An online manifold learning approach for model reduction of dynamical systems (Q2927840)

This paper deals with the reduced-order modeling of large-scale nonlinear problems where both the dats sets and the dynamics are systematically reduced. During each iteration cycle, an approximate solution is calculated in a low-dimensional subspace, providing may snapshots to construct an information matrix. The paper under review also discusses the truncation error produced by the subspace iteration using reduced models and provides an error bound. The capability of this method to solve a high-dimensional system is illustrated by linear and nonlinear equations. The method developed in this paper can also be used as a posterior error estimator for other reduced models.