Linear model reduction using SPOD modes

Abstract: The majority of model reduction approaches use an efficient representation of the state and then derive equations to temporally evolve the coefficients that encode the state in the representation. In this paper, we instead employ an efficient representation of the entire trajectory of the state over some time interval and solve for the coefficients that define the trajectory on the interval. We use spectral proper orthogonal decomposition (SPOD) modes, in particular, which possess properties that make them suitable for model reduction and are known to provide an accurate representation of trajectories. In fact, with the same number of total coefficients, the SPOD representation is substantially more accurate than any representation formed by specifying the coefficients in a spatial (e.g., POD) basis for the many time steps that make up the interval. We develop a method to solve for the SPOD coefficients that encode the trajectories in forced linear dynamical systems given the forcing and initial condition, thereby obtaining the accurate representation of the trajectory. We apply the method to two examples, a linearized Ginzburg-Landau problem and an advection-diffusion problem. In both, the error of the proposed method is orders of magnitude lower than both POD-Galerkin and balanced truncation applied to the same problem, as well as the most accurate solution within the span of the POD modes. The method is also fast, with CPU time comparable to or lower than both benchmarks in the examples we present.