TL;DR
Researchers have developed DarkRayNet, a neural network emulator that predicts cosmic-ray antideuteron fluxes from dark matter annihilation in seconds rather than days. By accounting for complex uncertainties in particle production and galactic travel, the tool reveals that upcoming experiments like GAPS are perfectly positioned to detect dark matter in mass ranges where current data is most sensitive.
What Problem Does This Paper Address?
Searching for dark matter (DM) is like looking for a needle in a haystack, but cosmic-ray antideuterons ($\bar{d}$) might be a "smoking gun." Unlike antiprotons or positrons, antideuterons produced by standard astrophysical processes (like cosmic rays hitting interstellar gas) are extremely rare at low energies. If we see them there, it’s a strong hint of DM annihilation.
However, predicting exactly what this signal should look like is a computational nightmare. It involves:
- Coalescence: Simulating how an antiproton and antineutron "stick" together to form a nucleus.
- Propagation: Solving complex diffusion equations to model how these particles travel through the Galaxy's magnetic fields and solar winds.
Traditional Monte Carlo simulations and diffusion solvers (like GALPROP) are too slow for the massive parameter scans needed to account for theoretical uncertainties. DarkRayNet bridges this gap by emulating these processes using machine learning.
Methods
The authors built a tool-chain centered around a Recurrent Neural Network (RNN).
- Why an RNN? Energy spectra are sequential data. RNNs are designed to handle sequences, allowing the model to "understand" the correlation between neighboring energy bins, which leads to much smoother and more accurate spectral predictions than standard dense networks.
- Input/Output: The network takes DM mass, annihilation channels (like $b\bar{b}$ or $W^+W^-$), coalescence parameters, and galactic propagation variables as inputs. It outputs the predicted energy spectrum at Earth.
- Physics Updates: They didn't just automate old models; they improved the underlying physics. They included a rescaling factor for $\Lambda_b$ baryons (which significantly boost antideuteron production) and implemented a spatial separation constraint in the coalescence model—ensuring the constituent particles are actually close enough in space to bind.
- Marginalization: They integrated the emulator with
pbarlike, a tool that uses existing AMS-02 antiproton data to constrain the allowed DM models, ensuring their antideuteron predictions remain consistent with what we already know.
Datasets
- Training Data: Generated using a combination of MadDM 3 and Pythia 8.2 for the initial particle production (source spectra) and GALPROP v56 for the journey through space.
- Calibration Data: They used historical data from the ALEPH experiment at LEP ($e^+e^-$ collisions) to fit the "coalescence momentum" ($p_c$), which determines the probability of antideuteron formation.
- Experimental Benchmarks: The study evaluates the sensitivity of three major experiments: AMS-02 (currently on the ISS), the upcoming GAPS (balloon-based), and the future AMS-100 mission.
Key Insights and Ideas
- The $\Lambda_b$ Shoulder: A major finding is the importance of $\Lambda_b$ baryons. Because they decay into antiprotons and antineutrons with very low relative momentum, they are "antideuteron factories." Standard simulations often underestimate this, so the authors introduced a rescaling factor ($r_{\Lambda_b}$) to match experimental data.
- Speed vs. Accuracy: DarkRayNet achieves a relative error of only about 1% compared to full simulations but runs in a fraction of a second. This allows for "on-the-fly" statistical analysis that was previously impossible.
- Complementarity is Key: Antideuterons are most useful when paired with antiproton limits. The paper shows that even if antiprotons don't show a clear signal (due to high backgrounds), antideuterons could still provide a definitive discovery, particularly for DM masses below 20 GeV.
Key Results
- Coalescence Momentum: The authors derived a best-fit value of $p_c = 210^{+27}_{-25}$ MeV from LEP data.
- AMS-02 Sensitivity: When constrained by antiproton limits, AMS-02 is sensitive to DM signals primarily in two windows: below 20 GeV and between 100–200 GeV.
- GAPS Potential: The GAPS experiment is shown to be a powerful independent probe for the sub-20 GeV region, potentially confirming tentative events reported by the AMS-02 collaboration.
- Future Outlook: The future AMS-100 mission will be a game-changer, with sensitivity more than three orders of magnitude better than current experiments, potentially probing DM masses up to the TeV scale.
Limitations and Future Work
- Solar Modulation: The study uses the "force-field approximation" for how the Sun's magnetic field affects incoming cosmic rays. While standard, this is a simplification of a very dynamic process.
- Backgrounds for AMS-100: For the ultra-sensitive AMS-100, the "background" from standard astrophysical processes (secondary antideuterons) will finally become significant. Future work will need to model these backgrounds with the same level of precision as the DM signal.
- Coalescence Physics: The model still relies on a simplified momentum-based coalescence. Moving toward more advanced quantum mechanical wave-function models is a suggested next step for the community.
Availability: The DarkRayNet package is open-source and available on GitHub.
Summary updated