7–11 Sept 2026
Cefalù
Europe/Rome timezone

Decoding and Forecasting Non-linear Variability Patterns in X-ray Binaries Using Koopman Operator Theory

Not scheduled
20m
Cefalù

Cefalù

Cefalù, province of Palermo, Sicily. Sala delle Capriate in the Town Hall (Piazza Duomo).
Poster

Speaker

Kaya Mori (Columbia University)

Description

The variability observed in stellar-mass black hole and neutron star binaries is highly complex and non-linear, exhibiting QPOs, stochastic and aperiodic components. Short-term variability, such as rapid flares, is commonly attributed to magnetic reconnection, plasmoid ejection and MHD turbulence, while long-term variability reflects changes in accretion states, jets, coronae, and their interactions. Decoding these non-linear variabilities can help us understand the dynamics of accretion disks, coronae, and relativistic jets, and potentially connect X-ray observations to GRMHD simulations. However, conventional Fourier-based methods, while powerful for detecting periodic signals in the frequency domain, have limitations in capturing the underlying non-linear evolution and state transitions. In this talk, I will introduce a new time-domain framework for accreting compact objects based on Koopman operator theory (KOT) and its modern, data-driven implementation, extended dynamic mode decomposition (EDMD). In this approach, the non-linear variability is embedded in a higher-dimensional linear space, where the Koopman operator decomposes light curve data into independently evolving dynamical modes with well-defined eigenvalues and eigenfunctions. KOT/EDMD is a well-developed yet actively evolving data-driven framework for analyzing and modeling complex nonlinear dynamical systems, with a growing range of real-world applications across science and engineering. I will present promising pilot-study applications of KOT/EDMD to X-ray lightcurve data of several BH and NS binaries, demonstrating its potential to provide new diagnostics for determining the physical processes that drive X-ray variability, and to predict future state transitions. By applying these modern time-domain analysis methods to multi-wavelength light-curve data, we may improve the interpretability and predictability of the many tones of accretion.

Author

Kaya Mori (Columbia University)

Co-authors

Mr Eric Miao (Columbia University) Prof. Reshmi Mukherjee (Barnard College) Dr Ruo-Yu Shang (Center for Computational Neuroscience at Flatiron institute)

Presentation materials