Local Optimizer

The performance of the gradient based methods strongly depends on the initial values supplied. Several optimization runs with different initial guesses might be necessary if no a priori knowledge (e.g., the result of a process simulation) about the dopant concentration profile is applied. Fig. 5.16

Figure 5.16: Progress of the gradient based optimizer.
\begin{figure}\centering\psfig{file=pics/gradres-rotated, width=0.75\linewidth}\par\end{figure}

shows the evolution of the target values for a certain initial guess. In this example the optimizer was stopped at a local minimum. Care must be taken to provide physically sound bounds for all parameters to avoid simulation failures.