Improving Sensitivity to Weak Pulsations with Photon Probability Weighting

All gamma-ray telescopes suffer from source confusion due to their inability to focus incident high-energy radiation, and the resulting background contamination can obscure the periodic emission from faint pulsars. In the context of the Fermi Large Area Telescope, we outline enhanced statistical tests for pulsation in which each photon is weighted by its probability to have originated from the candidate pulsar. The probabilities are calculated using the instrument response function and a full spectral model, enabling powerful background rejection. With Monte Carlo methods, we demonstrate that the new tests increase the sensitivity to pulsars by more than 50% under a wide range of conditions. This improvement may appreciably increase the completeness of the sample of radio-loud gamma-ray pulsars. Finally, we derive the asymptotic null distribution for the H-test, expanding its domain of validity to arbitrarily complex light curves.

💡 Research Summary

The paper addresses a fundamental limitation of high‑energy γ‑ray telescopes: because they cannot focus photons, source confusion and background contamination often drown out the periodic signals from faint pulsars. Focusing on the Fermi Large Area Telescope (LAT), the authors develop enhanced statistical tests for pulsation that assign each photon a weight equal to the probability that it originated from the candidate pulsar. These probabilities are derived from the instrument response functions (IRFs) together with a full spectral model of both the pulsar and all background components (Galactic diffuse, isotropic, and nearby point sources).

The authors begin by reviewing the classic Z²ₘ test and the H‑test, both of which use only the photon arrival phase. They show that when the photon weights wᵢ are set to the source‑origin probabilities, the test statistic Q = 2 T ∑ₖ(α̂ₖ²+β̂ₖ²) becomes a score test that is locally most powerful. The weighted Z²ₘ statistic is defined as

 Z²ₘ,w = (2 / ∑ᵢ wᵢ²) ∑ₖ=1^m (α̂ₖ²+β̂ₖ²)

and retains the χ²(2m) null distribution, provided the sample is large enough (≥ 50 photons for 3σ significance). The H‑test, which automatically selects the optimal number of harmonics, is similarly generalized to a weighted form H_w = max_i