Geometry Optimisation of Large Charged Systems

[Marcella Iannuzzi]

Hi Olivia, 

GAPW should not be much more expensive, in particular if  a reduced cutoff 
can be used.

Coming back to your original e-mail. Indeed, the optimisation of the 
geometry of a liquid water box makes little sense, in my opinion.
The internal pressure of liquid water is obtained from the fluctuations at 
finite temperature. A frozen configuration is close to useless.

Finally, to answer the question of Rolf: no, there is no automatic run-stop 
if SCF does not converge. 
In most of the cases, one does neither want nor need such a stop. It is 
true that for pathological systems it could prevent wasting of time. 

regards,
Marcella


On Friday, July 17, 2015 at 2:44:27 PM UTC+2, Olivia Lynes wrote:
>
> Hi Matt,
>
> Thanks for that info. Is this just a problem that's specific to Mg? I'll 
> see what happens with the GAPW method, do you know if it's much more 
> expensive than the standard GPW method?
>
> Thanks
>
> Olivia
>
> On Thursday, July 16, 2015 at 9:48:22 PM UTC+1, Matt W wrote:
>>
>> Hi Olivia,
>>
>> Mg2+ is nearly pathological if you use the default q10 pseudo and the GPW 
>> method. It needs a plane wave cutoff of around 1200 Ry or so to start 
>> getting reasonable answers. If you are sticking with non-hybrid 
>> functionals, like PBE, then try using the GAPW method (QS/METHOD section) 
>> and see if that helps. A cutoff of around 350 Ry should give well converged 
>> numbers then.
>>
>> Matt
>>
>> On Thursday, July 16, 2015 at 11:41:44 AM UTC+1, Olivia Lynes wrote:
>>>
>>> Hi all,
>>>
>>> I'm having a problem with a few of my geometry optimisations not 
>>> converging after several thousand steps and having looked at the energy 
>>> changes through the main output file it's not decreasing but is fluctuating 
>>> a lot in a certain range. I'm at a loss to why as I've got similar 
>>> calculations completed with similar input and just a different box size and 
>>> they have converged without this fluctuation in a few hundred steps.
>>>
>>> I've attached the input, coordination files and the first hundred or so 
>>> steps of the output file.
>>>
>>> These are geometry optimisations of an Mg 2+ ion in a box of 64 waters, 
>>> with a cell size of 13.41 angstroms, so the overall system is charged 2. 
>>> I've used the GTH DZVP basis sets and potentials. 
>>>
>>> Having done a cell optimisation on just the 64 waters, we introduced the 
>>> Mg ion and charged the system. In lieu of being able to do cell 
>>> optimisations on a charged system, then geometry optimisations are being 
>>> done at varying box sizes by 10ths of an angstrom as a sort of manual cell 
>>> optimisation to try to find both a minimum energy structure and a minimum 
>>> energy box size. 
>>>
>>> I've done similar calculations using Ca and Sr and haven't had this 
>>> issue with any of my jobs, they all converge in a few hundred steps. 
>>>
>>> Is this potentially a problem using charged systems? Is it necessary for 
>>> me to put a counter ion in which would let me do a standard cell 
>>> optimisation? 
>>>
>>> Or is this approach completely wrong and we would be better off doing 
>>> MD? 
>>>
>>> The aim is to do ab initio MD on these systems to look at the 
>>> coordination of the ion in a bulk of water, so reaching a minimum energy 
>>> structure and box size isn't essential but it would be nice to have the 
>>> regular DFT data to back it up. 
>>>
>>> Thanks in advance for any insights.
>>>
>>> Olivia
>>>
>>>
>>>
>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20150717/47676dfd/attachment.htm>