Machine learning only performs two learning steps for Ag bulk

[ritu_tomar]

Hey everyone,

I am facing issues when performing machine learning calculations. For Ag bulk, the learning cycle only performs two ionic steps and skips all the further ones. Also , a common issue I noted is that even if I set
TEBEG=100
TEEND = 800

the temperature fluctuates outside this range as well. I am not sure how to fix it.

I am attaching my POSCAR, OUTCAR and ML_AB files here.

[jonathan_lahnsteiner2]

Dear ritu_tomar ,


It can happen, that the first two MD steps are sufficient to learn enough basis set configurations
to execute machine learning MD steps for some time. But I agree it is suspicious in your
case that the MD runs without further learning for 10000 steps.
If you want to pick up more reference configurations I would recommend you to check the machine learning theory page
about error thresholds
wiki/index.php/Machine_learning_force_field:_Theory
But usually the default settings should suffice.

It is completely normal that temperatures are fluctuating around your input equilibrium
temperature in molecular dynamics. So when you set TEBEG and TEEND the system will start with an
equilibrium temperature of TEBEG and heat up to TEEND during your simulation.
The actual temperature of every simulation step will be above or below the actual equilibrium value.
Maybe check the molecular dynamics tutorial to gain further insight.
wiki/index.php/Liquid_Si_-_Standard_MD

I hope this helps.
All the best Jonathan