# Investigating the dark matter signal in the cosmic ray antiproton flux with the machine learning method

## TL;DR

This study uses machine learning to analyze cosmic ray antiproton data, revealing that the detection of dark matter signals strongly depends on the interaction and propagation models, with some models indicating potential signals around 100 GeV to 1 TeV.

## Contribution

It introduces a machine learning approach to efficiently explore complex parameter spaces in dark matter signal analysis from cosmic ray data.

## Key findings

- Most models show no significant dark matter signal above 2σ.
- The EPOS-LHC model suggests a >3σ signal at ~1 TeV.
- A significant signal at ~100 GeV is found in the diffusive reacceleration model, but diminishes with charge-dependent solar modulation.

## Abstract

We investigate the implications on the dark matter (DM) signal from the AMS-02 cosmic antiproton flux. Global fits to the data are performed under different propagation and hadronic interaction models. The uncertainties from the injection spectrum, propagation effects and solar modulation of the cosmic rays are taken into account comprehensively. Since we need to investigate extended parameter regions with multiple free parameters in the fit, the machine learning method is adopted to maintain a realistic time cost. We find all the effects considered in the fitting process interplay with each other, among which the hadronic interaction model is the most important factor affecting the result. In most hadronic interaction and CR propagation models no DM signal is found with significance larger than $2\sigma$ except that the EPOS-LHC interaction model requires a more than $3\sigma$ DM signal with DM mass around $1\,\mathrm{TeV}$. For the diffusive reacceleration propagation model there is a highly significant DM signal with mass around $100\,\mathrm{GeV}$. However, the signal becomes less than $1\sigma$ if we take a charge dependent solar modulation potential in the analysis.