constrained MD simulation crashes. Why ?

[Matt W]

Hi Manuel,

I haven't tried these type of calculations, but I think the problem is 
_not_ in the electronic structure like you are assuming, but in the 
constraints.

If I understand correctly you are trying to grow a collective variable very 
quickly

TARGET_GROWTH [angstrom*fs^-1] 0.1

I suspect this is far too big and might give the massive forces.

I think you should make a smaller system that you can run quickly. Then 
check that it works without constraints. Then try with restraints rather 
than constraints and gradually increase the restraint size. When you get a 
feel for the system you can try and use constraints if needed. It will be 
easier to help if you can post this sort of test system and explain clearly 
what you are trying to do.

HTH,

Matt




On Friday, June 10, 2016 at 8:12:30 AM UTC+1, Manuel Corral wrote:
>
> Hi,
>
> I've been trying hard to sort it out by myself but I still don't get it. I 
> increased the print level tag to high and then I realized that, after the 
> second SCF cycle, the forces between a couple of atoms are too high
>
> 155      1      Al        -80.38251986   215.26927889   352.92629495 
> 433      2      O          80.39131459  -215.30119051  -353.02822305 
> All interatomic forces are all right after the first cycle. This sharp 
> increase concerns these two atoms after the second SCF cycle. 
> *I rerun the calculation decreasing the time step. Surprisingly, things 
> get worse.*
>
> It would be helpful to have the atomic coordinates of the system before 
> the calculation crashes. The file with the atomic positions has only the 
> coordinates of the first step, that is, of the system at the beginning of 
> the calculation. The file is not updated before the crash, even though the 
> input file asks cp2k to do so after every md step. 
>
> I've read lots of things about the outer scf cycle and the preconditioner. 
> I tried everything. I still don't have a clue about what's going on.
>
> Please, is there anyone who could help me ?
> Manuel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20160610/2c607dc1/attachment.htm>