Systematics in the XENON1T data: The 15-keV anti-axion

The XENON1T collaboration Aprile et al. [2020] has found an excess of electron recoil events in their Science Run 1 data below ∼7 keV with a spectral shape consistent with that expected from a solar-axion-induced signal. The claimed statistical significance of the solar-axion model over the null hypothesis is 3.5 σ. In this work we provide suggestive evidence for mismodeling in the electron recoil data that may decrease the local significance of the axion model. To reach this conclusion, we search for a signal with the spectral template of the solar axion model, but shifted to higher (unphysical) energies above ∼7 keV. We find that the distribution of significances found from this side-band analysis does not follow the expected chi-square distribution. For example, we find a high-significance feature in the data, with a global p-value p ≈ 0 . 026 , when the solar axion model is shifted upwards in energy by ∼15 keV and allowed to have a negative normalization. We argue that such statistical tests, if performed a priori, provide a data-driven way to test and potentially account for systematic uncertainties on the background model in low-threshold dark matter experiments.