Re: [Wien] Extraction of atomic sphere potential
Thank you so much for your help. Best regards, Artem Tarasov ___ 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] Strategy for a large slab with SP and SOC
Your 4 points are not really recommended in the first place. If it is a scf convergence problem (which I doubt): grep:DIS case.scf . Does it look like divergence ? You need to find which eigenvalue causes the ghostband, from which atom and angular momentum. See *scf2* and *output2* files. Once you know this, look into case.scf1 to see how the LOs and energy parameters for this state are set and you probably have to modyfy case.in1(c). PS: Use init_lapw -prec 1n at the beginning, maybe with -ecut 0.999 . PS: I would NOT include HDLOs, if I had ghostbands. Mixing with PRATT helps only in very few cases, not really recommended for "normal" calculations. PPS: I hope you use runsp_c_lapw for something like WTe2 ? Am 17.06.2023 um 00:28 schrieb pluto via Wien: Dear Prof. Blaha, dear All, Thank you for the comment on slab strategy, this helps a lot. I have more specific question: for a large WTe2 slab (60 atoms), which is a material of low-symmetry that has a polarity also in the out-of-plane direction, I am getting ghostbands in lapw2 after few iterations. What is a good strategy to fix this? I was thinking of: 1. init_lapw -hdlo 2. Low mixing (like 0.05) with PRATT 3. Decrease RMT (from first tests, with RMT 2.5 ghostbands seem to appear after around 3 iterations, with 2.2 after many iterations) 4. Increase RKmax 3 and 4 are probably computationally expensive... I did several tests without SOC, I was typically using something like: init_lapw -sp -b -numk 100 -hdlo -fermit 0.002 Maybe other settings are critical? Bulk calculation converges very easily (first without and then with SOC) with default settings like init_lapw -sp -b -numk 2000 bulk bands look like the literature, and are practically the same with RMT 2.2 and RMT 2.5. Best, Lukasz On 2023-06-16 16:45, Peter Blaha wrote: No,this is not a good strategy. From a converged non-spin-polarized calculation you cannot come (easily) to a spin-polarized solution. So 1) is only good if you want to quote how much more stable a SP solution is compared to a non-SP. 2 + 3 is a good practice. You gain insight how large are the changes and on what atoms due to SO coupling. In terms of efficiency for large cases, I'd in particular preconverge with a course k-mesh and later on refine. --- Every runsp cycle starts with a case.clmsum/up/dn file. These files can come from an initialization, but of course also from any prior scf calculation (eg. with lower k-mesh or without SO). Of course, a restore_lapw ... gives you all files necessary to run another scf cycle. -NI would keep old broyden files, but after a "save_lapw" they are gone anyway. -NI is useful if you want to continue a scf, because eg. the first runsp stopped after 40 cycles and did not reach convergence yet. Am 16.06.2023 um 10:44 schrieb pluto via Wien: Dear All, I just would like to confirm the step-by-step convergence strategy for the large slab with SP and SOC (it refers in general to spin-momentum locked non-magnetic TMDC, but can be any other material). Is the following correct: 1. Converge without SP and without SOC, and save_lapw e.g. as CONV_NO_SP_NO_SOC so it can be used in another directory or on another computer for the next steps 2. Use this as a starting point to converge with SP, and save_lapw as CONV_W_SP_NO_SOC (one can also restore_lapw in another directory and start there) 3. Use this as a starting point to converge with SP and with SOC (and save_lapw to have it for the future) I often start with step 3 right away, but I think for a really large system this might be really inefficient. How does the program know to use the starting density from the previous step? Does restore_lapw creates the necessary files when I transfer to the new directory? Is -NI or some other setting in run_lapw important here? At the moment I am using an older cluster with many cores and use k-parallel. Still didn't manage with MPI, but maybe it is not needed for what I want because my klist file is typically 50-80 k-points, depending on the symmetry of the system. I use the QTL program quite a lot so having it parallellized would sometimes speed the things up a bit for me. Best, Lukasz ___ 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 -- --- Peter Blaha, Inst. f. Materials Chemistry, TU Vienna, A-1060 Vienna Phone: +43-158801165300 Email:
Re: [Wien] Strategy for a large slab with SP and SOC
In addition to what Peter said, pay careful attention to how you setup the slab calculation. WTe2 has a band gap of 1.0 eV, and it may be less with PBE. I suggest checking your bulk band gap first. With a small gap surface, TEMPS is needed and getting the positions right (-min) and a valence neutral surface will matter. You will need to converge density and positions without soc. If you don't setup the surface right GIGO and if will be horribly unstable. (STIFF in case.inm may help.) With GIGO the physics is wrong so your calculations will be meaningless. --- Professor Laurence Marks (Laurie) Department of Materials Science and Engineering Northwestern University www.numis.northwestern.edu "Research is to see what everybody else has seen, and to think what nobody else has thought" Albert Szent-Györgyi On Sat, Jun 17, 2023, 09:01 Peter Blaha wrote: > Your 4 points are not really recommended in the first place. > > > If it is a scf convergence problem (which I doubt): grep:DIS case.scf > . Does it look like divergence ? > > You need to find which eigenvalue causes the ghostband, from which atom > and angular momentum. > > See *scf2* and *output2* files. > > Once you know this, look into case.scf1 to see how the LOs and energy > parameters for this state are set and you probably have to modyfy > case.in1(c). > > PS: Use init_lapw -prec 1n at the beginning, maybe with -ecut 0.999 . > > PS: I would NOT include HDLOs, if I had ghostbands. Mixing with PRATT > helps only in very few cases, not really recommended for "normal" > calculations. > > PPS: I hope you use runsp_c_lapw for something like WTe2 ? > > Am 17.06.2023 um 00:28 schrieb pluto via Wien: > > Dear Prof. Blaha, dear All, > > > > Thank you for the comment on slab strategy, this helps a lot. > > > > I have more specific question: for a large WTe2 slab (60 atoms), which > > is a material of low-symmetry that has a polarity also in the > > out-of-plane direction, I am getting ghostbands in lapw2 after few > > iterations. What is a good strategy to fix this? > > > > I was thinking of: > > > > 1. init_lapw -hdlo > > 2. Low mixing (like 0.05) with PRATT > > 3. Decrease RMT (from first tests, with RMT 2.5 ghostbands seem to > > appear after around 3 iterations, with 2.2 after many iterations) > > 4. Increase RKmax > > > > 3 and 4 are probably computationally expensive... > > > > I did several tests without SOC, I was typically using something like: > > > > init_lapw -sp -b -numk 100 -hdlo -fermit 0.002 > > > > Maybe other settings are critical? > > > > Bulk calculation converges very easily (first without and then with > > SOC) with default settings like > > > > init_lapw -sp -b -numk 2000 > > > > bulk bands look like the literature, and are practically the same with > > RMT 2.2 and RMT 2.5. > > > > Best, > > Lukasz > > > > > > > > > > On 2023-06-16 16:45, Peter Blaha wrote: > >> No,this is not a good strategy. > >> > >> From a converged non-spin-polarized calculation you cannot come > >> (easily) to a spin-polarized solution. > >> > >> So 1) is only good if you want to quote how much more stable a SP > >> solution is compared to a non-SP. > >> > >> 2 + 3 is a good practice. You gain insight how large are the changes > >> and on what atoms due to SO coupling. > >> > >> > >> > >> In terms of efficiency for large cases, I'd in particular preconverge > >> with a course k-mesh and later on refine. > >> > >> --- > >> > >> Every runsp cycle starts with a case.clmsum/up/dn file. > >> > >> These files can come from an initialization, but of course also from > >> any prior scf calculation (eg. with lower k-mesh or without SO). Of > >> course, a restore_lapw ... gives you all files necessary to run > >> another scf cycle. > >> > >> -NI would keep old broyden files, but after a "save_lapw" they are > >> gone anyway. -NI is useful if you want to continue a scf, because eg. > >> the first runsp stopped after 40 cycles and did not reach convergence > >> yet. > >> > >> Am 16.06.2023 um 10:44 schrieb pluto via Wien: > >>> Dear All, > >>> > >>> I just would like to confirm the step-by-step convergence strategy > >>> for the large slab with SP and SOC (it refers in general to > >>> spin-momentum locked non-magnetic TMDC, but can be any other material). > >>> > >>> Is the following correct: > >>> > >>> 1. Converge without SP and without SOC, and save_lapw e.g. as > >>> CONV_NO_SP_NO_SOC so it can be used in another directory or on > >>> another computer for the next steps > >>> 2. Use this as a starting point to converge with SP, and save_lapw > >>> as CONV_W_SP_NO_SOC (one can also restore_lapw in another directory > >>> and start there) > >>> 3. Use this as a starting point to converge with SP and with SOC > >>> (and save_lapw to have it for the future) > >>> > >>> I often start with step 3 right away, but I think for a really large > >>> system this might be really inefficient. > >>> > >>> How does the program
