[DFTB-Plus-User] Unreasonably large number of steps in simultaneous geometry and cell optimization
Alexander Maryewski
A.Maryewski at skoltech.ru
Fri Dec 10 08:17:14 CET 2021
Dear Bálint Aradi,
Thank you for your suggestion! I tried the new relaxation driver after it was merged on master branch, and it certainly outperforms the old one by at least an order of magnitude in terms of relaxation steps.
Looks like the problem is no more. Again, I'm grateful to you for your assistance.
Best regards,
Alexander A. Maryewski
________________________________
From: DFTB-Plus-User <dftb-plus-user-bounces at mailman.zfn.uni-bremen.de> on behalf of Bálint Aradi <aradi at uni-bremen.de>
Sent: Sunday, December 5, 2021 11:40:56 PM
To: dftb-plus-user at mailman.zfn.uni-bremen.de
Subject: Re: [DFTB-Plus-User] Unreasonably large number of steps in simultaneous geometry and cell optimization
Dear Alexander Maryewski,
You are absolutely right, the current geometry optimizers of DFTB+ have
often very poor performance when lattice optimization is involved.
We are working on the problem, and hope to merge a pull request with
significantly improved geometry optimisers within the next few days.
Using the new rational function based driver (Rational), I get a
convergence for lattice and atom positions in only 174 steps for your
system. As you are using a development version already, you may be
interested to try it yourself:
https://github.com/dftbplus/dftbplus/pull/862
Some other notes to your input:
- You use the damped electrostatics, specify Hubbard U derivatives, but
did not switch on the third order correction (ThirdOrderFull = Yes). I
guess, that's just by mistake. (If you had not switched off the parser
errors with IgnordedUnprocessedNodes = Yes, you would have obtained an
error message about the superfluous HubbardDerivatives block...)
- The SCC tolerance of 1e-8 is really tight. You can speed up the
calculation by choosing the default value instead. That should still
give forces accurate enough to get a proper geometry convergence and
could spare you a few scc cycles in each step.
Best regards,
Bálint
On 04.12.21 23:11, Alexander Maryewski wrote:
> Dear DFTB+ users and developers,
>
>
> I'm experiencing a problem with unreasonably slow convergence for
> simultaneous geometry & cell optimization. The initial structure for
> optimization (organic molecular crystal) is obtained from PBE relaxation
> with MBD correction in FHI-aims, so it is supposed to be in proximity of
> DFTB minimum; the system itself is quite simple, with no flexible
> dihedral angles. The final structure is essentially identical to the
> initial one, the volume change is very small. However, the relaxation
> takes 3448 geometry steps and 91 lattice steps, which is, in my opinion,
> too much for such negligible change.
>
>
> I'm attaching the initial and final structures, as well as the input
> specification. I must note that changing the relaxation driver to BFGS
> or CG led to even bigger number of steps, while increasing the max step
> size had virtually no effect on number of steps. The problem also
> persists for all other types of dispersion corrections (D3, D4, and
> MBD), as well as for DFTB1, DFTB2, and xTB Hamiltonian. I'm using a
> development version of DFTB+ (commit 65b0275).
>
>
> Thank you in advance for you help!
>
>
> Input:
>
> Geometry = VaspFormat {
> <<< "POSCAR"
> }
>
> Driver = FIRE {
> MovedAtoms = 1:-1
> MaxForceComponent = 1e-4
> LatticeOpt = Yes
> MaxSteps = -1
> MaxAtomStep = 0.94486
> MaxLatticeStep = 0.94486
> }
>
>
> Hamiltonian = DFTB {
> SCC = Yes
> SCCTolerance = 1e-08
> MaxSCCIterations = 2000
> SlaterKosterFiles = Type2FileNames{
> Prefix = /home/amarjewski/Downloads/3ob-3-1/
> Separator = "-"
> Suffix = ".skf"
> }
> HCorrection = Damping {
> Exponent = 4.2
> }
>
>
> HubbardDerivs {
> C = -0.1492
> S = -0.11
> N = -0.1535
> }
> MaxAngularMomentum = {
> C = p
> S = d
> N = p
> }
> KPointsAndWeights = SupercellFolding {
> 2. 0. 0.
> 0. 2. 0.
> 0. 0. 1.
> 0.5 0.5 0.5
> }
> Dispersion = TS {
> RangeSeparation = 1.05
> }
> }
>
> Options {
> WriteResultsTag = Yes
> TimingVerbosity = -1
> }
> ParserOptions {
> IgnoreUnprocessedNodes = Yes
> }
>
>
> Initial structure:
>
> EA0
> 1.0000000000000000
> 0.000000 0.000000 -7.055000
> -0.000000 -10.130900 -0.000000
> -13.452700 0.000000 -0.000000
> C N S
> 32 16 12
> Direct
> 0.020520 0.692620 0.600950
> 0.042030 0.587050 0.180350
> 0.542030 0.412950 0.819650
> 0.042030 0.912950 0.680350
> 0.030900 0.137560 0.263670
> 0.530900 0.862440 0.736330
> 0.530900 0.637560 0.236330
> 0.030900 0.362440 0.763670
> 0.928500 0.241210 0.079410
> 0.428500 0.758790 0.920590
> 0.428500 0.741210 0.420590
> 0.928500 0.258790 0.579410
> 0.997970 0.671140 0.100580
> 0.497970 0.328860 0.899420
> 0.497970 0.171140 0.399420
> 0.997970 0.828860 0.600580
> 0.542030 0.087050 0.319650
> 0.531340 0.943920 0.822090
> 0.031340 0.056080 0.177910
> 0.520520 0.307380 0.399050
> 0.020520 0.807380 0.100950
> 0.972070 0.640130 0.506290
> 0.472070 0.359870 0.493710
> 0.472070 0.140130 0.993710
> 0.520520 0.192620 0.899050
> 0.983790 0.392900 0.586900
> 0.483790 0.607100 0.413100
> 0.483790 0.892900 0.913100
> 0.983790 0.107100 0.086900
> 0.031340 0.443920 0.677910
> 0.972070 0.859870 0.006290
> 0.531340 0.556080 0.322090
> 0.913600 0.774530 0.938670
> 0.884360 0.148230 0.570310
> 0.384360 0.851770 0.429690
> 0.384360 0.648230 0.929690
> 0.884360 0.351770 0.070310
> 0.534960 0.798690 0.663870
> 0.534960 0.701310 0.163870
> 0.034960 0.201310 0.336130
> 0.085280 0.976920 0.749030
> 0.413600 0.225470 0.061330
> 0.585280 0.023080 0.250970
> 0.034960 0.298690 0.836130
> 0.413600 0.274530 0.561330
> 0.085280 0.523080 0.249030
> 0.585280 0.476920 0.750970
> 0.913600 0.725470 0.438670
> 0.990480 0.028080 0.969370
> 0.490480 0.971920 0.030630
> 0.490480 0.528080 0.530630
> 0.990480 0.471920 0.469370
> 0.916110 0.621430 0.983970
> 0.416110 0.121430 0.516030
> 0.916110 0.878570 0.483970
> 0.101330 0.893130 0.206480
> 0.601330 0.106870 0.793520
> 0.601330 0.393130 0.293520
> 0.101330 0.606870 0.706480
> 0.416110 0.378570 0.016030
>
> Final structure:
>
> 60 F
> C N S
> 1 1 0.1618820919E-01 0.6953035972E+00 0.6028292686E+00
> 2 1 0.6054080320E-01 0.5816253594E+00 0.1889168374E+00
> 3 1 0.5605408032E+00 0.4183746406E+00 0.8110831626E+00
> 4 1 0.6054080212E-01 0.9183746373E+00 0.6889168379E+00
> 5 1 0.3894586478E-01 0.1463696753E+00 0.2863807959E+00
> 6 1 0.5389458648E+00 0.8536303247E+00 0.7136192041E+00
> 7 1 0.5389458626E+00 0.6463696788E+00 0.2136192044E+00
> 8 1 0.3894586258E-01 0.3536303212E+00 0.7863807956E+00
> 9 1 0.9194815724E+00 0.2560876082E+00 0.9025739993E-01
> 10 1 0.4194815724E+00 0.7439123918E+00 0.9097426001E+00
> 11 1 0.4194815742E+00 0.7560876117E+00 0.4097426009E+00
> 12 1 0.9194815742E+00 0.2439123883E+00 0.5902573991E+00
> 13 1 0.1005024217E+01 0.6667714686E+00 0.1023961561E+00
> 14 1 0.5050242166E+00 0.3332285314E+00 0.8976038439E+00
> 15 1 0.5050242175E+00 0.1667714722E+00 0.3976038438E+00
> 16 1 0.1005024218E+01 0.8332285278E+00 0.6023961562E+00
> 17 1 0.5605408021E+00 0.8162536268E-01 0.3110831621E+00
> 18 1 0.5307570847E+00 0.9362083091E+00 0.8076337958E+00
> 19 1 0.3075708471E-01 0.6379169088E-01 0.1923662042E+00
> 20 1 0.5161882092E+00 0.3046964028E+00 0.3971707314E+00
> 21 1 0.1618820762E-01 0.8046963992E+00 0.1028292686E+00
> 22 1 0.9622657415E+00 0.6386852028E+00 0.5001196801E+00
> 23 1 0.4622657415E+00 0.3613147972E+00 0.4998803199E+00
> 24 1 0.4622657378E+00 0.1386852066E+00 0.9998803195E+00
> 25 1 0.5161882076E+00 0.1953036008E+00 0.8971707314E+00
> 26 1 0.9767711350E+00 0.3817261045E+00 0.5960237041E+00
> 27 1 0.4767711350E+00 0.6182738955E+00 0.4039762959E+00
> 28 1 0.4767711330E+00 0.8817261080E+00 0.9039762953E+00
> 29 1 0.9767711330E+00 0.1182738920E+00 0.9602370469E-01
> 30 1 0.3075708476E-01 0.4362083057E+00 0.6923662040E+00
> 31 1 0.9622657378E+00 0.8613147934E+00 0.1196804560E-03
> 32 1 0.5307570848E+00 0.5637916943E+00 0.3076337960E+00
> 33 2 0.9108197374E+00 0.7815481119E+00 0.9219522648E+00
> 34 2 0.8628543644E+00 0.1347135404E+00 0.5859737071E+00
> 35 2 0.3628543644E+00 0.8652864596E+00 0.4140262929E+00
> 36 2 0.3628543630E+00 0.6347135438E+00 0.9140262920E+00
> 37 2 0.8628543630E+00 0.3652864562E+00 0.8597370800E-01
> 38 2 0.5489472848E+00 0.7895246992E+00 0.6367080419E+00
> 39 2 0.5489472805E+00 0.7104753046E+00 0.1367080424E+00
> 40 2 0.4894728477E-01 0.2104753008E+00 0.3632919581E+00
> 41 2 0.1132682966E+00 0.9872521695E+00 0.7577586812E+00
> 42 2 0.4108197374E+00 0.2184518881E+00 0.7804773523E-01
> 43 2 0.6132682966E+00 0.1274783045E-01 0.2422413188E+00
> 44 2 0.4894728050E-01 0.2895246954E+00 0.8632919576E+00
> 45 2 0.4108197421E+00 0.2815481158E+00 0.5780477357E+00
> 46 2 0.1132682994E+00 0.5127478277E+00 0.2577586806E+00
> 47 2 0.6132682994E+00 0.4872521723E+00 0.7422413194E+00
> 48 2 0.9108197421E+00 0.7184518842E+00 0.4219522643E+00
> 49 3 0.9682605312E+00 0.3602122790E-01 0.9694575748E+00
> 50 3 0.4682605312E+00 0.9639787721E+00 0.3054242521E-01
> 51 3 0.4682605362E+00 0.5360212317E+00 0.5305424258E+00
> 52 3 0.9682605362E+00 0.4639787683E+00 0.4694575742E+00
> 53 3 0.9256425787E+00 0.6018522640E+00 0.9799177228E+00
> 54 3 0.4256425811E+00 0.1018522679E+00 0.5200822773E+00
> 55 3 0.9256425811E+00 0.8981477321E+00 0.4799177227E+00
> 56 3 0.9347257894E-01 0.8945740691E+00 0.2184145744E+00
> 57 3 0.5934725789E+00 0.1054259309E+00 0.7815854256E+00
> 58 3 0.5934725786E+00 0.3945740725E+00 0.2815854253E+00
> 59 3 0.9347257856E-01 0.6054259275E+00 0.7184145747E+00
> 60 3 0.4256425786E+00 0.3981477360E+00 0.2008227717E-01
> 0.0000000000E+00 0.0000000000E+00 0.0000000000E+00
> 0.3796409345E-10 -0.2823521794E-10 -0.6775993499E+01
> 0.4331511054E-08 -0.1000091620E+02 -0.3952259833E-11
> -0.1248479890E+02 0.7282280671E-08 0.2555546997E-10
>
>
> Best regards,
>
> Alexander A. Maryewski
>
>
> _______________________________________________
> DFTB-Plus-User mailing list
> DFTB-Plus-User at mailman.zfn.uni-bremen.de
> https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user
>
--
Dr. Bálint Aradi
Bremen Center for Computational Materials Science, University of Bremen
http://www.bccms.uni-bremen.de/cms/people/b-aradi/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20211210/bc00804c/attachment-0001.htm>
More information about the DFTB-Plus-User
mailing list