Manuals >User's Guide >Optimizing
Print version of this Book (PDF file) 		prevnext

Optimization Algorithms

The choice of optimization algorithms depends upon:

    • the goal of the optimization, and
    • the nature of the model equations involved.

The optimization algorithm is selected with the Algorithm drop-down list on the Extract/Optimize table, described in Selecting the Optimizer Algorithm.

The following table includes a short description of each optimization algorithm.

Table 22 IC-CAP Optimization Algorithms
Algorithm
Description
Levenberg-Marquardt
 Non-linear search method with least-squares error function.
Random
 Random search method with stochastic gradient error function.
Hybrid (Random/LM)
 Combination of Random and Levenberg-Marquardt algorithms and error functions.
Sensitivity Analysis
Single-point or infinitesimal sensitivity analysis of a design variable. Prints partial derivatives with respect to each parameter.
Random (Gucker)1
 Random search method with least-squares error function.
Gradient1
 Gradient search method with least-squares error function.
Random Minimax1
 Random search method with minimax error function.
Gradient Minimax1
 Gradient search method with minimax error function.
Quasi-Newton1
 Quasi-Newton search method with least-squares error function.
Least Pth 1
 Quasi-Newton search method with least Pth error function.
Minimax1
 Two-stage, Gauss-Newton/Quasi-Newton method with minimax error function.
Hybrid (Random/Quasi-Newton)1
 Combines the Random and Quasi-Newton search methods.
Genetic1
 Direct search method using evolving parameter sets.

1 Uses full working precision of 15 digits during simulation and error calculation while optimizing. The IC-CAP Status window displays results based on the WORKING_PRECISION variable, which is 6 by default. At the end of the optimization, RMS and MAX error are calculated at the default precision. Therefore, results displayed at the end of the optimization may differ from those obtained during optimization steps.
prevnext