scientific article; zbMATH DE number 3874432

From MaRDI portal
Publication:3339957





The paper deals with the Gaussian nonlinear regression model: \(y=\eta (\theta)+\epsilon (y\in R^ N\); \(\theta\in U\subset R^ m\), U open; \(\epsilon \sim N(0,\Sigma)\), \(\Sigma\), \(n\times n\), known, non singular). \({\hat\theta }\) is the least squares estimate for \(\theta\), i.e.: \[ \| y-\eta ({\hat\theta })\|^ 2=_{\theta\in U}\| y-\eta (\theta)\|^ 2. \] Under convenient assumptions, an approximative probability density of the least squares estimate is given, the level of approximation being related to a certain level of regularity of the model (namely a measure of the probability that the sample goes beyond the nearest center of curvature of the mean-values manifold). The paper is devoted to the proof of a general theorem, after a heuristic approach for the case \(m=1\), \(N=2\), using differential geometry. As a consequence, confidence regions for \(\theta\) are determined. The computational aspects are presented on a concrete example (a special choice for \(\eta\) (\(\theta)\)).











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3339957)