This PR addresses an issue where ForceBalance detects convergence of the optimization after a single iteration due to too lax convergence criteria.
Here the criteria has been tightened, with the intention that ForceBalance will instead run for 12 steps before exiting. This should be sufficient time to reach convergence - in most previous optimisations of the FF non-bonded terms the objective function has reached a minimum by the 4th step, after which it fluctuates around an average value.
Description
This PR addresses an issue where ForceBalance detects convergence of the optimization after a single iteration due to too lax convergence criteria.
Here the criteria has been tightened, with the intention that ForceBalance will instead run for 12 steps before exiting. This should be sufficient time to reach convergence - in most previous optimisations of the FF non-bonded terms the objective function has reached a minimum by the 4th step, after which it fluctuates around an average value.
Status