The ABINIT Users Mailing List ( CLOSED )

[Matthieu Verstraete <mjv500@york.ac.uk>]

> kpgsph : BUG -
>  The variables ikg, mkmem, and mpw  must satisfy ikg<=(mkmem-1)*mpw,
>  while the arguments of the routine are
>  ikg =    3060, mkmem =      20, mpw =     159
>  Probable cause: Known error in invars1 for parallel spin-polarized case.
>  Temporary solution: Change the number of parallel processes.
>  Action : contact ABINIT group.
ooh, haven't seen this one yet. As usual without your input and a bit more 
verbose output we can't diagnose anything. Add platform compiler etc...

Is NiAl spin polarized?

> Therefore I decided to determine the elastic constants by calculation of the
> energies of strained samples.
Don Hamann has told us several times there is quite a large error on the 
end result. Using finite differences of stress tensors should be better 
(correct?).

> In these calculations the total energy should
> increase parabolically with increasing strain. But it doesn't. Depending on
> the parameters it usually increases somehow, but the curve is never a
> parabola. In some cases it even decreases a little.
> These are the parameters I already tried:
> ecut (280...480 eV)
20 Ha is not all that much for Ni

> ngkpt (2 2 2...10 10 10)
10x10x10 is again not all that much for tough metals. When you do strains 
the k and G grids shift and you are not guaranteed to have a monotonous 
variation of E wrt the grid density. You need very well converged grids 
(ecut nkpt) to get smooth results. ecutsm may help, but I'm not convinced.

> nband/fband 65/0.0d0
> occopt 3...4
> tsmear 0.03...0.07
this is quite high. with occopt 4 or 5 you can increase it a bit but 0.03 
is really a maximum.

> I tried with different pseudopotentials (Troullier-Martins) as well, namely
> 13al.pspnc, 28ni.pspnc / 13-Al.LDA.fhi, 28-Ni.LDA.fhi,  13-Al.GGA.fhi,
> 28-Ni.GGA.fhi.

> This is one of my infiles:
> acell 5.7970482729504687 5.767560060000 5.767560060000 angstrom
> ecut 480 eV
> ionmov 0
> ixc 1
careful here you are forcing LDA

> kptopt 1
> natom 16
> ngkpt 10 10 10
> nstep 15
> ntypat 2
> nband 65
> occopt 4
> tsmear 0.03
> toldfe 0.005 eV
could be much lower. How many Ha is this? not even 1.e-10

> chkprim 0
> znucl 13 28
> typat
> 1
...
> xangst 0.000000 0.000000 0.000000
> 0.000000 0.000000 2.883780
> 0.000000 2.883780 0.000000
> 0.000000 2.883780 2.883780
> 2.898524 0.000000 0.000000
> 2.898524 0.000000 2.883780
> 2.898524 2.883780 0.000000
> 2.898524 2.883780 2.883780
> 1.449262 1.441890 1.441890
> 1.449262 1.441890 4.325670
> 1.449262 4.325670 1.441890
> 1.449262 4.325670 4.325670
> 4.347786 1.441890 1.441890
> 4.347786 1.441890 4.325670
> 4.347786 4.325670 1.441890
> 4.347786 4.325670 4.325670

Are you sure these are correct? Shouldn't you be keeping the reduced 
coordinates fixed as a first step, and then relaxing them (ionmov 2) as a 
second step?

good luck...

Matthieu