Nara Moraes

Abstract

In nuclear theory, the characterization of transport parameters and phase transition properties of a new state of matter, the quark-gluon plasma (QGP), believed to have existed only in the earliest moments after the Big Bang, has been a central objective of scientists since the 1980s, when the development of quantum chromodynamics (QCD) predicted its existence. Since then, scientists have sought to reproduce the conditions necessary for its formation in large heavy-ion collision facilities, such as the LHC and RHIC. According to QCD, a smooth transition is expected to occur at energy densities of 1 GeV/fm³. More recent experiments reach center-of-mass energies of up to 5.2 TeV, sufficient to produce the conditions required for QGP formation. This state is estimated to last on the order of 10⁻²³ s, and therefore analyses of particle collision detector data can only probe properties of the matter after hadronization. The extraction of the physical parameters of the QGP is performed through the model-to-data comparison methodology, in which simulations of heavy-ion collisions are confronted with measured observables. Recent studies have achieved precise estimates of parameters such as the shear viscosity η/s using Bayesian inference, combined with Gaussian process emulators to efficiently explore the model parameter space. This approach accounts for multiple sources of uncertainty, including statistical and systematic uncertainties in the experimental data, uncertainties associated with the physical model and the emulation process, as well as correlations between different experimental observables. However, the correlation structure of these uncertainties has often been treated in a simplified manner, typically assuming independence between bins or employing limited approximations. In this work, we propose a systematic analysis of the impact of correlations in experimental uncertainties on the Bayesian inference of QGP parameters. In particular, we investigate how different structures of the covariance matrix of observables, including correlations between adjacent bins, global systematic effects, and varying correlation lengths, affect the shape of the posterior distribution, the width of marginal parameter distributions, and the correlation structure among them, with the aim of achieving a more robust and accurate characterization of the properties of the quark-gluon plasma.

Keywords: Quark-Gluon Plasma, Heavy-Ion Collisions, Bayesian Inference, Gaussian Processes, Correlated Experimental Uncertainties.

Posters

Upcoming
3rd Quantum Computing School & School on Quantum Simulation in the NISQ era

Event: 3rd Quantum Computing School & School on Quantum Simulation in the NISQ era

Date: 16 to 27 November, 2026 & 09 to 13 November, 2026

Simulation Code & Results Dataset:

Abstract: Correlations among experimental uncertainties can have a significant impact on statistical inference, particularly when observables are analyzed through multivariate likelihoods. In Bayesian parameter estimation, these correlations are encoded in covariance matrices and directly affect the geometry of the likelihood and, consequently, the resulting posterior distributions. However, estimating covariance structures from finite and high-dimensional datasets remains a challenging statistical problem. In this work, we investigate quantum machine learning approaches for learning classical covariance matrices and their application to Bayesian inference. Parameterized quantum circuits are employed as covariance estimators, with the goal of reconstructing correlation structures from synthetic datasets. The study is initially performed using quantum circuit simulation and datasets with controlled correlation patterns, providing a benchmark in which the true covariance matrix is known. The reconstructed covariance matrices are subsequently incorporated into multivariate likelihood functions to investigate how errors in covariance estimation propagate to Bayesian posterior distributions. Quantum estimates are then compared with classical covariance estimators in terms of estimation accuracy and their impact on the posterior distributions of inferred model parameters.

Event: PHENOEXP Workshop

Date: 10 a 12/08/2026

Simulation Code & Results Dataset: RHICDetectorsAuAu200_uncertainty_correlation_study

Abstract: Bayesian inference has become the state-of-the-art framework for fitting theoretical models to experimental data from heavy-ion collisions at LHC and RHIC accelerator facilities, enabling both parameter estimation and uncertainty quantification in the extraction of quark-gluon plasma (QGP) parameters. The QGP is a deconfined state of strongly interacting matter under extreme conditions, expected to have existed in the first moments of the Universe and known to be produced in heavy-ion collisions at accelerator facilities in the first years of the 21st century. In this model-to-data comparison approach, the likelihood combines the data with the associated uncertainties from both the measurements and the theoretical model. In the case of observables such as rapidity distributions, where the measured signal can receive contributions from the central and forward detector regions, bin-to-bin correlations may carry significant information. Since experimental correlations are often not provided by detector collaborations, current approaches neglect them. As a first step, we test the impact of modeled correlation structures on the inferred QGP parameter posteriors. As a second step, we aim to explore how these correlations can be informed by the uncertainty information provided by the collaborations themselves. Ultimately, we investigate whether incorporating these correlations in the likelihood has a significant impact on the inferred QGP parameter posteriors.

Presentations

Upcoming
32nd Quark Matter

Event: 32nd Quark Matter

Date: 21–27 Mar 2027

Status: Pending acceptance

Simulation Code & Results Dataset:

Abstract: Experimental measurements in heavy-ion collisions are expected to have systematic and statistical uncertainties; however, event-by-event correlations of systematic uncertainties are typically not made available, and in almost all cases only diagonal error bars for the binned observables are provided, with little to no guidance for the off-diagonal components of the covariance matrix. Bayesian inference requires the full uncertainty matrix, and the final results depend on these correlations. We perform a multi-pronged investigation of this issue. We present statistical and machine-learning methods to extract information about how much of the error bar is correlated versus uncorrelated, including, when multiple experimental samples are available, a comparative investigation of classical and quantum-computing approaches to correlation inference. We validate these methods using known synthetic data to quantify their performance and limitations. We then quantify the effect of uncertainty correlations in the experimental covariance matrix on various real examples of Bayesian analyses of soft-sector observables. These analyses aim to provide a better understanding of the impact of experimental uncertainty correlations on posterior distributions, informing their treatment in Bayesian inference and, more generally, in model-to-data comparisons.

Upcoming
XV LASNPA

Event: XV LASNPA

Date: 07 to 10 December, 2026

Simulation Code & Results Dataset:

Abstract: Bayesian inference has become the state-of-the-art framework for fitting theoretical models to experimental data from heavy-ion collisions at the LHC and RHIC accelerator facilities, enabling both parameter estimation and uncertainty quantification in the extraction of quark-gluon plasma (QGP) properties. The QGP is a deconfined state of strongly interacting matter that is formed under extreme conditions, expected to have existed only during the first moments of the Universe (and possibly in the cores of high-density stars), and known to be produced in heavy-ion collisions at accelerator facilities since the first years of the 21st century. Although remarkable, the QGP state created in heavy-ion collisions lasts for approximately 10−23 s, and its properties must be inferred through a complex set of theoretical models that take into account the whole chain of events, including the collision initial conditions, relativistic hydrodynamics, particlization, and hadronic transport, simulated sequentially and finally compared to the experimental hadronic observables measured in the final state. As a benchmark of its success, the model-to-data comparison approach using Bayesian inference predicted the shear viscosity with unprecedented precision and uncertainty quantification, as demonstrated by Bernhard et al. in 2019. In this model-to-data comparison approach, the likelihood combines the data with their associated uncertainties from both the experimental measurements and the theoretical models. During the parameter constraint process, in the case of observables such as pseudorapidity distributions, the measured data range is composed of contributions from the central and forward detector regions, which correspond to different detector systems, and thus experimental uncertainty correlations may carry significant information. Since experimental correlations are often not provided by the detector collaborations, current approaches neglect them in the likelihood. We verified that the addition of arbitrary experimental correlation models, such as compound symmetry and RBF, can shift the inferred posteriors by more than 0.8σ. We are now using machine learning methods to infer the underlying correlation structure from the experimental uncertainties and to investigate how incorporating these correlation structures into the likelihood impacts the inferred QGP parameter posteriors.

Upcoming
IFUSP Graduate Symposium

Event: IFUSP Graduate Symposium

Date: 19–23 October, 2026

Title: Study of the effect of correlated experimental uncertainty on Bayesian Inference of quark-gluon plasma properties

Abstract: Bayesian inference has become the state-of-the-art framework for fitting theoretical models to experimental data from heavy-ion collisions at the LHC and RHIC accelerator facilities, enabling both parameter estimation and uncertainty quantification in the extraction of quark-gluon plasma (QGP) properties. The QGP is a deconfined state of strongly interacting matter that is formed under extreme conditions, expected to have existed only during the first moments of the Universe (and possibly in the cores of high-density stars), and known to be produced in heavy-ion collisions at accelerator facilities since the first years of the 21st century. Although remarkable, the QGP state created in heavy-ion collisions lasts for approximately 10⁻²³ s, and its properties must be inferred through a complex set of theoretical models that take into account the whole chain of events, including the collision initial conditions, relativistic hydrodynamics, particlization, and hadronic transport, simulated sequentially and finally compared to the experimental hadronic observables measured in the final state. As a benchmark of its success, the model-to-data comparison approach using Bayesian inference predicted the shear viscosity with unprecedented precision and uncertainty quantification, as demonstrated by Bernhard et al. in 2019 [1]. In this model-to-data comparison approach, the likelihood combines the data with their associated uncertainties from both the experimental measurements and the theoretical models. In the parameter constraint process, for the case of observables such as pseudorapidity distributions, the measured data range includes contributions from the central and forward detector regions, which correspond to different detector systems, and thus experimental uncertainty correlations may carry significant information. Since experimental correlations are often not provided by the detector collaborations, current approaches neglect them in the likelihood [2]. We verified that the addition of arbitrary experimental correlation models, such as compound symmetry and RBF, can shift the inferred posteriors by more than 0.8σ. We are now using machine learning methods [3] to infer the underlying correlation structure from the experimental uncertainties and to investigate how incorporating these correlation structures into the likelihood impacts the inferred QGP parameter posteriors.

References:

1. J. E. Bernhard, J. S. Moreland, S. A. Bass, Nat. Phys. 15, 1113–1117 (2019).
2. A. Mankolli et al. (JETSCAPE Collaboration), Phys. Rev. C 114, 014905 (2026).
3. V. P. Soloviev, B. Adhikari, arXiv:2604.05637 [quant-ph] (2026).

Event: RETINHA XXXIII

Date: 24 a 26/06/2026

Simulation Code & Results Dataset: AlicePbPb2760_uncertainty_correlation_study

Abstract: In nuclear theory, the characterization of transport parameters and phase transition properties of a new state of matter, the quark-gluon plasma (QGP), believed to have existed only in the earliest moments after the Big Bang, has been a central objective of scientists since the 1980s, when the development of quantum chromodynamics (QCD) predicted its existence. Since then, scientists have sought to reproduce the conditions necessary for its formation in large heavy-ion collision facilities, such as the LHC and RHIC. According to QCD, a smooth transition is expected to occur at energy densities of 1 GeV/fm³. More recent experiments reach center-of-mass energies of up to 5.2 TeV, sufficient to produce the conditions required for QGP formation. This state is estimated to last on the order of 10⁻²³ s, and therefore analyses of particle collision detector data can only probe properties of the matter after hadronization. The extraction of the physical parameters of the QGP is performed through the model-to-data comparison methodology, in which simulations of heavy-ion collisions are confronted with measured observables. Recent studies have achieved precise estimates of parameters such as the shear viscosity η/s using Bayesian inference, combined with Gaussian process emulators to efficiently explore the model parameter space. This approach accounts for multiple sources of uncertainty, including statistical and systematic uncertainties in the experimental data, uncertainties associated with the physical model and the emulation process, as well as correlations between different experimental observables. However, the correlation structure of these uncertainties has often been treated in a simplified manner, typically assuming independence between bins or employing limited approximations. In this work, we propose a systematic analysis of the impact of correlations in experimental uncertainties on the Bayesian inference of QGP parameters. In particular, we investigate how different structures of the covariance matrix of observables, including correlations between adjacent bins, global systematic effects, and varying correlation lengths, affect the shape of the posterior distribution, the width of marginal parameter distributions, and the correlation structure among them, with the aim of achieving a more robust and accurate characterization of the properties of the quark-gluon plasma.

Reports

Title: Study of the effect of correlated experimental uncertainty on Bayesian Inference of quark-gluon plasma properties

Advisor: Prof. Matthew William Luzum

Project: 132734 / 2025-7 CNPq

Period: December 2025 to June 2026

Project