An overdtermined system of linear algebraic equations is considered. In accordance with initial problem all the components of it's solution should be non-negative. Errors in the right-hand side and the ill-conditioned system matrix result in non-positive solution of the least-square method. Smoothing and statistical regularizations would be of no use in this case.
Let the error vector of the right-hand side be be random and have a normal distribution with a known covariance matrix. We consider deviation from the least-square solution probability a quality criterion for the obtained non-negative solution. Existence of such a solution has been prooved. Finding this solution can be reduced to finding a minimum of a quadratic form.
Mail to Webmaster www@www-sbras.nsc.ru |
|Home Page| |English Part| |
Go to Home |