Reported cycles increase mean flux forecast error in an 11-source gamma-ray test with Gaussian OU models and uniform 30-day averages
We tested whether 30-day bin averaging changes later cycle forecasts for 11 reported gamma-ray candidates. We reused a signed NASA Fermi LAT data snapshot and a prior linear-flux comparison. Each Gaussian Ornstein-Uhlenbeck (OU) rule fitted bins ending by 2017-11-01 UTC. It then predicted selected bins ending by 2026-10-01 UTC without updates. We compared constant means with added sine curves at the reported periods. With uniform bin averages, the mean relative increase in mean squared error was 14.17% for fixed spectral slopes and 10.35% for free slopes. Errors rose for 10 of 11 and 9 of 11 sources. The mean score changes from point models were -0.70 and -0.90 percentage points. These small changes were concentrated in S4 1144+40. A lower time-scale bound changed the free-slope difference interval to include zero. The mean cycle score stayed negative in all tested settings. Averaging lowered the noise-only fitted OU time scale in 20 of 22 fits. The median averaged/point ratio was about 0.77 in each product. In 100 simulated records with a true 30-day time scale, the median fits were 46.9 days for points and 29.3 days for averages. We preserve data, code, paired predictions, figures, and independent numerical checks. A clean offline run reproduced the numerical results. The covariance is exact only for a Gaussian OU process with uniform time weighting. Actual LAT flux estimates use uneven exposure. The result concerns a fixed sine prediction rule and selected detections. It does not exclude transient cycles or binary black holes.
Fermi LAT · gamma rays · blazars · time averages · Ornstein-Uhlenbeck process · correlated noise · forecast comparison · public data · reproducible research
References
6- usesReported cycles increase 2017–2026 log-flux forecast error against OU noise in fixed-index data for 10 of 11 selected gamma-ray sources sci:sha256:39860ae189b8ba5f4102844896a62d900b370b33a31a91807baadc8b63ca475f
We reuse its signed data snapshot, 11-source selection, source matches, periods, quality cuts, and linear-flux comparison. We reproduce its point predictions and add uniformly averaged covariance.
- usesdoi:10.3847/1538-4357/ab910d
Table 1 supplies all 11 candidates and reported periods. The current LCR data use a different energy band and time-bin width.
- usesdoi:10.3847/1538-4365/acbb6a
The Fermi LAT Light Curve Repository supplies the public monthly photon-flux estimates. The package preserves exact NASA responses and provenance.
- doi:10.1093/mnras/stag1141
This work tests longer gamma-ray records, period persistence, and later peak predictions. Our test compares frozen predictions with correlated noise and changes the time-bin model.
- doi:10.3847/1538-3881/aa9332
This work describes the celerite Gaussian process framework. The later celerite2 TermConvolution documentation shows that boxcar integration is an established method.
- doi:10.1088/0004-637X/698/1/895
This work gives an earlier astronomical use of the OU model. We use OU covariance as a statistical reference for gamma-ray flux.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.