Ineffective FIXED_ATOMS constraints in QS MD

[Marco Caccin]

Hi Marcella,

Thanks a lot for the prompt reply! 
As you suggested, switching from AD_LANGEVIN to NOSE solved the issue. I do 
not strictly need to use an adaptive Langevin thermostat for my 
calculations, but I chose it so that I made only one choice between my 
different calculations (QM/MM md have Langevin), and also because a 
colleague of mine was using AD_LANGEVIN for a very similar (although 
unconstrained) system.
The system I'm trying to simulate is silica crack tip under tensile load 
with water molecule(s) in the gap (hence constrained atoms at the boundary) 
at constant temperature

The thermostat causing the issue also explains why the RMS displacement of 
constrained atoms is about one order of magnitude smaller than the 
unconstrained ones. May it be just a velocity of the Langevin part of the 
AD_LANGEVIN thermostat slipping through the code and producing the bug? 

Best regards, 

Marco

On Wednesday, 14 January 2015 07:34:46 UTC, Marcella Iannuzzi wrote:
>
>  Hi 
>
> the MD with constraints works fine for me, in both the NVE and the NVT 
> ensemble with a standard thermostat.
> The AD_LANGEVIN thermostat gives problems, instead, and it is possibly not 
> working properly with such constraints.
> Why are you using this thermostat? 
> Kind regards
> Marcella
>
> On Tuesday, January 13, 2015 at 5:20:28 PM UTC+1, Marco Caccin wrote:
>>
>> Dear all,
>>
>> I am trying to run Quickstep MD with some fixed atoms positions (set in 
>> MOTION / CONSTRAINT / FIXED_ATOMS), but the positions do change over time.
>> Curiously enough, constrained atoms have zero velocity at frame 0 and 
>> zero force at every timestep, which would suggest that the constraint is 
>> somehow applied but ineffective.
>>
>> This behaviour seems to be machine and version independent:
>> tested cp2k versions: 2.7 (Development Version) git: 0826abf, 2.6 
>> (Development Version) svn:14226, 2.5.1 svn:13632 
>> tested architectures/compilers: gfortran 4.8 on Darwin for cp2k 2.7, xlf 
>> 14.1 on BlueGene/Q for cp2k 2.5 and 2.6
>>
>> The attachments are a minimal example of the problem, showing 5 timesteps 
>> of dynamics of 3 atoms of Si of which atom 1 is (or rather, should be) 
>> fixed.
>>
>> Is this problem known, and is there any workaround? I know there is for 
>> GEO_OPT runs as in 
>> https://groups.google.com/forum/#!topic/cp2k/KAMhME81IbM, but I haven't 
>> found anything applicable to MD
>>
>> Kindest regards,
>>
>> Marco
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20150114/f866c7e0/attachment.htm>