A rational Krylov method for solving time-periodic differential equations (Q2479417): Difference between revisions

Consider the multidimensional linear differential equation \(y' + A y = F\) with a given matrix \(A\) and a given vector-valued function \(F\) subject to the periodicity condition \(y(0) = y(T)\). Such equations arise, e.g., in the semidiscretization of parabolic partial differential equations, and it is useful to have efficient numerical methods for their solution, in particular if the dimension of the system is large. The authors assume all eigenvalues of \(A\) to have positive real parts. This condition is satisfied in many concrete applications. The approach for the solution of the problem consists of three steps: (1) Find a function \(v\) that solves the given differential equation subject to the initial condition \(v(0) = 0\), (2) compute \(y_0 = y(0) = y(T) = v(T) + w(T)\) where \(w(T) = \exp(-T A) (I-\exp(-T A))^{-1}\), and (3) solve the initial value problem \(y' + A y = F\), \(y(0) = y_0\). Steps 1 and 3 are straightforward, and so the authors concentrate on an efficient way for the computation of the quantity \(w(T)\) required in step 2. To this end, a restricted denominator rational Krylov method is introduced and investigated. Theoretical error estimates are given and numerical results supporting the conclusions are presented.