Using metadynamics to train a machine-learned force field

It can be tricky to model transition states using [:Category:Transition_states#Static_methods | static methods] methods. Sometimes, this is insufficient and more time-consuming [:Category:Transition_states#Dynamic_methods | dynamic methods] must be used. By using [:Category:Advanced molecular-dynamics sampling | advance MD methods] in combination with [:https://vasp.at/wiki/Category:Machine-learned_force_fields | machine-learned force fields], the cost can be significantly reduced. [Metadynamics] is one such method, applying a biased potential on selected geometric parameters, to describe rare events.

ADD BULK - maybe some of the visual plots from Georg: /fsc/home/csheldon/test/metadynamics_mlff/mlff_run/spilling_factor/with_penaltypot/short_metadynamics/georg