A new error in variables model for solving positive definite linear system using orthogonal matrix decompositions (Q285053): Difference between revisions

The paper deals with the estimation of a positive definite solution to an overdetermined linear system with multiple right hand sides vectors. It considers that an error in the measured data and target matrices appears. The presented approach is based on a newly defined error function. The necessary and sufficient optimality conditions are derived and the direct algorithm to compute the solution is proposed. Two comparisons are presented: with respect to the interior point method and the method based on the quadratic programming. It is shown that the presented method leads to smaller standard deviation of error entries and smaller effective rank as desired by control problems. More precisely, the following problem is discussed: \[ \min_{X\succ 0} \text{tr} ((DX-T)^T(D-TX^{-1})), \] where \(D,T\in \mathbb R^{{m\times n}}\), \(m\geq n\), are given and a positive definite matrix \(X\in \mathbb R^{n\times n}\) is to be computed. \(\text{tr}\) stands for trace of a matrix. The simplified form of the problem is the following: \[ DX\cong T. \] It is called the EVI problem. Algorithm 1 is based on the orthogonal decomposition and the spectral decomposition. Algorithm 2 uses only the spectral decomposition. The most sophisticated Algorithm 3 is based on the spectral decomposition, rank determination, linear system solving, and the Cholesky decompositon. The complexity of computations is analysed and numerical experiments conclude the paper (with observations mentioned above). The paper is well written, instructive and easily understandable.