When doing QOCT with the conjugate-gradient optimization scheme, the gradient is
computed thanks to a forward-backwards propagation. For debugging purposes, this
gradient can be compared with the value obtained "numerically" (i.e. by doing
successive forward propagations with control fields separated by small finite
In order to activate this feature, set OCTCheckGradient to some non-zero value,
which will be the finite difference used to numerically compute the gradient.