Constraints on sub-GeV dark matter scattering on electrons with COSINE-100

Using 2.82 years of data from the COSINE-100 experiment, researchers set the most stringent constraints to date on sub-GeV dark matter scattering with electrons in NaI(Tl) crystals, finding no excess events and establishing new upper limits on the scattering cross-section for both light and heavy vector boson mediators.

The Gist

The Invisible Needle in a Cosmic Haystack: COSINE-100's New Search

Imagine the universe is filled with a mysterious, invisible fog called Dark Matter. We know it's there because its gravity holds galaxies together, but we've never actually "seen" a single particle of it. For decades, scientists have been looking for these particles by assuming they are heavy, slow-moving, and bump into atomic nuclei (like a bowling ball hitting a pin). But what if the dark matter is actually light, fast, and tiny—like a swarm of invisible mosquitoes? And what if, instead of hitting the nucleus, these "mosquitoes" bump into the electrons orbiting the atom?

This is the story of the COSINE-100 experiment, a team of scientists in South Korea who built a giant, ultra-sensitive trap to catch these light dark matter particles. Here is how they did it, explained simply.

1. The Trap: A Crystal Ball in a Deep Cave

The scientists didn't use a net; they used crystals. Specifically, they used 8 large crystals of Sodium Iodide (NaI), which glow when hit by energy. You can think of these crystals as giant, super-sensitive fireflies.

• The Location: To keep out the "noise" of cosmic rays (particles raining down from space), they buried their experiment deep underground in a mountain in South Korea. It's like trying to hear a whisper in a library, but the library is 700 meters (about 2,300 feet) underground.
• The Goal: They wanted to see if a dark matter particle would hit an electron inside the crystal. If it did, the electron would get a tiny "kick," causing the crystal to flash with a very faint light.

2. The Challenge: Hearing a Pin Drop in a Storm

The problem with looking for light dark matter is that the "kick" it gives to an electron is incredibly small. It's like trying to hear a pin drop in the middle of a rock concert.

• The Noise: The crystals are constantly buzzing with tiny background signals from natural radioactivity and electronic noise.
• The Innovation: In the past, the experiment could only "hear" signals that were relatively loud (above 1 keV of energy). But light dark matter whispers so quietly that it might only produce a signal equivalent to 0.7 keV (a tiny fraction of a keV).
• The Solution: The team used a clever trick called Deep Learning (a type of artificial intelligence). They trained a computer to act like a super-smart bouncer. The computer learned to distinguish between the "noise" (static) and the "signal" (the real flash from a dark matter hit). This allowed them to lower their "hearing threshold" and listen for much quieter whispers than before.

3. The Search: Two Scenarios

The scientists didn't know exactly how the dark matter would interact, so they tested two main theories, like checking for two different types of invisible ghosts:

1. The "Heavy Mediator" (The Contact): Imagine the dark matter and the electron bumping into each other directly, like two billiard balls hitting.
2. The "Light Mediator" (The Long-Range): Imagine the dark matter and electron are connected by a long, invisible rubber band. The dark matter pulls on the electron from a distance.

They ran their experiment for 2.82 years, collecting data from 172.9 "kilogram-years" of crystal exposure. That's a lot of time and a lot of crystal watching!

4. The Result: No Ghosts Found (Yet)

After analyzing millions of data points, the result was clear: They found no extra flashes.

• The Verdict: The number of flashes they saw matched exactly what they expected from natural background noise. There was no "excess" that could be blamed on dark matter.
• The Silver Lining: Even though they didn't find the particle, they learned something very important. By not finding it, they were able to draw a line in the sand. They said, "If dark matter exists, it cannot be this heavy or interact this strongly."
  • They ruled out a huge range of possibilities for how dark matter might interact with electrons.
  • Their limits are now the strictest rules for this type of experiment using Sodium Iodide crystals.

5. Why This Matters: The "DAMA" Mystery

There is another experiment called DAMA/LIBRA that has claimed to see a signal of dark matter for years. However, other experiments haven't been able to confirm it. Some scientists think the DAMA signal might be caused by light dark matter hitting electrons.

The COSINE-100 team checked this specific possibility. Their results completely rule out the idea that the DAMA signal is caused by this specific type of light dark matter. It's like solving a mystery by proving the suspect couldn't have been at the scene.

6. What's Next? The "Super-Trap"

The team isn't giving up. They are already planning an upgrade called COSINE-100U.

• The Upgrade: They plan to make the crystals even better at collecting light and lower the temperature to make them more sensitive.
• The Goal: They want to lower the threshold even further, down to the level of a single "photoelectron" (the smallest possible unit of light).
• The Analogy: If the current experiment is a microphone that can hear a whisper, the new one will be a stethoscope that can hear a heartbeat from across the room. They also plan to look for a "seasonal rhythm" in the data (since Earth moves around the Sun, the "wind" of dark matter hitting us changes throughout the year), which is a different way to find the signal without needing to know exactly what the background noise looks like.

Summary

The COSINE-100 experiment acted like a high-tech, ultra-sensitive ear, listening for the faintest whisper of light dark matter in a deep underground cave. They used AI to filter out the noise and listened for nearly three years. While they didn't find the dark matter particle, they successfully proved that it doesn't behave in the specific ways they tested. This clears the path for future, even more sensitive experiments to continue the hunt for the universe's most elusive ghost.

Technical Summary

1. Problem Statement

The existence of Cold Dark Matter (CDM) is well-supported by astrophysical observations, yet the particle nature of Dark Matter (DM) remains unknown. While Weakly Interacting Massive Particles (WIMPs) interacting with nucleons have been the primary focus of direct detection, the lack of positive signals and increasingly stringent constraints have shifted attention toward sub-GeV DM candidates.

A promising alternative involves DM-electron scattering, where light DM particles (mass $< 1$ GeV) transfer enough energy to ionize atomic electrons. This interaction requires detectors with very low energy thresholds. While silicon-based (e.g., SENSEI, DAMIC-M) and liquid noble gas experiments (e.g., XENONnT, PandaX-4T) have set strong limits, NaI(Tl) scintillators had not previously published a dedicated search for this specific interaction model, despite their relevance to the long-standing DAMA/LIBRA annual modulation anomaly. This paper addresses the gap by searching for sub-GeV DM scattering on electrons using the COSINE-100 experiment's NaI(Tl) crystals.

2. Methodology

Experimental Setup

• Detector: The COSINE-100 experiment, located at the YangYang underground laboratory in South Korea (700 m water equivalent depth).
• Target: 8 ultrapure NaI(Tl) crystals (total 106 kg). For low-energy analysis, only 5 crystals (61.3 kg effective mass) were used due to superior light yield and lower noise.
• Exposure: A dataset spanning 2.82 years, totaling 172.9 kg-year.
• Threshold Reduction: The analysis utilized a reduced energy threshold of 0.7 keV (8 photoelectrons), achieved via advanced noise rejection techniques, significantly improving sensitivity to sub-GeV DM.

Data Analysis & Signal Processing

• Event Selection:
  • Focused on single-hit events (DM interactions are unlikely to trigger multiple crystals).
  • Employed a Multilayer Perceptron (MLP) deep learning algorithm to discriminate between PMT-induced noise and genuine scintillation signals. This allowed the threshold to be lowered from 1.0 keV to 0.7 keV while maintaining high efficiency.
  • Applied cuts to remove muon-induced events (using plastic scintillator vetoes) and electronic noise.
• Background Modeling:
  • Constructed a comprehensive background model based on Geant-4 simulations and fits to data above 6 keV.
  • Dominant background sources near the threshold (0.7–2 keV) include $^3$H $\beta$-decay and $^{210}$Pb (Compton scattering and $\beta$-decay), accounting for ~75% of events.
  • Systematic uncertainties (event selection efficiency, source location, energy resolution parameters) were treated as nuisance parameters in the fit.
• Theoretical Modeling:
  • Mediators: Two benchmark scenarios were studied: a heavy vector boson (contact interaction) and a light vector boson (long-range interaction).
  • Ionization Factors ($f_{ion}$): Crucially, the authors utilized a relativistic calculation of atomic ionization factors for Sodium and Iodine. This is essential for the keV energy range, as non-relativistic approximations underestimate the event rate by up to an order of magnitude near the threshold.
  • Halo Model: Assumed the Standard Halo Model for DM velocity distribution.

Statistical Analysis

• A Bayesian analysis using Markov Chain Monte Carlo (MCMC) was performed to fit the observed energy spectrum (0.7 keV to ~7.8 keV).
• The fit included the expected DM signal spectra for various DM masses (30 MeV to 5 GeV) and background components.
• No excess events were found; 90% Confidence Level (CL) upper limits on the DM-electron scattering cross-section ($\sigma_e$) were derived.

3. Key Contributions

1. First Dedicated NaI(Tl) Search: This is the first published search for sub-GeV DM-electron scattering specifically using NaI(Tl) crystals, filling a critical gap in the global experimental landscape.
2. Relativistic Ionization Modeling: The paper emphasizes the necessity of using relativistic Hartree-Fock approximations for calculating ionization factors in NaI(Tl) at keV energies. It demonstrates that non-relativistic models significantly underestimate sensitivity in this regime.
3. Threshold Optimization: Successfully demonstrated the capability to analyze data down to 0.7 keV (8 photoelectrons) in NaI(Tl) using machine learning, overcoming previous noise limitations.
4. DAMA/LIBRA Constraints: The results provide the most stringent constraints to date for NaI(Tl) targets, directly testing the parameter space suggested by the DAMA/LIBRA annual modulation anomaly under the DM-electron scattering hypothesis.

4. Results

• Observation: No statistically significant excess of events over the expected background was observed in the 2.82-year dataset.
• Exclusion Limits (90% CL):
  • Light Mediator: Excludes cross-sections above $6.4 \times 10^{-33} \text{ cm}^2$ for a DM mass of 0.25 GeV.
  • Heavy Mediator: Excludes cross-sections above $3.4 \times 10^{-37} \text{ cm}^2$ for a DM mass of 0.4 GeV.
• Impact on DAMA/LIBRA: The derived limits completely exclude the best-fit regions for the DAMA/LIBRA modulation signal if interpreted as DM-electron scattering mediated by a vector boson (both light and heavy).
• Comparison: While silicon and liquid noble gas experiments currently hold tighter limits for very low masses (<50 MeV) due to lower thresholds, COSINE-100 provides the strongest constraints specifically for the NaI(Tl) target material.

5. Significance and Future Prospects

• Validation of NaI(Tl) Sensitivity: The study proves that NaI(Tl) crystals, traditionally used for WIMP-nucleon searches, are viable for sub-GeV DM-electron searches when thresholds are sufficiently low and relativistic effects are accounted for.
• Resolution of DAMA Anomaly: By ruling out the DM-electron scattering interpretation of the DAMA/LIBRA signal for vector mediators, the paper narrows the possible explanations for the DAMA anomaly, suggesting that if the signal is real, it likely arises from a different interaction mechanism or background.
• Future Outlook (COSINE-100U): The paper outlines plans for the upgraded COSINE-100U experiment. By improving light yield (via new encapsulation and cryogenic operation at -33°C), the threshold could be lowered to ~0.2 keV (5 photoelectrons).
  • This would enhance sensitivity to DM masses in the 10–50 MeV range, potentially reaching cross-sections below $6 \times 10^{-39} \text{ cm}^2$.
  • A future analysis using annual modulation techniques below the current threshold is proposed to further mitigate background uncertainties.

In conclusion, this work establishes the most stringent constraints on sub-GeV DM-electron scattering for NaI(Tl) targets to date, effectively closing the parameter space for this specific interaction model as an explanation for the DAMA/LIBRA signal, while paving the way for even more sensitive searches in the upcoming COSINE-100U era.