Energy jumps in initial phase of BFGS convergence

[Matt W]

Try altering the model hessian 
<https://manual.cp2k.org/trunk/CP2K_INPUT/MOTION/GEO_OPT/BFGS.html#list_USE_MODEL_HESSIAN> value 
to false. You could also try setting USE_RAT_FUN_OPT  in the same section 
to .TRUE.

These values have altered between different versions of the code.

It could also be that your forces are not accurate enough (I've not looked 
at your input, cutoff is the likely candidate) which messes up the 
approximate hessian construction.

Matt

On Thursday, May 3, 2018 at 8:51:20 AM UTC+1, Eduard Elias wrote:
>
> Hi All,
>
> I have a problem minimizing the geometry of my system. I am trying to 
> minimize a relatively large system (264 atoms), but using the BFGS 
> minimizer it doesn't seem to work. Every time I start the optimization the 
> initial phase of the run shows large positive jumps in energy before the 
> energy smoothly decreases:
>
> I use a Restart with the default parameters but this doesn't solve this 
> initial phase. Furthermore, I am stuck with a walltime of 24h on my 
> computer cluster so I cannot simply just let it run for a longer amount of 
> time. I have tried using the CG optimizer but, although it always minimizes 
> the energy from the start, it is extremely slow. I was wondering if someone 
> knows how to restart the geometry optimization in a way that it immediately 
> starts to decrease in energy? I attached the input files to this message. 
> Thanks in advance!
>
> With kind regards,
> Eduard Elias
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20180503/de728dee/attachment.htm>