Conceptual
Browse Instruct Curate

ResKoopNet: Learning Koopman Eigenpairs by Minimizing the Spectral Residual

A neural-network method that computes Koopman operator eigenpairs by directly minimizing the spectral residual, instead of filtering a precomputed spectrum as ResDMD does. This overcomes the 'spectral inclusion' problem and EDMD's spectral pollution, recovering a more precise and complete Koopman spectrum — including continuous spectra of high-dimensional nonlinear systems — with theoretical guarantees.