It is now more than a decade since an excess emission in GeV gamma rays in the Galactic centre region has been detected after accounting for well known astrophysical backgrounds from cosmic rays interacting with the interstellar medium. While there was a plethora of attempts over time to unravel the nature of this Galactic Center Excess (GCE), they largely converged to contrast two particular interpretations: a population of faint and unresolved millisecond pulsars and dark matter annihilation. Despite convincing arguments in favour of the conventional astrophysical explanation, it eludes a robust observational confirmation given the available gamma-ray data.
In this talk, we report on our approach based on Bayesian neural networks to perform a detailed gamma-ray analysis of this region in an attempt to address the origin of the GCE. Using simulations of the gamma-ray sky in the Galactic centre region, the network is trained to separate the detected gamma rays into components based on templates of background and GCE emission, which is modelled as a combination of dark annihilation and faint millisecond pulsars. Imperfections in our background model are inspected visually and the model complexity increased iteratively after applying the network to the data.
We confront the performance of the network with a more traditional maximum likelihood fitting approach and demonstrate that the network's prediction for the background templates is comparable to the likelihood method, while having an added advantage of accounting for a millisecond pulsar template in a self-consistent way.
In terms of the GCE, we find that the network is capable of detecting the sub-threshold pulsar population, while the dark matter contribution to the GCE is largely degenerate with other faint components in the region (notably the possible low-latitude part of the Fermi bubbles) and therefore - at the moment - remains challenging to distinguish.
|Collaboration / Activity||CNRS, LAPTh|
|First author||Christopher Eckner|