where the random variable depends on the parameter vector u. Even if this function is smooth the smoothness is lost as soon as the expectation is computed as a finite sample mean.
The optimizers follow the ideas in Numerical Recipes with some modifications. If this works for you be thankful. Multidimensional optimization is a nontrivial undertaking and a high quality implementation requires a significant amount of effort and expertise. I have no expertise in this field.
We have only a single application in mind: parameter optimization for exercise triggers used to price Bermudan style options. Miraculously the code which we have here can do that.
Definition in file Optimizer.h.
Go to the source code of this file.