Bug? The conserved quantity is NOT conserved after flexible cell MD is restarted

[Matt W]

Hi,

(I could well be wrong here, this is sort of anecdotal rather than 
rigorously checking the details) I think that the conserved quantity for 
the CSVR thermostat is quite a complicated thing compared to other 
thermostats. To my memory, the proper conserved quantity requires sums over 
information from all previous states in a non-trivial way that is tough to 
implement efficiently (i.e. adding new information at every step without 
complete recomputation).

After all that waffle, I suspect the 'problem' is the CSVR thermostat, and 
it is not actually an indication of bad behaviour, but a question of 
efficiency in calculating the conserved quantity (i.e. dynamics are 
correct, but it is hard to be sure without a conserved quantity). Maybe try 
Nose-Hoover after equilibriating with CSVR to be sure?

Matt

On Sunday, August 14, 2016 at 4:41:29 AM UTC+1, Rustam wrote:
>
> Hi Matthias,
>
> Thanks for the great advice! I tried changing the basis set to the one 
> compatible with Mg q2 and low cutoff. Unfortunately, the problem remains as 
> shown in the new figure.
>
> On a different note, what basis sets would you recommend for solid state 
> systems? All-electron basis sets? Or do you think I should generate a pair 
> of potential/basis for the problem at hand?
>
> Thanks again,
> Rustam
>
> PS. The input file is also attached.
>
>
> On Saturday, August 13, 2016 at 2:34:04 AM UTC-4, Matthias Krack wrote:
>>
>> Hi Rustam
>>
>> in the first place I would suggest to use a compatible pair of basis set 
>> and potential for Mg. The Mg TZV2P-GTH basis set was generated for a Mg 
>> GTH-*-q10, but a GTH-BLYP-q2 is employed. Unfortunately, the GTH_BASIS_SET 
>> file provides only generic labels. I will fix that. Please, run the case 
>> with a valid combinations of potential and basis set and report if the 
>> error persists. Note, that the Mg q10 potential requires a very large 
>> cutoff and for the Mg q2 potential there is only a DZVP basis set available 
>> in the file BASIS_SET.
>>
>> Best regards
>>
>> Matthias
>>
>> PS: The basis sets in GTH_BASIS_SETS are usually not well suited for 
>> (dense) solid state systems as they have been generated mainly for 
>> molecular systems.
>>
>> On Saturday, 13 August 2016 07:56:39 UTC+2, Rustam wrote:
>>>
>>> Hi guys,
>>>
>>> As Amandine Chouanard mentioned in a recent post  
>>> <https://groups.google.com/forum/#!topic/cp2k/mwieBkpEjns>the conserved 
>>> quantity shows discontinuities when a flexible cell MD simulation is 
>>> restarted. I used her input file and found that restart works fine with NVT 
>>> and NPT_I simulations, but NOT with NPT_F (see attached figure). The 
>>> discontinuities in NVT_F the conserved quantity are larger if restarts are 
>>> less frequent (i.e. trajectory pieces are longer).
>>>
>>> It seems that the conserved quantities in NPT_I and NPT_F simulations 
>>> are computed by the same subroutine in 
>>> motion/md_conserved_quantities.F:390. So it seems that the restart file 
>>> contains ALL data on the current state of the barostat and thermostat to 
>>> ensure a smooth restart.
>>>
>>> Is it possible that this has something to do with the CELL_REF setting? 
>>> Something along the lines that the grids before and after the restart are 
>>> different? I tried printing the GRID info but it seems to keep the number 
>>> of points constant...
>>>
>>> I will appreciate any suggestions/comments,
>>> Rustam
>>>
>>>
>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20160816/77fa4fbf/attachment.htm>