[Wien] :FCHECK in Wien2k 13.1 vs. 14
Peter Blaha
pblaha at theochem.tuwien.ac.at
Mon Nov 30 17:59:50 CET 2015
The problem of :FCHECK has two sides:
If it is big for a well converged calculation, it means that the RMTs
are not well chosen.
If it gets too big during MSR1a, it means that the minimization might
advance badly since the forces are too wrong, or it could also mean that
minimization moves on very fast and efficiently, but you got off the
Born-Oppenheimer surface a bit. So in essence sometimes even big :FCHECK
can minimize efficiently.
The difference between 13.1 and 14.2 are due to two reasons, a) a
"push", which I added (mixing of 0.2 for a well converged case) and b) a
new version of mixer.
While I still think that overall, both, the "push" AND the "new mixer"
are better, there are always counter examples and maybe the new scheme
is a little bit too greedy for your particular case. If 13.1 works well,
it should be ok to use this mixer.
However, you should now check if the final forces with 13.1 are really
small (so that it was "justified" that the mixer switched from MSR1a to
MSR1 and stopped optimizing the positions (you probably need a few more
normal scf iterations to get a lower :DIS and better converged forces to
get a convincing answer) - otherwise 13.1 stopped to early and was
"stagnating";
or if eventually due to this push there will be some further movement of
atoms and therefore convergence will take longer.
On the other hand, if you are saying that 14.2 does not converge after
100s of cycles, but 13.1 does ...?
Then stay for the moment with 13.1
Peter
On 11/30/2015 04:40 PM, Laurence Marks wrote:
> Yes....and no.
>
> In 14.2 Peter added a "Push" for the 1st iteration if :DIS is small,
> some lines starting with
> #check :dis of first cycle and rerun mixer with increased mixing
> after the execution of mixer
>
> This was added to avoid stagnation where it just takes a few small steps
> and stops.
>
> In your case this "disturbed" 14.2 and it is adjusting the greed down to
> try and stabilize. In 13.1 this push is missing, so it starts with a
> small greed and finds what it thinks might be a solution.
>
> The next version does not have this push (it is done inside mixer), but
> still needs work.
>
> Unfortunately both cases are really bad -- :FCheck should be close to
> 1.0 or less, and your values indicate something really wrong. For
> reasons I don't understand MSR1a does not behave well with non-centro
> symmetric structures. I don't know why this is.
>
> N.B., MSR1a is not perfect, and there are cases where it does not want
> to behave well. Whenever the model is susceptible to ghostbands it is
> not ideal. I am currently trying a few things, but.....
>
> N.N.B., Peter's explanation of why :FCHECK can be large is that the
> linearization of different atoms is unbalanced. Maybe he has ideas about
> how to cure this, it is beyond me.
>
>
> On Mon, Nov 30, 2015 at 9:00 AM, Elias Assmann <elias.assmann at gmail.com
> <mailto:elias.assmann at gmail.com>> wrote:
>
> Hi List,
>
> I reply to this thread
> <http://www.mail-archive.com/wien%40zeus.theochem.tuwien.ac.at/msg13276.html>.
> This is still about the same set of oxide calculations.
>
> Laurence Marks alerted me that :FCHECK (sum of forces) was large (see
> below). This is the same over a set of similar calculations (somewhat
> different structures, GGA+U vs. plain GGA), and does not seem to get
> better even after hundreds of calculations.
>
> I have been using the latest Wien2k (14.2), but I did similar things
> before with 13.1 and I did not remember having this problem, at least
> to this extent. So I compiled 13.1 (same options as 14.2) and
> switched one of my calculations to it for 10 iterations.
>
> Indeed, it seems that 13.1 is more “well-behaved” in this case than
> 14.2 — in any case they are different. Below, I include some output
> of the two versions, starting from the same charge density in both
> cases.
>
> As you can see, the two versions are the more or less the same in the
> first iteration, but diverge from there; and 13.1 seems to converge,
> in contrast to 14.2. Also, 13.1 seems to be already satisfied with
> the structure (it switches MSR1a off), while 14.2 continues
> minimizing. The greed (on whose smallness Prof. Marks also commented)
> is pretty similar.
>
> To me, that suggests that I should go on with these calculations using
> the older Wien. Is there any reason that doing so might be “unsafe”?
>
>
> Elias
>
>
> 14.2 :FCHECK: 11.008001043 12.164812176 -14.416571844
> :FCHECK: 249.814119002 264.664202568 -694.164686632
> :FCHECK: 97.292437031 103.207655853 -256.561081883
> :FCHECK: 126.184534662 136.666423509 -39.224005265
> :FCHECK: 176.349427337 188.815368048 36.914398561
> :FCHECK: 50.400659984 67.046955772 76.978472788
> :FCHECK: -74.640170442 25.157628216 121.498217602
> :FCHECK: -82.337983720 28.388321933 121.952351989
> :FCHECK: -130.382154150 24.859863620 204.399962720
> :FCHECK: -39.381668995 -31.379111345 91.507308943
>
> 13.1 :FCHECK: 11.009473749 12.153718710 -14.482142314
> :FCHECK: 10.707248409 11.500977675 -16.514184574
> :FCHECK: 10.289877753 10.742969135 -10.026151044
> :FCHECK: 7.566779454 6.128096382 -3.725471547
> :FCHECK: 3.572302068 1.087140350 6.011501945
> :FCHECK: -4.275691705 -7.052038873 31.588782325
> :FCHECK: -6.385420320 -8.403344519 33.874614737
> :FCHECK: -6.106651082 -8.096761064 35.566398852
> :FCHECK: -5.627436387 -7.565248768 32.843232377
> :FCHECK: -3.074299958 -4.674255384 27.987425362
>
> 14.2 :ENE : *WARNING** TOTAL ENERGY -222784.46215529
> :ENE : *WARNING** TOTAL ENERGY -222784.46926026
> :ENE : *WARNING** TOTAL ENERGY -222784.46462435
> :ENE : *WARNING** TOTAL ENERGY -222784.46263442
> :ENE : *WARNING** TOTAL ENERGY -222784.46223935
> :ENE : *WARNING** TOTAL ENERGY -222784.46214440
> :ENE : *WARNING** TOTAL ENERGY -222784.46218336
> :ENE : *WARNING** TOTAL ENERGY -222784.46213858
> :ENE : *WARNING** TOTAL ENERGY -222784.46236760
> :ENE : *WARNING** TOTAL ENERGY -222784.46181283
>
> 13.1 :ENE : ********** TOTAL ENERGY -222784.46212826
> :ENE : ********** TOTAL ENERGY -222784.46212374
> :ENE : ********** TOTAL ENERGY -222784.46212843
> :ENE : ********** TOTAL ENERGY -222784.46207306
> :ENE : ********** TOTAL ENERGY -222784.46209495
> :ENE : ********** TOTAL ENERGY -222784.46204241
> :ENE : ********** TOTAL ENERGY -222784.46199458
> :ENE : ********** TOTAL ENERGY -222784.46200681
> :ENE : ********** TOTAL ENERGY -222784.46200118
> :ENE : ********** TOTAL ENERGY -222784.46201120
>
> (It appears that the warnings come from the forces. As mentioned
> before there is not :WAR tag.)
>
> 14.2 :DIS : ( 0.0048283 for atom 7 spin 1) 0.0015628
> :DIS : ( 0.1175699 for atom 5 spin 1) 0.0137938
> :DIS : ( 0.0692919 for atom 5 spin 1) 0.0083222
> :DIS : ( 0.0098508 for atom 7 spin 1) 0.0017036
> :DIS : ( 0.0167530 for atom 5 spin 1) 0.0022822
> :DIS : ( 0.0278296 for atom 7 spin 1) 0.0038446
> :DIS : ( 0.0366785 for atom 5 spin 1) 0.0048138
> :DIS : ( 0.0371346 for atom 5 spin 1) 0.0047958
> :DIS : ( 0.0559135 for atom 5 spin 1) 0.0068198
> :DIS : ( 0.0157960 for atom 5 spin 1) 0.0035943
>
> 13.1 :DIS : ( 0.0048362 for atom 7 spin 1) 0.0015593
> :DIS : ( 0.0022134 for atom 12 spin 2) 0.0012247
> :DIS : ( 0.0052633 for atom 5 spin 1) 0.0014252
> :DIS : ( 0.0015732 for atom 5 spin 1) 0.0007705
> :DIS : ( 0.0046946 for atom 7 spin 1) 0.0010640
> :DIS : ( 0.0018880 for atom 8 spin 2) 0.0011590
> :DIS : ( 0.0017283 for atom 5 spin 1) 0.0008169
> :DIS : ( 0.0009736 for atom 8 spin 2) 0.0006944
> :DIS : ( 0.0018888 for atom 5 spin 1) 0.0007013
> :DIS : ( 0.0009991 for atom 7 spin 1) 0.0003925
>
> 14.2 :MIX : PRATT REG: 1.00E-06 GREED: 0.200
> :MIX : PRATT REG: 1.00E-06 GREED: 0.200 Reduce 0.25 0.20
> :MIX : MSE1a REG: 1.88E-06 GREED: 0.120 Newton 1.00 0.01
> :MIX : MSE1a REG: 2.17E-06 GREED: 0.083 Newton 1.00 0.01
> :MIX : MSE1a REG: 3.19E-06 GREED: 0.050 Newton 1.00 0.01
> :MIX : MSE1a REG: 4.08E-06 GREED: 0.034 Newton 1.00 0.03
> :MIX : MSE1a REG: 5.17E-06 GREED: 0.022 Newton 1.00 0.00
> :MIX : MSE1a REG: 5.20E-06 GREED: 0.021 Newton 1.00 0.06
> :MIX : MSE1a REG: 7.51E-06 GREED: 0.018 Newton 1.00 0.14
> :MIX : MSE1a REG: 4.52E-06 GREED: 0.017 Newton 1.00 0.10
>
> 13.1 :MIX : PRATT REG: 1.00E-06 GREED: 0.025
> :MIX : MSR1a REG: 1.00E-06 GREED: 0.030 Newton 1.00 0.03
> :MIX : MSE1a REG: 1.59E-06 GREED: 0.030 Newton 1.00 0.18
> :MIX : MSE1a REG: 2.61E-06 GREED: 0.042 Newton 1.00 0.29
> :MIX : MSE1a REG: 3.72E-06 GREED: 0.057 Newton 1.00 0.54
> :MIX : MSE1a REG: 3.58E-06 GREED: 0.093 Newton 1.00 0.46
> :MIX : PRATT REG: 1.00E-06 GREED: 0.025
> :MIX : MSR1 REG: 1.00E-06 GREED: 0.050 Newton 1.00 0.05
> :MIX : MSE1 REG: 1.60E-06 GREED: 0.050 Newton 1.00 0.22
> :MIX : MSE1 REG: 2.41E-06 GREED: 0.055 Newton 1.00 0.40
>
> On 11/03/2015 02:47 PM, Laurence Marks wrote:
> > I am not sure exactly what the problem(s) were, but from you
> case.scf it
> > was "right" that there were Warnings
> >
> > a) :FCHECK (bottom of case.scf) was large. This is the sum of all the
> > forces, and should be small. Particularly for cells without inversion
> > one can get bad, highly asymmetric densities in which case MSR1a can
> > have problems.
> >
> > b) The greed is small. Too small a value can be as bad as too
> large. I
> > have struggled with this for years and failed to find a strong ansatz
> > for this, although I believe the next release of the mixer will
> be better.
>
>
>
> --
> Elias Assmann
> Institute of Theoretical and Computational Physics
> TU Graz ⟨https://itp.tugraz.at/⟩
>
>
>
>
>
> --
> Professor Laurence Marks
> Department of Materials Science and Engineering
> Northwestern University
> www.numis.northwestern.edu <http://www.numis.northwestern.edu>
> Corrosion in 4D: MURI4D.numis.northwestern.edu
> <http://MURI4D.numis.northwestern.edu>
> Co-Editor, Acta Cryst A
> "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
> SEARCH the MAILING-LIST at: http://www.mail-archive.com/wien@zeus.theochem.tuwien.ac.at/index.html
>
--
P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-165300 FAX: +43-1-58801-165982
Email: blaha at theochem.tuwien.ac.at WIEN2k: http://www.wien2k.at
WWW: http://www.imc.tuwien.ac.at/staff/tc_group_e.php
--------------------------------------------------------------------------
More information about the Wien
mailing list