[Wien] spin-orbit (PBE and mBJ) for perovskites

Peter Blaha pblaha at theochem.tuwien.ac.at
Tue Dec 10 08:46:13 CET 2019


Don't go above EMAX=10 for SO calculations with RLOs.

On 12/9/19 9:32 PM, Mikhail Nestoklon wrote:
> Dear Prof. Blaha,
> Thank you for clarification. I think, now I more or less understand what 
> is happening.
> It seems, this issue is very easy to miss. Do you have any practical 
> recommendation how to spot this kind of problem in the calculations of 
> non-trivial material with few atoms?
> Sincerely yours,
> Mikhail Nestoklon
> 
>     Понедельник, 9 декабря 2019, 19:02 +03:00 от Peter Blaha
>     <pblaha at theochem.tuwien.ac.at>:
>     Hi,
> 
>     Fist of all: The RLOs (p-1/2)-LOs were originally designed to improve
>     the SO splitting of semicore states. There it makes a big difference and
>     is essential to converge E-tot in a meaningful way.
> 
>     (By meaningful I do not mean the absolute number as you tested it, but
>     an E-tot difference of 2 calculations at eg. 2 volumes).
> 
>     This can be best seen if you compare these semicore eigenvalues with the
>     eigenvalues of case.outputst, where the fully-relativistic splitting can
>     be found and compared.
> 
>     You can also add it in a case like Pb (usually, Pb does not have
>     semicore p-states), but the effect MUST be small.
> 
>     Most importantly, you MUST NOT have a too large E-max, because when the
>     spurious LO-eigenvalues of the APW+lo method enter in the spectrum at
>     high energy, you may get a problem as indicated below:
> 
>     I checked this for a free Pb-atom (in a big box).
> 
>     lapwso with Emax=10 gives:
> 
>                 -1.5996189 -1.5996189 -1.5996189 -1.5996189 -1.4111046
>                 -1.4111046 -1.4111046 -1.4111046 -1.4110328 -1.4110328
>                 -0.8489490 -0.8489490 -0.3028265 -0.3028265 -0.2040009
>                 -0.2040009 -0.2040009 -0.2040009 -0.0106114 -0.0106114
> 
>     while with EMAX=15 I get 6 spurious ghostbands ("bad Pb-5p states):
>                 -8.2276000 -8.2276000 -3.1336676 -3.1336676 -3.1336591
>                 -3.1336591 -1.6001880 -1.6001880 -1.6001880 -1.6001880
>                 -1.4113465 -1.4113465 -1.4113464 -1.4113464 -1.4112735
>                 -1.4112735 -0.8489490 -0.8489490 -0.2886641 -0.2886641
>                 -0.2000960 -0.2000960 -0.2000960 -0.2000960 -0.0106114
> 
>     The SO-splitting of 6p states in the Pb atom is 0.109 Ry, while the
>     "good" lapwso calculations yield 0.099, but the "bad ones only 0.088
>     Without the RLO the SO splitting is 0.092, so still better than the
>     "bad" calculation.
> 
>     Eventually one can avoid this switching back to LAPW for the Pb-p
>     states.
> 
>     Peter Blaha
> 
>     On 12/9/19 3:30 PM, Mikhail Nestoklon wrote:
>      > Dear Dr. Tran,
>      > Thank you for the suggestion. Indeed, for CsPbCl3 I get very similar
>      > values (0.70 for PBE-SO and 1.67eV for TB-mBJ) and if I add RLO on Pb
>      > for CsPbI3 at least for PBE-SO I have something close to value
>     given in
>      > Jishi.
>      > However, now I have a general question. How to understand that I
>     need to
>      > add RLO in this situation? Should I just try to add RLO on heavy
>     atoms
>      > and if the result changes prefer the values given by +RLO
>     calculations?
>      > Or I had to suspect that I need to add RLO when realized that the
>     result
>      > does not converge until de>15?
>      > Sincerely yours,
>      > Mikhail Nestoklon
>      >
>      > Пятница, 6 декабря 2019, 0:39 +03:00 от "Tran, Fabien"
>      > <fabien.tran at tuwien.ac.at </compose?To=fabien.tran at tuwien.ac.at>
>     </compose?To=fabien.tran at tuwien.ac.at>>:
>      >
>      > Hi,
>      >
>      > Some time ago I did calculations on CsPbCl3 and I could reproduce
>      > Jishi's results reasonably well except the one in the 1st
>      > column("GGA") which is probably wrong. I got 0.70 eV (0.71 eV from
>      > Jishi) for "GGA+SOC", 1.68 eV (1.59 eV from Jishi) for TB-mBJ and
>      > 2.86 eV (2.83 eV from Jishi) for "present". The only special
>      > requirement that was needed is a p1/2 LO on Pb. I used RKMAX=9 and
>      > de=10 in case.in1. The bug reported here
>      >
>      >
>     http://zeus.theochem.tuwien.ac.at/pipermail/wien/2019-November/029882.html
>      >
>      > is active but has very small effect of 0.01 eV for GGA+SOC. Try
>      > CsPbCl3 instead (struct and inso are attached).
>      >
>      > FT
>      >
>      >
>     ------------------------------------------------------------------------
>      > *From:* Wien <wien-bounces at zeus.theochem.tuwien.ac.at
>     </compose?To=wien%2dbounces at zeus.theochem.tuwien.ac.at>> on behalf of
>      > Mikhail Nestoklon <nestoklon at mail.ru </compose?To=nestoklon at mail.ru>>
>      > *Sent:* Thursday, December 5, 2019 8:21 PM
>      > *To:* A Mailing list for WIEN2k users
>      > *Subject:* [Wien] spin-orbit (PBE and mBJ) for perovskites
>      > Dear wien2k community,
>      > I plan to do some DFT calculation of inorganic perovskites using
>      > WIEN2k (19.1 with some patches except the last one for RLO).
>      > I’ve started from attempt to reproduce the values from Jishi et al.,
>      > JPCC 118, 28344 (2014), but can not get the numbers given in Table 2
>      > even for cubic CsPbI_3. The difference seem to stem from the
>     spin-orbit.
>      > What I did:
>      > Using the file in attachment (I use the lattice constant given in
>      > table 1 and R_MT indicated in the text) I do
>      > $ init_lapw -b -vxc 13 -rkmax 9.0 -lvns 6 -fftfac 4 && x kgen
>      > (14x14x14) && run_lapw -ec 0.00001 -cc 0.0001
>      > I use -lvns 6, the difference with the result using standard value
>      > is small, in Jishi et al. it is mentioned that l_{max}=10 which is
>      > default, assuming they might have meant lvns I tried lvns=10, the
>      > difference with lvns=6 is close to zero. I also tried to put Pb p3/2
>      > orbitals to core (if I put P1/2 to valence the result is strange,
>      > but this is a separate question) and see no difference.
>      > For k-mesh I check the convergence and see that for 14x14x14 the
>      > total energy convergence is below 1mRy and GAP is below 1 meV which
>      > is fine, so I am using this k mesh.
>      > Then I check the RKMAX convergence. For Rmt*Kmax =9 without
>      > spin-orbit I get the same number as in Table 2, but I see that this
>      > number is not fully converged: if I increase RKMAX further the total
>      > energy decreases for 8mRy and gap increases for almost 6 meV. I find
>      > it acceptable to use RKMAX=12: only then it is <1mRy and ~1meV for
>      > total energy and gap respectively. The gap with these numbers is
>      > 1.329meV which is 5 meV different from Table 2, this difference is
>      > probably acceptable.
>      > However, when I switch on the spin-orbit, the difference is huge.
>      > With the default values (I only increase llmax, however it does not
>      > change much) I get GAP about 0.269meV, which is almost 4 times
>      > different from the value given in Jishi et al. Table 2. As the SO
>      > value depends on de (in case.in1), I increase this parameter and
>      > check the value of GAP and also spin-orbit splitting of conduction
>      > and valence band (difference between energies of Gamma_8 and Gamma_7
>      > in R point). GAP and spin-orbit in conduction band fully (below
>      > 1meV) converge only when de=15. Still, the band gap is 0.259eV which
>      > is too far from value given in Jishi et al.
>      > The RLO (I tried to add it on Cs, as for Pb it should be useless
>      > [1]) does not help
>      > With the same parameters I do the mBJ run (TB-mBJ). As expected, the
>      > gap increases, but up to 0.722 eV which is two times different from
>      > the value given in Jishi et al.
>      > I did not try to change R_MT: I use the value given in their paper.
>      > The fact that I do reproduce the number given for PBE without
>      > spin-orbit indicates that I hardly did any mistake in the structure.
>      > However, the difference in the numbers with spin-orbit is too large
>      > to be explained by unconverged results, both on my and their side.
>      > Could you help me to understand, how can I reproduce their results?
>      > Thank you in advance.
>      > Sincerely yours,
>      > Mikhail Nestoklon
>      > [1]
>      >
>     https://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/msg17828.html
>      > _______________________________________________
>      > Wien mailing list
>      > Wien at zeus.theochem.tuwien.ac.at
>     </compose?To=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
>      >
>      >
>      > _______________________________________________
>      > Wien mailing list
>      > Wien at zeus.theochem.tuwien.ac.at
>     </compose?To=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
>      >
> 
>     --
> 
>                                             P.Blaha
>     --------------------------------------------------------------------------
>     Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
>     Phone: +43-1-58801-165300 FAX: +43-1-58801-165982
>     Email: blaha at theochem.tuwien.ac.at
>     </compose?To=blaha at theochem.tuwien.ac.at> WIEN2k: http://www.wien2k.at
>     WWW: http://www.imc.tuwien.ac.at/TC_Blaha
>     --------------------------------------------------------------------------
>     _______________________________________________
>     Wien mailing list
>     Wien at zeus.theochem.tuwien.ac.at
>     </compose?To=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
> 
> 
> _______________________________________________
> 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
> 

-- 

                                       P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-165300             FAX: +43-1-58801-165982
Email: blaha at theochem.tuwien.ac.at    WIEN2k: http://www.wien2k.at
WWW:   http://www.imc.tuwien.ac.at/TC_Blaha
--------------------------------------------------------------------------


More information about the Wien mailing list