Analysis and computation of limit cycles for systems with separable nonlinearities (Q960227)

The authors are concerned with a feedback system defined by \[ y=Hu,\qquad u=\Phi( y) , \] where \(H\) is a bounded, causal, linear time-invariant operator and \(\Phi\) is a memoryless nonlinear operator defined by a time-domain relation \[ ( \Phi( y) ) ( t) =\varphi( y( t) ) , \] \(\varphi:\mathbb{R}^{m}\to\mathbb{R}^{m}\) is a \(C^{1},\) and it is also assumed that both nonlinearity and its derivative are (at least locally) Lipschitz continuous. The system can be viewed as a feedback interconnection of a stable LTI plant with a memoryless, possibly multivariable, nonlinearity. The purpose of the paper is to study existence, uniqueness, and robustness of limit cycles (non-constant isolated periodic solutions). Furthermore, a computational framework that may be given both time-domain and frequency-domain realizations is provided for finding limit cycles. It also allows one to verify that the real system has indeed a limit cycle despite of model uncertainties and computation errors. The authors start by defining the return difference operator \(F:C( 1) \times\mathbb{R} _{+}\to C( 1) ,\) \[ F( z) =y-H[ T] \Phi( y) ,\tag{1} \] where \(H[ T] \) is a linear operator defined by the transfer function \(H( s/T) \) with corresponding impulse response function \(h( tT) ,\) and \(C( 1) \) stands for the set of 1-periodic continuous functions equipped with the norm \[ \| v\| _{C( 1) }=\max_{t\in[ 0,1] }| v( t) | . \] Then the principal results in the paper are derived using a linearization of the operator \(F\) in (1) and a right inverse of the resulting linear operator. It is discussed how the choice of a right inverse of an infinite-dimensional operator defined by a linearization of the dynamical system in a neighborhood of a limit cycle affects the accuracy of the analysis. Two alternative ways of computing the pseudo-inverse, which is the right inverse that provides sharpest bounds in the estimates used in the paper, are suggested. The Newton iteration scheme that has both frequency and time-domain implementations is proposed for computing limit cycles. First, its convergence properties are discussed (the Newton iteration converges to the manifold but not to a particular limit cycle). Then robustness results which confirm that the real system has a limit cycle, even if the exact model is not known and the Newton iteration may be terminated before an exact solution has been found, are obtained. Two examples, the well-known Van der Pole equation and the Wien bridge oscillator, are provided to illustrate the use of the Newton iteration for computing limit cycles. Remarkably, the analysis of the latter equation demonstrates that the suggested computational framework can be successfully employed for determination of both stable and unstable limit cycles.