Re: [Wien] Spin-orbit coupling SCF not converging
As Peter has already responded, TETRA is not appropriate for 2D structures. You got away with it without SOC, as that exactly solves the eigenproblem for RKMAX etc. However, SOC uses the finite set of lapw1 eigenvectors. There will therefore be something similar to telegraph noise as which eigenvectors are used changes with iterations. This leads to ill-conditioning which is amplified by the use of TETRA which I suspect can also introduce something similar to telegraph noise. The total conditioning of a particular case is a product of the condition numbers of different parts. As nband--> infinity the SOC calculation should be better conditioned. Small values of Beta and the total step are clear indicators that the problem is ill-conditioned. I would guesstimate that values smaller than 0.5 are an indicator of ill-conditioning if they persist; an occasional value is OK. Small GREED may also be an indicator of problems, particularly near convergence. However, GREED is more complicated and can legitimately be small far from the fixed-point solution. Wikipedia seems to have a reasonable page, https://en.wikipedia.org/wiki/Condition_number . I have never tried to calculate these for different parts of Wien2k, my understanding is somewhat empirical experience. Peter has more experience so has a better feel, although I am not sure he has ever tried to actually calculate the numbers. (A good project for someone.) On Wed, Nov 27, 2019 at 9:54 AM Luigi Maduro - TNW wrote: > I am using WIEN2k_19.1 and Using grep MULTISECANT *.scfm I get the > following: > > > > * MULTISECANT MIXING VER9 RELEASE 10.4.0 > > > > For the input of the SCF calculation with SOC the output of a SCF > calculation without SOC was used (with TETRA). For the SCF calculation with > SOC the following criteria were used: RKmax = 7.0, 21 k-points in IBZ, > charge convergence of 0.001e and energy convergence of 0.0001 Ry. These are > the same criteria as the original SCF calculation without SOC. > > > If I understood correctly, then looking at the size GREED and Beta should > be sufficient for determining if the calculation has converged, if so then > how small is too small for these parameters? > > > > > > *From:* Laurence Marks [mailto:laurence.ma...@gmail.com] > *Sent:* dinsdag 26 november 2019 18:19 > *To:* A Mailing list for WIEN2k users > *Cc:* Luigi Maduro - TNW > *Subject:* Re: [Wien] Spin-orbit coupling SCF not converging > > > > What version of Wien2k are you using, particularly the mixer (grep > MULTISECANT *.scfm)? > > > > Your calculations are "starving to death". The step size is so small (both > the GREED and Beta) that it is bouncing around on numerical noise. It may > well have already converged to the limits of the noise/conditioning in your > calculation, which is linked to RKMAX and the k-mesh and also (Peter's > response) from TETRA. The iterative diagonalizations also introduce some > noise. > > > > For the specific case I would remove the prior history (rm *.broyd*) and > continue it. > > > > On Tue, Nov 26, 2019 at 11:01 AM Luigi Maduro - TNW > wrote: > > Hello Laurence, > > > > This is the result I get when using Check-mixing (this is with the > thinnest slab model, and using SCALA with Emax = 10.0 Ry) > > > > > > :DIRQ : |MSR1|= 1.472E-06 |PRATT|= 3.852E-03 ANGLE= 79.0 DEGREES > > :DIRT : |MSR1|= 1.516E-06 |PRATT|= 4.100E-03 ANGLE= 79.2 DEGREES > > :MIX : MSE1 REGULARIZATION: 9.15E-04 GREED: 0.00200 Newton 1.00 > 0.0004 > > :ENE : ** TOTAL ENERGY IN Ry = -58196.30065156 > > :DIS : CHARGE DISTANCE ( 0.0046033 for atom7 spin 1) > 0.0009683 > > :PLANE: PW TOTAL 6.0026 DISTAN 3.20E-03 5.33E-02 % > > :CHARG: CLM/ATOM 74.0417 DISTAN 5.58E-04 7.54E-04 % > > :RANK : ACTIVE 14.44/16 = 90.26 % ; YY RANK 14.44/16 = 90.25 % > > :DIRM : MEMORY 16/12 SCALE 1.000 RED 2.57 PRED 0.95 NEXT 0.95 BETA > 0.05 > > :DIRP : |MSR1|= 1.024E-06 |PRATT|= 3.198E-03 ANGLE= 102.9 DEGREES > > :DIRQ : |MSR1|= 4.046E-06 |PRATT|= 1.005E-02 ANGLE= 82.5 DEGREES > > :DIRT : |MSR1|= 4.174E-06 |PRATT|= 1.054E-02 ANGLE= 84.0 DEGREES > > :MIX : MSE1 REGULARIZATION: 1.33E-03 GREED: 0.00500 Newton 1.00 > 0.0004 > > :ENE : ** TOTAL ENERGY IN Ry = -58196.30346394 > > :DIS : CHARGE DISTANCE ( 0.0012891 for atom8 spin 1) > 0.0002073 > > :PLANE: PW TOTAL 6.0026 DISTAN 1.51E-03 2.51E-02 % > > :CHARG: CLM/ATOM 74.0417 DISTAN 2.13E-04 2.88E-04 % > > :RANK : ACTIVE 15.31/16 = 95.68 % ; YY RANK 15.31/16 = 95.72 % > > :DIRM : MEMORY 16/12 SCALE 1.000 RED 0.39 PRED 0.95 NEXT 0.95 BETA > 0
Re: [Wien] Spin-orbit coupling SCF not converging
As general information, Beta is a scaling of the predicted step. Earlier versions of the mixer used an algorithm based upon improvements for this, in the more recent version (improved in the next one, 10.5) it is estimated from the previous history. A value of 0.05 means that the mixer thinks that the predicted step is very inaccurate. The specific case (deleted for size limits on emails) was taking very small steps (e.g. "Newton 1.00 0.0016") which are ~1E-3 of the a Pratt step, which is already small as the calculation is closed to converged (:DIS < 1D-3). With a GREED of 0.002 this means that the unpredicted step is also small, and both are ~1D-6 (:DIRT : |MSR1|= 2.768E-06). Very small steps are susceptible to numerical noise and ill-conditioning. All mixers employ Simplex differentiation which is a form of numerical differentiation so susceptible to such problems. I have not analyzed in detail myself the conditioning/noise of SOC in Wien2k. Peter's recommendations are almost certainly ones which improve the conditioning. -- Professor Laurence Marks Department of Materials Science and Engineering Northwestern University www.numis.northwestern.edu Corrosion in 4D: www.numis.northwestern.edu/MURI Co-Editor, Acta Cryst A "Research is to see what everybody else has seen, and to think what nobody else has thought" Albert Szent-Gyorgi ___ 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] Spin-orbit coupling SCF not converging
What version of Wien2k are you using, particularly the mixer (grep MULTISECANT *.scfm)? Your calculations are "starving to death". The step size is so small (both the GREED and Beta) that it is bouncing around on numerical noise. It may well have already converged to the limits of the noise/conditioning in your calculation, which is linked to RKMAX and the k-mesh and also (Peter's response) from TETRA. The iterative diagonalizations also introduce some noise. For the specific case I would remove the prior history (rm *.broyd*) and continue it. On Tue, Nov 26, 2019 at 11:01 AM Luigi Maduro - TNW wrote: > Hello Laurence, > > > > This is the result I get when using Check-mixing (this is with the > thinnest slab model, and using SCALA with Emax = 10.0 Ry) > > > > > > :DIRQ : |MSR1|= 1.472E-06 |PRATT|= 3.852E-03 ANGLE= 79.0 DEGREES > > :DIRT : |MSR1|= 1.516E-06 |PRATT|= 4.100E-03 ANGLE= 79.2 DEGREES > > :MIX : MSE1 REGULARIZATION: 9.15E-04 GREED: 0.00200 Newton 1.00 > 0.0004 > > :ENE : ** TOTAL ENERGY IN Ry = -58196.30065156 > > :DIS : CHARGE DISTANCE ( 0.0046033 for atom7 spin 1) > 0.0009683 > > :PLANE: PW TOTAL 6.0026 DISTAN 3.20E-03 5.33E-02 % > > :CHARG: CLM/ATOM 74.0417 DISTAN 5.58E-04 7.54E-04 % > > :RANK : ACTIVE 14.44/16 = 90.26 % ; YY RANK 14.44/16 = 90.25 % > > :DIRM : MEMORY 16/12 SCALE 1.000 RED 2.57 PRED 0.95 NEXT 0.95 BETA > 0.05 > > :DIRP : |MSR1|= 1.024E-06 |PRATT|= 3.198E-03 ANGLE= 102.9 DEGREES > > :DIRQ : |MSR1|= 4.046E-06 |PRATT|= 1.005E-02 ANGLE= 82.5 DEGREES > > :DIRT : |MSR1|= 4.174E-06 |PRATT|= 1.054E-02 ANGLE= 84.0 DEGREES > > :MIX : MSE1 REGULARIZATION: 1.33E-03 GREED: 0.00500 Newton 1.00 > 0.0004 > > :ENE : ** TOTAL ENERGY IN Ry = -58196.30346394 > > :DIS : CHARGE DISTANCE ( 0.0012891 for atom8 spin 1) > 0.0002073 > > :PLANE: PW TOTAL 6.0026 DISTAN 1.51E-03 2.51E-02 % > > :CHARG: CLM/ATOM 74.0417 DISTAN 2.13E-04 2.88E-04 % > > :RANK : ACTIVE 15.31/16 = 95.68 % ; YY RANK 15.31/16 = 95.72 % > > :DIRM : MEMORY 16/12 SCALE 1.000 RED 0.39 PRED 0.95 NEXT 0.95 BETA > 0.05 > > :DIRP : |MSR1|= 3.562E-07 |PRATT|= 1.508E-03 ANGLE= 72.0 DEGREES > > :DIRQ : |MSR1|= 1.479E-06 |PRATT|= 3.835E-03 ANGLE= 80.1 DEGREES > > :DIRT : |MSR1|= 1.522E-06 |PRATT|= 4.121E-03 ANGLE= 79.5 DEGREES > > :MIX : MSE1 REGULARIZATION: 9.97E-04 GREED: 0.00200 Newton 1.00 > 0.0004 > > :ENE : ** TOTAL ENERGY IN Ry = -58196.29908151 > > :DIS : CHARGE DISTANCE ( 0.0007022 for atom7 spin 1) > 0.0001661 > > :PLANE: PW TOTAL 6.0026 DISTAN 8.27E-04 1.38E-02 % > > :CHARG: CLM/ATOM 74.0417 DISTAN 1.39E-04 1.87E-04 % > > :RANK : ACTIVE 14.56/15 = 97.05 % ; YY RANK 14.56/15 = 97.07 % > > :DIRM : MEMORY 15/12 SCALE 1.000 RED 0.64 PRED 0.95 NEXT 0.97 BETA > 0.05 > > :DIRP : |MSR1|= 1.020E-06 |PRATT|= 8.274E-04 ANGLE= 53.5 DEGREES > > :DIRQ : |MSR1|= 4.021E-06 |PRATT|= 2.497E-03 ANGLE= 54.9 DEGREES > > :DIRT : |MSR1|= 4.149E-06 |PRATT|= 2.630E-03 ANGLE= 54.9 DEGREES > > :MIX : MSE1 REGULARIZATION: 9.23E-04 GREED: 0.00289 Newton 1.00 > 0.0016 > > :ENE : ** TOTAL ENERGY IN Ry = -58196.29797368 > > :DIS : CHARGE DISTANCE ( 0.0051635 for atom7 spin 1) > 0.0011664 > > :PLANE: PW TOTAL 6.0026 DISTAN 4.15E-03 6.91E-02 % > > :CHARG: CLM/ATOM 74.0417 DISTAN 6.89E-04 9.31E-04 % > > :RANK : ACTIVE 13.83/16 = 86.45 % ; YY RANK 13.82/16 = 86.38 % > > :DIRM : MEMORY 16/12 SCALE 1.000 RED 4.97 PRED 0.97 NEXT 0.94 BETA > 0.06 > > :DIRP : |MSR1|= 7.019E-07 |PRATT|= 4.150E-03 ANGLE= 88.8 DEGREES > > :DIRQ : |MSR1|= 2.677E-06 |PRATT|= 1.240E-02 ANGLE= 83.7 DEGREES > > :DIRT : |MSR1|= 2.768E-06 |PRATT|= 1.308E-02 ANGLE= 84.1 DEGREES > > :MIX : MSE1 REGULARIZATION: 1.18E-03 GREED: 0.00220 Newton 1.00 > 0.0002 > > :ENE : ** TOTAL ENERGY IN Ry = -58196.30074694 > > > > > > Cheers, > > Luigi > > PhD candidate > Kavli Institute of Nanoscience > > Department of Quantum Nanoscience > > Faculty of Applied Sciences > > Delft University of Technology > > > -- Professor Laurence Marks Department of Materials Science and Engineering Northwestern University www.numis.northwestern.edu Corrosion in 4D: www.numis.northwestern.edu/MURI Co-Editor, Acta Cryst A "Research is to see what everybody else has seen, and to think what nobody else has thought" Albert Szent-Gyorgi ___ 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] Spin-orbit coupling SCF not converging
A general comment about convergence, as there have been a few recent questions. Convergence is similar to paddling a canoe down a river from the mountains to the sea. Sometimes the water is rushing by and you (the mixer) have to try and avoid the rocks (ghost bands); sometimes you go over a waterfall and the problem changes (electronic phase transition); sometimes you are on a meandering river and do not seem to be making progress. Check-mixing is designed to give some idea about what is going on, as a simple grep on :DIS, for instance, does not reveal enough. _ Professor Laurence Marks "Research is to see what everybody else has seen, and to think what nobody else has thought", Albert Szent-Gyorgi www.numis.northwestern.edu On Tue, Nov 26, 2019, 07:21 Luigi Maduro - TNW wrote: > Hello there WIEN2k users, > > > > I am having a convergence problem with a supercell calculation of a slab > of MoS2 including Spin-Orbit Coupling (SOC). The supercell is made by > cutting in the plane of MoS2 which leads to either only edge Mo atoms or > edge S atoms. A vacuum parallel to the edges is introduced (about 18 > angstroms) that separates the repeating images. Converged SCF calculations > without SOC were found with this type of geometry for various widths. These > converged SCF calculations were then used as input for an SCF calculation > with SOC for the different widths. However, when including SOC only the > smallest width gave a converged SCF calculation, albeit after doubling the > amount of k-points and increasing nbands(more than twice what was > originally suggested when running init_so). The larger widths do not result > in a converged SCF calculation, even after substantially increasing nbands. > Is the solution to just keep on increasing the amount of k-points until I > do get converged SCF calculations or is the inclusion of the vacuum giving > problems for lapwso? > > > > > > As a side not, I have a general question on how to relate Emax and nbands > in case.in1. Up until recently I have been using SCALAPACK in WIEN2k and > now I have switched to using ELPA. In the case of MoS2, a system with large > spin-orbit coupling, the userguide recommends to increase the value of Emax > to up to 10 Ry for large SOC systems when running the init_so script. If > one is using ELPA then nbands should be increased. In the case of ELPA > should nbands be increased to twice the amount given in case.in1 when > initially running init_so? > > > > > > PhD candidate > Kavli Institute of Nanoscience > > Department of Quantum Nanoscience > > Faculty of Applied Sciences > > Delft University of Technology > > > ___ > Wien mailing list > Wien@zeus.theochem.tuwien.ac.at > > https://urldefense.proofpoint.com/v2/url?u=http-3A__zeus.theochem.tuwien.ac.at_mailman_listinfo_wien&d=DwICAg&c=yHlS04HhBraes5BQ9ueu5zKhE7rtNXt_d012z2PA6ws&r=U_T4PL6jwANfAy4rnxTj8IUxm818jnvqKFdqWLwmqg0&m=XX2No0AfEa4DjARfeJMqG19OtvlKCdEDrTXWDvLOmqo&s=WGaKK3zLkOCpwqEGHZ8_GDByOyFmnUuDzbYZU5bAppU&e= > SEARCH the MAILING-LIST at: > https://urldefense.proofpoint.com/v2/url?u=http-3A__www.mail-2Darchive.com_wien-40zeus.theochem.tuwien.ac.at_index.html&d=DwICAg&c=yHlS04HhBraes5BQ9ueu5zKhE7rtNXt_d012z2PA6ws&r=U_T4PL6jwANfAy4rnxTj8IUxm818jnvqKFdqWLwmqg0&m=XX2No0AfEa4DjARfeJMqG19OtvlKCdEDrTXWDvLOmqo&s=pdoMtEVvzwPB9V7bLKYqxvdVDi48PKMm7effnVG7774&e= > ___ 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] Spin-orbit coupling SCF not converging
lapwso should work for surface slabs. Is it a magnetic system ?? Probably not. My guess: these slabs get metallic and you used TETRA ??? The additionally degeneracy may cause problems. For a 2D system always use TEMP(S) and some smearing (conservative: 2 mRy; to get improved scf-convergence increase it (temporarily) to 10 mRy For such large cells I'd start with a small Emax (NBAND) and after convergence recheck with a larger value (it depends what quantities do you want to calculate with SO). On 11/26/19 2:21 PM, Luigi Maduro - TNW wrote: Hello there WIEN2k users, I am having a convergence problem with a supercell calculation of a slab of MoS2 including Spin-Orbit Coupling (SOC). The supercell is made by cutting in the plane of MoS2 which leads to either only edge Mo atoms or edge S atoms. A vacuum parallel to the edges is introduced (about 18 angstroms) that separates the repeating images. Converged SCF calculations without SOC were found with this type of geometry for various widths. These converged SCF calculations were then used as input for an SCF calculation with SOC for the different widths. However, when including SOC only the smallest width gave a converged SCF calculation, albeit after doubling the amount of k-points and increasing nbands(more than twice what was originally suggested when running init_so). The larger widths do not result in a converged SCF calculation, even after substantially increasing nbands. Is the solution to just keep on increasing the amount of k-points until I do get converged SCF calculations or is the inclusion of the vacuum giving problems for lapwso? As a side not, I have a general question on how to relate Emax and nbands in case.in1. Up until recently I have been using SCALAPACK in WIEN2k and now I have switched to using ELPA. In the case of MoS2, a system with large spin-orbit coupling, the userguide recommends to increase the value of Emax to up to 10 Ry for large SOC systems when running the init_so script. If one is using ELPA then nbands should be increased. In the case of ELPA should nbands be increased to twice the amount given in case.in1 when initially running init_so? PhD candidate Kavli Institute of Nanoscience Department of Quantum Nanoscience Faculty of Applied Sciences Delft University of Technology ___ 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] Spin-orbit coupling SCF not converging
For the convergence, please run "Check-mixing" (for recent versions) and paste the result. This may give an idea. It may be that the calculation is slowly changing the spin/orbital momentum and progressing towards the solution. You may have a tunnel or even spiral convergence. For SOC, the accuracy depends upon the number of bands used (nband with ELPA). I suggest testing the convergence with a smaller problem (e.g. bulk MoS2) then transferring this to your problem, e.g. chose nband such that it gives states to the same energy above Ef. _ Professor Laurence Marks "Research is to see what everybody else has seen, and to think what nobody else has thought", Albert Szent-Gyorgi www.numis.northwestern.edu On Tue, Nov 26, 2019, 07:21 Luigi Maduro - TNW wrote: > Hello there WIEN2k users, > > > > I am having a convergence problem with a supercell calculation of a slab > of MoS2 including Spin-Orbit Coupling (SOC). The supercell is made by > cutting in the plane of MoS2 which leads to either only edge Mo atoms or > edge S atoms. A vacuum parallel to the edges is introduced (about 18 > angstroms) that separates the repeating images. Converged SCF calculations > without SOC were found with this type of geometry for various widths. These > converged SCF calculations were then used as input for an SCF calculation > with SOC for the different widths. However, when including SOC only the > smallest width gave a converged SCF calculation, albeit after doubling the > amount of k-points and increasing nbands(more than twice what was > originally suggested when running init_so). The larger widths do not result > in a converged SCF calculation, even after substantially increasing nbands. > Is the solution to just keep on increasing the amount of k-points until I > do get converged SCF calculations or is the inclusion of the vacuum giving > problems for lapwso? > > > > > > As a side not, I have a general question on how to relate Emax and nbands > in case.in1. Up until recently I have been using SCALAPACK in WIEN2k and > now I have switched to using ELPA. In the case of MoS2, a system with large > spin-orbit coupling, the userguide recommends to increase the value of Emax > to up to 10 Ry for large SOC systems when running the init_so script. If > one is using ELPA then nbands should be increased. In the case of ELPA > should nbands be increased to twice the amount given in case.in1 when > initially running init_so? > > > > > > PhD candidate > Kavli Institute of Nanoscience > > Department of Quantum Nanoscience > > Faculty of Applied Sciences > > Delft University of Technology > > > ___ > Wien mailing list > Wien@zeus.theochem.tuwien.ac.at > > https://urldefense.proofpoint.com/v2/url?u=http-3A__zeus.theochem.tuwien.ac.at_mailman_listinfo_wien&d=DwICAg&c=yHlS04HhBraes5BQ9ueu5zKhE7rtNXt_d012z2PA6ws&r=U_T4PL6jwANfAy4rnxTj8IUxm818jnvqKFdqWLwmqg0&m=XX2No0AfEa4DjARfeJMqG19OtvlKCdEDrTXWDvLOmqo&s=WGaKK3zLkOCpwqEGHZ8_GDByOyFmnUuDzbYZU5bAppU&e= > SEARCH the MAILING-LIST at: > https://urldefense.proofpoint.com/v2/url?u=http-3A__www.mail-2Darchive.com_wien-40zeus.theochem.tuwien.ac.at_index.html&d=DwICAg&c=yHlS04HhBraes5BQ9ueu5zKhE7rtNXt_d012z2PA6ws&r=U_T4PL6jwANfAy4rnxTj8IUxm818jnvqKFdqWLwmqg0&m=XX2No0AfEa4DjARfeJMqG19OtvlKCdEDrTXWDvLOmqo&s=pdoMtEVvzwPB9V7bLKYqxvdVDi48PKMm7effnVG7774&e= > ___ 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