Dear Andrei, Thank you very much for the detailed answer.
Indeed, I discussed “total spin polarization”, which, as I assumed, should be mainly independent throughout the calculation method for the given system. Moreover, since in my system (Bi2Te3+Co) the spin polarization comes mainly from the Co atom, then I may conclude that the total spin polarization of the system corresponds to the magnetization of the Co atom. Luckily, Bi2Te3 system, itself, possess zero total spin polarization at all tested parameters (pseudopotentials, basis). However, the “opposite” values of magnetization for the Co atom are confused, i.e. either the Co atom have a strong magnetic moment (4μB), or does not have a magnetic moment at all (0μB). Concerning the convergence problem. I don't think this is critical one in my case. Basically, I have no problem with convergence and all calculations were properly converged (10^-3 or 10^-4 for DM and H). Moreover, latest version of Siesta (from “master brunch”) allows you to monitor the total spin polarization value during SCF cycles. So, as I noticed, the total spin polarization value converges very quickly to some value (for example: 4, 3 or 2) within 20-30 SCF cycles and then it almost does not change (only within first or second decimal place) until the end of the calculations. Thus, k-mesh, mixing schema and weights, electronic temperature etc. do not affect at the final result if the convergence is achieved. I have also tried more hard convergence criteria (like 10^-5 or 10^-6 for DM and H), but this is just a waste of time, because the main parameters of the system already almost not change after 10^-3. I’m grateful for the idea that it is necessary to compare the Siesta results against an all-electron method. Perhaps, it really can provide some kind of reference point for magnetism in my system. I’m not sure that all-electron method will be able to calculate such large system, but most likely a smaller system may be sufficient for estimation of magnetism in such system. Thank you one more time. With kind regards, Dmitriy Muzychenko From: Andrei Postnikov [SIESTA-L] Puzzle in Spin Polarized calculation in Siesta > Dear Dmitriy, > > as I understand, you are talking about total spin polarisation, which should > normally > be reproducible throughout the methods. Otherwise, in terms of local spin > moments, > expressed via Mulliken charges or whatever, the differences in the > definition > between methods (or, between basis sets) may substantially add to confusion. > > > This said, I'd suggest you to try to separate, as possible sources of worry, > the "local" issues (pseudopotentials, basis) from "global" ones (convergence > in a large system with – probably – narrow bands). Your P.S. gives an > impression that > you are trying everything possible in different directions at the same time. > For setting the "local" issues, I'd suggest a reasonable compact case, > including > your cobalt or whatever, to check against an all–electron method. Here > you'd check > the magnetism or whatever is essential for your problem to the accuracy > needed, > settle the XC etc., and then do not touch these issues anymore. > > This settled, the instability of results may come from problematic > convergence, > as you have a large system and probably narrow bands which swap back and > forth > across the Fermi level. If you have no gap but a metal, then in a large > system > you need to be very careful about the dense enough k-mesh, mixing schema > and weights, electronic temperature etc., to see that your system really > converges > and not fluctuates. > > Good luck, best regards > > Andrei > > -- prof. Andrei Postnikov -- > University of Lorraine - Laboratoire de Chimie/Physique - A2MC > ICPM, 1 Bd Arago - BP 95823, F-57078 Metz Cedex 03, France > ------------------------------------------------------------------------ > > ----- Dmitriy Muzychenko <d...@spmlab.phys.msu.su> a écrit : > > Dear Siesta Users and Developers > > > > I’m really confused by the results of the Spin Polarized calculation in > Siesta. > > > > My system is Bi2Te3 slab with single Co atom substitutions of a Bi atom > (typically concentration: 1 Co atom per 490-1100 Bi+Te atoms). > > > > The confusion arises when I start to analyze the spin properties of this > system. Depending on the pseudopotential or/and basis set I observe > completely different value of total spin polarization (S_up-S_down). For > example, numerous calculations have resulted in the following list of total > spin polarization values: 4.33, 4.19, 4.00, 3.99, 3,77, 3.52, 3.00, 2,99, > 2,90, 2.63, 2.59, 2.00, 1.99, 1.53 and 0.00. I note, that these values are > for exactly the same system, just calculated with a different pseudo(s) or > basis set(s). > > > > Such a wide range of total spin (from 4.33 up to 0.00) raise a logical > question, what value can I trust and why? The question is rather not in > asking for some “perfect set” of parameters, but in what the general > strategy(rule) for analyzing the results should I choose in order to filter > out clearly artifact results and then focused on the most reliable one? > > > > I would be grateful for any help, advise or some “rule” that will > allow to filter out incorrect results. > > > > Thank you in advance for any advice. > > > > With kind regards, > > Dmitriy Muzychenko > > > > > > > > P.S. I believe that it is useless to attach here any pseudo or basis that > I used, since I have tried a lot of possible combinations of pseudo and > basis (pseudo and basis: from paper [Comp. Mat. Sci. 98 (2015) 372]; from > Siesta Web-Database; my own generated/tested pseudo and basis). I have also > allowed Siesta prog to choose basis automatically, as well as used > “gen-basis” utility for basis or choose basis manually... Each attempt > gives some new value of total spin polarization (perhaps, it can only be > noted that the most common numbers are: 4.00, 2.00 and 0.00). Orbital > population analysis shows that in all cases polarization is comes mainly > from d-shell of Co atom, with the minor extra polarization of surrounding > atoms. I have used GGA PBE approximation with Spin Polarization as well as > with SOC with the similar results. For the Co species I have tried also > semicore (3s2 3p6 3d7 4s2) configuration with close to the same total spin > variation from S=4.33 to 0.00 depending on basis set or parameters used for > pseudopotential generation. Version of Siesta that I used is 4.1.b3/b4 or > latest code from “master brunch”. >
-- SIESTA is supported by the Spanish Research Agency (AEI) and by the European H2020 MaX Centre of Excellence (http://www.max-centre.eu/)
