Dear David, 0.4 is probably large enough such that pseudo-convergence should not happen. Nevertheless, if you want to be sure that it's ok, you can restart the calculation (after having done save_lapw) with the default mixing parameters in case.inm (MSEC1 with 0.2).
F. Tran On Sun, 9 Jan 2011, David wrote: > Professo Blaha, > > Thank you very much for your reply. Since the beginning of my calculation I've > been taking care of the mixing factor as you recommended. Now the calculation > seems to be approaching to convergence. > > However I'm a little concerned now about what you call "pseudo-convergence" > how can I evaluate if my calculation came to a full convergence or a > pseudo-convergence? > > By the way the mixing factor I'm using right now is 0.4 with the PRATT mixing > scheme, is this such a small mixing factor that could lead to > pseudo-convergence? > > Thank you very much for you help Dr. Blaha. > > David Guzman > Department of Physics > California State University - Los Angeles > > On 01/09/2011 01:16 AM, Peter Blaha wrote: > > Convergence of mBJ calculations is sometimes a bit tricky since not only the > > charge density but also the kinitic energy density (or the "vresp"-files) > > determine the potential. > > > > The recommended way is to start out with a PRATT mixing (with eg. 0.2) > > and once the :DIS has settled a bit (decreased from the initially big > > values, but then not changing much, usually after ~10 iterations) one can > > graduadely increase the mixing factor (to eg. 0.3 and 0.4). Maybe > > :DIS increases a little, but as long as there are no oszillations in ENE or > > ::GAP this is fine. > > Eventually, one may intermediately "save" the calculations (to have a > > restarting point) and continue with MSEC1, since once the vresp-files are > > partially converged, density-mixing as done with MSEC1 should be fine. > > Please be aware, that in mBJ-calculations not only the density (clmsum) is > > mixed, but that the mixing-factor in case.inm also mixes the > > exchange-potential in lapw0. Thus, there is a danger of "pseudo-convergence" > > when the mixing factor is too small. > > > > Am 09.01.2011 00:31, schrieb David: > > > Dear Professor Blaha and Wien users, > > > > > > I'm running mBJ calculations for CuZrTiO5, spin-polarized, it has 16 > > > atoms in the unit cell, and im running 300 k-points. My ultimate goal is > > > to get the band structure of this crystal. > > > > > > I haven't got any problems regarding code errors, in fact up to now > > > everything has run very smooth. My question though is the following: the > > > has has run 58 cycles now and the charge distance does not seem to > > > reduce significantly, in fact for the first 45 cycles or so it reduced > > > very slow, but now it's going up again, I was wondering if this is a > > > common behaviour? Also I was wondering if there is any way I can make it > > > converge faster, I know this is a large calculation, but may be some of > > > you guys have suggestions. > > > > > > Some details about my calculation: > > > > > > 1. I run a regular spin-polarized scf cycle using the MSEC1 with a > > > mixing factor of 0.2 and it converged in 18 cycles (convergence criteria > > > was charge: 0.00001 e) > > > > > > 2. I followed the steps for mBJ calculations from the manual, I used > > > PRATT with a mixing factor of 0.2 and the convergence criteria is > > > charge: 0.00001 e > > > > > > 3. Attached are the case.struct and case.inst if it helps > > > > > > Please if you required any other information in order for me to get some > > > help, I can send it > > > > > > Thank you very much > > > > > > David Guzman > > > Department of Physics > > > California State University - Los Angeles > > > > > > > > > > > > _______________________________________________ > > > Wien mailing list > > > Wien at zeus.theochem.tuwien.ac.at > > > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien > > > > _______________________________________________ > Wien mailing list > Wien at zeus.theochem.tuwien.ac.at > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien >
