A steady reference point for chaotic neutrino flavor changes

Neutrinos help exchange energy and lepton number between matter and radiation in collapsing stars and neutron-star mergers. How those particles are distributed among different flavors matters for heating and for the composition of expelled material. Yet their collective quantum evolution can unfold on scales too small for global astrophysical simulations to resolve directly.

A preprint by Jiabao Liu, Lucas Johns and Hiroki Nagakura, submitted on November 8, 2026, tackles one part of this problem: predicting the outcome of collisional flavor instability. Its central finding is that a stationary mathematical solution can provide a useful estimate of average flavor populations even when the actual evolution remains oscillatory or chaotic.

When collisions do more than damp motion

The instability arises when differences between neutrino and antineutrino collision rates amplify quantum coherence through collective neutrino self-interactions. Coherence describes the quantum relationships between flavors; flavor populations describe how much of each flavor is present.

The authors model two flavors, labeled electron and x, together with their antineutrino counterparts. Their equations include collective self-interactions and a simplified collision prescription. Collisions restore populations toward specified target values while damping coherence. The model neglects vacuum and matter contributions to flavor evolution.

The main construction assumes a spatially uniform gas with no preferred direction. Within that setting, the researchers seek states whose flavor populations remain constant while their coherences rotate together. Viewed in a frame rotating at the same rate, these homogeneous collective-precession states are fixed points.

Finding them requires solving just two self-consistency equations, even when the model contains many neutrino energies and energy-dependent collision rates. Each energy responds to a shared collective field; the combined responses must reproduce that field.

A fixed point need not be the destination

The researchers then ask whether the evolving system actually approaches these states. For detailed examples, they use a single energy bin and vary only the electron-neutrino collision rate between runs, keeping all parameters fixed during each evolution.

The outcomes differ sharply. Some trajectories approach collective precession with stationary populations. Others retain repeating population oscillations. As the collision rate changes, a repeating pattern splits into two, then four, and progressively more distinct maxima before irregular behavior appears.

Numerical calculations resolve eight period doublings, reaching a stable cycle with 256 maxima before repetition. The spacing of successive doubling thresholds approaches the Feigenbaum constant, approximately 4.669, a characteristic numerical pattern associated with period-doubling routes to chaos.

Irregularity alone would not establish chaos. The authors also measure how nearby trajectories separate. Positive growth persists when the integration step is reduced and across ten separate late-time intervals, providing strong numerical evidence for deterministic chaos in the tested model.

Another result cautions against equating a slowly growing instability with a small eventual effect. Depending on the solution branch, weak initial growth can precede a finite change in flavor populations. In some cases, the eventual outcome also depends on how the system was prepared.

Predicting averages rather than every fluctuation

Despite this complicated motion, the independently calculated fixed-point populations remain useful reference values. Across 1,594 trajectory records in the sampled periodic and chaotic regimes, their largest relative discrepancy from time-weighted mean occupations is 8.03%. The same solution branch is used throughout, without fitting a different prescription to each regime.

This is an empirical result for those tests, not a universal error bound. Stationary populations and long-time averages satisfy the same linear collision-balance constraints, but nonlinear correlations prevent them from being exactly equal.

The authors also report selected one-dimensional periodic-box calculations in which spatially averaged populations remain close to the fixed-point prediction, with maximum relative errors around 10%. That agreement concerns populations, not the complete quantum state.

What remains to be established

The fixed-point construction supports general energy spectra, but the exact transfer of the demonstrated chaotic dynamics to multiple energies requires energy-independent collision rates and compatible target spectra. It does not establish the same behavior for arbitrary realistic spectra.

This theoretical preprint does not calculate a changed supernova explosion or heavy-element yield. Instead, it supplies a candidate ingredient for models that approximate unresolved flavor evolution. Incorporating it requires care: simply adding relaxation toward these populations to unchanged classical collisions could count population restoration twice. Testing that coupling in realistic transport remains a separate task.