Total least-squares regularization of Tykhonov type and an ancient racetrack in Corinth (Q848588)

The work shows a connection between total least-squares (TLS) and an iterative least-squares solution in the nonlinear Gauss-Helmert model, with emphasis on the case where the lambda-weighted R-norm of the full parameter vector is minimized during the iteration. As a consequence, far more TLS problems can be regularized in Tykhonov's sense than has been suggested, since they were designed exclusively for the errors-in-variables (EIV) model. Most other alternatives do not really regularize in accordance with Tykhonov's principle (as claimed), due to the different error propagation that, unfortunately, is absent from many of those publications. A variation of the Golub/Hansen/O'Leary's total least-squares (TLS) regularization technique, based on the hybrid approximation solution (HAPS) within a nonlinear Gauss-Helmert Model, is introduced in the the beginning of the paper. By applying a traditional Lagrange approach to a series of iteratively linearized Gauss-Helmert models, a new iterative algorithm is developed. In practice, it can generate the Tykhonov regularized TLS solution, provided that some care is taken to do the updates properly. The algorithm actually parallels the standard TLS approach, as it is recommended in some of the geodetic literature, but unfortunately all too often in combination with erroneous updates that would still show convergence, although not necessarily to the (unregularized) TLS solution. Here, a key feature is that both standard and regularized TLS solutions result from the same computational framework, unlike the existing algorithms for Tykhonov-type TLS regularization. In the Section 1, the TLS approach is introduced along with the nonlinear Gauss-Helmert model, before combining it with Tykhonov regularization in the next Section 2. Finally, Section 3 shows a comparison of the performance of the developed new algorithm with existing ones in a variety of examples taken from the literature before presenting, in Section 4, the new algorithm. It is applied to a problem from archeology. The given numerical example shows how the empirical root mean squared errors (RMSE) of the estimated parameters could be significantly reduced by application of regularization, and the results suggest that the Tykhonov regularization parameter value, that generates a minimum empirical (RMSE), may be a good practical choice. Although widely accepted in the community, the empirical RMSE is to be used as quality indicator with some caution as it is, to a large part, informed by the data at hand. Conclusions and an outlook on further work are given in Section 5.