Estimation of solutions of random sets of simultaneous linear algebraic equations of high order (Q2577559)

The authors consider the following problem. Assume that the system of linear algebraic equations \(Ax=b\), where \(A\) is an \(n\times N\) matrix and \(b\in\mathbb R^N\), has a solution \(x\in\mathbb R^n\). Then one wishes to estimate the solution \(x\) of the above problem by means of a single realization of \(R\) and \(y\), where, now, \(R=A+\xi\) and \(y=b+\omega\), \(\xi\) and \(\omega\) being random, in the class of solutions given by \(\mu(\eta)=\Gamma\,R^T\,y\), where \[ \Gamma=\int_0^{\infty} (R^T R+tI)^{-1}\,d\eta(t), \] and \(\eta(t)\) is a function of bounded variation. The asymptotic relation between the spectra of the matrices \(R^TR\) and \(A^TA\) is studied. Then, an explicit expression for the quadratic risk is obtained, under suitable assumptions. Numerical simulations are also reported.