What is the expected behavior?
In #583 we added sqrt_lorentzian that partly induces complex number in the fit function, but this is blocking #564 since uncertainties package cannot compute error propagation of complex numbers. It should be possible we limit numbers in real value for curve fitting.
