Physicists at CERN have directly measured proton-oxygen collisions at 9.62 TeV per nucleon pair, closely recreating an important type of interaction that occurs when extremely energetic cosmic rays strike Earth\u2019s atmosphere.The ATLAS experiment analyzed 246 million selected collisions containing more than 5.1 billion reconstructed tracks and found that no major cosmic-ray interaction model successfully reproduced all the observed particle behavior.The measurements could improve simulations used to determine the energies, masses and possible origins of cosmic rays that are too energetic to measure directly with instruments in space.<\/p>\n
Earth is constantly being bombarded by particles from space, but the most energetic arrivals are difficult to understand because scientists rarely observe them directly.<\/p>\n
Instead, researchers usually detect the vast showers of secondary particles created when a cosmic ray smashes into a molecule high in the atmosphere. Working backward from those showers requires computer models that attempt to reconstruct the original particle and the collision that started everything.<\/p>\n
There is a problem. Different models can produce substantially different answers.<\/p>\n
Physicists working with CERN’s ATLAS experiment<\/a> have now recreated one of those atmospheric collisions under controlled conditions by smashing protons into oxygen nuclei inside the Large Hadron Collider. The measurements, reported in Physical Review Letters<\/a>, provide unusually direct data for improving the models used to interpret high-energy cosmic rays.<\/p>\n The ATLAS Calorimeter is a massive, multi-layered subsystem of the ATLAS Experiment at CERN’s Large Hadron Collider. (CREDIT: CERN, License: CC-BY-4.0) <\/p>\n \u201cWe now have a better understanding of how particles from objects more exotic than the Sun interact with our atmosphere,\u201d said Jesse Liu, an assistant professor of physics at New York University<\/a> who led the research.<\/p>\n Cosmic rays create enormous particle cascades<\/p>\n The Northern Lights offer a familiar example of particles interacting with Earth’s atmosphere, although auroras involve far lower energies than the cosmic rays examined in the new experiment.<\/p>\n Auroras form when charged particles associated with the Sun enter the upper atmosphere and interact with gases. High-energy cosmic rays can carry vastly more energy and may originate from violent astrophysical environments, including supernova remnants.<\/p>\n When one of these particles enters the atmosphere, it can collide with the nucleus of an oxygen or nitrogen atom<\/a>. That initial impact creates new particles, which collide again and produce still more particles.<\/p>\n The result is an extensive air shower that can spread across a large area before reaching detectors on the ground.<\/p>\n At sufficiently high energies, the incoming cosmic rays are so rare that direct measurements by satellites become impractical. Scientists therefore infer their properties from these cascades, making the accuracy of the underlying collision models critical.<\/p>\n Photo taken by the ATLAS experiment of a collision between a proton and oxygen at the Large Hadron Collider, which recreates a cosmic rainstorm of particles\u2014shown as yellow lines. (CREDIT: CERN\/ATLAS Collaboration) <\/p>\n Those models depend heavily on accelerator measurements. Yet previous collider experiments have generally used collision systems that do not closely reproduce a cosmic-ray proton hitting an atmospheric nucleus.<\/p>\n The LHC fired protons directly into oxygen<\/p>\n The CERN experiment was designed to close that gap.<\/p>\n During a special run in July 2025, the Large Hadron Collider<\/a> accelerated oxygen nuclei to 3.4 teraelectronvolts per nucleon while sending 6.8 TeV protons in the opposite direction. Their collisions produced a center-of-mass energy of 9.62 TeV per nucleon pair.<\/p>\n The setup closely resembles a high-energy proton cosmic ray striking an oxygen nucleus in Earth’s atmosphere.<\/p>\n ATLAS recorded data corresponding to 634 inverse microbarns of integrated luminosity. After event selection, the researchers analyzed about 246 million collisions containing approximately 5.11 billion reconstructed particle tracks.<\/p>\n \u201cThese data are unique, and in our measurements, we were able to show that our previous models for this type of interaction are actually very inaccurate,\u201d said Cigdem Issever of Humboldt University<\/a> of Berlin.<\/p>\n The measured proton-oxygen and proton-air cross sections are compared with leading theoretical models and previous cosmic-ray observatory results across a wide range of collision energies. Error bars include combined experimental and luminosity uncertainties. (CREDIT: Jesse Liu et al, Physical Review Letters) <\/p>\n The researchers measured how frequently interactions occurred along with the number, direction and transverse momentum of charged particles produced in the collisions.<\/p>\n The experiment recreated a 49-PeV cosmic ray collision<\/p>\n The laboratory collision corresponds to an extraordinary cosmic-ray energy when viewed from the perspective of a stationary atmospheric nucleus.<\/p>\n The researchers calculated that the proton-oxygen collisions<\/a> were equivalent to a cosmic-ray proton carrying about 49 petaelectronvolts, or 49 quadrillion electronvolts, hitting Earth’s atmosphere.<\/p>\n ATLAS measured a fiducial proton-oxygen cross section of 396 millibarns, with experimental and luminosity uncertainties. A cross section represents the probability of a particular type of particle interaction occurring.<\/p>\n The researchers then used the measurement to infer an inelastic proton-air cross section of 406 millibarns, with additional uncertainty arising from the theoretical conversion from oxygen to the nitrogen-oxygen mixture of Earth’s atmosphere.<\/p>\n That result was compatible with earlier cosmic-ray measurements at similar energies, providing an important connection between accelerator physics and observations made from air showers.<\/p>\n Primary charged-particle distributions in the fiducial region for data (black points), where the shaded band denotes the combined statistical and systematic uncertainties. The x value in each bin corresponds to the bin centroid. (CREDIT: Jesse Liu et al, Physical Review Letters) <\/p>\n The measurements also favored the lower end of the cross sections predicted by several widely used interaction models.<\/p>\n No simulation reproduced everything correctly<\/p>\n The most revealing result emerged when the researchers compared detailed particle production with seven models used in particle and cosmic-ray<\/a> physics.<\/p>\n None reproduced all the measurements.<\/p>\n The models disagreed particularly strongly over particle multiplicity, meaning how many charged particles were produced in each collision. At high multiplicities, some predictions differed from the data by approximately an order of magnitude.<\/p>\n Other measurements performed better. A model called Angantyr, implemented in the PYTHIA particle-physics simulation framework, came closest to reproducing the observed transverse-momentum and pseudorapidity distributions.<\/p>\n Several models successfully captured individual aspects of the collisions but failed elsewhere. Those differences matter because small errors during the first atmospheric collision<\/a> can propagate throughout an air shower and change what ground-based observatories eventually detect.<\/p>\n The findings therefore provide a powerful new dataset for retuning those simulations rather than demonstrating that any one existing model is universally correct.<\/p>\n Better models could reveal where cosmic rays come from<\/p>\n Improved atmospheric simulations have implications far beyond particle physics.<\/p>\n Scientists use air showers to estimate whether an incoming cosmic ray was a proton, helium nucleus, iron nucleus or another particle. They also need accurate models to reconstruct its original energy.<\/p>\n