Bayesian inference constraints on jet quenching across centrality, beam energy, and observable classes in LHC heavy-ion collisions

This study employs Bayesian inference to demonstrate that while jet quenching constraints across collision centrality are compatible, variations in beam energy and observable classes reveal distinct aspects of jet-medium interactions that limit predictive universality, motivating the use of leading-hadron-selected observables to bridge these gaps.

The Gist

The Great Jet Quenching Mystery: A Detective Story in a Particle Smasher

Imagine you are a detective trying to figure out how a specific type of "traffic jam" works inside a super-dense, super-hot fluid. But you can't see the fluid directly. Instead, you have to watch how fast-moving cars (particles) get slowed down, scattered, or stopped as they drive through it.

This is exactly what physicists are doing in this paper. They are studying heavy-ion collisions (smashing lead atoms together at nearly the speed of light) to understand the Quark-Gluon Plasma (QGP). Think of the QGP as a "primordial soup" of the universe's most fundamental building blocks, created for a split second in the lab.

Here is the breakdown of their investigation, using simple analogies:

1. The Crime Scene: The "Jet" and the "Soup"

When two lead atoms smash together, they create a tiny, incredibly hot drop of this "soup." Occasionally, a very fast particle (a "jet") is shot out of the collision.

• The Jet: Imagine a race car speeding through a thick mud pit.
• The Soup (QGP): The mud pit itself.
• Jet Quenching: As the race car drives through the mud, it loses speed and energy. The amount of energy it loses tells us how "thick" or "sticky" the mud is.

Physicists want to measure the "stickiness" (called the transport coefficient, $\hat{q}$) of this soup. But here's the problem: The mud isn't uniform. It changes depending on:

• Centrality: Did the cars crash head-on (central) or just graze each other (mid-central)? This changes the size of the mud pit.
• Beam Energy: How hard did they smash the atoms? This changes the temperature of the soup.
• The Observer: Are we watching the race car itself (inclusive jets), or just the fastest piece of debris flying off it (charged hadrons)?

2. The Detective's Tool: Bayesian Inference

The authors used a method called Bayesian Inference. Think of this as a "smart guess-and-check" game.

• They have a computer model (a virtual simulation) of the mud pit and the race cars.
• They have real data from giant detectors (ALICE, ATLAS, CMS) at the Large Hadron Collider (LHC).
• They ask the computer: "If we tweak the 'stickiness' of the mud, does our simulation look more like the real photos?"
• By running this millions of times, they narrow down the most likely value for the stickiness. This result is called the Posterior.

3. The Big Question: Is the Mud the Same Everywhere?

The main goal of this paper was to test Universality.

• The Hypothesis: "The rules of physics are the same everywhere. If we figure out the stickiness of the mud in a head-on crash, that same number should work for a side-swipe crash, or for a different temperature, or if we look at different pieces of debris."
• The Test: The team took their "smart guess" from the full dataset and tried to apply it to smaller, specific subsets of data (e.g., only head-on crashes, or only high-energy crashes) without re-running the guess-and-check game.

4. The Findings: The "One-Size-Fits-All" Myth

The results were surprising and nuanced. They found that the "one-size-fits-all" idea works partially, but not perfectly.

• Centrality (Head-on vs. Side-swipe):

  • Analogy: If you drive through a deep mud pit vs. a shallow one, the mud feels different.
  • Result: The "stickiness" numbers for head-on and side-swipe crashes were very similar. The model worked well for both. Verdict: Compatible.

• Beam Energy (Hotter vs. Cooler Soup):

  • Analogy: Hot honey flows differently than cold honey.
  • Result: When they tried to use the "hot soup" rules on "cooler soup" data (and vice versa), the predictions drifted apart. The model needed a slight adjustment. Verdict: Moderate shift.

• Observable Class (The Whole Car vs. The Fastest Debris):

  • Analogy: This is the most interesting part.
    • Inclusive Jets: Watching the whole car get stuck in the mud.
    • Charged Hadrons: Watching just the fastest piece of the car (the bumper) fly off.
  • Result: These two "observers" saw slightly different things. The "whole car" data suggested the mud was stickier than the "fastest debris" data suggested.
  • Why? The "fastest debris" tends to come from the edge of the collision (surface bias), while the "whole car" samples the deep middle. They are looking at different parts of the same event. Verdict: They overlap, but they pull in different directions.

5. The Prediction Test: Can You Use the Map for a New City?

The ultimate test of a good map is: If I draw a map of New York, can I use it to navigate Chicago?

The authors took the "map" (the model parameters) they learned from one type of data and tried to predict the other types of data without re-learning.

• Centrality: The map worked okay, but not perfectly. Predicting a side-swipe using a head-on map was a bit shaky.
• Beam Energy: The map was off. The "New York" map didn't fit "Chicago" well because the terrain (temperature) was too different.
• Observable Class: Surprisingly, the map worked quite well! Even though the "whole car" and "fastest debris" saw things differently, the model could translate between them reasonably well.

6. The Conclusion: We Need a Better Map

The paper concludes that while we have a good general understanding of the "stickiness" of the Quark-Gluon Plasma, our current model is a bit like a blurry photograph. It captures the main picture but misses the fine details.

• The Problem: The current model treats the "stickiness" as a single number, but it seems to depend on how you look at it (energy, angle, and what part of the particle you are watching).
• The Solution: The authors suggest we need new types of "observers." They propose looking at "Leading-Hadron Selected Jets."
  • Analogy: Instead of just watching the whole car or just the bumper, imagine watching the car while specifically tracking the driver's seat. This gives a middle-ground view that bridges the gap between the "whole car" and the "fastest debris."

In Summary:
This paper is a rigorous stress-test of our understanding of the universe's "primordial soup." It shows that while our current theories are strong, they aren't perfect yet. Different ways of looking at the data reveal slightly different truths, suggesting that the "rules of the road" in this subatomic world are more complex than we thought. To solve the mystery, we need to look at the collision from a new, hybrid angle.

Technical Summary

1. Problem Statement

Jet quenching in ultra-relativistic heavy-ion collisions serves as a primary probe for the transport properties of the Quark-Gluon Plasma (QGP), specifically the jet transport coefficient $\hat{q}$. However, extracting a universal value for $\hat{q}$ is challenging because measured suppression ($R_{AA}$) depends on a complex interplay of:

• Microscopic interaction mechanisms.
• Collision geometry and parton path length.
• Medium evolution (temperature history).
• Observable selection biases (e.g., charged hadrons vs. inclusive jets).

Previous global Bayesian calibrations have established a baseline, but it remains unclear whether the transport parameters inferred from one subset of data (e.g., central collisions or specific beam energies) are universally transferable to other subsets without refitting. The core question is whether a single effective transport description can consistently predict observables across different centralities, beam energies ($\sqrt{s_{NN}}$), and observable classes (hadrons vs. jets).

2. Methodology

A. Physics Model and Simulation

The authors utilize the JETSCAPE multistage framework to simulate PbPb collisions at $\sqrt{s_{NN}} = 5.02$ and $2.76$ TeV.

• Initial State: Generated using the TRENTo model for event-by-event entropy density profiles.
• Medium Evolution: The background medium evolves via MUSIC (2+1D viscous hydrodynamics) with fixed bulk parameters.
• Hard Scattering: Generated using PYTHIA within the JETSCAPE workflow.
• Energy Loss Model: A six-parameter effective model is calibrated. The transport coefficient $\hat{q}$ is parameterized as:
  $$\hat{q}_g(T, E, Q^2) = \hat{q}^{HTL}_g(T, E) \times F_6(E, Q^2)$$
  The six parameters ($\theta$) are:
  1. $Q_0$: Reference virtuality scale.
  2. $\tau_0$: Onset time for medium-induced energy loss (MATTER stage).
  3. $A, B$: Coefficients for logarithmic virtuality dependence in the denominator.
  4. $C$: Coefficient controlling large-$x_B$ suppression in the numerator.
  5. $\alpha_s$: Fixed strong coupling constant setting the overall normalization.

B. Bayesian Inference Framework

• Emulation: Due to the computational cost of full event simulations, a Gaussian Process (GP) emulator is trained on 50 Latin Hypercube Sampled (LHS) design points. The emulator uses Principal Component Analysis (PCA) to reduce the output space of charged-hadron and inclusive-jet $R_{AA}$.
• Likelihood: A multivariate normal likelihood compares emulator predictions to experimental data, incorporating both experimental (statistical + systematic) and emulator uncertainties.
• Sampling: Markov Chain Monte Carlo (MCMC) is used to sample the posterior distribution $p(\theta | y_{exp})$.

C. Subset Decomposition Strategy

To test universality, the authors perform three types of decompositions:

1. Centrality Split: Central (0–5/10%) vs. Mid-central (30–50/40–50%).
2. Beam Energy Split: 5.02 TeV vs. 2.76 TeV.
3. Observable Class Split: Charged-hadron only vs. Inclusive-jet only.

Universality Criteria:

1. Parameter Space: Subset posteriors for $\hat{q}/T^3$ should overlap significantly at a common reference point ($E_{ref}=100$ GeV, $T_{ref}=0.2$ GeV).
2. Observable Space (Cross-Prediction): A posterior calibrated on one subset must predict the other subset's data accurately without refitting.

3. Key Results

A. Global Calibration

The full dataset calibration constrains the parameters $Q_0$ and $\alpha_s$ most tightly ($Q_0 \approx 1.54 \pm 0.27$, $\alpha_s \approx 0.314 \pm 0.040$). Parameters governing the shape of the virtuality dependence ($A, B, C, \tau_0$) remain broadly unconstrained, indicating the current data primarily fixes the overall quenching scale rather than the detailed functional form.

B. Subset Decomposition (Parameter Space)

• Centrality: Central and mid-central posteriors show strong overlap ($\Delta\sigma_{68} \approx 0.02$). This suggests the transport coefficient is largely consistent across geometry classes.
• Beam Energy: A moderate shift is observed. The 5.02 TeV subset favors higher $\hat{q}$ values than the 2.76 TeV subset ($\Delta\sigma_{68} \approx -0.66$), though credible regions still overlap.
• Observable Class: A residual shift exists. Jet-only calibrations favor higher $\hat{q}$ and broader high-$\hat{q}$ tails compared to hadron-only calibrations ($\Delta\sigma_{68} \approx -0.41$).

C. Cross-Prediction (Observable Space)

This is the most critical finding. Posterior overlap does not guarantee predictive transferability.

• Centrality Transfer: Asymmetric.
  • Mid-central $\to$ Central: Works reasonably well (Overlap $\approx 0.45$, $\Delta\sigma_{68} \approx 0.85$).
  • Central $\to$ Mid-central: Fails significantly (Overlap $\approx 0.34$, $\Delta\sigma_{68} \approx 1.34$). The tight constraints from central events cannot flexibly describe the less-quenched mid-central data.
• Beam Energy Transfer: Both directions show persistent offsets ($\Delta\sigma_{68} > 1$), indicating that a single effective coefficient struggles to describe the different temperature histories and parton spectra of the two energies.
• Observable Class Transfer: Most stable.
  • Hadron $\to$ Jet and Jet $\to$ Hadron transfers show high overlap ($>0.6$) and low separation ($\Delta\sigma_{68} < 0.5$).
  • Despite the parameter shift, the inclusive $R_{AA}$ observables are robustly predicted across classes.

D. Sensitivity Analysis

The current dataset is most sensitive to $Q_0$ and $\alpha_s$. The shape parameters ($A, B, C$) have low sensitivity, explaining why they remain unconstrained. The sensitivity is dominated by central charged-hadron $R_{AA}$ bins.

4. Significance and Conclusions

1. Operational Definition of Universality: The paper establishes that "universality" in jet quenching must be tested in both parameter space (overlap) and observable space (predictive transfer). Overlap alone is insufficient.
2. Geometry and Energy Dependence: While a single effective transport coefficient can describe the average suppression, it fails to universally predict specific subsets (especially central-to-mid-central transfer and beam-energy shifts) without refitting. This suggests residual dependencies on path length and temperature history that the current effective model does not fully capture.
3. Hadron vs. Jet Compatibility: Surprisingly, the transfer between charged-hadron and inclusive-jet data is the most stable. This implies that while these observables probe different aspects of the shower (surface bias vs. full shower energy), the current effective parametrization captures the dominant suppression scale well enough to bridge them.
4. Future Directions: The authors propose leading-hadron-selected jet observables as a solution. These observables can tune the "leading fragment bias" within a reconstructed jet, potentially interpolating between hadron-biased and jet-inclusive constraints to resolve the residual tensions in the transport coefficient's scale dependence.

In summary, the work demonstrates that while global Bayesian calibrations provide a robust baseline, the extracted transport properties exhibit subtle but significant dependencies on collision geometry and beam energy that limit their predictive universality across all subsets.