[Pw_forum] vc-relax
Axel Kohlmeyer
akohlmey at cmm.chem.upenn.edu
Mon Feb 19 14:52:43 CET 2007
On Mon, 19 Feb 2007, Detlef Schmicker wrote:
DS> Thanks a lot for the response Paolo,
DS>
DS> I changed
DS> > > ecutwfc = 50 ,
DS> > > ecutrho = 300 ,
DS>
DS> but the behaviour does not seem better.
hallo detlef,
it would be easier for people that have not followed
this thread all along, to quote a few more lines, so
it is easier to see what this is about.
DS> I calculated for 76 hours now: The total energy is
76 hrs can be a lot or little, depending on what size
of a problem you have and how large/fast the machine
is you're running on. i have currently a big calculation
running that has been 'on the machine' for 4 weeks now
and in total consumed 1.5mio CPU hours. i'm quite pleased
with its progress and am optimistic it'll have advanced
enough in another week or so. proper speed and time are
quite relative in this business. ;-)
DS> going down as expected:
DS>
DS> ! total energy = -1961.72747371 Ry
DS> ! total energy = -1961.72856632 Ry
DS> ! total energy = -1961.73085136 Ry
DS> ! total energy = -1961.73414783 Ry
DS> ! total energy = -1961.73790040 Ry
DS> ! total energy = -1961.74093832 Ry
DS> ! total energy = -1961.74316902 Ry
DS> ! total energy = -1961.74619617 Ry
DS> ! total energy = -1961.74717438 Ry
DS> ! total energy = -1961.74824660 Ry
DS> ! total energy = -1961.75286729 Ry
DS> ! total energy = -1961.75525825 Ry
DS> ! total energy = -1961.75771758 Ry
DS> ! total energy = -1961.76046869 Ry
DS> ! total energy = -1961.76271513 Ry
DS> ! total energy = -1961.76306347 Ry
DS> ! total energy = -1961.76500771 Ry
DS>
DS>
DS> But the total force is not behaving as I expect it:
DS> Total force = 0.023362 Total SCF
DS> correction = 0.005194
DS> Total force = 0.020491 Total SCF
DS> correction = 0.003656
DS> Total force = 0.022102 Total SCF
DS> correction = 0.005813
DS> Total force = 0.020633 Total SCF
DS> correction = 0.004226
DS> Total force = 0.020646 Total SCF
DS> correction = 0.006542
DS> Total force = 0.031831 Total SCF
DS> correction = 0.007696
DS> Total force = 0.047256 Total SCF
DS> correction = 0.009570
DS> Total force = 0.044493 Total SCF
DS> correction = 0.007352
DS> Total force = 0.079400 Total SCF
DS> correction = 0.016914
DS> Total force = 0.080773 Total SCF
DS> correction = 0.014257
DS> Total force = 0.044399 Total SCF
DS> correction = 0.006424
DS> Total force = 0.048366 Total SCF
DS> correction = 0.004975
DS> Total force = 0.046974 Total SCF
DS> correction = 0.009519
DS> Total force = 0.041070 Total SCF
DS> correction = 0.009807
DS> Total force = 0.049513 Total SCF
DS> correction = 0.004709
DS> Total force = 0.089950 Total SCF
DS> correction = 0.010658
DS> Total force = 0.074462 Total SCF
DS> correction = 0.011620
DS>
DS> and the stress:
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -20.71
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -20.18
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -19.34
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -17.83
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -11.93
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -6.68
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -0.79
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -7.57
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -8.15
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -5.28
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -6.21
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -2.72
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -4.19
DS> total stress (Ry/bohr**3)
DS> (kbar) P= 0.91
DS> total stress (Ry/bohr**3)
DS> (kbar) P= 1.61
DS> total stress (Ry/bohr**3)
DS> (kbar) P= 3.21
DS> total stress (Ry/bohr**3)
DS> (kbar) P= -6.89
DS>
DS>
DS> and for completness I send the temperatures:
DS> Ekin = 0.00000000 Ry T = 0.0 K Etot =
DS> -1961.72747371
DS> Ekin = 0.00111722 Ry T = 3.8 K Etot =
DS> -1961.72744910
DS> Ekin = 0.00347429 Ry T = 7.8 K Etot =
DS> -1961.72737707
DS> Ekin = 0.00703939 Ry T = 13.2 K Etot =
DS> -1961.72710844
DS> Ekin = 0.01110756 Ry T = 19.3 K Etot =
DS> -1961.72679284
DS> Ekin = 0.01345837 Ry T = 24.6 K Etot =
DS> -1961.72747996
DS> Ekin = 0.01531120 Ry T = 29.1 K Etot =
DS> -1961.72785782
DS> Ekin = 0.01380786 Ry T = 31.7 K Etot =
DS> -1961.73238831
DS> Ekin = 0.01215649 Ry T = 32.9 K Etot =
DS> -1961.73501789
DS> Ekin = 0.01207252 Ry T = 33.8 K Etot =
DS> -1961.73617409
DS> Ekin = 0.01067002 Ry T = 34.0 K Etot =
DS> -1961.74219728
DS> Ekin = 0.00889187 Ry T = 33.7 K Etot =
DS> -1961.74636637
DS> Ekin = 0.01081347 Ry T = 33.9 K Etot =
DS> -1961.74690411
DS> Ekin = 0.01202862 Ry T = 34.5 K Etot =
DS> -1961.74844007
DS> Ekin = 0.01322190 Ry T = 35.2 K Etot =
DS> -1961.74949323
DS> Ekin = 0.01505569 Ry T = 36.3 K Etot =
DS> -1961.74800778
DS> Ekin = 0.01128462 Ry T = 36.4 K Etot =
DS> -1961.75372309
DS>
well, i assume you're running damped dynamics to optimize, right?
in that case it seems, that your structure is still relaxing
(hence the decrease of potential energy while the kinetic energy
stays on a high level). it might have helped to initialize
the velocities a little bit to have the system pick up speed
faster in the beginning. for as long as the potential/total
energy goes down, you actually _want_ some kinetic energy in
your system or else you'll be stuck in a local minimum.
for as long as you have your atoms move significantly, you'll
have significant total forces, the fact that stress tensor
is fluctuating a lot is to be expected for a dense system.
please note, that you can figure out the proper values for ecutwfc
and ecutrho with single point calculations (just enable printing
of forces and stress). most uspps work quite accurate for ecutwfc
30ry to 50ry, a choice of 35-40ry is usually conservative,
however to converge the stress tensor well (especially for GGA
functionals) you may need a 6 to 10 times higher density cutoff.
once you have those converged to an accuracy that you are comfortable
with, you can start the relaxation. it might be more efficient to
start the relaxation with less accurate choice of parameters and
then crank them up when you're getting close to the minimum.
DS> I tried several values for dt (50 -- 300) and did the
DS> calculation with lower
there are a bunch of strategies to make vc-relax work efficiently,
e.g. use the same mass for all atoms. the value of dt has to be
chosen to integrate the equations of motions in that case correctly,
i.e. it is usually determined by the smallest mass in the system.
note that also the fictitious mass for the cell dynamics is related
to that. check out the material at.
http://www.vlab.msi.umn.edu/events/lecture.shtml
DS> ecutwfc = 30 , ecutrho = 120 ,
DS> in the last weeks. I also calculated 5 times longer
DS> than the actual 78 hours, but nothing changed the
DS> behaviour, that the total force is not going down
DS> below approx. 0.02
DS>
DS> Perhaps somebody can help.
DS>
DS> Thanks a lot, Detlef
DS>
DS> P.S.: I have strong evidence for unusual mail problem
DS> with the list :-)
you have to make certain, that you always use the correct
sender address (not only the from line) which has to be the
one you are subscribed with. there are hundreds of people
on this list that have no problem. most reported problems
with mailing lists, where you have to be subscribed to be
able to post are related to this. these days also a lot of
mailing list mails are 'eaten' by overly eager spam filters
(since mailing lists are essentially using a similar distribution
mechanism as most spam does).
gruessle,
axel.
DS>
DS> --- Paolo Giannozzi <giannozz at nest.sns.it> schrieb:
DS>
DS> >
DS> > On Feb 15, 2007, at 18:23 , Detlef Schmicker wrote:
DS> >
DS> > > (In case an list administrator reads this, I can
DS> > send the
DS> > > error messages to him).
DS> >
DS> > unless you have solid evidence that it is not the
DS> > usual
DS> > problem (i.e. e-mail mismatch), please don't waste
DS> > your
DS> > time and mine
DS> >
DS> > > ecutwfc = 50 ,
DS> > > ecutrho = 190 ,
DS> >
DS> > not good: for USPP, ecutrho should be larger than
DS> > 4*ecutwfc
DS> >
DS> > Paolo
DS> > ---
DS> > Paolo Giannozzi, Democritos and University of Udine,
DS> > Italy
DS> >
DS> >
DS> > _______________________________________________
DS> > Pw_forum mailing list
DS> > Pw_forum at pwscf.org
DS> > http://www.democritos.it/mailman/listinfo/pw_forum
DS> >
DS>
DS>
DS>
DS>
DS>
DS>
DS>
DS> ___________________________________________________________
DS> Der frühe Vogel fängt den Wurm. Hier gelangen Sie zum neuen Yahoo! Mail: http://mail.yahoo.de
DS> _______________________________________________
DS> Pw_forum mailing list
DS> Pw_forum at pwscf.org
DS> http://www.democritos.it/mailman/listinfo/pw_forum
DS>
--
=======================================================================
Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu http://www.cmm.upenn.edu
Center for Molecular Modeling -- University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.
More information about the users
mailing list