Re: [Wien] Issue with different convergence results and QTL-B warnings
- The execution is always the same. I used the -in1new switch on a copy of the case directory to determine the correct energy parameters. For the proper calculations I manually changed the energy parameters in case.in1c - All the in-files are the same (except for the carbon part). So are the struct files - I did not perform relaxation. I'm trying to stick to the original structure of the molecule. - I use RKM=2.0 because the calculations are 3 times faster than for RKM=3.0 and it turns out that there is no significant difference between results (energy, J, MMI, gaps, ...) for RKM=2.0 and 3.0. Also changing RMT or k-sampling is unimportant (although for smaller RMT magnetic moments for Cr decrease about 0.2 Bohr magnetons). However this is based on my previous experience with PBE functional for these structures and it might not be true for B3LYP. The reason I use these RMT-s is because I wanted them to be the same as in Acta Phys. Pol. A. 127, 407 (2015) and Phys. Rev. B 73, 184431 (2006). - As for the magnetic moments they are almost constant and definitely hold their orientation. The same goes for partial charges - in both approaches they are the same. - Offtop: the relation J = (E_FM - E_AFM)/36 for Cr8 is obtained from a general Ising-like magnetic Hamiltonian: H=sum(J*S_{i}*S_{i+1}) where S are spins of Cr ions, S=+/- 3/2. For 8 Cr this gives 8 interactions between neighboring ions and for AFM configuration H=J * 8 * - 3/2 * 3/2=-18J=E_AFM and for FM configuration H=J * 8 * 3/2 *3/2=18J=E_FM If we take the difference we get E_FM-E_AFM=18J- -18J=18J+18J=36J Best Regards Michał Wojciechowski PHD student at Institute of Physics Department of Physics and Astronomy University of Zielona Góra, Poland - Oryginalna wiadomość - Od: "Lyudmila Dobysheva" Do: "A Mailing list for WIEN2k users" Wysłane: czwartek, 10 grudzień 2015 12:02:08 Temat: Re: [Wien] Issue with different convergence results and QTL-Bwarnings I am not an expert in B3LYP, so I am writing about only general considerations. The reason may be 1) different execution. For example, you have made, by accident, for the final calculation in first case -in1new, in second - usual, without this option. Check the executions in dayfile's, :log's, and everywhere where this info can be found. And check all parameters. Compare the in-files in the whole saved directories. Though system looks rather symmetric, there are 5 nonequivalent Cr atoms. Had you done force relaxation? Is struct files identical in both cases? in all four cases? 2) Maybe some parameter doesn't reach convergence, so small variations in calculation scheme (starting electron density) gives different results. Under suspicion, imho, RKM = 2.0 (as far as I remember, the recommendation for H-containing systems is RKM=3), very different RMT 2.40 and 0.83 Bohr for Cr and H, single k point. 3) maybe this exotic scheme of potential gives you really two different points? Look in case.scf other parameters, especially magnetic moments of Cr's, especially FM states (the difference is largest. For AFM it looks for me as not different). 3A) maybe this is a peculiarity of B3LYP scheme, and its iterative procedure allows to fix solution in some points that are not real minimum. Maybe users with experience with B3LYP can add something about this. 3B) Maybe this different solutions are really different minimums due to different Cr atoms with differently distributed moments among them. Offtop: > J = (E_FM - E_AFM)/36. Why 36? There is another number of atoms? Best wishes Lyudmila Dobysheva -- Phys.-Techn. Institute of Ural Br. of Russian Ac. of Sci. 426001 Izhevsk, ul.Kirova 132 RUSSIA -- Tel.:7(3412) 432045(office), 722529(Fax) E-mail: l...@ftiudm.ru, lyuk...@mail.ru (office) lyuk...@gmail.com (home) Skype: lyuka17 (home), lyuka18 (office) http://ftiudm.ru/content/view/25/103/lang,english/ -- ___ 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 ___ 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] Issue with different convergence results and QTL-B warnings
a) Interesting choice of RMTs, but probably not optimal. You should use setrmt as basis for your choice. It would reduce RMT(H) (0.6) and increase R(C) a bit, but still make Oxygen 15% larger than C. In particular the F-spheres are much too small at present. I'd use 2.0/1.6 or 1.9/1.7 for Cr and F spheres. RKMax=2.0 with your spheres is VERY small and I'd NOT trust any results with this setup. With my spheres you need RKMax 2.5-3.0. b) I agree, for such molecular materials, our default E-parameters might not be optimal and problems for all the small spheres F, O and C may appear. Even WITHOUT QTL-B warnings, the E-parameter of F and O may not be optimal ! As mentioned many times on the mailinglist, the way to check this is looking at case.scf1 and case.scf2. It is not enough just to look at case.in1c, as these numbers will be changed automatically in lapw1: 0.3 ---> EF-(or+) 0.2 -0.71 0.002 ---> it tries a search, but for small spheres it probably will actually use -0.71 because it does not find E-top You probably find in case.scf1 that the two C-s (l=0) E-parameter are very close to each other (using the default input)! a) Locate the fermi energy ! b) locate the mean energies of C-2s (and O-2s and F-2s) by looking in case.scf2 (:EPL and :EPH) c) change the FIRST line for l=0 of all the small atoms to this E-value in case.in1c (for C this might be -1.7 in your case). Leave the 0.30 untouched, as this will be automatically adjusted according to EF. Repeat this for ALL "small" atoms (except for H). - Otherwise: compare the magnetic states of your calculations. I'm rather surprised that also the FM solutions are different, but it could well be. We know that all "orbital-dependent" potentials (like LDA+U or hybrid-DFT) depend a lot on the starting electronic structure. Compare the :MMI and :QTL values. If your calculations are ok, then the total energy is a good measure, which state is most favorable. On 12/09/2015 05:57 PM, Michał Wojciechowski wrote: Dear Wien2k users! Currently I’m performing calculations for a chromium magnetic wheel - Cr_8F_8(O_2CH)_16. While I intend to get various quantities the primary one is magnetic interaction parameter J between chromium ions. To that end I require energies of two spin configurations - ferromagnetic (E_FM) and antiferromagnetic (E_AFM) - since J = (E_FM - E_AFM)/36. The case file is attached. I’m using RKM = 2.0 (NMATMAX = 50 000, so RKM is unreduced), RMT = 2.40, 1.20, 1.24, 1.00 and 0.83 Bohr for Cr, F, O, C and H, respectively, single k point since it is a molecule. Remaining parameters are default or as suggested by the program. I use two approaches. Approach 1. I start with PBE functional from a command line with: runsp_lapw -i 50 -ec 0.1 -cc 0.001 The calculations converge nicely with no warnings or errors. I then change the XC functional to B3LYP (option 47 in case.in0 and additional case.ineece file) and run the command: runsp_lapw -eece -i 50 -ec 0.1 -cc 0.001 The calculations converge nicely with no errors but there is a single warning: :WARN : QTL-B value eq. 11.12 in Band of energy -2.28970 ATOM= 27 L= 0 Atom 27 is carbon. The case.in1 entry for this atom is 0.303 0 (GLOBAL E-PARAMETER WITH n OTHER CHOICES, global APW/LAPW) 0 -0.71 0.002 CONT 1 00.30 0.000 CONT 1 10.30 0.000 CONT 1 Using -in1new I was able to determine, that using parameter -1.70 instead of 0.30 resolves the issue. I then changed entries for all carbon atoms to 0.303 0 (GLOBAL E-PARAMETER WITH n OTHER CHOICES, global APW/LAPW) 0 -0.71 0.002 CONT 1 0 -1.70 0.000 CONT 1 10.30 0.000 CONT 1 and continued the calculations. This time I got no warnings and obtain results: E_FM = -24370.92259156 Ry E_AFM = -24370.92652933 Ry J = 1.5 meV Approach 2. I starte with PBE functional and already modifie energy parameter in case.in1. After convergence is reached I continue with B3LYP functional. No warnings are present and I obtain the following results: E_FM = -24370.91735838 Ry E_AFM = -24370.92522898 Ry J = 3.0 meV So the issue is why are there such big differences? Which quantities should I check to determine the source of these discrepancies? I am running Wien version 13.1 on a one, single processor, multi-core Intel Xeon machine with openSUSE 13.1 operating system. Programs were build with Intel compilers and math libraries (Intel Composer 2013 SP1) and OpenMPI 1.6.5. Best Regards -- Michał Wojciechowski PHD student at Institute of Physics Department of Physics and Astronomy University of Zielona Góra, Poland ___ 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.Bl
Re: [Wien] Issue with different convergence results and QTL-B warnings
09.12.2015 20:57, Michał Wojciechowski wrote: Currently I’m performing calculations for a chromium magnetic wheel - Cr_8F_8(O_2CH)_16. J = (E_FM - E_AFM)/36. Approach 1. I start with PBE functional from a command line with: runsp_lapw -i 50 -ec 0.1 -cc 0.001 The calculations converge nicely with no warnings or errors. I then change the XC functional to B3LYP (option 47 in case.in0 and additional case.ineece file) and run the command: runsp_lapw -eece -i 50 -ec 0.1 -cc 0.001 The calculations converge nicely with no errors but there is a single warning: :WARN : QTL-B value eq. 11.12 in Band of energy -2.28970 ATOM= 27 L= 0 Atom 27 is carbon. Using -in1new I was able to determine, that using parameter -1.70 instead of 0.30 resolves the issue. I then changed entries for all carbon atoms and continued the calculations. This time I got no warnings and obtain results: E_FM = -24370.92259156 Ry E_AFM = -24370.92652933 Ry J = 1.5 meV Approach 2. I started with PBE functional and already modifie energy parameter in case.in1. After convergence is reached I continue with B3LYP functional. No warnings are present and I obtain the following results: E_FM = -24370.91735838 Ry E_AFM = -24370.92522898 Ry J = 3.0 meV I am not an expert in B3LYP, so I am writing about only general considerations. The reason may be 1) different execution. For example, you have made, by accident, for the final calculation in first case -in1new, in second - usual, without this option. Check the executions in dayfile's, :log's, and everywhere where this info can be found. And check all parameters. Compare the in-files in the whole saved directories. Though system looks rather symmetric, there are 5 nonequivalent Cr atoms. Had you done force relaxation? Is struct files identical in both cases? in all four cases? 2) Maybe some parameter doesn't reach convergence, so small variations in calculation scheme (starting electron density) gives different results. Under suspicion, imho, RKM = 2.0 (as far as I remember, the recommendation for H-containing systems is RKM=3), very different RMT 2.40 and 0.83 Bohr for Cr and H, single k point. 3) maybe this exotic scheme of potential gives you really two different points? Look in case.scf other parameters, especially magnetic moments of Cr's, especially FM states (the difference is largest. For AFM it looks for me as not different). 3A) maybe this is a peculiarity of B3LYP scheme, and its iterative procedure allows to fix solution in some points that are not real minimum. Maybe users with experience with B3LYP can add something about this. 3B) Maybe this different solutions are really different minimums due to different Cr atoms with differently distributed moments among them. Offtop: > J = (E_FM - E_AFM)/36. Why 36? There is another number of atoms? Best wishes Lyudmila Dobysheva -- Phys.-Techn. Institute of Ural Br. of Russian Ac. of Sci. 426001 Izhevsk, ul.Kirova 132 RUSSIA -- Tel.:7(3412) 432045(office), 722529(Fax) E-mail: l...@ftiudm.ru, lyuk...@mail.ru (office) lyuk...@gmail.com (home) Skype: lyuka17 (home), lyuka18 (office) http://ftiudm.ru/content/view/25/103/lang,english/ -- ___ 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