On the approximate solution of matrix differential-algebraic boundary-value problems by the least-squares method (Q2227223)

The authors S. Chiko, O. Nesmelova and M. Dzyuba make use of a generalized Green operator to give pseudo-solutions of an undetermined matrix differential-algebraic equation with linear boundary value conditions, continuing their work on similar topics, published since 2008. The matrix problem writes \[AZ'(t)=BZ(t)+F(t),\] \(A,B\) are time-depending \(\mathcal{C}^1\) matrices and \(F\) is also a vector of \(\mathcal{C}^1\) maps. The boundary condition reads \[\mathcal{L}Z(.)=\mathcal{U}.\] First the problem is restated to become a classical linear DAEs \[\Omega(t)z'(t)=\Theta(t)z(t)+M(F(t)).\] The case developed is the one where the previous system is not solvable w.r.t. the derivative (the invertible case is handled in a previous paper by the authors -- [Ukr. Math. J. 70, No. 2, 319--333 (2018; Zbl 1425.34033); translation from Ukr. Mat. Zh. 70, No. 2, 280--292 (2018)]). The authors replace it with an optimization problem to search for a solution of a least square problem. The constraint includes minimizing both the differential system and the boundary condition. The final example details the computations and the various constructions made to get the pseudo-solution, while the use of a control theory style of writing makes this paper a bit difficult to read.