LB-IC-CMA-ES with penalty bounds gets sometimes stuck

[nikohansen]

On the 20-D ellipsoid with 10 integer variables and bounded in [-0.5, 1.5] (hence binval(10) + elli(10)), LB-IC-CMA-ES gets stuck with ~10% chance beyond the bound like this:

It also happens with a lower bound of -0.499999.

Penalty weights from a different run with similar outcome:

    fun = cma.fitness_transformations.IntegerMixedFunction2(
                        lambda x: cma.ff.elli(x) + 1e-5 - 1e-12, int_idxs)    

Why? The update of gamma is too slow?