DarkRayNet: emulation of cosmic-ray antideuteron fluxes from dark matter

Jan Heisig, Michael Korsmeier, Michael Krämer, Kathrin Nippel, Lena Rathmann

TL;DR

This paper introduces DarkRayNet, a novel neural network emulator designed to rapidly predict cosmic-ray antideuteron fluxes from dark matter annihilation. By accounting for uncertainties in antideuteron production and propagation, DarkRayNet enables faster and more comprehensive searches for dark matter, with the potential to probe specific dark matter mass windows with upcoming experiments like GAPS.

What Problem Does This Paper Address?

The nature of dark matter (DM) remains one of the most significant mysteries in physics. Indirect detection, particularly through the observation of cosmic-ray antimatter, offers a promising avenue to probe DM annihilation. Antideuterons, due to their suppressed astrophysical background at low energies, are considered a "smoking gun" signature for DM. However, accurately predicting these fluxes is computationally intensive, involving complex simulations of DM annihilation, particle showering, and cosmic-ray propagation through the galaxy. Furthermore, significant uncertainties exist in these processes, particularly in the formation of antideuterons from antiprotons and antineutrons (coalescence) and in the models describing how cosmic rays travel through the galaxy. These uncertainties make it challenging to interpret experimental data and set robust limits on DM properties. This paper addresses this by developing a fast and accurate emulation tool that can handle these complexities and uncertainties, facilitating more efficient DM searches.

Methods

The core of this work is the development of DarkRayNet, a recurrent neural network (RNN) designed to emulate the complex process of antideuteron flux prediction. The authors employ a two-pronged approach:

  • Monte Carlo Simulations: They first use detailed Monte Carlo simulations (MadDM 3 and PYTHIA 8.2) to generate the initial antideuteron source spectra from DM annihilation. This process incorporates uncertainties in the coalescence model (e.g., the coalescence momentum $p_c$ and the $\Lambda_b$ baryon production rate) and considers various DM annihilation channels.
  • Neural Network Emulation: The simulated spectra, along with propagation parameters from two distinct models (INJ.BRK and DIFF.BRK), are used to train DarkRayNet. The network is structured to process input parameters (DM mass, branching fractions, coalescence parameters, and propagation parameters) and output the propagated antideuteron flux. The use of an RNN is particularly suited for handling the correlated energy bins of cosmic-ray spectra.
  • Marginalization: To account for uncertainties in propagation parameters, the authors implement a marginalization technique. This involves using a posterior distribution of propagation parameters and weighting the predicted fluxes accordingly, allowing for a more robust estimation of the DM signal's sensitivity.

Datasets

The training dataset for DarkRayNet is generated from a comprehensive set of Monte Carlo simulations. These simulations cover:

  • DM Parameters: A wide range of DM masses (2 GeV to 10 TeV) and branching fractions for various annihilation channels (qq, cc, bb, tt, gg, hh, W+W⁻, ZZ).
  • Coalescence Parameters: The coalescence momentum ($p_c$) and the $\Lambda_b$ production rate ($r_{\Lambda_b}$) are sampled within their relevant uncertainty ranges.
  • Propagation Parameters: Parameters from two established propagation models (INJ.BRK and DIFF.BRK) are sampled from fits to cosmic-ray proton and helium spectra, incorporating uncertainties derived from AMS-02 data.

The evaluation of the model's performance is done using a held-out portion of this simulated dataset. Experimental data from AMS-02 antiproton measurements are used in the marginalization process to constrain DM models.

Key Insights and Ideas

  • Antideuterons as a "Smoking Gun": The paper strongly reiterates that low-energy antideuterons are a highly sensitive and unique signature for DM annihilation due to their suppressed astrophysical background.
  • DarkRayNet's Efficiency: The development of DarkRayNet is a crucial step towards making DM indirect detection analyses computationally feasible, especially when dealing with numerous DM models and propagation scenarios. It drastically reduces computation time from hours to seconds.
  • Uncertainty Quantification is Key: The authors emphasize the importance of accounting for uncertainties in both the production (coalescence, $\Lambda_b$ rate) and propagation (two distinct models) of antideuterons. These uncertainties significantly impact the predicted fluxes and the resulting DM constraints.
  • Complementarity of Experiments: The study highlights how different experiments (AMS-02, GAPS, AMS-100) probe different energy ranges and thus offer complementary sensitivity to DM signals, particularly in specific mass windows.
  • Antiproton Limits are Crucial but Not Definitive: While antiproton limits are valuable for constraining DM models, they are not always sufficient to rule out DM signals, especially in certain mass ranges or for specific DM models. Antideuteron searches provide an independent and complementary probe.

Key Results

  • DarkRayNet Performance: The neural network emulator demonstrates high accuracy, with a median relative deviation of transformed fluxes below $6 \times 10^{-4}$ and a resulting relative error of $\mathcal{O}(10^{-2})$ in the actual flux.
  • DM Sensitivity Windows: For the bb annihilation channel, AMS-02 shows sensitivity to DM masses below 10 GeV, and in a small window between 10-20 GeV (for DIFF.BRK). For the Higgs portal model, sensitivity extends to higher masses (above 200 GeV).
  • GAPS and AMS-100 Potential: GAPS is shown to be competitive with AMS-02 for low-energy antideuterons, while AMS-100 offers significantly superior sensitivity, potentially probing DM masses up to the TeV scale.
  • Impact of $\Lambda_b$ Production: The uncertainty in the $\Lambda_b$ baryon production rate has a noticeable impact on the predicted antideuteron flux, particularly for small DM masses, underscoring the need for precise measurements of this process.

Limitations and Future Work

  • Coalescence Model: While the authors use a well-calibrated coalescence model, it is still a phenomenological model with inherent uncertainties, particularly regarding the $p_c$ parameter.
  • Propagation Model Simplifications: The use of two specific propagation models (INJ.BRK and DIFF.BRK) and a simplified halo size ($z_h=4$ kpc) represents a conservative choice, and further exploration with more complex models could refine the results.
  • Secondary Antideuteron Background: While the focus is on DM signals, the paper acknowledges that for AMS-100, the secondary antideuteron background becomes significant at higher DM masses, requiring careful consideration in future analyses.
  • Future Work: The authors suggest that DarkRayNet is a powerful tool for performing extensive DM parameter scans and for future interpretations of experimental data. Further work could involve extending the emulator to other cosmic-ray species or incorporating more sophisticated propagation models. The availability of the DarkRayNet code on GitHub facilitates such future developments.

Summary updated

Abstract

Abstract Cosmic-ray antimatter, particularly low-energy antideuterons, serves as a sensitive probe of dark matter annihilating in our Galaxy. We study this smoking-gun signature and explore its complementarity with indirect dark matter searches using cosmic-ray antiprotons. To this end, we develop the neural network emulator D̅ arkRayNet , enabling a fast prediction of propagated antideuteron energy spectra for a wide range of annihilation channels and their combinations. We revisit the Monte Carlo simulation of antideuteron coalescence and cosmic-ray propagation, allowing us to explore the uncertainties of both processes. In particular, we take into account uncertainties from the Λ b production rate and consider two distinctly different propagation models. Requiring consistency with cosmic-ray antiproton limits, we find that AMS-02 shows sensitivity to a few windows of dark matter masses only, most prominently below 20 GeV. This region can be probed independently by the upcoming GAPS experiment. The program package D̅ arkRayNet is available on GitHub, https://github.com/kathrinnp/DarkRayNet .
BibTeX
@article{heisig2024darkraynet,
  author = {Jan Heisig and Michael Korsmeier and Michael Krämer and Kathrin Nippel and Lena Rathmann},
  title = {{DarkRayNet: emulation of cosmic-ray antideuteron fluxes from dark matter}},
  journal = {Journal of Cosmology and Astroparticle Physics},
  year = {2024},
  doi = {10.1088/1475-7516/2024/11/017},
}
Figure