Lessons from the first JUNO results

TL;DR

This work analyzes JUNO's initial reactor-neutrino data to assess its impact on the large-mass-splitting $|Δm^2_{3\ell}|$ and the neutrino mass ordering, by reprocessing the JUNO spectrum with detailed signal, background, and systematics modeling and by integrating external constraints from global oscillation fits. The authors reproduce JUNO's $Δm^2_{21}$ and $θ_{12}$ and then show that JUNO combined with NuFIT-6.1 yields a modest preference for Normal Ordering with $Δχ^2_{IO-NO}$ around 3, corresponding to a $p$-value of roughly 2%. A full global fit including JUNO data boosts this to $Δχ^2_{IO-NO} ≈ 4.6$–$9.4$ depending on atmospheric data inclusion, indicating no definitive MO determination yet. Monte Carlo studies quantify the chance probability and show the result is within the expected statistical range, while robustness studies reveal that extreme systematic shifts could affect the MO significance but remain unlikely given JUNO's calibration constraints. Overall, JUNO's early results demonstrate significant sensitivity to the oscillation framework when combined with global data, and future data releases will clarify the MO question.

Abstract

First results from the JUNO reactor neutrino experiment already determine with world-leading precision the small neutrino squared-mass splitting $Δm^2_{21}$ and the mixing angle $θ_{12}$. In this article we perform an exploratory study beyond these, taking advantage of the first JUNO data release to discuss its sensitivity to the large squared-mass splitting, $Δm^2_{3\ell}$. When combined with constraints from global oscillation data, this may already contain some information on the neutrino mass ordering. Indeed, we find that the combination of the complementary $Δm^2_{3\ell}$-determinations gives a slight preference for Normal Ordering, with a p-value for Inverted Ordering of 2%-2.6% ($2.2σ$-$2.3σ$). We study the robustness of this result with respect to potential systematic uncertainties and statistical fluctuations. Taken at face value, a full global analysis of oscillation data including the publicly available JUNO information and data leads to a preference for Normal Ordering with $Δχ^2 = 4.6$ and 9.4 without and with Super-K and IceCube-24 atmospheric neutrino data, respectively.

Figures (7)

• Figure 1: Predicted spectra (dashed) compared to the official JUNO ones JUNO:2025gmd (grey). Left: based on our own predicted spectrum, normalized to match the normalization of their un-oscillated spectrum. Right: bin-per-bin rescaled to match the JUNO un-oscillated spectrum.
• Figure 2: Left: Determination of $\Delta m^2_{12}$ and $\sin^2\theta_{12}$ for the two configurations cnf 1 and cnf 2, compared to the JUNO results (black dashed line). Contours are for $1\sigma$, $2\sigma$, and $3\sigma$ (2 dof). Right: Best-fit reactor neutrino spectra without pull shifts (lower histograms) and reactor neutrino+background spectra with pull shifts (higher histograms) for cnf 1 (upper) and cnf 2 (lower).
• Figure 3: Impact of first JUNO data on the global determination of $\Delta m^2_{21}$ and $\sin^2 \theta_{12}$. We show the $1\sigma$, $2\sigma$, and $3\sigma$ allowed regions (2 dof) without (black) and with (colored) JUNO data.
• Figure 4: Dependence of $\chi^2_\text{JUNO}$ on $\Delta m^2_{3\ell}$ for cnf 1 and 2 (two left panels) and on $\Delta m^2_{ee}$ with an extended range for cnf 2 (right panel). In the left panels, dashed and dotted curves correspond to $\Delta\chi^2_\text{JUNO}$ (cnf 2) combined with the determination of $|\Delta m^2_{3\ell}|$ from NuFIT-6.1 global data without and with SK-ATM, respectively.
• Figure 5: Monte Carlo simulation of the $\Delta \chi^2_\mathrm{IO-NO}$ distribution for JUNO combined with the NuFIT-6.1 constraint on $|\Delta m^2_{3\ell}|$ without SK-ATM (left panel) and with SK-ATM (right panel). Black vertical lines indicate the values obtained by the observed data. The green shaded part in the left (right) panel contains 67.5% (69.5%) of the histogram for NO. Dashed curves show the Gaussian approximation based on the $T_0$-value from the Asimov data set.