Study: brain sits near—not at—critical point
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Rubén Calvo Ibáñez and colleagues at Universidad de Granada published a study in Physical Review Letters showing that widely used signatures of brain criticality—including power-law scaling in eigenvalue spectra and renormalization group analyses—can be statistical artifacts when neural signals are autocorrelated and data is limited.
- A deliberate counterexample model with zero connectivity between brain regions still produced covariance statistics with fake power-law tails when driven by slow noise, demonstrating that autocorrelation alone can mimic criticality in systems with no genuine collective dynamics.
- The team's three-tool framework—time-shift randomization that shuffles each region's timeline independently, data pooling across participants, and exponent matching against recurrent model predictions—successfully separates true collective dynamics from sampling artifacts.
- Applied to the LEMON dataset (resting-state fMRI from 136 healthy participants, 183 brain regions, ~10-minute sessions), the framework returned an effective coupling strength of approximately 0.88—close to but safely below the critical threshold of 1.0.
- Time-shift randomization caused the near-critical signatures to collapse almost entirely, confirming the original signals reflected true collective dynamics rather than artifacts, and the extracted scaling exponents matched predictions from a linear recurrent firing-rate model.
- fMRI's slow BOLD signal and short recording sessions make it an 'ideal breeding ground' for spurious criticality signatures, the authors note, a vulnerability that likely affected prior studies in the field.
- The researchers suggest the brain's position slightly below criticality may preserve computational benefits like multiscale collective modes and strong amplification while avoiding instability from sitting exactly at the threshold.
Why it matters: This matters because it gives the brain-criticality field a practical tool to separate real collective dynamics from statistical noise, while also refining the core finding: the brain sits at ~0.88 coupling strength, not 1.0, suggesting it deliberately stays below the critical threshold for robustness. Researchers in neuroscience and reservoir computing—who have built AI architectures around the 'edge of chaos' idea—now have both a critique of prior methods and a reusable framework applicable beyond brain data.
