New method for Gamma/Hadron separation in HAWC using neural networks

Abstract

The High Altitude Water Cherenkov (HAWC) gamma-ray observatory is located at an altitude of 4100 meters in Sierra Negra, Puebla, Mexico. HAWC is an air shower array of 300 water Cherenkov detectors (WCD’s), each with 4 photomultiplier tubes (PMTs). Because the observatory is sensitive to air showers produced by cosmic rays and gamma rays, one of the main tasks in the analysis of gamma-ray sources is gamma/hadron separation for the suppression of the cosmic-ray background. Currently, HAWC uses a method called compactness for the separation. This method divides the data into 10 bins that depend on the number of PMTs in each event, and each bin has its own value cut. In this work we present a new method which depends continuously on the number of PMTs in the event instead of binning, and therefore uses a single cut for gamma/hadron separation. The method uses a Feedforward Multilayer Perceptron net (MLP) fed with five characteristics of the air shower to create a single output value. We used simulated cosmic-ray and gamma-ray events to find the optimal cut and then applied the technique to data from the Crab Nebula. This new method is tuned on MC and predicts better gamma/hadron separation than the existing one. Preliminary tests on the Crab data are consistent with such an improvement, but in future work it needs to be compared with the full implementation of compactness with selection criteria tuned for each of the data bins.