Section: New Results
Effective algebraic systems theory


The purpose of [97] is to present a survey on the effective algebraic analysis approach to linear systems theory with applications to control theory and mathematical physics. In particular, we show how the combination of effective methods of computer algebra $$ based on Gröbner basis techniques over a class of noncommutative polynomial rings of functional operators called Ore algebras $$ and constructive aspects of module theory and homological algebra enables the characterization of structural properties of linear functional systems. Algorithms are given and a dedicated implementation, called OreAlgebraicAnalysis , based on the Mathematica package HolonomicFunctions , is demonstrated.

As far as we know, there is no algebraic (polynomial) approach for the study of linear differential timedelay systems in the case of a (sufficiently regular) timevarying delay. Based on the concept of skew polynomial rings developed by Ore in the 30s, the purpose of [73] is to construct the ring of differential timedelay operators as an Ore extension and to analyze its properties. Classical algebraic properties of this ring, such as noetherianity, its homological and Krull dimensions and the existence of Gröbner bases, are characterized in terms of the timevarying delay function. In conclusion, the algebraic analysis approach to linear systems theory allows us to study linear differential timevarying delay systems (e.g. existence of autonomous elements, controllability, parametrizability, flatness, behavioral approach) through methods coming from module theory, homological algebra and constructive algebra.

Within the algebraic analysis approach to linear systems theory, in [98], we investigate the equivalence problem of linear functional systems, i.e., the problem of characterizing when all the solutions of two linear functional systems are in a onetoone correspondence. To do that, we first provide a new characterization of isomorphic finitely presented modules in terms of inflations of their presentation matrices. We then prove several isomorphisms which are consequences of the unimodular completion problem. We then use these isomorphisms to complete and refine existing results concerning Serre's reduction problem. Finally, different consequences of these results are given. All the results obtained are algorithmic for rings for which Gröbner basis techniques exist and the computations can be performed by the Maple packages OreModules and OreMorphisms .

In [99], we study algorithmic aspects of the algebra of linear ordinary integrodifferential operators with polynomial coefficients. Even though this algebra is not Noetherian and has zero divisors, Bavula recently proved that it is coherent, which allows one to develop an algebraic systems theory over this algebra. For an algorithmic approach to linear systems of integrodifferential equations with boundary conditions, computing the kernel of matrices with entries in this algebra is a fundamental task. As a first step, we have to find annihilators of integrodifferential operators, which, in turn, is related to the computation of polynomial solutions of such operators. For a class of linear operators including integrodifferential operators, we present an algorithmic approach for computing polynomial solutions and the index. A generating set for right annihilators can be constructed in terms of such polynomial solutions. For initial value problems, an involution of the algebra of integrodifferential operators then allows us to compute left annihilators, which can be interpreted as compatibility conditions of integrodifferential equations with boundary conditions.

Recent progress in computer algebra has opened new opportunities for the parameter estimation problem in nonlinear control theory, by means of integrodifferential inputoutput equations. In [102], we recall the origin of integrodifferential equations. We present new opportunities in nonlinear control theory. Finally, we review related recent theoretical approaches on integrodifferential algebras, illustrating what an integrodifferential elimination method might be and what benefits the parameter estimation problem would gain from it.


Computational real algebraic geometric approach:

In [74], we present a symbolicnumeric method for solving the ${H}_{\infty}$ loopshaping design problem for low order singleinput singleoutput systems with parameters. Due to the system parameters, no purely numerical algorithm can indeed solve the problem. Using Gröbner basis techniques and the Rational Univariate Representation of zerodimensional algebraic varieties, we first give a parametrization of all the solutions of the two Algebraic Riccati Equations associated with the ${H}_{\infty}$ control problem. Then, following some works on the spectral factorization problem, a certified symbolicnumeric algorithm is obtained for the computation of the positive definite solutions of these two Algebraic Riccati Equations. Finally, we present a certified symbolicnumeric algorithm which solves the ${H}_{\infty}$ loopshaping design problem for the above class of systems.

In [58], the asymptotic stability of linear differential systems with commensurate delays is studied. A classical approach for checking that all the roots of the corresponding quasipolynomial have negative real parts consists in computing the set of critical zeros of the quasipolynomial, i.e., the roots (and the corresponding delays) of the quasipolynomial that lie on the imaginary axis, and then analyzing the variation of these roots with respect to the variation of the delay. Based on solving algebraic systems techniques, a certified and efficient symbolicnumeric algorithm for computing the set of critical roots of a quasipolynomial is proposed. Moreover, using recent algorithmic results developed by the computer algebra community, we present an efficient algorithm for the computation of Puiseux series at a critical zero which allows us to finely analyze the stability of the system with respect to the variation of the delay.

In [59], we present new computer algebra based methods for testing the structural stability of $n$D discrete linear systems (with $n\ge 2$). More precisely, we show that the standard characterization of the structural stability of a multivariate rational transfer function (namely, the denominator of the transfer function does not have solutions in the unit polydisc of ${\u2102}^{n}$) is equivalent to fact that a certain system of polynomials does not have real solutions. We then use stateoftheart algorithms of the computer algebra community to check this last condition, and thus the structural stability of multidimensional systems.
