[Wien] Problems with convergence with LDA+U calculations including spin-orbit coupling

Wed Apr 27 13:33:00 CEST 2011

Dear Profs. Blaha and Marks,

Thanks for your answer, I really appreciated it: taught me a few differences
between LAPW and pseudopotential calculations. The calculation actually
converged after a few more iterations. It looks like the first inclusion of
U is the most problematic - I started from U=1 eV and increased it to 4 eV
in steps of 1 eV, and the subsequent runs took quite fewer steps to
converge.

Thanks very much,

Marcos
Universidad de Cantabria, Spain

On Mon, Apr 25, 2011 at 9:34 PM, Laurence Marks <L-marks at northwestern.edu>wrote:

> As Peter said, do NOT simply reduce the greed, it does not work the
> same as it does in PP calculations.
>
> More to the point, you/we need more information to know what is going
> on. Assuming that you are using the latest version (or similar) then
>
> grep -e :DIR -e GREED -e :FRMS -e :ENE -e :CHARG -e PRATT -e :DIS -e
> "MIXING SC" -e PLANE *.scf  | tail -40
>
> will give more information and what is really going on...
>
> On Mon, Apr 25, 2011 at 2:21 PM, Peter Blaha
> <pblaha at theochem.tuwien.ac.at> wrote:
> > The convergence behaviour is not very good, but it is not too bad either.
> >
> > Check also :MMT or some :MMI0xx
> >
> > For more complicated cases (metallic ??) it is well possible that one
> needs
> > more
> > than 40 iterations. Thus my suggestion is to keep going and submit
> another
> > job.
> >
> > Sometimes it is better to keep the history (use -NI switch), sometimes it
> > can also
> > be better to remove the history (rm *.broy*; which will be done
> > automatically when
> > -NI is not supplied).
> >
> > The mixing parameter has little importance with MSEC1, but still, a too
> > small mixing
> > can lead to too small steps and thus "pseudoconvergence" (this simply
> means,
> > that so
> > little new density is mixed, that eg. the total energy is not changing
> and
> > one thinks
> > the calculation is converged).
> >
> > PRATT is most likely "useless" for your problem.
> >
> > In LDA+U calculations it is easily possible to reach different states
> > (different
> > orbital occupations), which are stable and local minima. Once you have
> > reached
> > convergence, there is never a guarantee that this is the groundstate and
> > really
> > has the lowest energy.
> > The suggested procedure in the UG is NOT ALWAYS leading to the lowest
> state,
> > but often it will do.
> >
> >
> >> I have searched the list but I'm a bit confused with the posts, so I'll
> >> pose my questions here.
> >>
> >> I am running LDA+U with and without spin-orbit coupling for a structure
> in
> >> a cell with the following cell parameters (Bohr):
> >>
> >> P   LATTICE,NONEQUIV.ATOMS: 14127_P4/mbm
> >> MODE OF CALC=RELA
> >>  10.353171 10.353171 43.972470 90.000000 90.000000 90.000000
> >>
> >> and a cell sampling of 12x12x3, which gives 432 points in the BZ, 42 in
> >> the IBZ. My RKMAX is 8 - it was necessary to go to such high values to
> >> converge the spin moment of
> >> the system, as well as the energy down to 1 mRy. I followed the
> procedure
> >> stated in the manual: started with the LSDA calculation, then included
> >> spin-orbit coupling and now
> >> I am including the U gradually (well, at this point, just trying to...).
> I
> >> have to get to U=4 eV, so I'm growing it in steps of 1 eV.
> >>
> >> Convergence was very good for the LSDA+U calculations without SOC, and
> >> they have finished, already (that is, I have already gotten to the 4 eV
> in
> >> U). However, I am having
> >> problems converging the scf, with the charge and energy distances (as
> >> obtained by grep convergence case.dayfile) oscillating, and never
> attaining
> >> convergence, which I set
> >> to ec=0.00001 and cc=0.0001 (the numbers are appended by the end of this
> >> message). More than 40 SCF steps have passed.
> >>
> >> Usually, in pseudopotential calculations, one possible solution is to
> >> decrease the mixing weight of the charge densities for the following SCF
> >> step, and the other is to
> >> increase the smearing temperature. I would like to tweak with the mixing
> >> rather than increasing the temperature, if possible, because from my
> >> experience, increasing the
> >> temperature can sometimes lead to a non-magnetic state. However, I see
> >> from the mailing list that (at least in LAPW calculations), too small a
> >> mixing can lead to
> >> pseudo-convergence - which I don't understand what is. So, my questions
> >> are the following:
> >>
> >> 1) What is the best thing to do to make my calculation converge:
> increase
> >> the mixing, or to diminish it? Should I change the mixing for MSEC1, and
> if
> >> so, by which amount?
> >> Or, would it be good to change to PRATT, instead? I am quite lost with
> >> regards to that.
> >> 2) Would the application of SOC to the already converged LSDA+U
> >> calculation yield too different results from starting from a converged
> >> LSDA+SOC with later inclusion of U?
> >> 3) What is the pseudo-convergence mentioned in previous emails in the
> >> list?
> >>
> >> Of course, I will gladly provide more information, if needed.
> >>
> >> Best regards,
> >>
> >> Marcos
> >> Universidad de Cantabria, Spain
> >>
> >> =======================================
> >>
> >> Convergence of energy and charge distances:
> >>
> >> :ENERGY convergence:  0 0.00001 .0002941150000000
> >> :CHARGE convergence:  0 0.0001 .0088826
> >> :ENERGY convergence:  0 0.00001 .0000916950000000
> >> :CHARGE convergence:  0 0.0001 .0085946
> >> :ENERGY convergence:  0 0.00001 .0000428400000000
> >> :CHARGE convergence:  0 0.0001 .0112915
> >> :ENERGY convergence:  0 0.00001 .0001772500000000
> >> :CHARGE convergence:  0 0.0001 .0113780
> >> :ENERGY convergence:  0 0.00001 .0002456850000000
> >> :CHARGE convergence:  0 0.0001 .0104415
> >> :ENERGY convergence:  0 0.00001 .0001420300000000
> >> :CHARGE convergence:  0 0.0001 .0083852
> >> :ENERGY convergence:  0 0.00001 .0000500150000000
> >> :CHARGE convergence:  0 0.0001 .0084675
> >> :ENERGY convergence:  0 0.00001 .0001809700000000
> >> :CHARGE convergence:  0 0.0001 .0134776
> >> :ENERGY convergence:  0 0.00001 .0001993400000000
> >> :CHARGE convergence:  0 0.0001 .0112308
> >> :ENERGY convergence:  0 0.00001 .0002392700000000
> >> :CHARGE convergence:  0 0.0001 .0104208
> >> :ENERGY convergence:  0 0.00001 .0002088150000000
> >> :CHARGE convergence:  0 0.0001 .0069138
> >> :ENERGY convergence:  0 0.00001 .0001630050000000
> >> :CHARGE convergence:  0 0.0001 .0109876
> >> :ENERGY convergence:  0 0.00001 .0002176700000000
> >> :CHARGE convergence:  0 0.0001 .0012609
> >> :ENERGY convergence:  0 0.00001 .0001806300000000
> >> :CHARGE convergence:  0 0.0001 .0042825
> >> :ENERGY convergence:  0 0.00001 .0000720450000000
> >> :CHARGE convergence:  0 0.0001 .0045301
> >> :ENERGY convergence:  0 0.00001 .0000129050000000
> >> :CHARGE convergence:  0 0.0001 .0011304
> >> :ENERGY convergence:  0 0.00001 .0002283150000000
> >> :CHARGE convergence:  0 0.0001 .0047456
> >> :ENERGY convergence:  0 0.00001 .0003098100000000
> >> :CHARGE convergence:  0 0.0001 .0100820
> >> :ENERGY convergence:  0 0.00001 .0003708750000000
> >> :CHARGE convergence:  0 0.0001 .0087276
> >> :ENERGY convergence:  0 0.00001 .0000788950000000
> >> :CHARGE convergence:  0 0.0001 .0090876
> >> :ENERGY convergence:  0 0.00001 .0000849050000000
> >> :CHARGE convergence:  0 0.0001 .0093747
> >> :ENERGY convergence:  0 0.00001 .0000842350000000
> >> :CHARGE convergence:  0 0.0001 .0079561
> >> :ENERGY convergence:  0 0.00001 .0000553500000000
> >> :CHARGE convergence:  0 0.0001 .0122027
> >> :ENERGY convergence:  0 0.00001 .0000434050000000
> >> :CHARGE convergence:  0 0.0001 .0109352
> >> :ENERGY convergence:  0 0.00001 .0000227750000000
> >> :CHARGE convergence:  0 0.0001 .0122136
> >> :ENERGY convergence:  0 0.00001 .0000194300000000
> >> :CHARGE convergence:  0 0.0001 .0121261
> >> :ENERGY convergence:  0 0.00001 .0000292700000000
> >> :CHARGE convergence:  0 0.0001 .0131197
> >> :ENERGY convergence:  0 0.00001 .0000360650000000
> >> :CHARGE convergence:  0 0.0001 .0110031
> >> :ENERGY convergence:  0 0.00001 .0000300950000000
> >> :CHARGE convergence:  0 0.0001 .0115123
> >> :ENERGY convergence:  0 0.00001 .0000323600000000
> >> :CHARGE convergence:  0 0.0001 .0099887
> >> :ENERGY convergence:  0 0.00001 .0000147550000000
> >> :CHARGE convergence:  0 0.0001 .0110331
> >> :ENERGY convergence:  0 0.00001 .0001196200000000
> >> :CHARGE convergence:  0 0.0001 .0005253
> >>
> >>
> >>
> >> _______________________________________________
> >> Wien mailing list
> >> Wien at zeus.theochem.tuwien.ac.at
> >> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> >
> > --
> > -----------------------------------------
> > Peter Blaha
> > Inst. Materials Chemistry, TU Vienna
> > Getreidemarkt 9, A-1060 Vienna, Austria
> > Tel: +43-1-5880115671
> > Fax: +43-1-5880115698
> > email: pblaha at theochem.tuwien.ac.at
> > -----------------------------------------
> > _______________________________________________
> > Wien mailing list
> > Wien at zeus.theochem.tuwien.ac.at
> > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> >
>
>
>
> --
> Laurence Marks
> Department of Materials Science and Engineering
> MSE Rm 2036 Cook Hall
> 2220 N Campus Drive
> Northwestern University
> Evanston, IL 60208, USA
> Tel: (847) 491-3996 Fax: (847) 491-7820
> email: L-marks at northwestern dot edu
> Web: www.numis.northwestern.edu
> Chair, Commission on Electron Crystallography of IUCR
> www.numis.northwestern.edu/
> Research is to see what everybody else has seen, and to think what
> nobody else has thought
> Albert Szent-Gyorgi
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20110427/018a2067/attachment.htm>