[Q-e-developers] rVV10 non-local functional in PW

[Nicolas Mounet]

Dear Paolo,

Thanks a lot.
just to be sure I checked indeed a rVV10, multi-node, calculation on 
graphite, with and w/o the change, and the results are exactly the same 
(energy & stress).
So this was probably a non-issue if, as you said, one does reasonable 
things with the input.

cheers,
Nicolas

On 02/10/2017 06:05 PM, Paolo Giannozzi wrote:
> Slightly longer answer: it is hard to follow the logic of kernel-table
> file naming, but it seems to me that if the file name is not
> broadcast, the first processor will use what was saved to file; all
> the others will use the default value. This is good as long as you did
> not specify another kernel-table file name in input (I guess you
> didn't); a recipe for disaster, otherwise.
>
> Paolo
>
> On Fri, Feb 10, 2017 at 8:32 AM, Paolo Giannozzi <p.giannozzi at gmail.com> wrote:
>> Short answer: I don't know.
>>
>> Paolo
>>
>> On Thu, Feb 9, 2017 at 9:24 AM, Nicolas Mounet <nicolas.mounet at epfl.ch> wrote:
>>> Dear Paolo,
>>>
>>> in your opinion, is the second issue a real one, i.e. affecting the
>>> results of calculations with rVV10? My intuition is that no, because I
>>> presume that if the vdW table name is wrong somewhere, the code would
>>> crash since the kernel table file would not be the right one. But I'd
>>> better be sure of this - I have in mind the few thousands calculations I
>>> ran with rVV10, with QE 5.2...
>>>
>>> thanks and cheers,
>>> Nicolas
>>>
>>> On 02/09/2017 07:55 AM, Paolo Giannozzi wrote:
>>>> Hi Nicolas, thank you. I fixed the first issue. The old data file is
>>>> going to disappear so the second issue will disappear with it
>>>>
>>>> Paolo
>>>>
>>>> On Thu, Feb 9, 2017 at 1:28 AM, Nicolas Mounet <nicolas.mounet at epfl.ch> wrote:
>>>>> Dear all,
>>>>>
>>>>> I noticed what seem to be two (hopefully minor) issues related to the rVV10
>>>>> functional, in PW.
>>>>> (This time, I checked first the latest GIT version of the code...)
>>>>>
>>>>> The first one is in Modules/funct.f90, in the function get_dft_short. There
>>>>> is a misplaced "end if" around line 1039, resulting in a shortname that is
>>>>> never assigned to 'RVV10' even when inlc is equal to 3 (because it's inside
>>>>> another 'if inlc==2' ...).
>>>>> This is most probably really minor because apparently the only case when
>>>>> this function is called is when the dft is not defined from the input but
>>>>> from the pseudo, and I guess it never happens for non-local vdw functionals.
>>>>>
>>>>> The other issue is in PW/src/pw_restart.f90 : in line 1958, I think one
>>>>> should replace "if inlc==1 .OR. inlc==2" by "if inlc>0", because otherwise
>>>>> it excludes rvv10 (inlc=3), and the vdw_table_name is then not broadcast.
>>>>> I'm not sure how important this latter issue is: to me it seems the function
>>>>> initializing the kernel anyway redefines the vdW table name. But I'm not
>>>>> sure.
>>>>>
>>>>> I submit my fixes here (attached).
>>>>>
>>>>> cheers,
>>>>> Nicolas
>>>>>
>>>>> _______________________________________________
>>>>> Q-e-developers mailing list
>>>>> Q-e-developers at qe-forge.org
>>>>> http://qe-forge.org/mailman/listinfo/q-e-developers
>>>>>
>>>>
>>>>
>>>>
>>>
>>> --
>>> Nicolas Mounet
>>> Research Scientist
>>> EPFL STI IMX THEOS
>>> ME D2 1025
>>> Station 9
>>> CH-1015 Lausanne (Switzerland)
>>> Phone: +41 21 69 31159
>>> _______________________________________________
>>> Q-e-developers mailing list
>>> Q-e-developers at qe-forge.org
>>> http://qe-forge.org/mailman/listinfo/q-e-developers
>>
>>
>>
>> --
>> Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
>> Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
>> Phone +39-0432-558216, fax +39-0432-558222
>
>
>

-- 
Nicolas Mounet
Research Scientist
EPFL STI IMX THEOS
ME D2 1025
Station 9
CH-1015 Lausanne (Switzerland)
Phone: +41 21 69 31159