PeV Gamma Rays from Cygnus X-3 Test a Cornerstone of Physics

Special relativity rests on Lorentz invariance: physical laws remain unchanged under Lorentz transformations, which relate measurements made by observers moving at constant velocities relative to one another. Some proposed extensions of established physics allow small departures from this symmetry. This study asks whether particular departures could make the highest-energy gamma rays unstable before they reach Earth.

The researchers test superluminal Lorentz-invariance violation (LIV) scenarios described by modified dispersion relations. In these models, photons above certain energies could decay into an electron–positron pair or split into three lower-energy photons. Either process could strongly reduce the number of photons arriving at the highest energies, producing a sharp spectral cutoff.

The target is Cygnus X-3, a binary system about 9 kiloparsecs away. LHAASO observations found gamma rays from the source with energies extending up to 3.7 petaelectronvolts (PeV). One PeV is 10¹⁵ electronvolts—one thousand trillion electronvolts. In the paper’s table, the maximum reconstructed energy is 3.735 ± 0.411 PeV; that measurement is distinct from the stated reach of the detected photons. The observations came during four flaring periods recorded between 2019 and 2024.

Searching for a cutoff

The team analyzed data from LHAASO’s KM2A detector, which is sensitive to gamma rays above roughly 10 teraelectronvolts. They used a three-dimensional fit to event positions and reconstructed energies, comparing a no-cutoff hypothesis with hypotheses that impose a hard cutoff at different energies. The test was repeated under two assumed intrinsic source spectra: a photomeson-production model used as the baseline, and a power-law model to assess sensitivity to that assumption.

The analysis also accounts for absorption by radiation fields along the route to Earth. Cygnus X-3 lies in a crowded region of the sky, so the fit included nearby gamma-ray sources. Their spatial and spectral parameters were fixed to values from a prior study, rather than independently determined in this analysis. The team used simulations to calibrate its limits because event counts at the highest energies are limited and the test statistic does not follow a simple standard distribution there. An event-by-event, or unbinned, analysis provided a cross-check.

The data show no evidence for an LIV-induced cutoff. For the baseline source spectrum, the analysis places a 95% confidence lower limit of 3.05 PeV on the cutoff energy: lower cutoff energies are excluded at that confidence level within the tested setup. The unbinned check gives a consistent limit of 3.06 PeV. These values constrain a possible feature; they are not measurements of an observed cutoff.

What the limits mean

The authors translate the cutoff limit into bounds on LIV energy scales using the predictions for each process. For second-order superluminal LIV, the photon-decay channel gives a lower limit of about 9.09 × 10²⁴ electronvolts, while photon splitting gives about 1.73 × 10²⁶ electronvolts—around 10⁻² of the Planck energy scale. The paper describes the photon-splitting bound as improving previous limits by nearly an order of magnitude for this second-order superluminal scenario. It also reports a first-order limit of about 2.72 × 10³⁴ electronvolts.

These bounds apply to the specific modified-dispersion scenarios and photon interactions tested. They do not show that Lorentz invariance is violated, nor do they rule out every possible LIV model or other ways such physics might appear.

There are uncertainties in translating the observed spectrum into a cutoff constraint. The authors identify the assumed intrinsic spectrum as the main systematic: changing from the baseline model to a power law shifts the cutoff limit by about 9.5%. They report a total systematic uncertainty of about 9.6% on that limit. Background uncertainty is described as negligible. The paper notes limited simulated detector-response statistics as an additional uncertainty for the unbinned analysis, which is why the binned analysis is used as the baseline.

This account is based on version 1 of an arXiv preprint, dated November 11, 2026. A preprint is a public manuscript, not by itself evidence of peer review. The result is best understood as a model-dependent constraint from a particular source and analysis—not a general verdict on the foundations of relativity.