[Pw_forum] End of SCF - convergence NOT achieved

[Ben Palmer]

Hi Tom,

Sorry for the delay in replying.  I've been using antiferromagnetic 
state with -0.3, 0.3,-0.3,0.3 as starting points.

This is the link to my input files:

https://my.pcloud.com/publink/show?code=kZUTmvZXUrJtJhfNwjJBevtM8nudf6lUr1X

I think I've identified a couple of problems.  I was originally using 
ibrav=0 and had manually set the crystal axis.  When I changed this to 
either ibrav=6 or ibrav=14 it started to converge (this is with the 
atoms set in the exact FCC locations).

However, I've started to break the symmetry by varying the positions of 
the atoms slightly.  This all worked fine with nspin=0, but it seems a 
lot trickier now that magnetism is involved.  I tried a calculation this 
morning where I slightly moved just four of the 32 atoms.  There was no 
symmetry, but the calculation did finish.

Using exact locations, the calculation ran in about ~15 minutes with 20 
iterations (my input/output files are test_010.in and test_010.out).  
With the four slightly perturbed atoms the calculation completed in 3.75 
hours with 53 iterations (my input/output files are test_013.in and 
test_013.out).

If I perturb every atom slightly, it fails to converge after 250 iterations.

Thank you,

Ben


On 28/06/17 13:14, Tom Daff wrote:
> Hi Ben
>
> It would probably be helpful to see your input file, as there are many
> difficulties with iron and it could be any one of them!
>
> What magnetism are you using? FCC iron is most stable in antiferromagnetic
> double layer spin state (unless you are willing to look into spin-spiral),
> and around the equilibrium volume there is a crossover in stability of the
> high-spin and low-spin ferromagnetic states and I've found it difficult to
> get convergence if you are off with the starting spin.
>
> I've found that mixing_beta = 0.1, with the default mixing scheme gives me
> the most efficient convergence, and increasing the number of k-points can
> sometimes help too.
>
> Tom Daff
>
> University of Cambridge
>
>
>
> -----Original Message-----
> From: pw_forum-bounces at pwscf.org [mailto:pw_forum-bounces at pwscf.org] On
> Behalf Of Ben Palmer
> Sent: 28 June 2017 08:32
> To: PWSCF Forum <pw_forum at pwscf.org>
> Subject: [Pw_forum] End of SCF - convergence NOT achieved
>
> Hi Everyone,
>
> I'm trying to run a scf and vc-relax calculation for FCC Iron (nspin = 2).
> I've managed to do this for a 4 atom example (although the final scf
> calculation did fail: Error in routine ggen (138): smooth g-vectors missing
> !).  When I scale up to 32 atoms, it fails to converge.
>
> I've increased the number of steps to 250, used Plain, TF and local-TF
> modes, changed the mixing mode, beta and ndim.  I've also tried increasing
> nbnds and changing the starting magnetisation and celldm(1).
>
> The structure looks fine and I'm using the output from the 4 atom vc-relax
> calculation (scaled up 2x2x2).
>
> I wanted to ask if anyone has any suggestions on what I could try in order
> to get the calculation to converge?
>
> Thank you.
>
> Ben Palmer
>
> student @ University of Birmingham
>
>
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://pwscf.org/mailman/listinfo/pw_forum