The ABINIT Users Mailing List ( CLOSED )

["Anglade Pierre-Matthieu" <anglade@gmail.com>]

Dear Joe, dear Zaq

Thanks for your files. I will have a look and let you know when and if I have any suggestion about SCF convergence problems.

Regards

PMA

On 6/7/07, zaq Liu <iloveabinit@gmail.com> wrote:

Dear Prof. Joe,

Thank you very much  for your  pseudopotential .
Why do you use  5s2 5p6 6s2 4f8 as valence but not 5s2 5p6 6s2 4f7 5d1 ?
And I find your rc=2.1 ( if I understand correctly), why is it so small ?
The half of distance is often  2.8-3.2. The small rc need a high ecut,
I check it need about 30Ha.is it right ? while our rc=3.0 need 16 Ha.
In addition, when I use Custom rrkj rather than Vanderbilt, the SCF
loops can obtain convergence for both hcp Gd and GdN. If iprcel= 49
and diemix = 0.5 are applied , the convergence becomes much faster
(about 50 times).
The optimized lattice constance for GdN is less than the exp. values by 3%.

Best regards,

Bo Liu



2007/6/7, Josef Zwanziger <jzwanzig@dal.ca>:
> Hi,
>
> I've attached the input files for USpp2abinit I made for gadolinium,
> using as inspiration the Vanderbilt ultra-soft pseudopotential for Gd
> distributed with CASTEP. It uses as valence 5s2 5p6 6s2 4f8, with two
> projectors in each channel (s, p, f) and l=2 for the local channel.
> Using it with abinit 5.3.4, Gd metal SCF converges ok but needs a lot
> of empty states and is quite sensitive to tsmear (I use occopt = 4). GdN
> was easier to converge and also optimized its structure very well. On
> the other hand, Gd structure did not optimize. Because this
> pseudopotential uses 18 electrons in valence with two projectors, it is
> very flexible but consumes a lot of memory. I don't do research on
> f-block elements so I haven't really pursued this, but thought it might
> be of interest to supply a USPP-based PAW data set to the discussion to
> compare with Atompaw-based PAW.
>
> Joe
>
> On Wed, 2007-06-06 at 16:49 +0200, Anglade Pierre-Matthieu wrote:
> > Hi,
> >
> > What you are saying appears quite weird without having a look at your
> > input file. Would you mind telling us a little bit more (for instance
> > showing abinit's input and output) ?
> > By the way, if the SCF convergence of your system is really as
> > difficult as you describe it I'm very much interested in having a look
> > at it. I would be very grateful if you share with me pseudopotentials
> > and input file.
> >
> > Regards
> >
> > PMA
> >
> > On 6/5/07, zaq Liu < iloveabinit@gmail.com> wrote:
> >         Dear Dr. Marc Torrent,
> >
> >         Thank you very much for your suggestion.
> >         I find that rc= 3.2 seems to be reasonable for hcp Gd metal.
> >         3.2 is a little smaller than the half of the distance for two
> >         Gd ions (3.37).
> >         The relaxed lattice constant is very close to the exp. values
> >         (smaller by 2%)
> >         But during the calculation,the SCF loops need about 200-300
> >         times.
> >         Furthermore, when I calculate GdN and GdP, the SCF loops never
> >         reach
> >         convergence after 350 times. I try a number of variables
> >         diemac dielng
> >         and iprcel and iscf 14, 17 , but the situation does not
> >         change.
> >         Could you please give me some useful advice how to improve the
> >         atomic
> >         data for Gd?
> >         Thank you very much!
> >
> >         Best regards,
> >
> >         Bo Liu
> >
> >         2007/6/1, Marc Torrent <marc.torrent@cea.fr>:
> >         > Quick answer, with some key ideas:
> >         >
> >         > - you should verify your log derivatives and eventually
> >         change the
> >         > reference energy (especially for d, f electrons).
> >         > - your high cut-off is a sign of a small pseudization
> >         radius ; are you
> >         > sure of the rc=2. or 2.2 ? I think you should obtain
> >         reasonnable values
> >         > for ecut with rc=2.4.
> >         >
> >         > Marc Torrent
> >         >
> >         >
> >         > iloveabinit@gmail.com a écrit :
> >         > > Dear All,
> >         > >
> >         > > I generate the Gd atomicdata using atompaw code.
> >         > > The calculated lattice constants  of hcp Gd and GdN show
> >         > > a very large errors (smaller than the experimental values
> >         by about 15%).
> >         > > The LDA and GGA show the similar results. Also the ecut
> >         needed is about 28 Ha when the total energy difference is
> >         smaller than 1meV/atom. Is this ecut too high ? I know the
> >         common ecut for PAW is about 15 Ha.
> >         > >
> >         > > The input file is attached below. The attached file uses
> >         rc=2.0, I also test rc= 2.2 and get the same result. I am
> >         appreciated for any useful advice!
> >         > >
> >         > > Bo Liu
> >         > >
> >         > > INPUTFILE
> >         > >
> >         > > Gd 64  #1s2 2s2 2p6 3s2 3p6 3d10 4s2 4p6 4d10 (5s2 5p6 4f7
> >         5d1 6s2)
> >         > > LDA-PW scalarrelativistic loggrid 2001
> >         > > 6 5 5 4 0
> >         > > 4 3 7
> >         > > 5 2 1
> >         > > 0 0 0
> >         > > c
> >         > > c
> >         > > c
> >         > > c
> >         > > v
> >         > > v
> >         > > c
> >         > > c
> >         > > c
> >         > > v
> >         > > c
> >         > > c
> >         > > v
> >         > > v
> >         > > 3
> >         > > 2.0 2.0 2.0 2.0   #rc rpaw
> >         > > n
> >         > > y
> >         > > 2
> >         > > n
> >         > > y
> >         > > 3
> >         > > n
> >         > > y
> >         > > 5
> >         > > n
> >         > > Vanderbilt
> >         > > 4 0
> >         > > 2.0
> >         > > 2.0
> >         > > 2.0
> >         > > 2.0
> >         > > 2.0
> >         > > 2.0
> >         > > 2.0
> >         > > 2.0
> >         > > 0
> >         > >
> >         > >
> >         > >
> >         > >
> >         >
> >
> >
> >
> >
> > --
> > Pierre-Matthieu Anglade
> --
> Josef W. Zwanziger
> Prof. of Chemistry and
>  Canada Research Chair in NMR Studies of Materials
> Director, Atlantic Region Magnetic Resonance Centre
> Dept. of Chemistry
> Dalhousie University
> Halifax, NS B3H 4J3 Canada
> Tel: +1 (902) 494-1960
> Fax: +1 (902) 494-1867
> Net: jzwanzig@dal.ca
> Web: http://jwz.chem.dal.ca
>
>

-- 
Pierre-Matthieu Anglade