[Pw_forum] Two questions on a La2O3 slab: (1) inconsistent energy results; (2) "maximum number of steps has been reached"

noon at seas.ucla.edu noon at seas.ucla.edu
Sun Jun 1 17:56:40 CEST 2014


Thank you for your help and sorry about the delayed reply. It looks like
both increasing ecutrho and avoiding the use of assume_isolated = esm are
doing the trick of getting bfgs convergence.

Some comments and questions though:

1. *Degauss*. Just from my own trial-and-error simulations, it seems
degauss = 1D-4 or 1D-3 worked the best for me. I started using it when I
got previous "charge is wrong - smearing needed" crash errors. It's
possible you may be correct in larger values are better in that those past
simulations I did were when I was even more inexperienced; hence, there may
have been other issues I was unaware of. I'll experiment more with this.

2. *Mixing-beta*. My experience in reducing mixing-beta is that it
sometimes causes the simulation to crash with the message "error in routine
mix_rho (1): negative dr2". This seems to pop up again more frequently when
combined with an increase in mixing_rho.

3. *k-points*. I've struggled to understand these due to my lack of
background and accessible resources so I've basically been using gamma as a
lazy way out. I've done lengthy trial-and-error procedures before to
determine optimal values for n1, n2 and n3 but that was back when I had
other problems too. *Are optimal selection of the k-points often necessary
to properly converge; or can I still get the correct convergence with
gamma, just maybe slower/less efficiently*?

4. *electon_maxstep*. I use 800 out of caution. Though rare, I do come
close to not converging in some SCF cycles when I experiment with different
structures.

5. *version*. 5.0.1 but the system also recently got 5.0.3 as well. I
haven't switched yet.

Thanks again for your help.


--
Daniel Noon
UCLA Chemical and Biomolecular Engineering

On Mon, May 12, 2014 at 8:58 AM, Giuseppe Mattioli <
giuseppe.mattioli at ism.cnr.it> wrote:

> Dear Daniel
> I do not know the properties of your La oxide but now I can try to
> give you some help. There are several things you should try to change
> in your input file, but first of all the plane-wave cutoff
> (ecutwfc=30) seems to be too low for your PAW pseudopotentials. Did
> you try to reach convergence on the basis set (e.g., all forces very
> similar at the end of the first scf iteration on the same geometry)?.
> Moreover, if you do not indicate the density cutoff (ecutrho), a value
> of ecutwfc*4 is automatically assigned. This is a good choice for
> norm-conserving pseudopotential, a poor choice for ultrasoft
> pseudopotential, and a choice that should be checked in the case of
> PAW pseudopotential (your case).
>
> > degauss = 1D-4
>
> Very low. A higher value of gaussian smearing (0.05~0.01) helps the
> scf to reach the ground state. If your defective oxide is an insulator
> (probably it is not...), the energetics also should be fine with 0.01,
> you can restart the relax calculation near to the final convergence
> with this lower value. If the defective oxide puts electrons in the
> conduction band (very common for oxygen vacancies in oxides), you may
> try with the Marzari-Vanderbilt ('mv') smearing.
>
> You did not specify nspin, so nspin=1. This is probably a good choice
> for the intrinsic oxide, but what about the vacancy? You have two
> excess electrons (in a semiconductor slang), are they diamagnetic or
> paramagnetic? Forcing a diamagnetic unstable state could be a possible
> reason for the reported instabilities.
>
> > assume_isolated = esm,
> > esm_bc = bc1
>
> Why? Try before to converge your system without such electrostatic
> tricks. I've never used the esm part of the code, and I'm not even
> sure that it works properly. It is not, for sure, something to begin
> with!
>
> > conv_thr = 1D-7,
>
> In difficult cases you may try to start with lower values, and even to
> use the scf_must_converge keyword (with care...) to recover from a
> problematic starting geometry.
>
> > mixing_beta = 0.4,
>
> Lower values (down to 0.05~0.01) help scf convergence.
>
> nosym = .TRUE.,
>
> Why? symmetry speeds up the calculation.
>
> > mixing_mode = 'local-TF',
> > mixing_ndim = 8
> > upscale = 1D+3
>
> not necessary, if everithing else has been setup correctly
>
> K_POINTS (gamma)
>
> Your supercell is rather small, that is, large in the reciprocal
> space. I do not know if you are familiar with reciprocal space
> techniques (if you are not, I recommend a certain degree of
> improvement of your knowledge...:-)), but you probably need to check
> the convergence wrt the k-point sampling of the Brillouin zone.
> For example, use
>
> K_POINTS (automatic)
> n1, n2, n3, 0, 0, 0
>
> and raise n1, n2, n3 until the forces/eigenvalues/.../what you're
> interested in is converged. Of course, the more the k-points the
> heavier the calculation...
>
> > Maximum number of steps:
> > electron_maxstep = 800,
>
> Does the scf stop before reaching such a value? the nstep value (in
> &CONTROL list) must be assigned. the default value for relax
> calculations is 50. Does the bfgs procedure stop before reaching 50,
> complaining that "The maximum number of steps has been reached"? It
> would be an hapax, at least in my experience... By the way, which
> version of the code are you using?
>
> When you are more skilled, you will probably consider to treat your
> defective oxide at a DFT+U level. We will speak again about it...
>
> HTH and welcome to the QE community
>
> Giuseppe
>
> Giuseppe Mattioli
> ISM-CNR
> Italy
>
> Quoting noon at seas.ucla.edu:
>
> > Apologies. Beginning user and first time post. My name is Daniel and my
> > affiliation is University of California Los Angeles (UCLA).
> Specifically, I
> > am a Ph.D. candidate in the chemical engineering department at UCLA.
> >
> > The simulations for the La2O3 slab with a surface O-atom removed are all
> > identical (same 39 atom cell, same pseudopotentials, etc), except for
> > differing in one or more of the following: degauss, etot_conv_thr,
> > forc_conv_thr, mixing_ndim, upscale. The final energy results can vary
> > widely, like between -8330 Ry to -8360 Ry, far too large of a
> distribution
> > for the purposes the results are being used for. No such phenomenon
> happens
> > when that O-atom is kept intact, energies consistent to within about 0.5
> > Ry. I get the "maximum number of steps reached" issue (possibly
> non-issue?)
> > on all my simulations, O-atom intact or removed.
> >
> > I have a sample input file attached (converted from UNIX text to text
> > format viewable in Windows). Output file was too big; not sure what
> samples
> > to take from it that would help aside from previous general descriptions
> of
> > what's going on.
> >
> > Thanks for any help you may have to offer.
> >
> > --
> > Daniel Noon
> > UCLA Chemical and Biomolecular engineering
> >
> >
> > On Fri, May 9, 2014 at 8:52 AM, Giuseppe Mattioli <
> > giuseppe.mattioli at ism.cnr.it> wrote:
> >
> >>
> >> Dear unknown user
> >> First of all, please sign always your posts to this forum with your name
> >> and scientific affiliation, we care about it :-)
> >>
> >> Regarding your questions, it is impossible to help you without seeing
> (at
> >> least the relevant parts of) your input files. What do you mean when
> you say
> >> "separate simulations" of the same system? Did you change the cell or
> the
> >> vacuum region? Did you double the slab? Did you use different
> >> pseudopotentials?
> >> The second question can be also related to inconsistencies or errors in
> >> your input files. Again, it is impossible to give you advice without
> seeing
> >> them
> >>
> >> HTH
> >>
> >> Giuseppe
> >>
> >> On Friday 09 May 2014 07:57:20 noon at seas.ucla.edu wrote:
> >> > Hello,
> >> >
> >> > I've been trying out pw.x, with my first goal being to reproduce some
> of:
> >> > http://pubs.acs.org/doi/abs/10.1021/jp103604b. I've been running into
> >> some
> >> > issues and was wondering if I could get some advice:
> >> >
> >> > (1) When I run simulations of a La2O3 slab with a surface oxygen atom
> >> > removed (39 atom supercell), I get wildly inconsistent results
> >> (internally,
> >> > on QE, when doing multiple simulations) for the energy when doing
> >> "relax",
> >> > with a range of about 30 Ry. The distribution of resulting energies
> from
> >> > separate simulations seems to narrow when I tighten conv_thr, but I
> >> cannot
> >> > seem to get them to be acceptably consistent. Anyone familiar with
> this
> >> > issue and how to address it? I appear to have no such problem when
> there
> >> is
> >> > no oxygen atom removed, with energies consistent within about 0.5 Ry.
> >> >
> >> > (2) For all of my simulations of La2O3 slabs, regardless of what my
> >> inputs
> >> > are and as long as I do not run into the maximum number of iterations
> for
> >> > SCF or crash for another reason, I always eventually get the message
> "The
> >> > maximum number of steps has been reached" with the simulation
> stopping.
> >> > This message pops up even when it appears the number of SCF cycles and
> >> BFGS
> >> > steps do not reach nsteps. I notice that, during the simulations, it
> >> seems
> >> > to get to a point where the energy switches back and forth between two
> >> > values between SCF cycles, and then eventually it decides to stop with
> >> that
> >> > message. Is there a fix for this? Or is this something I don't need to
> >> even
> >> > worry about (i.e. maybe the final result is acceptable)?
> >> >
> >> > Appreciate any advice you may have to offer. Can provide more
> information
> >> > upon request. Thanks.
> >>
> >>
> >> ********************************************************
> >> - Article premier - Les hommes naissent et demeurent
> >> libres et ègaux en droits. Les distinctions sociales
> >> ne peuvent être fondèes que sur l'utilitè commune
> >> - Article 2 - Le but de toute association politique
> >> est la conservation des droits naturels et
> >> imprescriptibles de l'homme. Ces droits sont la libertè,
> >> la propriètè, la sùretè et la rèsistance à l'oppression.
> >> ********************************************************
> >>
> >>    Giuseppe Mattioli
> >>    CNR - ISTITUTO DI STRUTTURA DELLA MATERIA
> >>    v. Salaria Km 29,300 - C.P. 10
>
> >>    I 00015 - Monterotondo Stazione (RM)
> >>    Tel + 39 06 90672836 - Fax +39 06 90672316
> >>    E-mail: <giuseppe.mattioli at ism.cnr.it>
> >>    ResearcherID: F-6308-2012
> >>
> >> _______________________________________________
> >> Pw_forum mailing list
> >> Pw_forum at pwscf.org
> >> http://pwscf.org/mailman/listinfo/pw_forum
> >>
> >>
>
>
> --
> ********************************************************
> - Article premier - Les hommes naissent et demeurent
> libres et ègaux en droits. Les distinctions sociales
> ne peuvent être fondèes que sur l'utilitè commune
> - Article 2 - Le but de toute association politique
> est la conservation des droits naturels et
> imprescriptibles de l'homme. Ces droits sont la libertè,
> la propriètè, la sùretè et la rèsistance à l'oppression.
> ********************************************************
>
>     Giuseppe Mattioli
>     CNR - ISTITUTO DI STRUTTURA DELLA MATERIA
>     v. Salaria Km 29,300 - C.P. 10
>     I 00015 - Monterotondo Stazione (RM)
>     Tel + 39 06 90672836 - Fax +39 06 90672316
>     E-mail: <giuseppe.mattioli at ism.cnr.it>
>
> _______________________________________________
> 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/20140601/61124ebb/attachment.html>


More information about the users mailing list