A fully differential charged-particle phase-space measurement of +jets production in $pp$ collisions at TeV with the ATLAS detector
Using the OmniFold method on 140 fb⁻¹ of 13 TeV ATLAS data, this paper presents the first published unbinned, fully differential measurement of Z+jets production, providing particle-level Monte Carlo events and weights that enable the construction of arbitrary observables for future reinterpretation.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the author. For technical accuracy, refer to the original paper. Read full disclaimer
In the heart of Europe, beneath the border of France and Switzerland, lies the Large Hadron Collider, a machine that smashes protons together at speeds approaching the speed of light. When these tiny particles collide, they do not simply bounce off one another; they transform, creating a shower of new, short-lived particles that fly out in all directions. Physicists study these collisions to understand the fundamental rules that govern the universe, much like a mechanic might take apart a complex engine to see how its gears fit together. One of the most reliable ways to test these rules is to look at a specific event where a heavy particle called a Z boson is created and immediately decays into two muons, which are heavy cousins of the electron. Because the Z boson is so heavy, it often recoils against a spray of other particles, forming what physicists call a jet. By studying the shape and energy of these jets, scientists can probe the strong force that holds atomic nuclei together, testing whether our current theories of physics hold up under extreme conditions.
For decades, the standard way to analyze these collisions has been to sort the data into fixed categories, or bins, based on pre-chosen measurements. Imagine trying to describe a complex painting by only counting how many red pixels are in the top-left corner, how many blue ones are in the bottom-right, and so on. While this method works, it throws away a vast amount of information about the relationships between the different parts of the image. If a scientist later decides they want to know something about the texture of the brushstrokes, they cannot go back and look at the original painting; they only have the pixel counts, which are insufficient. This limitation meant that many interesting questions about the subatomic world remained unaskable because the data had already been compressed into a format that could not answer them.
A researcher using the ATLAS detector at the Large Hadron Collider has now broken this barrier. They have produced the first published measurement of a particle collision that is not sorted into bins at all. Instead of summarizing the data into a few histograms, they have released a complete, unsorted record of the collision, describing the motion of every single charged particle that flew out of the crash site. This record covers 140 inverse femtobarns of data collected from proton-proton collisions at a center-of-mass energy of 13 TeV. The result is a digital reconstruction of the event where every charged particle is listed with its speed and direction, accompanied by a weight that tells scientists how likely that specific configuration is to occur in nature. This approach allows anyone to ask new questions of the data years after the experiment is over, without needing to run the complex analysis again.
To achieve this, the researcher had to overcome a significant hurdle: the data from each collision is not uniform. Some crashes produce a few dozen particles, while others produce hundreds. Traditional computer methods struggle with this variability, but the team used a type of artificial intelligence known as a transformer network. These are the same kinds of systems that power modern language models, but here they were trained to understand the geometry of particle collisions. The computer was taught to recognize patterns in the spray of particles, learning to correct for the fact that the detector itself is not perfect and sometimes misses particles or measures their speed slightly wrong. By training on billions of simulated events before looking at the real data, the system learned to distinguish between the true physics of the collision and the quirks of the machine.
The power of this new method was demonstrated by applying it to four very different types of questions. First, the researcher looked at the total number of charged particles and the total energy they carried. They found that the computer simulations used to predict these collisions were often off by as much as fifty percent in certain regions, suggesting that our current models of how particles interact need significant tuning. Second, they examined the internal structure of the jets themselves. In the past, measuring the shape of a jet required a specific, rigid definition of what a jet was. With this new method, scientists can apply any definition they like to the same data, revealing subtle differences in how particles cluster together that were previously invisible.
The researcher also measured a quantity called the energy correlator, which tracks how energy is shared between pairs of particles at different angles. This measurement is sensitive to the transition between the world of quarks and gluons, which behave like free particles, and the world of protons and neutrons, where they are locked together. The data showed a clear pattern that matched theoretical predictions, confirming our understanding of how the strong force behaves at high energies. Finally, the researcher measured the intrinsic dimensionality of the jets. This is a concept from mathematics that asks how many coordinates are actually needed to describe a complex shape. They found that while a jet might seem to exist in a space with hundreds of dimensions, the actual physics constrains it to a much simpler, lower-dimensional shape. This discovery provides direct experimental evidence that the chaotic spray of particles is actually governed by a surprisingly simple underlying order.
Perhaps the most significant aspect of this work is not just the specific numbers found, but the way the data is shared. The entire dataset, consisting of millions of simulated events with their calculated weights, has been made public. This means that a scientist in a different field, or one working on a completely different theory, can download the data and construct their own measurements without ever needing to understand the complex detector or the original analysis code. It transforms a single experiment into a permanent, reusable resource for the entire scientific community. By moving from a fixed set of answers to a flexible, open-ended record, this work ensures that the questions asked today will not limit the discoveries made tomorrow. The experiment has not just measured the particles; it has preserved the full story of their creation for future generations to read.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.