A damped cycle model fitted before 2020 lowers ten-day forecast error in 2020–2026 MAXI data for Cygnus X-2

The known long cycles of Cygnus X-2 can help predict later MAXI measurements. We fit daily data before 2020. All model parameters stay fixed in the later test. The test has 1500 daily MAXI 2–20 keV measurements from 2020-01-23 to 2026-10-03. We compare Gaussian process models with ten-day forecasts. A model with a damped cycle reduces mean squared error by 10.72% against one Ornstein–Uhlenbeck process. The conditional 95% interval is 4.38% to 15.41% with 200 day loss blocks. This interval holds the training fits fixed. The reduction is 12.18% against a coherent cycle with an ordinary decay term. The reduction is 11.20% against two decay terms with no cycle. The fitted cycle period is 53.42 d. Its decay time is 28.13 d. These are model parameters. We do not estimate a physical period error. Three MAXI sensitivity checks also show a gain. Swift BAT 15–50 keV data show no comparable gain. The result supports a useful forecast model for the tested soft X-ray data. It does not identify the physical cause of the long cycles. The package preserves source data, hashes, code, forecasts, figures, and reproduction steps.

Cygnus X-2 · MAXI · Swift BAT · X-ray binaries · forecast validation · Gaussian processes · superorbital variability · reproducible research