[Wannier] Problem with use_ws_distance

[Dominik Gresch]

Hi Lorenzo,

I opened an issue on GitHub for this, and I'd suggest we continue the 
discussion there.

Best,

Dominik


On 16.02.2017 15:48, Lorenzo Paulatto wrote:
> On Thursday, February 16, 2017 1:34:42 PM CET Dominik Gresch wrote:
>> Thank you very much for your reply. Indeed, increasing the tolerance (in
>> my case, to about 7.d-4) fixes this issue. I changed the code such that
>> the tolerance can be given as an input parameter, see here:
>> https://github.com/greschd/wannier90/tree/irdist_ws_tol
>>
>> First of all, do you think this is useful / necessary in general? If
>> yes, are there other places in ws_distance.F90 where the tolerance
>> should be changed accordingly? From glancing over the code I would guess
>> maybe eps = 10 * irdist_ws_tol would make sense. Of course we could also
>> let the user set eps and use irdist_ws_tol = eps / 10 . Please let me
>> know what you think about this.
>
> All the thresholds are inside ws_distance, there should be nothing outside of
> this file, which is good.
>
> You are right that the various thresholds should be consistent, although I'm
> not sure it is trivial to do because some are measured in direct lattice
> units, other in fractional units.
>
> In particular :
> 1. in R_ws_sc_equiv, line 235, we are accepting a *relative* error of 10^-5 in
> cartesian coordinates (coordinate system is important for anisotropic
> lattices)
>
> 2. in ws_translated_dist, line 157 we require an *absolute* precision of 10^-6
> in fractional coordinates.
>
> If I remember correctly, I put this second check in order to spot bugs in the
> code, but now I think the check in ws_translate_dist could just be removed.
>
> I think it could be a good idea to have the threshold value in R_ws_sc_equiv
> be an input parameter, and maybe default to something larger than 1.d-5
>
> Also, I'm not sure that using a relative threshold, instead of an absolute
> value is justified, because it would require the centres of the WFs to be less
> accurately equivalent when using finer grids of k-points.
>