No label defined (Q4023731)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 108273
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | No label defined |
scientific article; zbMATH DE number 108273 |
Statements
23 January 1993
0 references
optimal experimental design
0 references
constrained nonlinear least squares problems
0 references
statistical analysis
0 references
numerically stable computation
0 references
parameter estimates
0 references
Gauss-Newton method
0 references
information matrices
0 references
approximate covariance matrix
0 references
test examples
0 references
0.7888122797012329
0 references
0.7810496687889099
0 references
In science as well as in industry, real life processes are increasingly simulated and optimized on computers. The process under investigation is described by a mathematical model, but often some of the model constants (so-called parameters) are unknown. Reliable parameter estimates can be calculated utilizing observations of the real life process. This approach yields nonlinear least squares problems which are in general restricted by constraints on the observed state variables or on the parameters. In order to arrive at reliable parameter estimates at minimal experimental effort, this paper presents a numerical procedure to compute an optimal experimental design for constrained least squares problems. In the first chapter, a generalized Gauss-Newton method is reported to determine the minimum of a constrained least squares problem. Special attention is focused on the solution of the linearized least squares problem and on a computationally convenient statistical analysis of the optimal parameter estimates. Based on this statistical analysis, an approximate covariance matrix of the parameter estimates which depends on the frequency factors for measurements to be taken at given observation points is proposed in the second chapter together with optimal experimental design criteria. This approach generalizes the concept of information matrices known from classical optimal experimental design theory. For constrained least squares problems, however, the reduction to the subspace of independent variables is done implicitly, since only the original parameters have a physical or technical meaning. The approximate covariance matrix of the parameter estimates is differentiable with respect to the frequency factors for measurements. An explicit formula of the derivatives and a numerically stable algorithm for their computation is given in the third chapter. Thus, approximate optimal experimental designs can be computed cheaply by optimization methods of higher order. In the fourth chapter, the special case of nonlinear boundary value problems is considered and the proposed procedure is illustrated in the fifth chapter by some test examples. Finally, an application taken from biology, the fast fluorescence rise of photosynthesis, demonstrates that parameter estimates based on an optimal experimental design are significantly better than those calculated on observations measured at equidistant time points.
0 references