Multiplicative perturbation bounds for multivariate multiple linear regression in Schatten \(p\)-norms (Q2029823)

The authors extend multivariate multiple linear regression (MLR) analyses to sketched MMLR in general Schatten \(p\)-norms by interpreting the sketched problem as a multiplicative perturbation. They present a geometric interpretation of the action of the sketching matrix \(S\) in terms of relevant subspaces and show that a key term in assessing the accuracy of the sketched MMLR solution can be interpreted as the tangent of a largest principal angle between these subspaces if \(S\) has orthonormal rows or if \(S\) preserves the rank. Furthermore, they generalize this interpretation to the operator norm difference between an orthogonal and oblique projector with the same range when \(S\) preserves rank. Their results are extensions of \textit{P. J. Maher} 's work on Schatten \(p\)-norms [Filomat 21, No. 1, 99--111 (2007; Zbl 1273.47003)] as well as of the analysis by \textit{J. T. Chi} and \textit{I. C. F. Ipsen}, Linear Algebra Appl. 624, 87--102 (2021; Zbl 07355214)] of the sketched MMLR problem in a Schatten \(p\)-norm.