Where does the neutrino Mass Ordering preference come from?
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: High Energy Physics: Phenomenology. The day in particle phenomenology: models, signatures and constraints.
Kai: We're looking at the paper: "Where does the neutrino Mass Ordering preference come from?".
Mira: Here's the gist of the paper. After the recent JUNO result, the comparison of the effective mass splittings measured in ν¯e disappearance at reactors and in νµ disappearance has become one of the leading handles on the neutrino mass ordering The combination disfavors the inverted ordering at 1.9σ This preference comes from a balance between several small preferences, with reactor data with T2K or with NOvA pointing to opposite orderings The Role of Mass Splitting Comparisons The mass ordering is determined by comparing the effective mass splittings measured in different channels, specifically comparing ν¯e disappearance at reactors, ∆m2ee, and νµ disappearance at accelerator and atmospheric experiments, ∆m2µµ For normal ordering ∆m2ee > ∆m2muu, while the opposite is true for inverted ordering This approach requires precise measurements of both splittings because a single effective splitting carries no information on the ordering The difference between these two effective splittings is of few percent, depending on mixing angles and the CP phase Data Sets Driving Preference The present preference for the ordering depends on which data sets drive it, as a shift of about one percent in a single experiment can change the outcome The analysis combines reactor experiments JUNO and Daya Bay with the accelerator experiments T2K and NOvA, together with MINOS, IceCube/DeepCore and their combination in NuFIT 6.1 The two most precise νµ disappearance experiments, when combined with reactor measurements, point to different orderings Tension Between Experiments The preference is driven by a tension between experiments that point to opposite orderings; for instance, when combined with reactors, NOvA disfavors IO at 1.4σ while T2K disfavors NO at 0.8σ The values of ∆m231 preferred by T2K and NOvA differ by about 3.5%, which is comparable to the difference between ∆m2ee and ∆m2muu when switching the mass ordering This difference does not come from the CP phase, as the cos δCP term changes ∆m2µµ by at most 0.4% Future Prospects and Limitations Further JUNO data will most likely not increase the mass ordering preference beyond 3σ because the comparison is then limited by the precision of ∆m2muu The extraction of ∆m2muu relies on neutrino energy reconstruction, which depends on the modeling of neutrino-nucleus interactions and experimental treatment More JUNO data will address both features that drive the present preference The combination of reactor and νµ disappearance data disfavors IO at 1.9σ The final preference is a balance between experiments that point to opposite orderings, leading to a mild IO rejection at 1.9σ when combining JUNO and Daya Bay with the sum of the T2K and NOvA χ2 curves The ordering is then decided by where the accelerator interval lies with respect to a fixed reactor value The preference for NO does not come from a set of consistent indications, as the two most precise accelerator experiments point to opposite orderings Appendix A: Statistical Method The statistical treatment involves computing TMO = min x χ2IO - min x χ2NO, which is positive when NO describes the data better The tension between experiments is computed using the parameter goodness-of-fit test T = min h χ2J + X i Gi - min χ2J - X i min Gi The CLs method is used to determine rejection, where CLs(IO) = pIO / (1 − pNO) Pseudo-experiments are generated to account for the non-parabolic nature of the JUNO χ2 and to assess statistical fluctuations in the global minimum The CLs method includes this information by dividing the probability of the tested ordering by the probability that the other ordering produces the observed outcome The significance is converted into a one-sided significance, nσ = Φ−1(1 − p), with Φ being the cumulative standard normal distribution The CLs method plays a role when data are unusual for both orderings, as it includes this information by dividing the probability of the tested ordering by the probability that the other ordering produces the observed outcome The CLs(NO) plays a role when the data favor IO The CLs(IO) grows when NO is typical, and CLs(NO) grows when IO is typical Table II Summary The results show that JUNO+Daya Bay+Acc+IC yields a CLs of 1.9σ for NO and 3.1σ for IO The tension in ∆m231 in each ordering is quantified by TMO, which is positive when NO describes the data better The CLs values are calculated based on the probability density distributions generated from pseudo-experiments under each ordering The observed TMO = 5.91 lies in the tail of the IO distribution, so that IO rarely produces it, with pIO = 0.013 (2.2σ) The CLs(IO) is 0.027 (1.9σ) Figure 7 Interpretation Figure 7 shows the two distributions for JUNO+Daya Bay+Acc+IC, where the observed TMO = 5.91 lies in the tail of the IO distribution, so that IO rarely produces it, with pIO = 0.013 (2.2σ) Since this value sits at the centre of the NO distribution, CLs only slightly weakens the rejection, with CLs(IO) = 0.027 (1.9σ) The observed TMO lies on the side of the tested ordering, so that IO is disfavored when it rarely produces it Table III Comparison The comparison with the JUNO collaboration shows that with Acc+IC, the IO pseudo-experiments reach the observed value less easily, and the rejection becomes 2.2σ The column J+DB+acc.
Kai: combines JUNO and Daya Bay with the accelerator curve of each row The comparison with Ref.
13: shows TMO = 4.85 and pIO = 0.018 (2.1σ) The column Acc+IC shows TMO = 5.91 and pIO = 0.013 (2.2σ) Final Conclusion on Preference The present preference for NO does not come from a set of consistent indications, as the two most precise accelerator experiments point to opposite orderings The values of ∆m231 preferred by T2K and NOvA differ by about 3.5%, which is comparable to the difference between ∆m2ee and ∆m2muu that distinguishes the two orderings The preference for NO is not robust, as a shift of this size in a single experiment can change the preferred ordering The next JUNO analysis will most likely reveal this tension to be a statistical fluctuation Table I Summary Table I shows that when T2K’s χ2 curve is moved to NOvA’s best fit, both agree with reactors in NO, the whole tension falls on IO, which then would be rejected at 1.8σ With T2K’s χ2 moved to NOvA’s best fit, both agree with reactors in NO, the whole tension falls on IO, which then would be rejected at 1.8σ The same effect enters the confidence intervals of JUNO, which we compute with the Feldman–Cousins construction The CLs(IO) is 1.9σ Figure 6 Interpretation Figure 6 shows that the distributions of T have a much longer tail than the χ2 expectation, populated by the pseudo-experiments in which the JUNO best fit jumps to a neighbouring minimum Read as √T, the tension of the full combination is 2.4σ in NO and 3.1σ in IO The same effect enters the confidence intervals of JUNO, which we compute with the Feldman–Cousins construction The CLs(IO) is 1.8σ Figure 4 Interpretation Figure 4 shows that as more data significantly reduces the reactor uncertainty on the mass splitting, the rejection of IO would only grow from 1.9σ to 2.8σ In the limit of a perfect reactor measurement, the rejection of IO would asymptote at 2.
Paper summary: Kai: So, we're looking at this paper called "Where does the neutrino Mass Ordering preference come from?". It's basically asking which specific data sets are actually pushing us toward one mass ordering or the other.
Mira: Exactly. The main thrust of the paper is that when you compare how we measure different things—the effective mass splittings in reactor experiments versus accelerator experiments—that comparison becomes a really strong way to decide if the neutrino masses are normal or inverted.
Kai: But it doesn't just pick one winner; it shows that the preference for one ordering, say inverted ordering, is actually quite mild at one point nine sigma when you combine all these different data sets together <ref:2611.13629#pg1>.
Lev: That's important because if the preference isn't strong, then we have to be careful about what we claim right now. It comes from a balance between several small preferences instead of one big signal that screams one way or the other, which is a lot to untangle for us.
Mira: Right. The authors point out that this preference doesn't come from just one measurement; it's this delicate trade-off between different experiments, like reactor data with T2K or NOvA pointing in opposite directions and then being balanced by other things.
Kai: It really highlights how much the final answer depends on which specific data set you let carry the most weight in your analysis. You can see a shift of about one percent in just one experiment can actually flip the preferred ordering entirely.
Lev: That's where my concern is—if that sensitivity is that high, then we need to be super careful about how much we trust any single experiment when we try to nail down the mass ordering.
Mira: And the paper lays out how they combine JUNO and Daya Bay with T2K and NOvA, along with MINOS, IceCube/DeepCore in NuFIT six point one to see this tension clearly <ref:2611.13629#pg3>. They show that these two most precise accelerator experiments, when paired with reactor measurements, actually point to different orderings.
Paper summary: Kai: That's the core of it then: the combination of reactor and neutrino disappearance data points in opposite directions when you look at them together.
Mira: And they quantify this tension by looking at how much the value of alpha, which is Delta m two three one, preferred by T2K and NOvA actually differs from what we see between the effective mass splittings, which is what drives the ordering choice.
Lev: That difference in alpha values between T2K and NOvA is about three point five percent, which they say is comparable to the difference you get when you switch from normal ordering to inverted ordering for those effective mass splittings.
Kai: So, it's like the two main data sources are pulling us in slightly different directions by a margin that's almost exactly the margin that separates normal from inverted scenarios.
Mira: And they show that this difference isn't actually coming from the CP phase term, because changing the CP phase only shifts alpha two three one by at most zero point four percent, which is way smaller than what we see here.
Lev: That means the tension isn't due to something we can easily model or measure with current tools like CP violation; it seems rooted in the structure of how these different experiments measure their respective splittings.
Kai: So, when you look at the results, they find a rejection of inverted ordering at one point nine sigma when combining JUNO and Daya Bay with the sum of the T2K and NOvA chi squared curves <ref:2611.13629#pg1,combining JUNO and Daya Bay with>.
Mira: That one point nine sigma figure comes from using their CLs method to see how much data favors one ordering over the other, including accounting for those non-parabolic features in the JUNO chi squared distribution <ref:2611.13629#pg1>.
Lev: The paper notes that because of these statistical fluctuations and modeling uncertainties, there's a tension between the experiments that is quantified by TMO, which is positive when NO describes the data better.
Kai: And when they look at Figure seven they see that the observed value for TMO is five point nine one, and this falls in the tail of the inverted ordering distribution, meaning inverted ordering rarely produces it with a probability of zero point zero one three or about two point two sigma significance.
Paper summary: Mira: Since that value sits right in the middle of where NO is typical, they say that the CLs for inverted ordering comes out to one point nine sigma, which is why we see that mild rejection there <ref:2611.13629#pg1>.
Lev: The paper also mentions that as more data comes in and reactor uncertainties get smaller, like with better JUNO data, the rejection of inverted ordering would actually grow from one point nine sigma up to two point eight sigma because the measurement record becomes more precise over time <ref:2611.13629#pg1>.
Kai: So the implication is that JUNO data will be crucial for pushing this preference further if we want to see a stronger signal on whether we're in normal or inverted ordering.
Mira: But they also state a clear limitation: the extraction of alpha two three one itself relies on neutrino energy reconstruction, which depends heavily on how well we model neutrino-nucleus interactions and the experimental treatment.
Lev: That is a major caveat because if our models for those interactions are off, then our error bars on alpha two three one get bigger, and that muddies all the comparison you're trying to make.
Kai: So, in short, the paper "Where does the neutrino Mass Ordering preference come from?" shows that the current preference is a delicate balance of several small preferences between different experiments.
Mira: And while it disfavors inverted ordering at one point nine sigma with this combination of data, they stress that this result is highly dependent on where you place the accelerator interval relative to a fixed reactor value <ref:2611.13629#pg1>.
Lev: The final conclusion is that the preference for normal ordering isn't super robust because the two most precise accelerator experiments point in opposite directions, and a shift of just about one percent in one experiment can change which ordering gets favored.
Kai: It’s not a simple yes or no yet; it’s a story about how different pieces of the puzzle interact and what kind of data we need to get to settle the question for sure.
Conclusion: Kai: So, this paper is looking at where we get our hints about whether neutrinos are normal or inverted ordering across all these different experiments.
Mira: Yeah, "Where does the neutrino Mass Ordering preference come from?" is a title that makes it clear they’re digging into those comparisons between reactor and accelerator data.
Kai: The authors combine data from JUNO, Daya Bay, T2K, and NOvA to see what the final picture looks like for normal versus inverted ordering.
Mira: It boils down to this comparison of effective mass splittings—how the neutrino masses affect electron neutrinos in reactors versus muon neutrinos in accelerators.
Kai: The authors find that when you put all these pieces together, they get a preference for normal ordering at about one point nine sigma, but it’s really a very thin line.
Mira: It means the preference isn't super strong; it’s just a mild rejection of the inverted case when you look at this specific combination of measurements.
Kai: The real kicker is that they show this preference isn't solid because there are these tensions between different experiments pointing in opposite directions, which makes the final result really sensitive to where you place those accelerator results.
Mira: So, while they see a rejection of inverted ordering at one point nine sigma, the authors themselves are careful to say that this preference might just be a statistical fluctuation if we don't get more data.
Kai: It sounds like JUNO’s future measurements are going to be key because they could push that rejection up to two point eight sigma if the reactor uncertainty gets tight enough.
Pedro A. N. Machado, *, Joãao Paulo Pinheiro
Fermi National Accelerator Laboratory · State Key Laboratory of Dark Matter Physics, Tsung-Dao Lee Institute & School of Physics and Astronomy · Key Laboratory for Particle Astrophysics and Cosmology (MOE) & Shanghai Key Laboratory for Particle Physics and Cosmology
hep-ph
Submitted: 2026-11-10
Updated: 2026-11-10
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 73/100
The gist: After the recent JUNO result, the comparison of the effective mass splittings measured in ν¯e disappearance at reactors and in νµ disappearance has become one of the leading handles on the…
Key concepts
- Effective Mass Splittings
- These are measurements of how neutrino masses contribute to flavor change in different disappearance channels. Comparing the splitting measured in electron neutrino disappearance ($ u_e$) at reactors ($ ext{∆}m_{2ee}$) with the splitting measured in muon neutrino disappearance ($ u_ au$) at accelerators ($ ext{∆}m_{2 au au}$) is crucial because a single measurement doesn't tell you the mass ordering.
- Mass Ordering
- This refers to whether neutrinos have a Normal Ordering (NO) or an Inverted Ordering (IO) for their mass squares. The comparison of the two effective mass splittings mentioned above is the primary method used to determine which ordering is favored by the experimental data.
- Tension Between Experiments
- This describes the conflict where different experiments point towards opposite neutrino mass orderings. For example, NOvA might disfavor IO at 1.4$ ext{--}1.5 ext{--} ext{sigma}$ while T2K disfavors NO at 0.8$ ext{--}1 ext{--} ext{sigma}$. This tension is a key driver in the final preference calculation.
- CLs Method
- This statistical method is used to determine the rejection of an ordering (like IO) by calculating the probability that it produces the observed data compared to another ordering (like NO). It accounts for statistical fluctuations and helps quantify how well each ordering fits all the combined experimental results.
Terminology
Summary
After the recent JUNO result, the comparison of the effective mass splittings measured in ν¯e disappearance at reactors and in νµ disappearance has become one of the leading handles on the neutrino mass ordering The combination disfavors the inverted ordering at 1. p. 09σ This preference comes from a balance between several small preferences, with reactor data with T2K or with NOvA pointing to opposite orderings
The Role of Mass Splitting Comparisons
The mass ordering is determined by comparing the effective mass splittings measured in different channels, specifically comparing ν¯e disappearance at reactors, ∆m2ee, and νµ disappearance at accelerator and atmospheric experiments, ∆m2µµ For normal ordering ∆m2ee > ∆m2muu, while the opposite is true for inverted ordering This approach requires precise measurements of both splittings because a single effective splitting carries no information on the ordering The difference between these two effective splittings is of few percent, depending on mixing angles and the CP phase
Data Sets Driving Preference
The present preference for the ordering depends on which data sets drive it, as a shift of about one percent in a single experiment can change the outcome The analysis combines reactor experiments JUNO and Daya Bay with the accelerator experiments T2K and NOvA, together with MINOS, IceCube/DeepCore and their combination in NuFIT 6.1 The two most precise νµ disappearance experiments, when combined with reactor measurements, point to different orderings
Tension Between Experiments
The preference is driven by a tension between experiments that point to opposite orderings; for instance, when combined with reactors, NOvA disfavors IO at 1.4σ while T2K disfavors NO at 0. p. 48σ The values of ∆m231 preferred by T2K and NOvA differ by about 3. p. 05%, which is comparable to the difference between ∆m2ee and ∆m2muu when switching the mass ordering This difference does not come from the CP phase, as the cos δCP term changes ∆m2µµ by at most 0. p. 34%
Future Prospects and Limitations
Further JUNO data will most likely not increase the mass ordering preference beyond 3σ because the comparison is then limited by the precision of ∆m2muu The extraction of ∆m2muu relies on neutrino energy reconstruction, which depends on the modeling of neutrino-nucleus interactions and experimental treatment More JUNO data will address both features that drive the present preference
Conclusion on Mass Ordering
The combination of reactor and νµ disappearance data disfavors IO at 1.9σ The final preference is a balance between experiments that point to opposite orderings, leading to a mild IO rejection at 1. p. 09σ when combining JUNO and Daya Bay with the sum of the T2K and NOvA χ2 curves The ordering is then decided by where the accelerator interval lies with respect to a fixed reactor value The preference for NO does not come from a set of consistent indications, as the two most precise accelerator experiments point to opposite orderings
Appendix A: Statistical Method
The statistical treatment involves computing TMO = min x χ2IO - min x χ2NO, which is positive when NO describes the data better The tension between experiments is computed using the parameter goodness-of-fit test T = min h χ2J + X i Gi - min χ2J - X i min Gi The CLs method is used to determine rejection, where CLs(IO) = pIO / (1 − pNO) Pseudo-experiments are generated to account for the non-parabolic nature of the JUNO χ2 and to assess statistical fluctuations in the global minimum The CLs method includes this information by dividing the probability of the tested ordering by the probability that the other ordering produces the observed outcome The significance is converted into a one-sided significance, nσ = Φ−1(1 − p), with Φ being the cumulative standard normal distribution The CLs method plays a role when data are unusual for both orderings, as it includes this information by dividing the probability of the tested ordering by the probability that the other ordering produces the observed outcome The CLs(NO) plays a role when the data favor IO The CLs(IO) grows when NO is typical, and CLs(NO) grows when IO is typical
Table II Summary
The results show that JUNO+Daya Bay+Acc+IC yields a CLs of 1.9σ for NO and 3. p. 01σ for IO The tension in ∆m231 in each ordering is quantified by TMO, which is positive when NO describes the data better The CLs values are calculated based on the probability density distributions generated from pseudo-experiments under each ordering The observed TMO = 5. p. 391 lies in the tail of the IO distribution, so that IO rarely produces it, with pIO = 0. p. 8 p. 8013 (2.2σ) The CLs(IO) is 0. p. 8027 (1.9σ)
Figure 7 Interpretation
Figure 7 shows the two distributions for JUNO+Daya Bay+Acc+IC, where the observed TMO = 5.91 lies in the tail of the IO distribution, so that IO rarely produces it, with pIO = 0.013 (2.2σ) Since this value sits at the centre of the NO distribution, CLs only slightly weakens the rejection, with CLs(IO) = 0. p. 8027 (1.9σ) The observed TMO lies on the side of the tested ordering, so that IO is disfavored when it rarely produces it
Table III Comparison
The comparison with the JUNO collaboration shows that with Acc+IC, the IO pseudo-experiments reach the observed value less easily, and the rejection becomes 2.2σ The column J+DB+acc. p. 8 combines JUNO and Daya Bay with the accelerator curve of each row The comparison with Ref. p. 8 [13] shows TMO = 4. p. 685 and pIO = 0. p. 6018 (2.1σ) The column Acc+IC shows TMO = 5. p. 891 and pIO = 0. p. 6013 (2.2σ)
Final Conclusion on Preference
The present preference for NO does not come from a set of consistent indications, as the two most precise accelerator experiments point to opposite orderings The values of ∆m231 preferred by T2K and NOvA differ by about 3.5%, which is comparable to the difference between ∆m2ee and ∆m2muu that distinguishes the two orderings The preference for NO is not robust, as a shift of this size in a single experiment can change the preferred ordering The next JUNO analysis will most likely reveal this tension to be a statistical fluctuation
Table I Summary
Table I shows that when T2K’s χ2 curve is moved to NOvA’s best fit, both agree with reactors in NO, the whole tension falls on IO, which then would be rejected at 1.8σ With T2K’s χ2 moved to NOvA’s best fit, both agree with reactors in NO, the whole tension falls on IO, which then would be rejected at 1. p. 38σ The same effect enters the confidence intervals of JUNO, which we compute with the Feldman–Cousins construction The CLs(IO) is 1. p. 69σ
Figure 6 Interpretation
Figure 6 shows that the distributions of T have a much longer tail than the χ2 expectation, populated by the pseudo-experiments in which the JUNO best fit jumps to a neighbouring minimum Read as √T, the tension of the full combination is 2.4σ in NO and 3. p. 21σ in IO The same effect enters the confidence intervals of JUNO, which we compute with the Feldman–Cousins construction The CLs(IO) is 1. p. 68σ
Figure 4 Interpretation
Figure 4 shows that as more data significantly reduces the reactor uncertainty on the mass splitting, the rejection of IO would only grow from 1.9σ to 2.8σ In the limit of a perfect reactor measurement, the rejection of IO would asymptote at 2. p. 4
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Ability to determine which data sets drive mass ordering preference by analyzing combinations of reactor experiments with accelerator data, specifically identifying that
the combination of reactor and νµ disappearance data points to opposite orderings.
-
Capability to quantify the tension between experimental results using a formula like TMO = min x χ2 IO - min x χ2 NO, allowing for a precise calculation of the rejection of an ordering in units of chi squared.
-
Skill in comparing the difference between effective mass splittings, specifically noting that
the values of the mass splitting preferred by these two experiments differ by about 3.5%, comparable to the difference expected between the two orderings.
-
Accuracy in predicting future sensitivity, such as determining that
further JUNO data would not bring the rejection of the inverted ordering beyond about 3σ, as the comparison is then limited by the precision of the νµ disappearance experiment.
-
Proficiency in simulating complex statistical scenarios using pseudo-experiments to assess whether observed tensions are genuine physical preferences or merely
statistical fluctuation
when dealing with non-parabolic chi squared profiles.
Sources
- Evidence for oscillation of atmospheric neutrinos
- Atmospheric neutrino oscillation analysis with neutron tagging and an expanded fiducial volume in Super-Kamiokande I-V
- Measurement of Atmospheric Neutrino Mixing with Improved IceCube DeepCore Calibration and Data Processing
- Direct Evidence for Neutrino Flavor Transformation from Neutral-Current Interactions in the Sudbury Neutrino Observatory
- Combined Analysis of all Three Phases of Solar Neutrino Data from the Sudbury Neutrino Observatory
- Solar neutrino measurements using the full data period of Super-Kamiokande-IV
- Simultaneous Precision Spectroscopy of $pp$, $^7$Be, and $pep$ Solar Neutrinos with Borexino Phase-II
- First Results from KamLAND: Evidence for Reactor Anti-Neutrino Disappearance
- Constraints on $\theta_{13}$ from A Three-Flavor Oscillation Analysis of Reactor Antineutrinos at KamLAND
- Precision measurement of reactor antineutrino oscillation at kilometer-scale baselines by Daya Bay
- Measurement of Reactor Antineutrino Oscillation Amplitude and Frequency at RENO
- First measurement of reactor neutrino oscillations at JUNO
- Precision constraints for three-flavor neutrino oscillations from the full MINOS+ and MINOS data set
- Results from the T2K experiment on neutrino mixing including a new far detector $\mu$-like sample
- Precision measurement of neutrino oscillation parameters with 10 years of data from the NOvA experiment
- Planck 2018 results. VI. Cosmological parameters
- DESI DR2 Results II: Measurements of Baryon Acoustic Oscillations and Cosmological Constraints
- Constraints on Neutrino Physics from DESI DR2 BAO and DR1 Full Shape
- Toward the discovery of matter creation with neutrinoless double-beta decay
Related papers
- MadNIS at NLO
- The NANOGrav 15-year data set: A Limit on Continuous Spin Gravitons
- DIS-secting the universe with high-energy neutrinos
- Crowdsourcing Gravitational Waves from Superradiant Axions
- Classification of g-modes for neutron stars with a strong transition: Novel universal relation including slow stable hybrid stars
- Scalar axion field of toroidal electromagnetic pulses