The ABINIT Users Mailing List ( CLOSED )

[avanderg <avanderg@mines.edu>]

Hello,
Thanks I have changed iscf to 3 and it is working tremendously better - 
it hasn't given any errors and it is significantly further then previously.
Thanks to all for the help.
Abram Van Der Geest



Prasenjit Sen wrote:

> Dear Abram:
>     A quick suggestion is try using iscf=3 instead of 5. See if that
> works.
>
>   Best,
>   Prasenjit Sen
>
> On Mon, 19 Jun 2006, avanderg wrote:
>
>  
>
> > Hello All,
> >
> > I have been trying to optimize the geometry of a three layer Ni slab
> > with a vacuum slab.  I have been unable to get this system to converge
> > for the geometry optimization.  I first ran convergence tests for the
> > system before geometry optimization to get reasonable values of ecut,
> > the k points and tsmear.  Since I have been trying to run the geometry
> > optimization I have changed many parameters but I still am having
> > problems with the convergence.  The error I have been seeing is:
> >
> > scfcge: WARNING -
> >  Potential-based CG line minimization has trouble to converge.
> >
> > and
> >
> >  scfcge : ERROR -
> >  Potential-based CG line minimization not converged after 13 restarts.
> >  Action : read the eventual warnings about lack of convergence.
> >  Some might be relevant. Otherwise, raise nband.
> >
> > Trying to take this into account (and the FAQ) I have used up to 130
> > bands for the 3 Nickel atoms (which only has 10 valence electrons each)
> > which seems extremely large to me.  I have also started to use
> > non-defalut values of nline and nnsclo.  I have also used different
> > convergence parameters using both tolvrs and toldfe as well for tolmxf.
> > I have also tried using the initial locations of the atoms as the
> > expected final result (from literature) and using the bulk spacing.  I
> > have also used both the Broyden and CG values for ionmov.  It basically
> > comes down to the fact that I have tried many things that have been
> > mentioned in past forum posts for surface calculations for transition
> > state metals.  These calculations also invariably take a large amount of
> > computation time (days).  I would really appreciate some help.  Below I
> > have included one of my most recent input files and I have also attached
> > my output file since my log file is so large (scfcge is the only warning
> > I am getting).
> >
> > Thank you
> >
> > Abram Van Der Geest
> >
> >
> > --INPUT--------
> > #I'm trying to get the ground stat of the Ni 111 surface
> > #optimizing the surface
> >
> > #kpts
> > kptopt   1
> > kptrlatt  9  0  0
> >          0  9  0
> >          0  0  1
> > shiftk   0.0  0.0  0.0
> >
> > #geo opt
> > optcell 0
> > tolmxf 1.0d-5
> > ionmov 2
> > ntime 15
> >
> > #metal
> > occopt 4    #use first cold smearing option
> > tsmear 0.003 # d electrons should have smaller tsmear
> >
> > nband 75
> >
> > #magnetized
> > nsppol 2    #treat spin
> > spinat 0.0 0.0 3.0   #break spin symmetry of atom 1
> >       0.0 0.0 3.0   #break spin symmetry of atom 2
> >       0.0 0.0 3.0   #break spin symmetry of atom 3
> >
> > #def of super cell
> > acell 2*3.25564 9.796022881
> > rprim    1.0 -1.0  0.0
> >         0.0  1.0 -1.0
> >         1.0  1.0  1.0
> > ckprim  0
> >
> > #def of atom type
> > ntypat 1   #one type of atom
> > znucl 28   #and it is Ni
> > natom  3   #number of atoms per supercell
> > typat  3*1  #the first three atoms are nickel
> >
> > xred 1/3 2/3 0.78065362  #the 1st atom is at origin
> >     2/3 1/3 0.0  #location of second atom
> >     1.0 1.0 0.21934638  #location of atom 3
> >
> > #Misc
> > toldfe 1.0d-5  #scf tolerance (total E)
> > prtwfk  1
> > getwfk -1      #get wfk from earlier run
> > ixc    1       #LDA approx
> > diemac 1.0d6   #preconditioner for a metal
> > nstep  100      #max number of scf cycles
> > ecut   25      #energy cut off for plane waves
> > diemix 0.05
> >
> > nline  6
> > nnsclo 3
> >
> >
> >    
>
>
>