📝 SummarXiv

Abstract

Dynamic Mode Decomposition (DMD) is a powerful, data-driven method for diagnosing complex dynamics. Various DMD algorithms allow one to fit data with a low-rank model that decomposes it into a sum of coherent spatiotemporal patterns. Nominally, each rank of the DMD model is interpreted as a complex, stationary spatial mode modulated by a single set of complex time dynamics (consisting of exponential growth/decay and oscillation), and an amplitude. However, the specifics of how these DMD components are interpreted do not appear to be consistent with the information actually present in the DMD decomposition or the underlying data. While there is a clear physical interpretation for the complex time dynamics, there is practically no guidance on the complex spatial modes. To resolve these issues, we introduce the phasor notation of the DMD model for conjugate pair DMD modes, which results in a strictly positive and real spatial pattern as well as spatiotemporal waveform. The phasor notation terms result in an interpretable DMD model that provides a more complete diagnoses of the model components, as demonstrated on a toy model. This DMD interpretation needs to be adjusted for DMD variants which alter the relationship between the DMD model and the data, such as those that window data in time. We derive the phasor notation terms for one such method, multi-resolution Coherent Spatiotemporal Scale-separation, and demonstrate the new terms by interpreting a multi-scale data set.