Can Dakota handle complex variables?

[sitadrost]

I'm thinking of using Dakota's optimisation under uncertainty capabilities for the design of a piezoelectric transducer, using a 1D model. The design variables and objective function values I'm interested in are real-valued, but the model equation contains complex numbers (mixed differential-delay system function in Laplace domain, where also some of the parameter values may be complex-valued).

Can Dakota handle complex-valued variables and parameters? And if so, do gradient-based methods with analytical gradients/Hessians generally perform better than black-box type methods?