Brain Turbulence — Whole-Brain Dynamics and Psychiatry
The math that describes a hurricane's energy cascade also describes a resting human brain. Same Kolmogorov exponents, same statistical signature, different substrate. A 10-minute fMRI scan, run through this lens, can guess whether a depressed patient will respond to an SSRI before they swallow the first pill — at AUC 0.70, against a clinical baseline of barely-better-than-chance.
Gustavo Deco's group at Pompeu Fabra in Barcelona built the framework. They treat the brain's blood-oxygen signal as a turbulent fluid and measure how correlated activity at large scales relates to correlated activity at small scales. The cascade exponents are the diagnostic. In responders, the cascade looks near-critical and regular. In non-responders, it's broken at baseline — the brain is too ordered, too local, information dies before it propagates.
How it works
The framework rests on the criticality hypothesis: the brain self-tunes near the phase boundary between two failure modes. Below it (subcritical), signals die out locally and global coordination fails. Above it (supercritical), signals cascade uncontrollably — the dynamical signature of epilepsy. At the critical point, dynamic range, information transmission, and computational repertoire all peak.
Empirical hooks: neural avalanches in fMRI and EEG follow power-law size and duration distributions with exponents near -3/2 and -2, matching the predictions of self-organized criticality from sandpile models. Closer-to-critical brains correlate with higher measured fluid intelligence.
Deco's contribution was importing Kolmogorov's 1941 cascade math directly. In a turbulent fluid, energy injected at large scales fragments into smaller eddies following the famous -5/3 spectral slope. Replace "energy at scale r" with "BOLD-signal correlation at distance r between brain regions," and the same statistical machinery applies. The cascade slope becomes a single number describing how well the brain moves information across spatial scales.
The clinical result
The March 2025 Molecular Psychiatry paper (Deco, Kringelbach and colleagues) ran the framework on 76 unmedicated MDD patients and 123 healthy controls. Patients did an 8-week trial of escitalopram or duloxetine. The pre-treatment turbulence metric predicted responders at ROC-AUC 0.70, p = 0.02.
The number sounds modest. The clinical context is what makes it interesting:
| Predictor | AUC for SSRI response |
|---|---|
| Symptom severity scales | ~0.55 |
| Prior treatment history | ~0.55–0.60 |
| Genetic markers (CYP2D6, etc.) | ~0.55 |
| Pre-treatment turbulence cascade | 0.70 |
Current depression treatment is sequential trial-and-error. The average patient gets meaningful relief only after 2–3 medications over 6–18 months. Each failed trial is months of continued illness. A baseline scan that re-routes non-responders directly to TMS, ketamine, or psychotherapy — without burning a quarter of a year on an SSRI that won't work — is the first concrete payoff of the criticality framework outside the lab.
The same machinery, applied to traumatic brain injury (Frontiers in Neuroinformatics, 2024, Deco et al.), discriminates vegetative state from minimally conscious state. The criticality signature appears to track consciousness level as a continuous variable, not a binary.
What's contested
Three live disputes:
Is the brain actually critical, or just looks it? Power-law statistics can arise from non-critical mechanisms — neural data with subsampling, finite-size effects, and slow drift can mimic critical exponents. Touboul and Destexhe (2010) and follow-ups argued much of the avalanche literature failed strict statistical tests for true criticality. The field has tightened methods since, but the question is not closed.
Does AUC 0.70 generalize? The Molecular Psychiatry cohort was 76 patients at one site, one scanner, two drugs. Psychiatric biomarker history is littered with predictors that hit 0.70+ in discovery cohorts and collapse to 0.55 in replication. Multi-site replication with a pre-registered protocol has not happened.
Is "consciousness ≈ criticality" a theory or a slogan? Integrated Information Theory (Tononi) and the criticality framework agree the critical state maximizes information integration, but neither makes a quantitative prediction the other doesn't. Critics including Scott Aaronson have argued IIT's measure φ is incomputable in practice and unconstrained by data. Criticality-as-consciousness inherits that critique.
Why this has to do with other realms
The brain turbulence finding extends Kolmogorov's -5/3 cascade exponent into biology, a domain Kolmogorov himself never considered. The same statistical structure governs jet engine exhaust, the gas clouds that collapse into stars, weather at 10 km altitude, and BOLD signal propagation across 86 billion neurons. That's at least eight orders of magnitude in spatial scale carrying the same math. See concept turbulence for the physics history — Navier-Stokes existence and smoothness remains a Clay Millennium Problem, and the brain may now be one of its odder testbeds.
The neuromorphic angle is sharper. If biological brains hit peak compute at the critical point, the design target for tech neuromorphic computing is not stable convergence but tuned instability. No production neuromorphic chip — Intel Loihi 2, IBM NorthPole, SpiNNaker — is currently optimized for criticality. The hypothesis says they're leaving performance on the table by aiming for the wrong stable point.
An open question
If anhedonia is a subcritical brain state and awe is a brief supercritical excursion, what does the full criticality spectrum of human emotion look like — and which existing drug, taken at the wrong dose, has been pushing patients across the wrong threshold for decades?
Key sources
- Deco G., Sanz Perl Y., Kringelbach M. L. et al. (2025) — Molecular Psychiatry paper on pre-treatment turbulence predicting SSRI response. The landmark clinical result.
- Deco G., Kringelbach M. L. (2020) — Cell Reports "Turbulent-like Dynamics in the Human Brain." Original framework paper introducing Kolmogorov cascade analysis to fMRI.
- Beggs J. M., Plenz D. (2003) — Journal of Neuroscience "Neuronal Avalanches in Neocortical Circuits." The empirical foundation for the criticality hypothesis in cortex.
- Touboul J., Destexhe A. (2010) — PLOS One critique arguing power-law statistics in neural data don't necessarily imply criticality. The honest skeptic case.
- Kolmogorov A. N. (1941) — the original cascade exponent papers (translated in Proceedings of the Royal Society A, 1991). To verify: exact citation of the English-translated reprints.
- Tononi G. (2008) — "Consciousness as Integrated Information," Biological Bulletin. The IIT framework criticality-consciousness arguments lean on.
Further reading
- The Hidden Spring by Mark Solms (2021) — a working neuropsychoanalyst's case for consciousness as a graded, brainstem-rooted phenomenon. Pairs well with criticality-as-continuum.
- Other Minds by Peter Godfrey-Smith — for the question of whether criticality could be substrate-independent enough to apply to octopus nervous systems.
- Deco's lab page at Pompeu Fabra (upf.edu/web/cns) — preprints and code for the turbulence framework, including the Python pipeline for fMRI cascade analysis.
- Sean Carroll's Mindscape episode with Anil Seth on consciousness measurement — useful framing of where biomarker claims should be skeptically pressed.
- Per Bak, How Nature Works (1996) — the popular introduction to self-organized criticality. Dated but still the cleanest explanation of why sandpiles, earthquakes, and brains might share a math.
See Also
- concept turbulence — the parent physics concept; Navier-Stokes and the Kolmogorov cascade
- concept frisson — musical chills as a brief supercritical excursion (same criticality framework, different trigger)
- concept overview effect — DMN suppression and awe; possibly the high-energy mirror image of depressive subcriticality
- concept gut brain axis — microbiome as an upstream knob on the brain's dynamical state
- tech neuromorphic computing — criticality as an unexploited design principle for spiking chips
- concept distributed cognition — octopus arms as candidate distributed-criticality substrates
- concept holographic principle — another case of the same math showing up across wildly different physical scales