A simple enhanced sampling approach using a bias potential learned from a 
nonequilibrium work distribution and the Jarzynski equality.

LAMMPS, CP2K, and PLUMED inputs are provided. For each system, the workflow is 
as follows:
1. Perform a number of steered MD (SMD) runs along the reaction coordinate 
   of choice
2. Fit a neural network (NN) to the approximate free energy surface with the 
   nn.py script
3. Calculate the true free energy surface from a biased run, using the NN as 
   bias potential, as a reweighted histogram

A matching nn.py script is provided for each system, and depends on scikit-learn 
and numpy. For the HBr system we also have inputs for the HLDA analysis, for 
which you'll need G. Piccini's scripts from https://github.com/piccig/HLDA