A statistically robust framework for detecting and classifying hysteresis patterns in astrophysical spectral evolution

Abstract

Loop-like patterns between spectral parameters are frequently interpreted as evidence of hysteresis in time-dependent astrophysical emission processes. Such patterns have been reported in hardness-intensity diagrams (HID) of accreting black-hole X-ray binaries during state transitions, in the radio-to-X-ray correlation plane during outbursts, in solar activity indices over the solar cycle, and in the spectral energy distribution of active galactic nuclei during flaring episodes. HID often exhibit apparent loops, whose orientation encodes the relative timescales of particle acceleration and radiative cooling. Visual inspection, however, does not provide a statistically controlled detection method. We develop a statistically robust and empirically calibrated framework for detecting, quantifying, and classifying hysteresis patterns in ordered two-dimensional data with measurement uncertainties. The framework provides the normalised signed area $A_\mathrm{norm}$ enclosed by chronologically ordered points in the plane as the primary detection statistic, computed using the shoelace formula. We define open and closed area estimators, introduce cancellation diagnostics for multi-loop structures, and propagate measurement uncertainties via Monte Carlo sampling. Statistical significance is assessed using null ensembles generated by time-order randomization, physically motivated autoregressive surrogate models, and Fourier phase-randomized surrogates. We validate the method on synthetic blazar flare trajectories, and demonstrate its application to an XMM-Newton observation of Markarian~421 during a December 2023 flaring episode, where we confirm a CCW hysteresis loop with $A_\mathrm{norm} = +0.64$ that is robust against measurement noise but does not reach formal significance against stochastic null models, possibly due to the open trajectory geometry.