[QE-users] hp.x code overestimates Hubbard U value

Thu Sep 12 14:46:59 CEST 2019

Dear Soumyadeep,

Your input files look fine to me.

I would try to compute U in a self-consistent way. Just take the value you got from a one-shot calculation and plug it in the SCF input and then redo the HP calculation. Repeat this procedure until U does not change any longer. On top of that, you may perform a relaxation of the structure. By doing so, you will obtain U and the structure which are consistent with each other. More about this is discussed in the papers which I mentioned in my previous email.

Best,

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: Soumyadeep <soumyadeep at rrcat.gov.in>
Sent: Thursday, September 12, 2019 2:36:00 PM
To: Timrov Iurii
Cc: Quantum ESPRESSO users Forum
Subject: Re: [QE-users] hp.x code overestimates Hubbard U value

Dear Iurii,

On 12-09-2019 17:33, Timrov Iurii wrote:
> Dear Soumyadeep,
>
> Unfortunately I cannot open the input file "LiFeAs-0GPa.scf". Could
> you please share it via Google Drive or Dropbox? Is you system
> magnetic, metallic?
>
Sorry for this I will upload it again (if problem continues please check
run_hp_NCPP file). The system is non-magnetic and metallic.

> The value of U is very sensitive to the type of localized functions
> which are used, so we have Wannier in cRPA vs atomic/ortho-atomic in
> LRT. That may make a big difference, I think.
>
I agreed. Since both are very different from each other.

> Another aspect, comparison of methods cRPA vs constrained DFT (LRT).
> There were some attempts to obtain same U with these two methods, but
> nobody succeeded to so far, to the best of my knowledge. In
> particular, here is an important paper concerning this issue: Phys.
> Rev. B 74, 125106 (2006). In Fig. 2 of this paper you can see that
> cRPA gives the value of U which is very sensitive to the fact which
> electrons are allowed to screen the d-d interactions. So in your case,
> I would try to understand how U was computed with cRPA (if the authors
> provide enough details in the publication which you cited), i.e. which
> electrons were allowed to screen. In the LRT approach instead, there
> is no this kind of choice which electrons are allowed to screen: when
> one perturbs e.g. d electronic states of Fe then all other electrons
> in the system are allowed to screen (this is done via the
> self-consistent solution of modified Kohn-Sham equations). To
> conclude, comparing cRPA with LRT is not an easy task (at least I do
> not know any publication which clearly makes such a comparison, apart
> from the one which I mentioned above).
Again you are correct. I think in cRPA they allowed only d-electrons to
screen since they used them to generate wannier functions.

>
> Best regards,
>
> 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: Soumyadeep <soumyadeep at rrcat.gov.in>
> SENT: Thursday, September 12, 2019 12:56:03 PM
> TO: Quantum ESPRESSO users Forum
> CC: Timrov Iurii
> SUBJECT: Re: [QE-users] hp.x code overestimates Hubbard U value
>
> Dear Iurii,
>    Thanks you very much for the detailed explanation.
>
>    >> ..."U" lies between 3.0-3.5 eV...
>    The determined value of U is calculated using cRPA method - Wannier
>
> function basis - Norm conserving PP's
> (https://journals.jps.jp/doi/abs/10.1143/JPSJ.79.044705).
>
>    When I do it using LRT I get almost doubled? I have checked it
> w.r.t
> different PP/Exchange-correlation functional here is the results,
>    NCPP --> 6.43, USPP --> 6.85, PBESOL --> 6.81, PAW --> 6.83
>
>    > different oxidation states
>    I think oxidation state is same in both cases. Fe^2+ for Fe-besed
> superconductors
>
>    > What is important is to use U (for calculations of band
> structure,
> DOS, phonons, etc.) with the same parameters which which it was
> computed.
>    Yes agreed. First of all I like to calculate it for reported values
> in
> literature then apply it for my calculations.
>
>    Input files are attached (NCPP) for your consideration, that may
> help
> you to find out my mistakes. I am just a beginner of such calculation
> and don't think without my mistake such large difference can occur
> though U is not an universal parameter.
>
> with many thanks and best regards
> Soumyadeep
>
> On 12-09-2019 14:35, Timrov Iurii wrote:
>> Dear Soumyadeep,
>>
>> Hubbard U is not a universal parameter. If you compute Hubbard U
> from
>> first principles, e.g. using linear response theory (LRT) [PRB 71,
>> 035105 (2005); PRB 98, 085127 (2018)] then the value of U depends on
>> many factors:
>>
>> - Manifold of localized functions (atomic, ortho-atomic, Wannier,
>> etc.);
>>
>> - Pseudopotentials;
>>
>> - Oxidation state;
>>
>> - Exchange-correlation functional (LDA, PBE, PBEsol, etc.);
>>
>> - Environment (e.g. U for Mn-3d in MnO and BiMnO3 is not exactly the
>> same).
>>
>> For example, check the paper by H. Kulik and N. Marzari, J. Chem.
>> Phys. 129, 134314 (2008). In particular, check the appendix in this
>> paper, and you will see that U is very sensitive to the oxidation
>> state.
>>
>> Another aspect is whether U was computed in "one-shot" or
>> "self-consistently". If you want to learn more about the
>> self-consistent U, I recommend these papers: H. Hsu et al., PRB 79,
>> 125124 (2009); C. Ricca et al., PRB 99, 094102 (2019); M. Cococcioni
>> and N. Marzari, PRM 3, 033801 (2019).
>>
>>> ..."U" lies between 3.0-3.5 eV...
>>
>> Where do these values come from? Are they empirical? Or were they
>> computed with cRPA? Please note that  comparing U is not very
>> meaningful (in my understanding), especially if you do not specify
> how
>> it was determined. As I wrote above, the value of U computed from
>> first principles depends on many factors. So if you compute it using
>> different methods (LRT vs cRPA), different localized functions
> (atomic
>> vs Wannier), different oxidation states, etc. then you will not get
>> the same U. What is important is to use U (for calculations of band
>> structure, DOS, phonons, etc.) with the same parameters which which
> it
>> was computed.
>>
>> I recommend to have a look at the hands-on about calculation of U
>> using Quantum ESPRESSO (hp.x code):
>>
>> https://agenda.ethernet.edu.et/event/33/
>>
>> Hope this helps! If something is not precise in my explanations, it
>> would great if other experts correct me.
>>
>> Best regards,
>>
>> 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 <users-bounces at lists.quantum-espresso.org> on behalf of
>> Soumyadeep <soumyadeep at rrcat.gov.in>
>> SENT: Wednesday, September 11, 2019 4:35:37 PM
>> TO: Users
>> SUBJECT: [QE-users] hp.x code overestimates Hubbard U value
>>
>> Dear QE users,
>>
>>    I am getting a overestimated value of Hubbard parameters using
> hp.x
>>
>> executable of Quantum Espresso. My material is iron based
>> superconductors (LiFeAs) for which "U" lies between 3.0-3.5 eV, but
>> using this code I get it around 6.4-6.5 eV.
>>
>>    Can anyone find out possible mistakes?
>>
>> with many thanks and best regards
>> Soumyadeep
>> -------------------------------------------------------------------
>> Soumyadeep Ghosh,
>> Senior Research Fellow,
>> Homi Bhabha National Institute (HBNI),
>> Raja Ramanna Centre for Advanced Technology, Indore, India-452013
>> Mob: (+91)9424664553
>> User Lab: 0731244-2580
>> Email: soumyadeepghosh35 at gmail.com, soumyadeep at rrcat.gov.in
>> -------------------------------------------------------------------
>> _______________________________________________
>> Quantum ESPRESSO is supported by MaX
>> ( [1]www.max-centre.eu/quantum-espresso [1])
>> users mailing list users at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users
>>
>>
>> Links:
>> ------
>> [1] http://www.max-centre.eu/quantum-espresso
>> _______________________________________________
>> Quantum ESPRESSO is supported by MaX
>> (<http://>www.max-centre.eu/quantum-espresso [2])
>> users mailing list users at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users
>
>
> Links:
> ------
> [1] http://
> [2] http://www.max-centre.eu/quantum-espresso
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20190912/70275caf/attachment.html>