Fitting linear relationships. A history of the calculus of observations 1750-1900 (Q1279603)

This bock gives historical and mathematical descriptions of the fitting of linear relationships by the procedures of least absolute deviations and the minimax absolute deviations and by the method of least squares. The description of historical backgrounds for the corresponding periods and the lifes of some scientists are also given. Astronomical, geodetic and navigation problems were the first to require solutions of the fitting of linear relationships. First, practicioners determined the values of the \(q\) unknown parameter constants by selecting a set of \(q\) equations from the \(n\) avaiable. As an alternative to the method of selected points, Boscovich minimized the sum of the absolute errors subject to an adding-up constraint. Laplace's discussion of the Boscovich procedure was an alternative to the minimax procedure. Legendre publicated in 1805 his description of the method of least squares. Gauss made significant contributions to the method of least absolute errors in 1809. In the same year he described a fitting procedure based on the principle of maximum probability and the normal law. In 1774 Laplace adopted an inverse (or Bayesian) approach to the fitting problem. Gauss derived in 1809 the normal law from the Principle of the Arithmetic Mean and the Principle of Maximum Probability. Adrain developed the normal law independently of Gauss. In 1818 Laplace developed an asymptotic theory for the least absolute errors procedure and proposed an alternative derivation of the method of least squares. In 1823 Gauss gave a very full treatment of the subject. He studied also the computational problems. Gauss solved (1828) the problem of fitting a linear model subject to a set of linear constraints. Practical variants of his procedures were developed by Doolitle and Cholesky. Donkin (1844) identified the concept of potential energy as a suitable analogy for the statistical concept underlying the method of least squares. Bravais (1846) adopted a geometrical approach. Laplace (1816) developed a sequential least squares method. Thiele (1897, 1903) gave a method based on the orthogonal-triangular decomposition. Edgeworth (1888, 1923) developed the methods for determining two constants in such a way that they minimize the sum of the absolute deviations.