A computer program was written for estimating components in a mixture by so
lving an overdetermined system of linear equations by Chebyshev's method. A
n overdetermined system of linear equations contains more equations than un
knowns. If the equations are uncertain, then, in general, no exact solution
exists. A feasible solution can nevertheless be found by minimising the to
tal error of the equations. In Chebyshev's method the problem is transforme
d to a linear programming problem, which is solved. The program was tested
on two artificial mixtures containing 4 and 10 rock types, respectively. Th
e computation was based on geochemical XRF analyses of the components and m
ixtures. Best results were obtained by removing weakly fitting elements fro
m the model in 2 or 3 steps, so that the number of elements then became 2 o
r 3 times the number of components. In the best solution for the mixture of
4 components the error, defined as the largest difference between the comp
uted and true compositions, was 4%. For 10 components the error was 8.5%. I
nvestigation of the influence of missing components on the residuals of the
equations showed the largest absolute residual to increase drastically whe
n more than 40% of the components were missing from the model. The suitabil
ity of the program was also tested on fine fractions of till, for which the
proportions of glacially ground rock types and sediments were estimated. (
C) 1998 Elsevier Science Ltd. All rights reserved.