Ok, so trunk-458 is way more limited in the number of available vdW
functionals, in comparison with trunk-468.  Problem is, my collaborators
*do* want to test different functionals for their calculations.

So I followed Mike's advice: I changed a bit some rc's, and now only for
two of the vdW functionals ghost states are detected for pseudos with the
vdW functionals present in trunk-462: C09 and LMKLL. I tried changing the
rc's for C09 in an automated and systematic way (sweeping all possible
values of rc for the 1s and 2p channels in a range of values) and none of
them was able to pass the ghost state detection. I will start the process
for the LMKLL functional now, but I suspect the same will happen.

Could there be a bug in the XC library, or somewhere else?

Best regards,

Marcos

On Wed, Sep 2, 2015 at 5:05 PM, Marcos Veríssimo Alves <
[email protected]> wrote:

> Hi Mike and Camps,
>
> That's weird. Perhaps a change in the XC library of 462? Thanks a lot for
> sharing. I'll use 458 to make the tests.
>
> Cheers,
>
> Marcos
>
> On Wed, Sep 2, 2015 at 4:54 PM, Michael R. C. Williams <
> [email protected]> wrote:
>
>> Hi Marcos,
>>
>> I have also had some trouble with the ghost state warning when using vdW
>> pseudos in trunk-462. In some cases, pseudos that I had used with trunk-458
>> without getting the warning caused the warning in trunk-462. I found that
>> small changes to the rc values used to generate the pseudos sometimes
>> resulted in a pseudo that didn’t lead to the ghost state warning (e.g., if
>> something like an rc of 1.20 led to a pseudo that caused the warning,
>> sometimes a pseudo with rc of 1.19 didn’t). However, I never figured out
>> exactly what was going on.
>>
>>
>> Mike
>>
>> On Sep 2, 2015, at 1:17 PM, I. Camps <[email protected]> wrote:
>>
>> Hello,
>>
>> I generated the vdW pseudos with trunk-458 with the DRSLL and LMKLL
>> functionals without the ghost problem for H, C, O and N. I used the INP
>> files from SIESTA pseudopotentials webpage.
>>
>>
>> []'s,
>>
>> @mps
>>
>> On Wed, Sep 2, 2015 at 4:06 PM, Marcos Veríssimo Alves <
>> [email protected]> wrote:
>>
>>> Hi all,
>>>
>>> I am trying to generate pseudos for H with the vdW functionals and use
>>> them in vdW calculations using trunk-462 for the H2 molecule. However, I am
>>> running into ghost states when I try running the calculation for the H2
>>> molecule with the DRSLL, LMKLL, C09 and BH functionals.
>>>
>>> I tried using my own generated C09 pseudo, and also a standard PBE
>>> pseudo (downloaded from the siesta pseudo database), with all the vdW
>>> functionals, and I find that for both pseudos, DRSLL, LMKLL, C09 and BH
>>> functionals. This leads me to believe that it is not really a problem of
>>> the pseudo, even because I tried changing the rc for the channel with the
>>> ghost state for H (2p).
>>>
>>> Now my collaborators have called my attention to an undocumented flag
>>> that allows one to ignore ghost states and proceed normally with the
>>> calculations. As far as I know, this is a quite dangerous thing to do. Is
>>> there any rule of thumb for deeming a certain system safe, or is it just
>>> the results of the calculations of the properties of the system of interest
>>> being correct that would allow me to conclude this?
>>>
>>> Cheers,
>>>
>>> Marcos
>>>
>>
>>
>>
>

Responder a