On expansions for nonlinear systems, error estimates and convergence issues

TL;DR

This paper reviews classical and introduces a new mixed expansion for solving nonlinear differential equations, providing error estimates and analyzing convergence issues, with implications for control theory and numerical methods.

Contribution

It introduces a novel mixed expansion inspired by quantum mechanics, offers rigorous error estimates for nonlinear ODEs, and investigates convergence properties including counterexamples.

Findings

New error estimates involving weak Sobolev norms.

Counterexamples showing failure of Magnus expansion convergence.

Open problem posed for Sussmann's expansion convergence.

Abstract

Explicit formulas expressing the solution to non-autonomous differential equations are of great importance in many application domains such as control theory or numerical operator splitting. In particular, intrinsic formulas allowing to decouple time-dependent features from geometry-dependent features of the solution have been extensively studied. First, we give a didactic review of classical expansions for formal linear differential equations, including the celebrated Magnus expansion (associated with coordinates of the first kind) and Sussmann's infinite product expansion (associated with coordinates of the second kind). Inspired by quantum mechanics, we introduce a new mixed expansion, designed to isolate the role of a time-invariant drift from the role of a time-varying perturbation. Second, in the context of nonlinear ordinary differential equations driven by regular vector fields, we give rigorous proofs of error estimates between the exact solution and finite approximations of the formal expansions. In particular, we derive new estimates focusing on the role of time-varying perturbations. For scalar-input systems, we derive new estimates involving only a weak Sobolev norm of the input. Third, we investigate the local convergence of these expansions. We recall known positive results for nilpotent dynamics and for linear dynamics. Nevertheless, we also exhibit arbitrarily small analytic vector fields for which the convergence of the Magnus expansion fails, even in very weak senses. We state an open problem concerning the convergence of Sussmann's infinite product expansion. Eventually, we derive approximate direct intrinsic representations for the state and discuss their link with the choice of an appropriate change of coordinates.