Localized Dynamic Mode Decomposition with Temporally Adaptive Partitioning

Abstract: Dynamic Mode Decomposition (DMD) is a widely used data-driven algorithm for predicting the future states of dynamical systems. However, its standard formulation often struggles with poor long-term predictive accuracy. To address this limitation, we propose a localized DMD framework that improves prediction performance by integrating DMD's strong short-term forecasting capabilities with time-domain decomposition techniques. Our approach segments the time domain of the dynamical system, independently constructing snapshot matrices and performing localized predictions within each segment. We first introduce a localized DMD method with predefined partitioning, which is simple to implement, and then extend it to an adaptive partitioning strategy that enhances prediction accuracy, robustness, and generalizability. Furthermore, we conduct an error analysis that provides the upper bound of the local and global truncation error for our method. To demonstrate the effectiveness of our approach, we apply it to four benchmark problems: Burgers' equation, the Allen-Cahn equation, the nonlinear Schrodinger equation, and Maxwell's equations. Numerical results show that our method significantly improves both predictive accuracy and computational efficiency.