[Wannier] Question about searcg_shell: cell too long?

[Jonathan Yates]

On 13 Jun 2008, at 23:14, Han Hsu wrote:
>
> I am trying to calculate MLWFs of LaCoO3. When using wannier90.x - 
> pp to get
> nnkp file, an error occur:
>
> Unable to satisfy B1 with any of the first  12 shells
>  Your cell might be very long, or you may have an irregular MP grid
>  Try increasing the parameter search_shells in the win file  
> (default=12)
>
> So I increase the number of shells. Same thing happened for  
> search_shells =
> 13, ... 196. Then, when I set search_shells = 197, suddenly thousands
> nearest neighbors are found, and I got following error message
>
> kmesh_get: something wrong, found too many nearest neighbours

Han,

  This is subtle problem, and unfortunately the error message is  
misleading.

The first two kpoint shells are actually sufficient to satisfy the so- 
called B1 condition. However, there is not enough precision in the  
specified lattice vectors, and so we don't quite hit the target  
accuracy for solving B1. I realise that specifying more sig. fig. in  
the lattice vectors is tough - but a perfectly reasonable solution is  
to lower the B1 tolerance: set
  kmesh_tol=0.00001 (this is 10x > than the default)
  and it'll work (no need to alter search_shells from the default 12)  
Note that lowering this tolerance will has no effect on the accuracy  
of the wannier functions.

Thanks for pointing this out. In the future we should do two things:

1- Make a script to automatically create w90 input from pwscf inputs  
- and hence always include enough sig. figs.

2- The kmesh routine could be smarter; detect when we have only just  
failed to satisfy B1 and offer a more helpful error message.

  Yours
    Jonathan



>

--
Dr Jonathan Yates         |    Theory of Condensed Matter Group
Corpus Christi College    |    Cavendish Laboratory
Cambridge, CB2 1RH, UK    |    Cambridge, CB3 OHE, UK
email jry20 at cam.ac.uk     |    Tel +44 (0)1223 337461