The clearest clue is at the start
Data from relativistic heavy-ion collisions point, within a detailed model, to a plasma that began with both light and strange quark sectors below equilibrium. The clearest imbalance was in the strange sector: the inferred initial strange-quark fugacity had a median of 0.31, compared with 0.69 for light quarks. In this analysis, fugacity expresses a flavor's starting abundance relative to equilibrium: 1 means equilibrium, and less than 1 means undersaturation.
That distinction was more convincing than the proposed difference in equilibration speed. The posterior probability that the initial strange fugacity was below the light one was about 0.90, with robustness variants ranging from 0.85 to 0.97. The probability that strange quarks equilibrated more slowly was about 0.71, ranging from 0.65 to 0.78, a preference the analysis described as weak.
The inferred times, meaning the model's estimates for how long each sector takes to approach equilibrium, pointed in the same direction, but their uncertainty was substantial. The median strange-quark equilibration time was 4.13 fm/c, with a 68% credible interval reported as plus 3.11 and minus 1.46 fm/c. The light-quark median was 2.92 fm/c, with plus 2.80 and minus 1.96 fm/c. Those broad ranges are why the study does not treat slower strange equilibration as established.
How the model reads the collisions
The conclusion comes from a forward model, meaning a simulation chain that turns assumed collision conditions into predicted observables. It combines fluctuating initial conditions, viscous hydrodynamics, the conversion of fluid into particles, hadronic rescattering and a non-equilibrium fugacity sector for light and strange quarks.
To map the model's sensitivity, the flavor-independent study generated 10,000 minimum-bias events at midrapidity. It applied a pseudorapidity cut of 0.5 and sorted results into 10% centrality bins defined by total initial entropy.
In those simulations, delayed equilibration was associated with nearly unchanged charged multiplicity and reduced transverse flow, the collective motion of the produced matter. Separating light and strange equilibration redistributed identified-hadron yields according to the valence flavors in those hadrons. These are model-dependent sensitivity results, not direct causal estimates from experiments.
The Bayesian calibration was anchored to midrapidity Au+Au data at a center-of-mass energy of 200 GeV, spanning 0% to 80% centrality. That calibration scope is important when reading the inferred values: they describe what this model can learn from that dataset, rather than a measurement made independently of the model.
The inference also covered only part of the model design. All 14 Au+Au-relevant parameters were varied in the design, but only 8 were sampled and 6 were fixed because the design was too sparse for a joint calibration. The sampled parameters were given independent uniform priors.
The researchers trained a Gaussian-process emulator, a statistical approximation of the model's outputs, across a Latin-hypercube design of 500 points. Approximately 250 completed simulations were available for training. After standardizing the observables, they retained 9 principal components, which explained 99% of the design variance.
A useful result with sharp limits
Validation was broadly reassuring but not perfect. Cross-validation gave a median emulator error of 0.28 design-spread units. Standardized residuals had a width of 1.16 against a unit-normal target, indicating mild overconfidence; the analysis corrected this by globally inflating emulator uncertainty.
In a separate leave-one-out closure test, 100 held-out design points produced broadly nominal credible-interval coverage, especially for the chemical equilibration parameters. The light initial fugacity showed slight over-coverage at the narrower interval.
Once calibrated, the model's posterior predictive bands, which show the range of outputs allowed by the inferred parameters, narrowed and generally followed measured trends in particle yields, transverse energy, mean transverse momenta and charged flow. The fit still missed some details: kaon mean transverse momentum was about one standard deviation below the data in central and mid-central collisions, central Xi yields were underpredicted and central transverse energy was overpredicted.
The study also kept Lambda and proton observables outside the likelihood. Those comparisons showed opposite mismatches: Lambda was underpredicted in central collisions, while proton production was overpredicted at all centralities. In the most central comparison, the posterior proton value was near 39, against a measured value of approximately 32.
Taken together, the calibration supports an initially undersaturated flavor sector more clearly than it supports a hierarchy of equilibration times. The initial strange sector was more strongly below equilibrium than the light sector, but the reported probability of slower strange equilibration remained weak. The result is therefore a model-based constraint within the calibrated Au+Au setup, not a universal equilibration clock.
The document is identified as arXiv version 1 and a PhD dissertation dated 2026. It reports support from the U.S. Department of Energy, NERSC and the OSG Consortium through the grants, contracts and awards listed in its acknowledgements.
Paper data and sources
Original title: Constraining Light and Strange Flavor Equilibration in Relativistic Heavy-Ion Collisions
Authors: Andrew Gordeev
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-27
DOI: Not available
Original paper · Full text