Nonflow suppression in flow analysis with a maximum likelihood estimator

This paper demonstrates that a maximum likelihood estimator effectively mitigates non-flow effects and addresses detector acceptance deficiencies in flow analysis, offering a compelling alternative to standard methods like particle correlation and event plane techniques.

The Gist

Imagine you are at a massive, chaotic dance party inside a giant, invisible ballroom. This ballroom is actually a Quark-Gluon Plasma (QGP), a super-hot soup of particles created when heavy atoms smash into each other at nearly the speed of light.

Physicists want to know: Is the dance floor moving in a coordinated way? Do the dancers (particles) have a collective rhythm, swirling together in specific patterns? This "collective rhythm" is called Flow.

However, there's a problem. Some dancers aren't following the main rhythm.

• The "Non-Flow" Noise: Some dancers are just hanging out with their best friends (particle decay), some are pushing off each other to conserve momentum, and some are just randomly bumping into neighbors. These interactions create "noise" that makes it look like the whole room is dancing in a pattern when it's actually just a few people messing around.

For a long time, scientists used standard tools (like taking a group photo and counting heads) to measure the dance. But these tools often get confused by the noise, leading to inaccurate results.

The New Tool: The "Maximum Likelihood Estimator" (MLE)

This paper introduces a new, smarter tool called the Maximum Likelihood Estimator (MLE). Think of MLE not as a simple head-counter, but as a super-detective or a sophisticated music producer.

Instead of just counting, the MLE asks: "If I assume a specific rhythm is happening, how likely is it that I would see exactly this arrangement of dancers?" It tries to find the "true rhythm" that makes the observed chaos look the most probable.

The Experiment: Two "Toy" Scenarios

To test if this detective is good at ignoring the noise, the authors created two fake scenarios (Toy Models) to simulate the "bad dancers":

1. The "Parent-Child" Scenario (Particle Decay):
   Imagine a parent dancer splits into two children. The children stay close together, mimicking a coordinated move, but they aren't part of the main group rhythm.

   • The Test: The MLE had to figure out the main group's rhythm while ignoring these parent-child pairs.
   • The Result: The MLE did a better job than the old methods. It realized, "Ah, these two are just a family unit, not the main dance," and filtered them out more effectively.

2. The "Push-and-Pull" Scenario (Momentum Conservation):
   Imagine a group of dancers who, because of the laws of physics, must push off each other so the total "push" of the group remains zero. This creates a fake pattern that looks like a flow but is actually just physics balancing itself out.

   • The Test: The MLE had to ignore this global balancing act.
   • The Result: Again, the MLE was superior. It managed to see the true underlying flow even when the "push-and-pull" noise was strong.

Why is MLE Better?

The authors compared MLE to the old "standard methods" (like the Event Plane method and Particle Correlation). Here is the analogy:

• Old Methods (The Group Photo): If you take a photo of the dance floor and try to guess the rhythm by looking at who is standing next to whom, you might get tricked by the parent-child pairs or the push-and-pull groups. You might think the whole room is spinning when only a few people are.
• MLE (The Music Producer): The MLE listens to the entire song at once. It doesn't just look at pairs; it looks at the probability of the whole arrangement. It's like a producer who can hear the bass line (the true flow) even when there's a lot of static and random clapping (the noise) in the background.

The "Broken Camera" Bonus

The paper also tested what happens if your camera (the detector) is broken and misses some parts of the dance floor.

• Old Methods: If the camera misses the left side of the room, the old methods get confused and think the dance is lopsided.
• MLE: The MLE is smart enough to say, "I know my camera is broken on the left, so I will weigh the data from the right side differently to compensate." It can fix the picture even with a broken lens.

The Bottom Line

This paper proves that the Maximum Likelihood Estimator is a powerful new tool for physicists.

• It is better at ignoring the "noise" (non-flow effects) caused by particle decay and momentum conservation.
• It is more flexible and can handle broken detectors.
• It gives a clearer, more accurate picture of the "perfect liquid" dance of the Quark-Gluon Plasma.

In short, if the old methods were like trying to hear a whisper in a noisy room with your hands over your ears, the MLE is like putting on noise-canceling headphones that let you hear the whisper perfectly.

Technical Summary

1. Problem Statement

In relativistic heavy-ion collisions, flow harmonics ($v_n$) are crucial observables used to characterize the anisotropic expansion of the Quark-Gluon Plasma (QGP). However, extracting these harmonics is complicated by non-flow effects—correlations between particles that do not arise from the collective hydrodynamic expansion of the medium. Common non-flow sources include:

• Particle decays: Resonance decays producing correlated daughter particles.
• Momentum conservation: Global conservation laws inducing back-to-back correlations.
• Jet fragmentation and HBT interferometry.

Traditional methods for extracting flow, such as particle correlation (cumulants) and the event plane method, often struggle to fully suppress these non-flow contributions, particularly in small collision systems or low-multiplicity events where statistical fluctuations are significant. While higher-order cumulants suppress uncorrelated signals, they may not be sufficient for all non-flow scenarios. The authors investigate whether the Maximum Likelihood Estimator (MLE), a statistical inference tool, offers a more robust alternative for isolating genuine collective flow from non-flow artifacts.

2. Methodology

The study employs a comparative analysis using Monte Carlo simulations to test the efficacy of MLE against standard methods (particle correlation and event plane).

A. The MLE Framework

Instead of relying on multi-particle correlators, MLE treats flow harmonics as parameters ($\theta$) of an underlying probability distribution.

• Likelihood Function: For an event with $M$ particles, the likelihood is constructed based on the one-particle azimuthal distribution function $f_1(\phi; \theta)$:
  $$L(\theta) = \prod_{j=1}^{M} f_1(\phi_j; \theta)$$
  where $f_1(\phi) = \frac{1}{2\pi} [1 + \sum 2v_n \cos n(\phi - \Psi_n)]$.
• Optimization: The estimator finds the parameters $\hat{\theta}_{MLE}$ that maximize the log-likelihood $\ell = \sum \log f_1$.
• Advantage: MLE is asymptotically normal and unbiased, theoretically offering superior accuracy for large sample sizes compared to other estimators.

B. Toy Models for Non-Flow

To rigorously test the method, the authors constructed two "toy models" that artificially exaggerate specific non-flow effects:

1. Toy Model I (Particle Decay):
   • Simulates a scenario where a fraction ($P_{dcy}$) of primary particles decay into two daughter particles.
   • The decay angle follows a Gaussian distribution centered on the parent particle's angle.
   • This introduces localized, short-range correlations.
2. Toy Model II (Momentum Conservation):
   • Simulates global momentum conservation by enforcing transverse momentum conservation within small, independent subsets of particles (rather than the whole event).
   • This mimics the effect of momentum conservation in small systems or mini-jets, creating global constraints that distort flow signals.

C. Revised MLE Strategy

Recognizing that standard MLE assumes independent particle emission (ignoring the decay mechanism), the authors proposed a Revised MLE. This approach expands the parameter space to include decay parameters ($P_{dcy}$, $\sigma$) directly in the likelihood function, allowing the estimator to explicitly account for the non-flow mechanism.

D. Detector Acceptance Correction

The study also tested MLE's ability to correct for non-uniform detector acceptance by introducing a weighting factor $w(\phi) = 1/s(\phi)$ into the log-likelihood function, where $s(\phi)$ is the acceptance function.

3. Key Results

A. Performance on Toy Model I (Decay)

• Standard MLE vs. Traditional Methods: In simulations with decay probabilities ($P_{dcy}$) up to 50% and decay widths ($\sigma$) up to 30°, the standard MLE consistently provided estimates of $v_2$ closer to the true input values than the event-plane or particle correlation methods.
• Underestimation Trend: All methods tended to underestimate $v_2$ as non-flow increased, but MLE showed the smallest deviation.
• Revised MLE Success: When the parameter space was expanded to include decay parameters ($P_{dcy}, \sigma$), the Revised MLE significantly improved accuracy. For example, at $M_0=500$, the Revised MLE improved the $v_2$ estimation by ~7.7% over standard particle correlation and ~5.4% over standard MLE.

B. Performance on Toy Model II (Momentum Conservation)

• Suppression of Non-Flow: In scenarios where momentum conservation was enforced on small subsets (simulating strong non-flow), all methods underestimated the true flow. However, MLE remained the closest to the input value across various multiplicities ($M$) and subset sizes ($M_{sub}$).
• Robustness: As the subset size decreased (increasing non-flow strength), the performance gap between MLE and traditional methods widened, with MLE demonstrating superior suppression of the non-flow bias.

C. Comparison with Existing Literature

• The authors compared their results with previous studies (e.g., STAR collaboration) regarding particle pair emissions. They clarified that the direction of the bias (overestimation vs. underestimation) depends on whether the non-flow pairs are correlated with the event plane and the opening angle.
• In their specific decay model (correlated with the event plane), the non-flow effect led to an underestimation of flow, contrasting with some literature where small opening angles caused overestimation. MLE handled this underestimation better than correlators.

D. Detector Acceptance

• The MLE framework successfully corrected for significant detector acceptance deficiencies (both step-function and complex sinusoidal variations). By applying the weighted log-likelihood, the MLE recovered the correct flow distribution, whereas uncorrected methods failed to predict the flow accurately.

4. Key Contributions

1. Validation of MLE for Non-Flow: The paper provides the first systematic demonstration that MLE is an effective tool for mitigating non-flow effects in flow analysis, outperforming standard correlation and event-plane methods in toy models.
2. Revised MLE Formulation: The authors introduced a novel extension of MLE that incorporates specific physical mechanisms (like decay kinematics) directly into the likelihood function, significantly enhancing estimation accuracy.
3. Handling Detector Effects: The study confirms that MLE can naturally integrate acceptance corrections via weighting, offering a unified framework for handling both non-flow and detector inefficiencies.
4. Clarification of Bias Mechanisms: The work clarifies the interplay between opening angles, event-plane correlations, and the resulting bias in flow extraction, resolving apparent contradictions with previous literature.

5. Significance and Conclusion

The study concludes that Maximum Likelihood Estimation is a compelling alternative to traditional flow analysis methods. Its primary advantages are:

• Superior Accuracy: It provides more accurate estimates of flow harmonics in the presence of non-flow, particularly in low-multiplicity or small-system collisions where non-flow is dominant.
• Flexibility: The method allows for the inclusion of complex physical scenarios (decay, conservation laws) directly into the statistical model, which is difficult with standard correlator approaches.
• Robustness: It effectively handles detector acceptance issues without requiring separate correction steps.

While the study utilized toy models with exaggerated non-flow effects, the results suggest that MLE could be a vital tool for future analyses of small collision systems (e.g., p+Pb, d+Au) and high-precision measurements where isolating genuine collective behavior from non-flow backgrounds is critical. The authors plan to extend this work to real experimental data and systems with even smaller multiplicities.