On the algebraical and numerical theory of errors of observations and the combination of observations