Achieving a 0.49 keV threshold in the COSINE-100 dark matter experiment

Deep in the universe, a vast, invisible substance called dark matter is thought to hold galaxies together. It makes up about a quarter of everything that exists, yet it does not emit light, heat, or any signal we can easily catch. Scientists have spent decades trying to find it, proposing that it might be made of heavy, slow-moving particles that occasionally bump into ordinary atoms. When such a collision happens, it should leave a tiny flash of light, a whisper of energy that sensitive detectors can hear. The challenge is that these whispers are incredibly faint and easily drowned out by the constant static of the natural world, from radioactive rocks to cosmic rays. To hear them, researchers must build detectors deep underground, shielded by miles of rock, and listen with instruments so precise they can distinguish a single spark of light from the noise.

In a quiet valley in South Korea, a team of scientists has been listening for these whispers using a large array of crystals made of sodium iodide, a material that glows when struck by a particle. This experiment, known as COSINE-100, has been running for six years, collecting a massive amount of data. For a long time, the team could only trust signals that were strong enough to rise clearly above the background noise, which meant they missed the faintest, most elusive whispers. However, a new analysis of their full six-year dataset has changed the game. By using a sophisticated computer program to learn the difference between a real particle hit and a random glitch, the researchers have lowered their listening threshold to a level of 0.49 keV. This is a unit of energy so small that it represents the faintest signal the detector can reliably identify, allowing the team to hear the quietest possible interactions.

The key to this achievement was teaching a computer to recognize the unique shape of a light flash caused by a particle. In the past, the team relied on simpler methods that struggled when the signals were weak, often mistaking them for electronic noise or missing them entirely. For this new study, the researchers created millions of computer simulations of what a real particle hit should look like. They then trained a neural network, a type of artificial intelligence, to compare the actual data from the crystals against these simulations. The computer learned to spot subtle patterns in the way the light pulses rose and fell, distinguishing the slow, smooth decay of a real event from the sharp, erratic spikes of noise. This training was done separately for different groups of events, depending on how many distinct peaks of light appeared in the signal, ensuring the computer was as accurate as possible across the entire range of energies.

The result is a filter that is both sharp and sensitive. It keeps more than ten percent of the real particle signals at the new, ultra-low threshold while rejecting ninety-nine percent of the noise. To prove this method works, the team checked their results against known sources of light within the detector, such as radioactive atoms naturally present in the crystals. They confirmed that the computer was correctly identifying real events and that its performance remained steady over the entire six years of data collection. This stability is crucial, as it means the team can trust the data they are seeing, knowing that the detector has not drifted or changed its behavior over time.

With this new ability to hear fainter signals, the researchers have significantly improved their chances of finding dark matter, particularly the lighter versions of the hypothetical particles. Lighter particles move slower and create smaller flashes of energy, so they were previously hidden below the detection limit. Now, the team can look for them in a mass range between 2.0 and 2.8 GeV/c², a region of the universe that has remained largely unexplored by other experiments. Their analysis suggests that if these light dark matter particles exist and interact with protons in a specific way, the COSINE-100 detector should be able to see them. While the team has not yet found a definitive signal, their improved sensitivity allows them to set the strictest limits yet on where these particles might be hiding, effectively ruling out a wide swath of possibilities that other experiments could not reach.

This work also helps settle a long-standing debate in the field. Another experiment, DAMA/LIBRA, has reported seeing a yearly wobble in its data that it claims is caused by dark matter, but other detectors have not been able to confirm this. Because the two experiments use the same type of crystal, they should be able to compare notes directly. However, differences in how they measure energy had made a direct comparison difficult. By lowering their threshold to 0.49 keV and carefully accounting for how their crystals respond to energy, the COSINE-100 team has now reached the same low-energy starting point as DAMA/LIBRA. This allows them to test the other experiment's claim over the full range of energies where the signal was reported, without needing to make assumptions about how the particles behave.

The findings represent a major step forward in the search for the invisible universe. By pushing the boundaries of what their detectors can hear, the scientists have opened a new window into the low-mass realm of dark matter. They have shown that with the right tools and enough patience, even the faintest whispers of the cosmos can be distinguished from the silence. While the mystery of dark matter remains unsolved, this work proves that the search is far from over, and that the next breakthrough could come from listening just a little bit more closely.