Separable Nonlinear Least-Squares Parameter Estimation for Complex Dynamic Systems

Figure 1

The plot gives MSEs on a log scale (computed as averages over Monte-Carlo simulation runs) for linear parameters plotted against the quality of prior information. In panel (a), the comparison is on the basis of the noise level. The graph indicates that the performance of NLS worsens with decreasing quality of prior information. On the other hand, the performance of SLS is not affected by the quality of prior information, in agreement with the experimental design. Except for the rare case of high-quality prior information, where NLS is better, SLS clearly outperforms NLS. In panel (b), the comparison is based on the sample size. The overall pattern is similar to that in panel (a).

(a)

(b)