SemiCharmTag: a tool for Semileptonic Charm tagging

This paper introduces "SemiCharmTag," a novel tool for LHCb that utilizes secondary vertex hadron track tagging to effectively reject or isolate charm semileptonic backgrounds in dimuon Drell-Yan measurements at 13.6 TeV, significantly improving signal-to-background ratios while enabling the creation of unbiased charm-enriched samples.

The Gist

Imagine you are a detective trying to find a specific type of rare bird (the Drell-Yan signal) in a massive, noisy forest. The problem is that the forest is absolutely swarmed with a very similar-looking, common bird (the Charm background). In fact, for every rare bird you see, there are hundreds of the common ones. If you just look at the birds flying by, you'll never be able to tell them apart, and your count of the rare birds will be completely wrong.

This is the challenge faced by physicists at CERN's LHCb experiment. They want to study high-energy particle collisions to understand the universe, but their "rare birds" (prompt muons from Drell-Yan processes) are being drowned out by "common birds" (muons coming from the decay of heavy charm particles).

Here is how the paper introduces a new tool called SemiCharmTag to solve this problem, explained through everyday analogies.

The Problem: The "Noisy Forest"

In the world of particle physics, when protons smash together, they create a shower of particles.

• The Signal (Drell-Yan): These are "prompt" particles. They are born directly at the crash site (the primary vertex) and fly away immediately. They are the rare birds we want to count.
• The Background (Charm/Beauty): These are "non-prompt" particles. They are born from heavy particles (like charm quarks) that travel a tiny, tiny distance before decaying into the particles we see. It's like a bird that hatches in a tree a few meters away from the crash site and then flies into the crowd.

The problem is that the "common birds" (charm decays) are so numerous that they hide the "rare birds." Furthermore, we don't know exactly how many of each type of "common bird" there are, making it hard to mathematically subtract them.

The Solution: The "Sidekick" Detective (SemiCharmTag)

The authors developed a new method called SemiCharmTag. Instead of just looking at the bird (the muon) itself, this method looks at the bird's sidekick (a hadron track) that might be flying alongside it.

Think of it like this:

• The Rare Bird (Signal): When a prompt muon is born, it usually flies solo or with a very messy, random group of debris from the main crash. It doesn't have a specific "partner" it was born with.
• The Common Bird (Charm Background): When a charm particle decays, it usually spits out a muon and a specific partner particle (like a kaon or pion) at the exact same moment from the same tiny spot. They are a "tag team."

SemiCharmTag is a smart algorithm (a machine learning brain) that looks for these tag teams. It asks: "Is this muon flying with a partner that looks like it was born with it at a secondary location?"

The Two Strategies

The paper describes two ways to use this detective tool, depending on what you want to achieve:

1. The "Double-Tag" Strategy (The Bouncer)

Goal: To clean up the forest and let more rare birds through while kicking out the common ones.

Imagine a bouncer at a club. The bouncer checks every pair of birds.

• If a muon is flying with a "suspicious partner" (a track that suggests it came from a charm decay), the bouncer says, "No entry!" and kicks the whole pair out.
• If both muons in a pair pass the check (meaning neither has a suspicious partner), they are let in as "Signal."

The Result: This acts like a super-efficient filter. It manages to kick out about 78% of the background noise while keeping 81% of the real signal. It improves the signal-to-noise ratio by a factor of 4. It's like clearing the forest so you can finally see the rare birds clearly.

2. The "Single-Tag" Strategy (The Sample Collector)

Goal: To create a perfect "control group" of the common birds to understand them better.

Sometimes, you don't just want to filter; you want to study the common birds to understand their behavior so you can subtract them accurately later.

• The Single-Tag strategy is like a trap that specifically catches only the common birds (charm decays) and ignores the rare ones.
• It catches a muon that is definitely flying with a partner, confirming it's a "Charm" bird.
• Once caught, the scientists look at the other muon in the pair (the "probe") to see what it looks like.

The Result: This creates a pure sample of background data. It's like catching 1,000 common birds, studying their flight patterns, and then using that data to mathematically remove them from your final count of rare birds. It achieves a 21.4% efficiency in catching these background birds while barely letting any rare birds slip into the trap.

Why This Matters

Before this tool, scientists had to guess how many "common birds" were in the forest because they didn't know enough about how charm particles decay. This guesswork introduced huge errors.

SemiCharmTag changes the game by:

1. Reducing the noise: Making the rare signal stand out 4 times more clearly.
2. Data-driven truth: Instead of guessing the background, it builds a "pure sample" of the background directly from the data.

The Bottom Line

This paper presents a clever new way to separate the "signal" from the "noise" in particle physics. By looking at the "sidekick" particles that travel with the main particles, the LHCb team can now filter out the overwhelming background of charm decays. This allows them to study the rare, fundamental processes of the universe with much greater precision, even in the most crowded and chaotic parts of the collision data.

It's essentially giving the physicists a pair of smart glasses that allow them to ignore the crowd and focus on the stars they are trying to study.

Technical Summary

1. Problem Statement

The primary challenge addressed in this paper is the extraction of Drell-Yan (DY), thermal, and preequilibrium dilepton signals in proton-proton collisions at the LHC (specifically $\sqrt{s} = 13.6$ TeV).

• Signal vs. Background: In the invariant mass range of 2.9 to 5 GeV/$c^2$, the signal (prompt dileptons) is overwhelmed by a massive background from semileptonic decays of charm ($c\bar{c}$) and beauty ($b\bar{b}$) hadrons.
• Current Limitations:
  • The signal-to-background ratio ($S/B$) is extremely low (approx. 2.7% for charm and 1.3% for beauty before tagging).
  • Standard discrimination variables like the Impact Parameter (IP) are insufficient because the lifetimes of charm hadrons ($c\tau \approx 45\text{--}310$ $\mu$m) are short, making the separation from prompt vertices difficult.
  • Model Dependence: There is a critical lack of precise knowledge regarding charm fragmentation fractions (e.g., the ratio of baryons like $\Lambda_c^+$ to mesons) and semileptonic branching fractions. Relying on Monte Carlo simulations for background modeling introduces large systematic uncertainties.
  • Low $p_T$ Challenges: At low transverse momentum ($p_T < 10$ GeV/$c$), standard isolation criteria fail, and simulation mismodeling of parton showers and hadronization becomes significant.

2. Methodology: The SemiCharmTag Algorithm

The authors propose a novel, data-driven approach using a machine-learning-based tagging algorithm that exploits the topological differences between prompt leptons and those from heavy-flavor decays.

• Core Concept: Instead of relying solely on the lepton's own properties, the algorithm forms muon-hadron pairs within the same event. It analyzes the secondary vertex formed by a muon and a reconstructed hadron track.
• Discriminative Variables: The algorithm uses a Boosted Decision Tree (BDT) trained on XGBoost, utilizing variables such as:
  • Topological: Impact Parameter (IP), IP $\chi^2$, Distance of Closest Approach (DOCA) between muon and hadron, Flight Distance (FD).
  • Kinematic: Transverse momentum ($p_T$) of the hadron, Invariant mass of the muon-hadron pair.
  • Particle ID: Probability of the hadron being a Kaon ($PID_K$).
  • Angular: Direction of flight angle (DIRA).
• Training: The BDT is trained to distinguish three categories:
  1. Signal: Prompt Drell-Yan muons (paired with prompt hadrons from the primary vertex).
  2. Background 1: Muons from charm semileptonic decays (paired with the decay hadron).
  3. Background 2: Muons from beauty decays.

3. Key Contributions & Strategies

The paper presents two distinct strategies based on the same underlying algorithm:

A. Double-Tag Strategy (Background Rejection)

• Goal: Improve the $S/B$ ratio for signal extraction.
• Mechanism: A dimuon candidate is rejected if at least one of the two muons forms a muon-hadron pair with a BDT signal probability below a specific threshold. Both muons must pass the threshold to be classified as signal.
• Advantage: Maximizes background rejection power.

B. Single-Tag Strategy (Background Template Construction)

• Goal: Create unbiased, pure samples of charm background directly from data to serve as templates for signal extraction (avoiding simulation bias).
• Mechanism:
  1. Select one muon ("tag") with a high probability of originating from a charm decay (high BDT charm score, low beauty score).
  2. Use the properties of the other muon ("probe") to construct the background distribution.
  3. This isolates the kinematic properties of the probe muon in a sample enriched with charm decays.
• Advantage: Provides a data-driven method to model the shape of the charm background without relying on uncertain fragmentation fractions or branching ratios.

4. Key Results

The results are based on full LHCb simulations for Run 3 conditions ($\sqrt{s} = 13.6$ TeV).

• Performance of Double-Tag (Rejection):

  • Achieves a signal efficiency of 81%.
  • Rejects 78% of charm background and 74% of beauty background.
  • Improves the Signal-to-Background ratio by a factor of ~3.7 (charm) and 3.2 (beauty).
  • Bias Check: The selection introduces minimal bias (<5%) on the kinematic distributions ($p_T$, rapidity) of the remaining Drell-Yan signal, except at the edges of the acceptance.

• Performance of Single-Tag (Template Construction):

  • Achieves a charm efficiency of 21.4% on the single muon level.
  • Suppresses Drell-Yan (prompt) muons by a factor of ~90 (efficiency drops to 1.1%).
  • Purity: The resulting charm template has a prompt-prompt contamination ($C_{pp}$) of < 0.01%.
  • Validation: Closure tests confirm that the extracted IP distribution of the "probe" muon matches the true simulated charm distribution with no significant bias (consistent with unity within $2\sigma$).

5. Significance

• Enabling Low-Mass Measurements: This tool makes it feasible to measure Drell-Yan, thermal, and preequilibrium dileptons in the challenging 2–5 GeV/$c^2$ mass range, which is crucial for testing perturbative QCD and searching for gluon saturation.
• Data-Driven Systematics: By providing a method to extract pure background templates from data, SemiCharmTag significantly reduces the systematic uncertainties associated with the unknown fragmentation fractions of charm baryons and semileptonic branching ratios.
• Generalizability: While developed for LHCb, the methodology is transferable to any detector with sufficient vertex resolution, lepton identification, and hadron reconstruction capabilities.

In summary, SemiCharmTag offers a robust, machine-learning-driven solution to the "heavy-flavor background" problem in low-mass dilepton physics, enabling more precise measurements of fundamental QCD processes and the properties of strongly interacting matter.