ModelReductionofSparseSystems.PDF

Parallel Algorithms for Balanced Truncation Model Reduction of Sparse Systems? Jos′eM. Bad′?a1, Peter Benner2, Rafael Mayo1, and Enrique S. Quintana-Ort′?1 1 Depto. de Ingenier′?a y Ciencia de Computadores, Universidad Jaume I, 12.071–Castell′on, Spain; {badia,mayo,quintana}@icc.uji.es. 2 Fakult¨atf¨ur Mathematik, Technische Universit¨atChemnitz, D-09107 Chemnitz, Germany; benner@mathematik.tu-chemnitz.de. Abstract. We describe the parallelization of an e?cient algorithm for balanced truncation that allows to reduce models with state-space di- mension up to O(105). The major computational task in this approach is the solution of two large-scale sparse Lyapunov equations, performed via a coupled LR-ADI iteration with (super-)linear convergence. Experi- mental results on a cluster of Intel Xeon processors illustrate the e?cacy of our parallel model reduction algorithm. 1 Introduction Consider the continuous linear time-invariant (LTI) dynamical system in state- space form: x˙(t) = Ax(t) + Bu(t), t 0, x(0) = x0, (1) y(t) = Cx(t) + Du(t), t 0, ≥ where A Rn×n, B Rn×m, C Rp×n, D Rp×m, and x0 Rn is the initial state of the∈ system. Also,∈ the number∈ of states,∈ n, is known∈ as the state-space dimension (or the order) of the system, and the associated transfer function matrix (TFM) is de?ned by G(s) = C(sI A)?1B + D. Hereafter, we assume that the spectrum of the state matrix, A, is? contained in the open left half plane, implying that the system is stable. The model reduction problem requires ?nding a reduced-order LTI system, x?˙ (t) = A?x?(t) + B?u?(t), t 0, x?(0) =x ?0, (2) y?(t) = C?x?(t) + D?u?(t), t 0,