In VISTA/SFC,
different external optimizers can be integrated using tool agents as
described in Section 3.5.1.
For optimizing device performance parameters over a given input
variable space, a constrained optimizer with sequential quadratic
approximations
has been integrated. It minimizes the target
function, which can be assembled out of the input and output
values.
The gradient is calculated by evaluating finite differences,
and the Hessian matrix is built by an BFGS update.
For the calibration and fitting applications,
an optimizer based on the Levenberg-Marquardt
algorithm
[Kha95][PKSS95] is available.