Termination criteria: set a large or small value to effectively disable a criterion?

[larshanen]

In one of the code snippets on the termination criterion documentation page there's a comment on the ftol for NSGA-2:

# Very large value effectively disables this criterion

However, in my experiments I observed behavior that seems to suggest the opposite.

Keeping everything else constant (same data, termination settings, and NSGA-2 hyperparameters), I noticed that smaller ftol values lead to longer runs:

• ftol = 0.0001 → terminated after 962 generations
• ftol = 0.001 → terminated after 418 generations

The resulting Pareto-optimal solutions were better with the smaller ftol, which aligns with expectations if the algorithm is allowed to run longer.

This makes me wonder whether the comment might be misleading. Based on these results, it seems more accurate to say:

# Very small value effectively disables this criterion

Am I interpreting this correctly, or is there something I’m missing in how ftol is applied here? I think the answer will also apply to cvtol and xtol.

Thank you for maintaining this amazing toolbox!