* [QE-users] Different implementations of the Hubbard 'U' > between QE and VASP > > Hello all, > > While performing PBE+U calculations on a 2-D material, I wanted to try > using NCPPs and then PAW PPs. For the former, I obviously tried QE and for > the latter, VASP. The electronic bandstructures look very similar (the > parameters are completely the same, except for plane wave cutoffs, as you > might imagine) but the band gap is different. > With NCPPs, the Hubbard 'U' seems to suppress the band gap (compared to > U=0) whereas for VASP calculation, it seems to increase the band gap > (compared to U=0). Is there a qualitative explanation for this? > > I am very sorry if this question is not totally appropriate for the forum > but I just want to know if conceptually, there are some differences in > implementation that I am missing out. > > Best, > Shivesh Sivakumar > University of Washington > Seattle, WA-98105 > ___ > Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso) > users mailing list users@lists.quantum-espresso.org > https://lists.quantum-espresso.org/mailman/listinfo/users -- Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche, Univ. Udine, via delle Scienze 208, 33100 Udine, Italy Phone +39-0432-558216, fax +39-0432-558222 ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
Re: [QE-users] Different implementations of the Hubbard 'U' between QE and VASP
Dear Shivesh, > While performing PBE+U calculations on a 2-D material, I wanted to try using > NCPPs and then PAW PPs. For the former, I obviously tried QE and for the > latter, VASP. In Quantum ESPRESSO, DFT+U can be used with any type of pseudopotentials (NC PPs, US PPs, PAW PPs). So your motivation to switch to VASP because of PAW PPs looks strange to me. However, it is important to note that DFT+U for PAW PPs in QE and VASP is not implemented in the same way. There are some technical differences (in QE the projections are done on atomic/ortho-atomic orbitals while in VASP on \beta projector functions). This aspect will be discussed briefly in the paper which we are working on (Timrov, Marzari, Cococcioni). > The electronic bandstructures look very similar (the parameters are > completely the same, except for plane wave cutoffs, as you might imagine) but > the band gap is different. Which parameters are completely the same? Do you refer to a k points sampling, smearing, lattice parameter? Or Hubbard U? > With NCPPs, the Hubbard 'U' seems to suppress the band gap (compared to U=0) > whereas for VASP calculation, it seems to increase the band gap (compared to > U=0). Is there a qualitative explanation for this? Which values of U do you use in the two codes? How did you choose them? Please note that the Hubbard U parameter is not a universal parameter. If you compute it from first principles (e.g. using linear response theory, cRPA, or other approach), its value depends on many factors (type of the Hubbard manifold, pseudopotentials, functional, oxidation state, etc.). So if you use the same value of U in two different codes (i.e. different implementations of DFT+U), with different PPs, with different Hubbard manifolds, etc., then the effect of the Hubbard correction with same U will be different. The correct procedure is the following: to compute Hubbard parameters from first-principles for a given set of computational parameters (Hubbard manifold, PPs, functional, etc.), and use it consistently with exactly the same set of parameters. In QE, Hubbard parameters can be computed using the HP code. Hubbard U is not global, and hence it is not portable (i.e. you cannot compute it in one code with one set of parameters, and then use it in another code with another set of parameters; even in the same code you cannot compute with one set of parameters and then use it in the same code with another set of parameters). Greetings, Iurii -- Dr. Iurii Timrov Postdoctoral Researcher STI - IMX - THEOS and NCCR - MARVEL Swiss Federal Institute of Technology Lausanne (EPFL) CH-1015 Lausanne, Switzerland +41 21 69 34 881 http://people.epfl.ch/265334 From: users on behalf of Shivesh Sivakumar Sent: Wednesday, May 6, 2020 12:35:30 AM To: Quantum ESPRESSO users Forum Subject: [QE-users] Different implementations of the Hubbard 'U' between QE and VASP Hello all, While performing PBE+U calculations on a 2-D material, I wanted to try using NCPPs and then PAW PPs. For the former, I obviously tried QE and for the latter, VASP. The electronic bandstructures look very similar (the parameters are completely the same, except for plane wave cutoffs, as you might imagine) but the band gap is different. With NCPPs, the Hubbard 'U' seems to suppress the band gap (compared to U=0) whereas for VASP calculation, it seems to increase the band gap (compared to U=0). Is there a qualitative explanation for this? I am very sorry if this question is not totally appropriate for the forum but I just want to know if conceptually, there are some differences in implementation that I am missing out. Best, Shivesh Sivakumar University of Washington Seattle, WA-98105 ___ Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso) users mailing list users@lists.quantum-espresso.org https://lists.quantum-espresso.org/mailman/listinfo/users
