Explore constrained optimization

This direction would explore how we can use symmetries in the mean-field algorithm.

One first action is to use symmetries to reduce the parameter space by removing symmetry equivalent states. For example in cases where spins pointing in different directions are equivalent. Reducing the parameter space would be beneficial for convergence.

Another direction is to explore specifically breaking or not breaking particular symmetries in the starting guess and then looking at the symmetry breakings which the mean field solutions have.

In this direction, we can explore how and whether we can use ~qsymm to aid us in this exploration. (And maybe even interface the code with qsymm).