Section: New Results
Keywords : polynomial matrix, matrix fraction, matrix multiplication, asymptotic complexity, rank, nullspace, determinant, reduced form, inversion, Padé approximants.
Algorithms and Software for High Performance Linear Algebra
We have pursued our study of asymptotically fast algorithms for the most basic operations on polynomial matrices, with an emphasis on reductions to the multiplication problem. The target matrices are typically n× n of degree dwith univariate entries in for an arbitrary commutative field. Among the studied operations are computing the rank , a nullspace basis, the determinant or a reduced form ; there is also the task of computing the matrix of fractions which is equal to the inverse of a generic polynomial matrix  . In  we highlight the role played by two problems when designing asymptotically fast algorithms for any of the operations above: computing minimal bases of some matrix Padé approximants, and expanding/reconstructing polynomial matrix fractions . We show that reducing these two problems to polynomial matrix multiplication implies the same kind of reductions for all the other operations, hence cost estimates in operations in where is the exponent of scalar matrix multiplication, and with and two real constants. The latter type of reduction has been developed in  for establishing that a polynomial nullspace basis (almost minimal) can be computed using about the same number of operations as for multiplying two polynomial matrices. These studies, and an algorithm we propose for computing the Kalman form  , may be useful for the treatment of multivariable linear systems in control.
Part of these theoretical developments yield software components in the LinBox library (see § 5.6 ).