⚛️ high-energy experiments

Simultaneous efficiency measurements of $b$ - and $c$ -jets in $t\bar{t}$ events from $\sqrt{s}=13.6$ TeV $pp$ collision data collected with the ATLAS detector

This paper presents simultaneous measurements of $b$ -jet identification and $c$ -jet misidentification efficiencies for the GN2 tagger using 56 fb $^{-1}$ of $\sqrt{s}=13.6$ TeV ATLAS $t\bar{t}$ data, achieving precisions better than 2% and 5% respectively by leveraging known branching fractions and kinematic likelihood reconstruction.

Original authors: ATLAS Collaboration

Published 2026-07-08

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: ATLAS Collaboration

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 authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the Large Hadron Collider (LHC) as the world's most powerful particle smasher. When it fires protons at each other, they explode into a shower of smaller particles, many of which clump together to form "jets." Some of these jets come from heavy, short-lived particles called bottom quarks (b-jets), while others come from slightly lighter charm quarks (c-jets).

The ATLAS experiment at CERN is like a giant, ultra-sensitive camera trying to take a picture of these explosions. But here's the problem: the camera's software (the "tagger") isn't perfect. Sometimes it sees a charm jet and mistakenly thinks, "Ah, that's a bottom jet!" This is called a "mistag."

This paper is essentially a quality control report for that camera software. The team wanted to know:

How good is the software at correctly spotting a bottom jet?
How often does it falsely accuse a charm jet of being a bottom jet?

They used a specific type of particle collision called a top-quark pair event ( $t\bar{t}$ ) as their "test lab." Think of a top quark as a very unstable parent that almost always splits into a bottom quark and a W boson. The W boson then splits again.

In some cases, the W boson splits into a lepton (like an electron or muon) and a neutrino. This is the "clean" side of the event.
In other cases, the W boson splits into two quarks (a charm and a strange quark). This is the "messy" side.

The Analogy of the Detective
Imagine you are a detective trying to identify a suspect (the bottom jet) in a crowd.

The "Clean" Side: You know for a fact that one person in the crowd must be the suspect because they came from a specific, reliable source (the leptonically decaying top quark). You check the camera's software against this known suspect to see how well it identifies them. This measures the efficiency (how often it gets it right).
The "Messy" Side: On the other side of the room, you know there is a group of people who are definitely not the suspect, but they look a bit similar. Specifically, you know the W boson on this side produces a charm quark about 33% of the time. You use this known group to see how often the camera software gets confused and says, "That's the suspect!" when it's actually a charm quark. This measures the mistagging rate.

What They Did
The researchers took data from 2022 and 2023, representing 56 "inverse femtobarns" of collisions (a fancy way of saying a massive amount of data). They used a new, advanced AI tool called GN2 (a "single-transformer" neural network) to do the tagging.

They didn't just guess; they built a mathematical model to simultaneously solve for two things at once:

How often the AI correctly spots the real bottom jets.
How often the AI falsely spots charm jets as bottom jets.

They compared their real-world data against computer simulations. Since the computer simulations aren't perfect, they calculated "Scale Factors" (SFs). Think of these as correction coefficients. If the computer simulation says the AI is 90% accurate, but the real data shows it's actually 95%, the scale factor is 1.05. Scientists apply this factor to future simulations to make them match reality.

The Results

Precision: They measured the efficiency of spotting bottom jets with a precision better than 2% in most cases. For spotting charm jets (mistags), the precision was better than 5%.
Consistency: They checked their results against a different method (using events where both top quarks decay into leptons) and found the numbers matched perfectly. This gives them high confidence that their measurements are correct.
The "GN2" Advantage: The paper notes that this new AI tagger is better at distinguishing charm jets from bottom jets than older tools, making the measurement of these "mistags" harder but more valuable.

In Summary
This paper doesn't discover a new particle or a new force of nature. Instead, it provides a calibration manual for the ATLAS experiment's software. By precisely measuring how often the software confuses charm jets for bottom jets, and how often it correctly finds bottom jets, the team ensures that every future experiment using this data is built on a solid, accurate foundation. It's the difference between a blurry photo and a sharp, high-definition image of the subatomic world.

Technical Summary: Simultaneous Efficiency Measurements of $b$ - and $c$ -jets in $t\bar{t}$ Events from $\sqrt{s}=13.6$ TeV $pp$ Collision Data

Problem and Motivation
In analyses at the Large Hadron Collider (LHC), identifying jets originating from bottom ( $b$ ) or charm ( $c$ ) quarks is critical for studies ranging from Higgs boson decays to top-quark pair production. While Monte Carlo (MC) simulations provide estimates of tagging efficiencies, they are imperfect. Consequently, data-driven scale factors (SFs)—the ratio of efficiency in data to that in simulation—are required to correct simulation weights. Previous ATLAS measurements of $b$ -jet SFs utilized dileptonic $t\bar{t}$ events, while $c$ -jet mistagging SFs were measured using semileptonic $t\bar{t}$ events. However, these were often treated as separate measurements. This paper addresses the need for a simultaneous measurement of $b$ -jet tagging and $c$ -jet mistagging efficiencies using the latest GN2 tagger, a single-transformer-based graph neural network, to improve the precision and consistency of flavor tagging calibration for Run 3 data.

Methodology
The analysis utilizes 56 fb $^{-1}$ of proton-proton collision data collected by the ATLAS detector at a center-of-mass energy of $\sqrt{s}=13.6$ TeV during 2022 and 2023. The event selection targets semileptonic $t\bar{t}$ events, characterized by exactly one isolated electron or muon, missing transverse momentum, and specific jet multiplicity requirements (typically 3–4 jets, with allowances for higher multiplicities under specific kinematic conditions).

A kinematic likelihood technique is employed to reconstruct the $t\bar{t}$ decay topology, mapping reconstructed jets to the decay products of the top quarks and $W$ bosons without relying on pre-existing $b$ -tagging information. This allows for the identification of:

Leptonic top-jet: The $b$ -jet from the top quark decaying leptonically ( $t \to W b \to \ell \nu b$ ).
Hadronic top-jet: The $b$ -jet from the top quark decaying hadronically.
$W$ -jets: The two jets arising from the hadronic decay of the $W$ boson ( $W \to q\bar{q}'$ ), which predominantly contain $c$ -quarks (branching fraction $\approx 33\%$ ) and light-flavor quarks.

The analysis employs a simultaneous fit to determine the SFs for $b$ -jets ( $b_t(i)$ ) and $c$ -jets ( $c_t(i)$ ) as a function of jet transverse momentum ( $p_T$ ) across six GN2 tagging intervals (ranging from 0–65% to 90–100% efficiency in simulation).

$b$ -jet SFs: Constrained by comparing the fraction of leptonic top-jets falling into each GN2 interval against MC expectations. Light-flavor jet backgrounds in this channel are normalized using the invariant mass of the lepton and the leptonic top-jet ( $m_{j\ell}$ ).
$c$ -jet SFs: Constrained by analyzing the GN2 discriminant of the tagged $W$ -jets in events where the other $W$ -jet is untagged and both top-jets are tightly tagged. The known $W \to c\bar{s}$ branching fraction allows the extraction of $c$ -jet mistagging rates.
Combined Fit: A total $\chi^2$ minimization combines the semileptonic measurements with an independent $b$ -jet SF measurement from dileptonic $t\bar{t}$ events (using a prior) to constrain the $b$ -jet efficiencies further.

Systematic uncertainties are evaluated by varying MC modeling parameters (parton shower, hadronization, ISR/FSR), jet energy scale/resolution, and background normalizations. The largest uncertainties arise from $t\bar{t}$ modeling (specifically the choice of parton shower/hadronization model) and jet reconstruction.

Key Results

$b$ -jet Tagging Efficiency: The measured SFs show no strong dependence on jet $p_T$ in the range 20–250 GeV. For the tightest tagging interval (0–65%), the total uncertainty reaches better than 2% for jets with $p_T > 60$ GeV. The results are consistent with independent measurements from dileptonic $t\bar{t}$ events.
$c$ -jet Mistagging Efficiency: Several measurements in the four loosest tagging intervals achieve a precision better than 5%. The SFs generally do not exhibit a strong $p_T$ dependence, though the tightest interval (0–65%) shows a higher efficiency in data than in MC simulation.
Simultaneous Determination: The fit successfully extracts both $b$ -jet and $c$ -jet SFs simultaneously, with the statistical correlation between the two being small compared to systematic correlations. The normalisation factors for light-flavor backgrounds and signal yields are found to be consistent with expectations.

Significance and Claims
The paper claims to be the first to report simultaneous measurements of $b$ -jet tagging and $c$ -jet mistagging SFs using $t\bar{t}$ semileptonic events. This approach leverages the specific decay topology of $t\bar{t}$ events to constrain both efficiencies within a single framework, utilizing the GN2 tagger which offers improved performance over previous algorithms. The results provide essential calibration inputs for ATLAS analyses requiring precise flavor tagging, particularly those involving $b$ - and $c$ -quark identification in the high-luminosity Run 3 era. The paper modestly notes that while the measurements are precise, the uncertainties are dominated by systematic effects related to MC modeling, particularly the parton shower and hadronization models.

More like this