The ABINIT Users Mailing List ( CLOSED )

["Cote Michel" <michel.cote@umontreal.ca>]

Dear Yukihiro,

One possible explanation might be that your calculated forces are not precise 
enough. From your input, I see that you are asking for a convergence in 
toldfe of 10^-6 which implies that the calculated forces are not very 
accurate, maybe 10^-3. This means that the Broyden algorithm does not have 
accurate information of the potential surface to minimise the system to the 
desire tolmxf of 5x10^-5 (default). In others words, the procedure is 
affected by the noise in the calculated forces. 

When you do a structural relaxation, it is always preferable to use the 
convergence criterion toldff which should be smaller that the criterion of 
the Broyden procedure tolmxf. See tutorial 1.3 for a longer explanation.

Michel

*************************************************************** 
  Michel Cote                          tel: +1 (514) 343-5628     
  Professeur adjoint                   fax: +1 (514) 343-2071               
  Département de physique
  Université de Montréal
  C.P. 6128, succ. Centre-ville
  Montréal (Québec) H3C 3J7          Michel.Cote@umontreal.ca
  Canada              http://www.phys.umontreal.ca/~michel_cote
***************************************************************
  
  
                          
 

> -----Message d'origine-----
> De : okuno@osa.sci.jri.co.jp [mailto:okuno@osa.sci.jri.co.jp]
> Envoyé : 1 octobre, 2003 06:55
> À : forum@abinit.org
> Objet : [abinit-forum] On the Optimization of the atom potisions
> 
> 
> Dear Abinit Users.
> 
> I'm now trying to optimize the atom positions 
> of the LaGaO_{3} whose  space group is Pnma and contains 
> 20 atoms in the unit cell.
> 
> I'm use abinit input option 
> ionmov=2 and ntimes=100 but the results are 
> not conversed. 
> 
> Abinit outputs WARNING like..
> 
> fconv : WARNING -
>   ntime=  100 was not enough Broyd/MD steps to converge gradients:
>   max grad (force/stress) = 2.2787E-04 > tolmxf= 5.0000E-05 
> ha/bohr (free atoms\
> )
> 
> the results (max gfad(force/stress)) seems ossilate and not 
> easy to conversed. 
> 
> Another try we take is to use molecular dynamics with damping 
> factor that 
> is 
> 
> ionmov=1, vis=400, dtion 200.0 and ntime= 500 and =1000.
> 
> But the result seems that the atoms are moving with time and 
> not converse
> to stable positions. 
> Are there some techniques that optimize the atom positions in 
> rather large 
> cell? 
> 
> Some body know the ideas to optimize atom positions such system, 
> it is my pleasure to teach me.
> 
> Sincealy.
> 
> Yukihiro Okuno
> 
> My input file for optimization is like below
> #____ Here begins abinit input file 
> 
> #ndtset 2
> 
> ionmov 2
> ntime  200
> #vis 400.0
> #dtion 200.0
> #ntime1 500.
> #ntome2 1000
> 
> 
> 
> 
> #Definition of the unit cell
> acell  10.365147  14.683171 10.436957
> rprim  1.0  0.0  0.0
>        0.0  1.0  0.0
>        0.0  0.0  1.0
> 
> #Definition of the atom types
> ntype 3
> znucl 57 31 8  # for La, Ga, O
> #Definition of the atoms
> natom 20
> type 2 2 2 2 1 1 1 1 3 3 3 3 3 3 3 3 3 3 3 3
> xred
>     0.00000   0.0       0.5
>     0.50000   0.5       0.0
>     0.00000   0.5       0.5
>     0.50000   0.0       0.0
>    -0.009900  0.250    -0.002900
>     0.490100  0.250     0.502900
>     0.009900  0.750     0.002900
>     0.509900  0.750     0.497100
>     0.498000  0.250     0.060000
>     0.998000  0.250     0.440000
>    -0.498000  0.750    -0.060000
>     0.002000  0.750     0.560000
>     0.229000  0.037000  0.762000
>     0.729000  0.463000 -0.262000
>    -0.229000  0.537000 -0.762000
>     0.271000 -0.037000  1.262000
>    -0.229000 -0.037000 -0.762000
>     0.271000  0.537000  1.262000
>     0.229000  0.463000  0.762000
>     0.729000  0.037000 -0.262000
> 
> #Definition of the planewave basis set
> #ecut1  40.0         # Maximal kinetic energy cut-off, in Hartree
> ecut  35.0
> #Definition of the k-point grid
> kptopt 1
> ngkpt 4 4 4
> 
> #Definition of the SCF procedure
> nstep  100          # Maximal number of SCF cycles
> toldfe 1.0d-6     # Will stop when, twice in a row, the difference
>                   # between two consecutive evaluations of 
> total energy
>                   # differ by less than toldfe (in Hartree)
> diemac 12.0       # Although this is not mandatory, it is worth to
>                   # precondition the SCF cycle. The model dielectric
>                   # function used as the standard preconditioner
>                   # is described in the "dielng" input 
> variable section.
>                   # Here, we follow the prescription for bulk silicon.
> 
> 
>