[Wien] incorrect band splitting when RLO added for a system with spatial inversion
Dear Wien2k users, I am facing a problem with symmetry preserved result when relativistic local orbital (RLO) is added to the spin-orbit coupling calculations. The system is bismuthene, a 2D layer of Bi atoms, in orthorombic lattice WITH inversion. Without RLO the bands are doubly degenerated, up to numerical precision below 1 micro eV, (correct result). By adding RLO for 5P_1/2 (found by autosearch to -8.08 Ry, close to the -8.3 Ry from lstart) the bands split. For the top valence band it can reach values of several meV (!), see plot in the attachment. It stays degenerate at Gamma point while at the other time-reversal invariant points (X, Y, and S point in the plot) it does not. My question is: How to improve precision? Do you have any idea what could be reason for and provide possible hint how to solve the problem? Another (perhaps minor) point: Autosearch finds RLO at -8.08 Ry while lstart finds for 5P_1/2 value -8.3 Ry, what is behind the 0.2 Ry renormalization? Note 1: Calculations done by version 14.2, XC_PBESOL and default basis type used, increasing Emax and Rmax do not improve results, the "ghost" splitting stays there. Note 2: Wien2k version 16 gives similar results. For intallation -- I had to patch the siteconfig script commenting line 148 if("$test" != "/") goto iloop otherwise it runs into the infinite iloop if the FFTWl_LIBS are not set up. Best regards, Martin Gmitra Uni Regensburg ___ Wien mailing list Wien@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
Re: [Wien] incorrect band splitting when RLO added for a system with spatial inversion
This is related to the fact, that the RLOs are implemented without symmetry constrains (and relativistic spherical harmonics basis) and for numerical reasons this leads to these small splittings. This is also noted when calculating EFGs, that one must not use RLOs. Why are you using RLOs and the Bi 5p states as valence anyway ? States at -8 Ry should usually be kept in the core, unless one has very good reasons not to do so (high pressure calculations). Treating them as core, they get the full relativistic treatment (have proper j-based radial wave functions) and this description is much better than a "band-treatment" with a scalar-relativistic basis (l-based), where the band-width is usually much smaller that spin-orbit effects. The Bi spheres should be sufficiently large and eventual small core leakage can be treated using the automatic way with .lcore. The change in energy between lstart and lapw1 is due to a shift in potential. In lstart we have an energy-zero = 0 at infinity, in lapw we have another energy zero. On 12/14/2016 07:58 AM, Martin Gmitra wrote: Dear Wien2k users, I am facing a problem with symmetry preserved result when relativistic local orbital (RLO) is added to the spin-orbit coupling calculations. The system is bismuthene, a 2D layer of Bi atoms, in orthorombic lattice WITH inversion. Without RLO the bands are doubly degenerated, up to numerical precision below 1 micro eV, (correct result). By adding RLO for 5P_1/2 (found by autosearch to -8.08 Ry, close to the -8.3 Ry from lstart) the bands split. For the top valence band it can reach values of several meV (!), see plot in the attachment. It stays degenerate at Gamma point while at the other time-reversal invariant points (X, Y, and S point in the plot) it does not. My question is: How to improve precision? Do you have any idea what could be reason for and provide possible hint how to solve the problem? Another (perhaps minor) point: Autosearch finds RLO at -8.08 Ry while lstart finds for 5P_1/2 value -8.3 Ry, what is behind the 0.2 Ry renormalization? Note 1: Calculations done by version 14.2, XC_PBESOL and default basis type used, increasing Emax and Rmax do not improve results, the "ghost" splitting stays there. Note 2: Wien2k version 16 gives similar results. For intallation -- I had to patch the siteconfig script commenting line 148 if("$test" != "/") goto iloop otherwise it runs into the infinite iloop if the FFTWl_LIBS are not set up. Best regards, Martin Gmitra Uni Regensburg ___ Wien mailing list Wien@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: bl...@theochem.tuwien.ac.atWIEN2k: http://www.wien2k.at WWW: http://www.imc.tuwien.ac.at/TC_Blaha -- ___ Wien mailing list Wien@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
Re: [Wien] incorrect band splitting when RLO added for a system with spatial inversion
Dear Peter, Thank you for your answer. The main reason to use the RLO is the effect close to the Fermi level, where the top valence and bottom conduction bands are strongly pushed towards each other due to spin-orbit coupling. In the attached figure you can find plot for the different cases. The blue line is for the 5P in valence without RLO, adding RLO, red line, the band inversion is strongly enhanced. When 5P are in core, the effect of band inversion is sort of in between the cases without RLO and with RLO, see green and violet dashed curves. Effect of .lcore is rather minor, compare the green and violet lines. The spheres we consider are 2.5 a.u., and 5P in the core gives charge leakage of 0.003020. For the larger spheres of 2.7 a.u., the core charge leakage reduces to 0.001118, but resulting band structure (using .lcore as well) is pretty the same as the violet one, not shown in the figure. Could you comment please what is the good criteria to decide whether the leakage is small or not? I would be very glad you could comment on the RLO effect to the band inversion and give your statement which calculations are more correct? Best regards, Martin Gmitra Uni Regensburg ___ Wien mailing list Wien@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
Re: [Wien] incorrect band splitting when RLO added for a system with spatial inversion
Is this all converged with respect to EMAX in case.in1 ??? So from your figure one would find that these bands at EF are affected by >0.1 eV when you add a p1/2 basis function at -8 Ry This sounds VERY strange to me and I don't know if one can explain this by an "orthogonality effect" (If the 5p-1/2 state is better described due to this basis, the 6p-1/2 must be orthogonal to them and thus are also affected). What I would do is to check what comes out by putting the 5p in core, but then try to add a RLO with an energy=0.3 (a 6p-RLO). Charge leakage does not seem to have an effect as seen in your plot or according to your statement that you get the same reults with a different RMT. On 12/15/2016 10:59 AM, Martin Gmitra wrote: Dear Peter, Thank you for your answer. The main reason to use the RLO is the effect close to the Fermi level, where the top valence and bottom conduction bands are strongly pushed towards each other due to spin-orbit coupling. In the attached figure you can find plot for the different cases. The blue line is for the 5P in valence without RLO, adding RLO, red line, the band inversion is strongly enhanced. When 5P are in core, the effect of band inversion is sort of in between the cases without RLO and with RLO, see green and violet dashed curves. Effect of .lcore is rather minor, compare the green and violet lines. The spheres we consider are 2.5 a.u., and 5P in the core gives charge leakage of 0.003020. For the larger spheres of 2.7 a.u., the core charge leakage reduces to 0.001118, but resulting band structure (using .lcore as well) is pretty the same as the violet one, not shown in the figure. Could you comment please what is the good criteria to decide whether the leakage is small or not? I would be very glad you could comment on the RLO effect to the band inversion and give your statement which calculations are more correct? Best regards, Martin Gmitra Uni Regensburg ___ Wien mailing list Wien@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: bl...@theochem.tuwien.ac.atWIEN2k: http://www.wien2k.at WWW: http://www.imc.tuwien.ac.at/TC_Blaha -- ___ Wien mailing list Wien@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
Re: [Wien] incorrect band splitting when RLO added for a system with spatial inversion
Thanks for replay! Considering the RLO for 6P (0.3 Ry), it strengthen the spin-orbit coupling effects rather significantly, comparable to results when considering RLO for 5P at -8Ry, see yellow and dark red curves in the attached plot. Can one understand this as the orthogonality effect as you have suggested? Important observation is that dispersions with RLO for 5P in valence and RLO for 6P (with 5P in core) is different (yellow and dark red curves). Anyhow, I am getting impression from your argumentation that the correct practice would be to take 5P for core + .lcore and add the RLO for the 6P unoccupied states, right? If one considers this case (results shown by dark red curve, RLO for 6P) as an benchmark, results obtained when enlarging Emax up to 45 Ry (blue plus symbols) can be assumed as converged (compare orange, green and blue curves) but hardly reach the case with RLO for 6P. There is still a difference of about 0.1 eV. What do you think about the observed differences? Do you think is there a sense to increase the Emax even more? (How to get even more bands without increasing Rkmax?) Best regards, Martin Gmitra Uni Regensburg ___ Wien mailing list Wien@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
Re: [Wien] incorrect band splitting when RLO added for a system with spatial inversion
Emax seems converged. This is no issue. Yes, from my point of view the "best" calculation should be to put 5p in core (+ .locre if necessary) and have the 6p RLO. It seems it has a rather large effect. To check my results I would: a) Calculate the qtl's with p-1/2 p3/2 splitting (x qtl -so) and check that within these bands there is significant character of p1/2 or p3/2. (Character-plot with band structure), which varies from Gamma-X explaining the importance of the p1/2 RLO. b) Do a calculation without symmetry (P1) and calculate the bands in different "equivalent" directions of the BZ. I'd like to see that there is no symmetry-bug related to the RLOs. c) I'd also would reconfirm, that after an scf calculation with SO but WITHOUT RLO, and just doing the bandstructure with and without RLO (non-selfconsistent) one gets more or less the same change (or is it an scf-effect ?). In addition, is the sum of eigenvalues (case.scf2) after lapw2 lower with RLO (check of variational principle). In any case, looks interesting What says experiment ??? Am 21.12.2016 um 21:50 schrieb Martin Gmitra: Thanks for replay! Considering the RLO for 6P (0.3 Ry), it strengthen the spin-orbit coupling effects rather significantly, comparable to results when considering RLO for 5P at -8Ry, see yellow and dark red curves in the attached plot. Can one understand this as the orthogonality effect as you have suggested? Important observation is that dispersions with RLO for 5P in valence and RLO for 6P (with 5P in core) is different (yellow and dark red curves). Anyhow, I am getting impression from your argumentation that the correct practice would be to take 5P for core + .lcore and add the RLO for the 6P unoccupied states, right? If one considers this case (results shown by dark red curve, RLO for 6P) as an benchmark, results obtained when enlarging Emax up to 45 Ry (blue plus symbols) can be assumed as converged (compare orange, green and blue curves) but hardly reach the case with RLO for 6P. There is still a difference of about 0.1 eV. What do you think about the observed differences? Do you think is there a sense to increase the Emax even more? (How to get even more bands without increasing Rkmax?) Best regards, Martin Gmitra Uni Regensburg ___ Wien mailing list Wien@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 -- -- Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna Phone: +43-1-58801-165300 FAX: +43-1-58801-165982 Email: bl...@theochem.tuwien.ac.atWIEN2k: http://www.wien2k.at WWW: http://www.imc.tuwien.ac.at/staff/tc_group_e.php -- ___ Wien mailing list Wien@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