The program generates free energy profiles of different types. Profile parameters can be set in profiles.json. Free energy profiles are related to the time-dependent probability distributions through the Fokker–Planck equation. The FP solver is implemented in a Fortran program, which returns the failure and success probability distributions. Additional parameters, such as the time step (dt), the number of simulation timesteps (timesteps), the number of timesteps for output, and the interval displayed in the output file, can be modified in solver.py. Simulation results are saved to .npz files for further analysis.

The results are analysed using Bayesian optimisation. The surrogate model is Gaussian process regression (GPR), and the acquisition function is expected improvement (EI). The acquisition function and the evaluator type can be changed in run_multiple_opt. The optimiser is implemented in multiple_bayesopt.py. Reference models are computed for different polymer lengths and are used for comparison in the loss function.

Upcoming tasks: Currently, the generated profiles are continuous functions. Instead, I will generate a realistic set of discrete data points (events) to see how the program handles them. The obtained results will then need to be characterized statistically.