[Pw_forum] BFGS relaxation ignoring forc_conv_thr
Leonid Kahle
leonid.kahle at epfl.ch
Wed Sep 20 16:02:55 CEST 2017
Thanks a lot for the fast answer,
as you say, after this warning is printed another step is performed, and
the message:
bfgs converged in 4 scf cycles and 3 bfgs steps
(criteria: energy < 1.0E-06 Ry, force < 1.0E-15 Ry/Bohr)
is printed after the last step. The keyword 'converged' is of course
misleading in such a case.
Should others encounter the same problem, I switched to
&IONS
ion_dynamics = 'damp'
/
from the default 'bfgs'. This makes it possible to reduce the force
components to 1.0D-7.
Leonid
On 09/20/2017 02:03 PM, Paolo Giannozzi wrote:
> If you look carefully in the output, you will notice this:
> WARNING: bfgs curvature condition failed, Theta= 0.815
> Message from routine bfgs:
> history already reset at previous step: stopping
> This kind of problem arises when one is very close to convergence and
> forces contain too much numerical noise for the minimization to be
> effective. Not sure why it performs a further step.
> Paolo
>
> On Wed, Sep 20, 2017 at 9:56 AM, Leonid Kahle <leonid.kahle at epfl.ch
> <mailto:leonid.kahle at epfl.ch>> wrote:
>
> Dear all,
>
> A relaxation of a rather large system (Li21 Si3 P3 S24) is shown
> as converged even though the force convergence threshold is not
> reached. I will paste the final output below, and the full input
> and output are attached. I put a very low threshold 1D-15 to prove
> my point here, but also more sensible thresholds are ignored.
>
> As explained in the documentation, but also in this forum (
> http://qe-forge.org/pipermail/pw_forum/2004-November/076156.html
> <http://qe-forge.org/pipermail/pw_forum/2004-November/076156.html>
> ) convergence of the relaxation is achieved when all the
> components of the force vector are smaller than forc_conv_thr (and
> energy criterion is met as well). In the example below this is
> obviously not the case, but bfgs is shown as converged without any
> warning.
>
> Can anyone help me find an explanation for this? I am using the
> parallel version of QE 6.1 compiled with intel compilers on an
> x86_64 architecture.
>
> Many thanks in advance,
>
> Leonid Kahle
>
> --
>
> Doctoral Assistant
> EPFL STI IMX THEOS
> ME-D2 1019 (Bâtiment ME)
> Station 9
> CH-1015 Lausanne (Switzerland)
>
>
> Forces acting on atoms (cartesian axes, Ry/au):
>
> atom 1 type 1 force = 0.00000940 0.00000005
> 0.00000759
> atom 2 type 1 force = 0.00003922 -0.00005066
> -0.00003891
> atom 3 type 1 force = -0.00000527 -0.00001786
> -0.00000047
> atom 4 type 1 force = 0.00000695 -0.00000525
> -0.00000490
> atom 5 type 1 force = 0.00000837 0.00000136
> 0.00000695
> atom 6 type 1 force = -0.00000139 0.00000800
> -0.00000768
> atom 7 type 1 force = -0.00000042 -0.00000074
> -0.00000888
> atom 8 type 1 force = -0.00002354 -0.00001155
> -0.00003861
> atom 9 type 1 force = 0.00001305 -0.00001546
> 0.00000555
> atom 10 type 1 force = -0.00001756 0.00002269
> -0.00001184
> atom 11 type 1 force = -0.00000255 -0.00001734
> 0.00001741
> atom 12 type 1 force = 0.00001174 0.00001759
> -0.00001735
> atom 13 type 1 force = -0.00000633 -0.00000979
> 0.00000484
> atom 14 type 1 force = -0.00001325 0.00001294
> -0.00000493
> atom 15 type 1 force = 0.00000782 -0.00002956
> -0.00000586
> atom 16 type 1 force = -0.00000317 0.00001363
> 0.00000878
> atom 17 type 1 force = -0.00002203 0.00000270
> -0.00000480
> atom 18 type 1 force = -0.00000998 0.00000531
> 0.00001788
> atom 19 type 1 force = 0.00000060 -0.00000321
> -0.00001756
> atom 20 type 1 force = -0.00000688 0.00002017
> 0.00000404
> atom 21 type 1 force = -0.00000129 -0.00000416
> -0.00001532
> atom 22 type 2 force = -0.00000319 -0.00000167
> 0.00000395
> atom 23 type 2 force = 0.00001103 0.00001040
> 0.00000626
> atom 24 type 2 force = 0.00000396 -0.00000186
> 0.00000201
> atom 25 type 3 force = -0.00001044 0.00000478
> -0.00000086
> atom 26 type 3 force = -0.00001291 0.00000415
> -0.00000411
> atom 27 type 3 force = 0.00001756 -0.00000296
> 0.00000913
> atom 28 type 4 force = -0.00001578 0.00000879
> 0.00000423
> atom 29 type 4 force = -0.00001227 0.00000946
> -0.00000839
> atom 30 type 4 force = 0.00001416 0.00001301
> 0.00001436
> atom 31 type 4 force = 0.00001660 -0.00000202
> 0.00000659
> atom 32 type 4 force = 0.00003452 -0.00001067
> 0.00001793
> atom 33 type 4 force = 0.00000938 -0.00001056
> 0.00000237
> atom 34 type 4 force = -0.00001496 -0.00000472
> 0.00000305
> atom 35 type 4 force = -0.00001625 -0.00000739
> -0.00000918
> atom 36 type 4 force = 0.00001063 0.00001606
> -0.00000310
> atom 37 type 4 force = -0.00000110 0.00000394
> -0.00000268
> atom 38 type 4 force = -0.00000163 0.00000039
> 0.00000700
> atom 39 type 4 force = -0.00000066 -0.00000016
> 0.00000132
> atom 40 type 4 force = -0.00000813 0.00000603
> 0.00000016
> atom 41 type 4 force = -0.00001613 0.00000196
> 0.00000515
> atom 42 type 4 force = -0.00000014 0.00000820
> -0.00000460
> atom 43 type 4 force = 0.00000091 0.00000952
> 0.00001063
> atom 44 type 4 force = 0.00000012 -0.00000246
> 0.00000554
> atom 45 type 4 force = 0.00000154 0.00000256
> 0.00001045
> atom 46 type 4 force = 0.00000846 0.00000608
> -0.00000078
> atom 47 type 4 force = 0.00000508 -0.00000261
> 0.00001345
> atom 48 type 4 force = 0.00000340 0.00000050
> 0.00000619
> atom 49 type 4 force = 0.00000354 -0.00000332
> 0.00000588
> atom 50 type 4 force = -0.00000428 0.00000637
> 0.00000854
> atom 51 type 4 force = -0.00000650 -0.00000066
> -0.00000641
>
> Total force = 0.000149 Total SCF correction =
> 0.000000
>
> bfgs converged in 4 scf cycles and 3 bfgs steps
> (criteria: energy < 1.0E-06 Ry, force < 1.0E-15 Ry/Bohr)
>
> End of BFGS Geometry Optimization
>
>
>
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org <mailto:Pw_forum at pwscf.org>
> http://pwscf.org/mailman/listinfo/pw_forum
> <http://pwscf.org/mailman/listinfo/pw_forum>
>
>
>
>
> --
> Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
> Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
> Phone +39-0432-558216, fax +39-0432-558222
>
>
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20170920/fc6756df/attachment.html>
More information about the users
mailing list