Probing unexplored spin-dependent dark matter-proton coupling with few-photoelectron threshold in COSINE-100

The COSINE-100 experiment reports world-leading constraints on spin-dependent dark matter-proton scattering in the 1.75–2.25 GeV/c² mass range and sets new sub-GeV limits via the Migdal effect by achieving a record-low detection threshold of 3–4 photoelectrons, despite observing no statistically significant annual modulation signal.

The Gist

The Big Picture: Hunting the Invisible Ghosts

Imagine the universe is filled with invisible "ghosts" called Dark Matter. We know they are there because they have gravity (they hold galaxies together), but they don't shine, reflect light, or talk to us normally. Scientists think these ghosts might occasionally bump into normal atoms, but it's incredibly rare.

The COSINE-100 experiment is like a giant, ultra-sensitive listening post buried deep underground in a mountain in South Korea. Its job is to listen for the faint "thud" of a Dark Matter ghost hitting a crystal.

The Problem: Too Much Static

For years, this listening post had a rule: "We only listen if the sound is loud enough to be heard over the background noise."

• The Old Rule: They only counted hits that created 8 "peaks" of light (photons).
• The Issue: Light Dark Matter ghosts are very weak. When they hit a crystal, they make a tiny, quiet "thud" that only creates 3 or 4 peaks of light.
• The Analogy: Imagine trying to hear a mouse squeak in a room where a vacuum cleaner is running. If you only listen for sounds louder than the vacuum, you'll never hear the mouse. The old rule meant they were missing the smallest, lightest ghosts.

The Breakthrough: Tuning the Microphone

This paper is about the team successfully turning down the volume on the "vacuum cleaner" (noise) so they could finally listen for the "mouse squeak" (the tiny Dark Matter hits).

They did this by:

1. Lowering the Threshold: They decided to listen for hits as small as 3 or 4 light peaks instead of 8.
2. Smart Filtering (The AI Detective): The problem with listening this quietly is that the detectors themselves make noise (like static on a radio). To fix this, they used a Machine Learning AI (a Multi-Layer Perceptron).
   • The Analogy: Imagine a bouncer at a club. The old bouncer just let anyone in with a loud voice. The new bouncer (the AI) looks at how the voice sounds. Is it a smooth, rhythmic beat (a real signal)? Or is it a jagged, chaotic screech (noise)? The AI learned to tell the difference between a real Dark Matter hit and a glitch in the machine.

The "Migdal Effect": The Domino Trick

The paper also used a clever trick called the Migdal effect to hunt for even lighter ghosts (ones lighter than a proton!).

• The Analogy: Imagine a heavy bowling ball (the nucleus) gets hit by a tiny pebble (Dark Matter). Usually, the bowling ball just wobbles a tiny bit. But sometimes, the wobble is so sudden that the tiny marbles sitting on top of the bowling ball (electrons) get flung off!
• Even if the Dark Matter is too light to move the heavy bowling ball enough to be seen, it might knock the tiny marbles off. This creates a bigger signal that the detector can see. This allowed them to hunt for Dark Matter that is 1,000 times lighter than what they could find before.

The Results: Silence is Golden

After listening carefully for years with their new, super-sensitive setup:

• Did they find the ghosts? No. They didn't hear any "mouse squeaks."
• Is that bad news? Actually, it's great news for science!
  • The Analogy: If you are looking for a specific type of rare bird in a forest, and you listen for a week and don't hear it, you haven't failed. You've proven that the bird doesn't live in that specific tree.
• What did they learn? They drew a new map of the forest. They proved that Dark Matter ghosts cannot be hiding in the "3 to 4 light peak" size range, specifically for masses between 1.75 and 2.25 GeV (a specific weight range). They also set the strictest rules yet for the super-light ghosts (15–58 MeV) using the Migdal trick.

Why This Matters

This experiment is a game-changer because it proved that NaI(Tl) crystals (the salt-like detectors they use) can be tuned to hear the faintest whispers in the universe.

• Before: They were looking for a shout.
• Now: They are listening for a whisper.

Even though they didn't find Dark Matter this time, they successfully closed the door on a whole new area of possibilities. This paves the way for the next generation of experiments (like COSINE-100U) which will be even more sensitive, potentially finally catching a glimpse of the universe's most elusive ghost.

Technical Summary

1. Problem Statement

Direct detection of Dark Matter (DM), specifically Weakly Interacting Massive Particles (WIMPs), faces significant challenges when searching for low-mass candidates (below a few GeV/$c^2$).

• Energy Threshold: Low-mass DM interactions produce nuclear recoils with energies below 1 keV. Traditional detectors often have energy thresholds too high to detect these faint signals.
• Noise Regime: Pushing detection thresholds down to the "few-photoelectron" (PE) regime (3–4 PE) introduces substantial background noise from photomultiplier tubes (PMTs), including phosphorescence (delayed light emission) and Cherenkov radiation, which can mimic genuine low-energy signals.
• Spin-Dependent (SD) Sensitivity: While many experiments focus on spin-independent interactions, there is a need for improved constraints on spin-dependent DM-proton couplings, particularly in the sub-GeV mass range where previous experiments have limited sensitivity.

2. Methodology

The COSINE-100 collaboration utilized data from their experiment located at the Yangyang Underground Laboratory (South Korea) to address these challenges.

• Experimental Setup:

  • Target: 106 kg of NaI(Tl) scintillating crystals. Sodium ($^{23}$Na) and Iodine ($^{127}$I) are proton-odd nuclei, making them ideal for probing spin-dependent interactions.
  • Shielding: The detector is shielded by 700 m of rock, a liquid scintillator veto, and layers of copper, lead, and plastic scintillators to suppress cosmic muons and environmental radiation.
  • Data Selection: The analysis focused on single-hit events (energy deposition in only one crystal) from a stable 4-year dataset (excluding the initial 2.4 years to avoid cosmogenic radionuclides). Only crystals 2, 3, 4, 6, and 7 were used due to noise or light yield issues in others.

• Pioneering Low-Threshold Analysis:

  • Threshold Reduction: Unlike previous COSINE-100 analyses that used an 8 PE threshold, this study lowered the trigger to 3 and 4 isolated peaks (reconstructed photoelectrons).
  • Event Definition: Events were defined by the number of clusters (NC) of isolated peaks in the waveform.
  • Signal Simulation: A detailed NaI(Tl) waveform simulation incorporating scintillation characteristics and single-PE pulse shapes was used to determine trigger efficiencies (62.5% for NC-3 and 81.0% for NC-4).

• Advanced Event Selection (Three-Step Process):
  To distinguish signal from the prevalent noise in the few-PE regime, a three-step selection was implemented:

  1. Deadtime Cut: Removed events following high-energy interactions to suppress phosphorescence. (Retained 97% signal, reduced noise by ~31.5% for NC-3).
  2. Cluster Charge Cut: Removed abnormally large single clusters caused by Cherenkov radiation in PMT glass. (Retained ~97% signal, reduced noise by ~10–15%).
  3. Machine Learning (MLP): A Multi-Layer Perceptron (MLP) trained on waveform kinematic variables (time differences between clusters, charge asymmetry between PMTs) was used to separate signal-like events (clusters within the 200 ns scintillation decay time) from noise-like events (clusters spread over wider windows due to phosphorescence).
     • Selection Criteria: Set to retain 50% of signal events, achieving ~91% (NC-3) and ~97% (NC-4) noise rejection.

• Analysis Strategy:

  • Annual Modulation Search: The team searched for the annual modulation signal predicted by the Standard Halo Model (SHM), fitting the event rate with a sinusoidal function (period = 365.25 days, phase = 152.5 days).
  • Background Modeling: Due to the complexity of the few-PE background, a phenomenological model was used, combining a constant baseline, an exponential decay term, and the sinusoidal modulation term.
  • Systematics: A sideband analysis (events just outside the signal region) was used to derive crystal-specific systematic uncertainties to account for rate fluctuations.
  • Migdal Effect: The analysis incorporated the Migdal effect, where atomic electrons are ionized/excited during a nuclear recoil, allowing sensitivity to even lower DM masses (sub-GeV).

3. Key Contributions

• First Few-PE Analysis in COSINE-100: Successfully implemented a detection threshold of 3 and 4 PE, a significant reduction from the previous 8 PE limit.
• Novel Noise Mitigation: Developed a robust, multi-layered event selection strategy combining deadtime cuts, charge cuts, and MLP-based machine learning to isolate genuine scintillation signals from PMT-induced noise in an unstudied regime.
• Phenomenological Background Modeling: Established a method to search for modulation signals in the few-PE regime where comprehensive physics-based background simulations are currently unfeasible.
• Migdal Effect Integration: Extended the sensitivity of NaI(Tl) detectors to the sub-GeV mass range by explicitly modeling the Migdal effect.

4. Results

• Annual Modulation: No statistically significant annual modulation signal was observed in either the NC-3 or NC-4 datasets. The modulation amplitudes were consistent with zero (e.g., $A = -0.002 \pm 0.026$ counts/kg/day for NC-3).
• New Constraints on SD DM-Proton Cross Section:
  • Standard Nuclear Recoils: The study set world-leading 90% Confidence Level (C.L.) upper limits on the spin-dependent DM-proton cross section in the 1.75–2.25 GeV/$c^2$ mass range, a region previously unexplored by other experiments.
  • Migdal Effect: By including the Migdal effect, the analysis established world-leading limits in the 15–58 MeV/$c^2$ mass range. This significantly extends the reach beyond the previous COSINE-100 3-year limit of 100 MeV/$c^2$ and surpasses limits set by Collar and XENON1T in this low-mass region.
• Comparison: The results are competitive with or superior to other leading experiments (PICO-60, PandaX-4T, CRESST, NEWS-G) in their respective mass ranges, particularly for spin-dependent interactions.

5. Significance

• Probing Unexplored Parameter Space: This work demonstrates that NaI(Tl) detectors, traditionally limited by higher thresholds, can effectively probe the sub-GeV dark matter parameter space when equipped with advanced noise rejection techniques.
• Validation of Low-Mass Searches: The successful isolation of signals in the few-PE regime validates the feasibility of using scintillators for low-mass DM searches, bridging a critical gap between heavy-target experiments and those using lighter targets.
• Foundation for Future Experiments: These results provide a critical foundation for the next-generation COSINE-100U experiment at Yemilab. The demonstrated capability to operate at low thresholds with high signal retention suggests that future upgrades with higher light yields will further enhance sensitivity to light dark matter.
• Methodological Advancement: The combination of machine learning and phenomenological modeling for background suppression in the few-PE regime offers a blueprint for future low-threshold dark matter experiments.