Non-converging optimization with Hilbert-Schmidt distance about krotov HOT 2 CLOSED

Based on our discussion, this is not a bug, nor is the functional non-convex. The problem is simply that the two functionals result in a significantly different optimization landscape that requires different values for λₐ for numerical stability. However, it is challenging to really get to the bottom of this. We've seen examples where λₐ for J_T_hs has to be large than for J_T_re, and the other way around.

While our understanding of the different optimization landscapes is still imperfect, the best resolution is to tone down the recommendation for J_T_hs for any optimization using density matrices and to point out that changing the functional may require adjusting λₐ.

from krotov.

I fully agree with this work-around.

from krotov.