Take a look at the new version of this page: [ beta version ]. Tell us what you think.

ABSTRACT

If one has a set of observables (z1, ··· , zm) which are bound in a relation with certain parameters (a1, ··· , an) by an equation &zgr;(z1, ··· , a1, ···) = 0, one frequently has the problem of determining a set of values of the ai which minimizes the sum of squares of differences between observed and calculated values of a distinguished observable, say zm. If the solution of the above equation for zm, zm = &eegr;(z1, ··· ; a1, ···) gives rise to a function &eegr; which is nonlinear in the ai, then one may rely on a version of Gaussian regression [1, 2] for an iteration scheme that converges to a minimizing set of values. It is shown here that this same minimization technique may be used for the solution of simultaneous (not necessarily linear) equations.

REFERENCES

• 1: COLLATZ, L. Numerische und graphische Methoden. Handbuch d. Physik, Band II, Berlin (1955).
• 2: HARTLEY, K.O. The modified Gauss-Newton method for the fitting of non-linear regression functions by least squares. Technometrics 8 (1961), 269-280.