CERN Accelerating science

A high-dimensional jet-powered measurement of the strong force

Top image: An event display of a high-momentum Z boson event in the ATLAS detector associated with a number of “jets”, which are complex clusters of particles created via interactions mediated by the strong force.

Precise measurements of fundamental particles at the Large Hadron Collider (LHC) are usually reported for just one or two particle properties at a time, as simultaneous measurements of many quantities with traditional techniques require an unreasonably large amount of data. On May 31st, 2024, however, the ATLAS Collaboration released a measurement of twenty-four quantities at the same time using a strategy enabled by machine learning. These two dozen measured properties describe events in which a Z boson is produced with a large momentum and accompanied by complex cascades of particles called “jets” that are sensitive probes of the strong force.

A novel public data format

This result has been released in a new format for LHC measurements: high-dimensional and unbinned datasets that can be used for a wide range of scientific and educational applications. Physicists can easily configure the datasets to produce traditional binned measurements of any of the measured properties, or arbitrary combinations of them, with full uncertainty covariances and customized binning. These results can be used to visualize differential cross-section spectra that quantify properties of the strong force, to perform hypothesis testing in the search of new physics phenomena, and to improve the modeling of particle physics simulation software. The measurement is public and user-friendly, with detailed examples in interactive Jupyter notebooks that can be run on the cloud using e.g. the CERN Service for Web-based ANalysis (SWAN) or Google Colab. These pedagogical and accessible examples can be used both in education and by the broader research community.

Figure 1: A screenshot from the public code that shows an example of how to modify the results to change the measurement bins or even construct new variables out of the twenty-four measured ones.

Measuring the strong force with jets

High-momentum Z boson production is of interest to particle physicists as it sheds light on the properties of the strong force. Crucially, the Z boson is produced alongside a number of “jets”, i.e. collimated groups of dozens of particles produced by fundamental particles like quarks and gluons in a process mediated by the strong force known as “hadronization”. The angle and energy distribution of these particles define the jet internal structure: a jet originating from a quark is expected to have a narrower profile with fewer particles than a jet originating from a gluon. Accurately assessing which jets originated from which objects can then be used to yield key information regarding the angle between the jets and the Z boson. Drawing connections between the dynamics of jets and the properties of the particles from which they originated can therefore help illuminate fundamental properties of the strong force and, by extension, the Standard Model.

Tens of thousands of neural networks

By training over 25,000 individual neural networks, ATLAS physicists were able to simultaneously measure twenty-four different properties of these high-momentum Z boson events. These networks were specially designed to reduce any subtle distortion effects from the ATLAS detector by learning a function that reweights particle physics simulations to closely resemble real data.

Figure 2: The method used to perform the twenty-four dimensional measurement, called OmniFold, trained tens of thousands of neural networks to reduce any subtle distortion effects from the ATLAS detector by learning a function that reweights particle physics simulations to closely resemble real data. It is based on multidimensional density reweighting.

Impact 

This result is impactful both as a new precision measurement of high-momentum Z boson events as well as an innovative and interactive public result from the ATLAS Collaboration. The measured datasets are accessible for anyone to use and easy to explore using Jupyter or Google Colab notebooks. The format of the published data lends itself to educational as well as research purposes, enabling anyone to directly compare measurements with theoretical predictions in minutes.

Further, this result shows that innovative applications of machine learning  are enabling ATLAS physicists to make unprecedented kinds of particle physics measurements, to explore lesser-known regimes of the Standard Model, and to invent new ways of presenting datasets for a new era of high-dimensional, unbinned measurements that can be applied for testing diverse theories both now and in the future.

Further information