Turbulence — The Last Unsolved Problem in Classical Physics
Stir cream into coffee. You have just performed a calculation no supercomputer on Earth can reproduce from first principles. The equations governing that swirl were written by Navier in 1822 and Stokes in 1845. Two centuries later, the Clay Mathematics Institute still offers $1 million to anyone who can prove their 3D solutions stay smooth — or produce a counter-example. The prize is one of six remaining Millennium Problems. Heisenberg reportedly said he would ask God about relativity and turbulence, and expected an answer only on relativity.
Turbulence sits at the centre of aviation, fusion, cardiology, climate, and (recently) psychiatry. We can describe its statistics, simulate its gross behaviour at enormous cost, and predict its averages. The equations resist us at the level of proof.
The cascade Kolmogorov wrote down in 1941
Andrey Kolmogorov's three 1941 papers proposed an energy cascade: large eddies, fed by stirring, break into smaller eddies, which break into smaller eddies, down to a dissipation scale (the Kolmogorov scale — roughly 0.1 to 1 mm in atmospheric air) where viscosity converts kinetic energy to heat. Between the input and dissipation scales, the energy spectrum follows a power law in wavenumber with exponent −5/3.
The k⁻⁵/³ spectrum has been measured in jet streams, ocean currents, wind-tunnel pipes, tokamak plasmas, and stellar convection zones. Few empirical laws in physics span this many orders of magnitude across this many substrates.
What Kolmogorov did not explain: why the cascade settles on −5/3, how the breaking proceeds locally, and the "anomalous dissipation" puzzle — energy keeps dissipating at the same rate as Reynolds number rises, even though viscosity (the agent of dissipation) effectively vanishes. The 2025 work by Sreenivasan and Schumacher in Annual Review of Fluid Mechanics lays out a taxonomy of what "solving turbulence" could even mean. Their categories range from engineering closure (already useful) to mathematical regularity (the Millennium Prize). The field has stopped pretending these are the same problem.
Why the equations resist proof
The Reynolds number Re = ρvL/μ controls the transition. Below ~2,300 in a pipe, flow is laminar. Above ~4,000, it goes turbulent. A commercial jet cruises at Re ~ 10⁷. The Navier-Stokes equations are nonlinear, and that nonlinearity does three things at once:
- Sensitive dependence on initial conditions. Tiny perturbations grow exponentially. Long-term prediction is impossible in principle, not just in practice. Weather is the canonical case.
- Scale coupling. Every scale of motion interacts with every other. You cannot study large-scale flow without the small scales feeding back.
- Possible finite-time blow-up. Nobody has proved that smooth solutions stay smooth forever in 3D. Terence Tao's 2016 result showed a modified version of the equations can blow up. The gap between his modification and the physical equations is the prize.
The closest 2025 advance came from Javier Gómez-Serrano at Brown collaborating with Google DeepMind (arXiv 2509.14185). They used physics-informed neural networks pushed to near-machine precision to find new families of unstable singularities in the Euler equations (the inviscid limit of Navier-Stokes). No stable singularity is believed to exist; cataloguing the unstable ones is a step toward proof or disproof. Prediction markets nudged upward, slightly.
Where it shows up
| Domain | What turbulence does | Specific 2024–25 finding |
|---|---|---|
| Aviation | Clear-air turbulence at cruise | Moderate-severe CAT up 55% since the 1970s; University of Reading 2024 shows 60–155% rises on North Atlantic, North Pacific, East Asia routes 1980–2021 |
| Cardiology | Damages endothelium at branch points | Atherosclerotic plaque clusters at the aortic arch and carotid bifurcation, where flow goes turbulent |
| Fusion | Heat leaks across magnetic field lines in tokamaks | ORNL Frontier simulations (2025) showed plasma rotation can worsen confinement — overturning a 30-year assumption |
| Psychiatry | Brain near-criticality dynamics | Molecular Psychiatry 2024: baseline brain turbulence amplitude predicts antidepressant response 8 weeks before treatment |
| Superfluids | Quantum vortex reconnection | FAMU-FSU June 2025: a universal law — vortices separate faster than they approach, producing energy bursts that mirror classical intermittency |
The brain result is the strangest. Gustavo Deco's group formalised "brain turbulence" as spatiotemporal variability in local synchronisation of neural signals, mathematically analogous to velocity-gradient variability in a fluid. Lower amplitude correlates with antidepressant non-response. The same energy-spectrum tools used on jet streams now sit in clinical psychiatry papers. Whether the analogy is mechanism or coincidence is genuinely open.
What's contested
Is the −5/3 spectrum truly universal, or an attractor? The quantum-turbulence result from FSU shows the same scaling in a system whose microphysics (quantised vortex filaments with circulation set by ℏ) is wholly different from a classical fluid. If the cascade emerges identically from incompatible substrates, the spectrum may be a statistical attractor independent of the underlying physics. That is a stronger claim than Kolmogorov made.
Are vortices the right primitive? University of Michigan researchers used explainable AI on wall-bounded turbulence in November 2025 and found that Reynolds stresses near and far from the wall, plus streaks at moderate distances, dominated the dynamics — while vortices contributed less than canonical wall-turbulence models assume. If replicated, this rewrites a textbook.
Can ML "solve" turbulence? Physics-informed neural networks, Fourier neural operators, and diffusion-based subgrid models all emulate turbulence statistics at a fraction of direct-simulation cost. None of them solves the equations. They generalise poorly outside their training distributions. A 2025 startup claim of an AI-assisted Navier-Stokes proof was not validated by the Clay Institute; prediction markets moved against it. The distinction between engineering tool and mathematical proof keeps tripping up press coverage.
Why this has to do with other realms
Turbulence is irreversibility in action. The Navier-Stokes equations are time-reversible at the level of symbols, yet a turbulent flow never spontaneously un-mixes. This is the same asymmetry that governs entropy and the concept arrow of time, expressed in a different substrate. The cascade transfers structure from large scales to small ones, and that one-way ticket is where the arrow lives.
The bridge to biology is sharper than it looks. Branching networks — coronary arteries, mycelium, river deltas — face the same problem of flow stability under nonlinear coupling. The brain result hints that whatever the cortex does to integrate information across regions, the mathematical signature looks like a turbulent cascade near criticality. If that holds, turbulence is not just a fluid phenomenon. It is a structural feature of any system that mixes information across scales.
An open question
If quantum and classical fluids produce the same Kolmogorov spectrum from incompatible microphysics, and if the brain near criticality produces the same statistical signatures from neither, then what exactly is a turbulent cascade? A property of fluids, or a property of information moving through scale-coupled networks of any kind?
Key sources
- Sreenivasan, K.R. & Schumacher, J. (2025). "What Is the Turbulence Problem, and When May We Regard It as Solved?" Annual Review of Fluid Mechanics — the synthesis paper that establishes the taxonomy of "solving."
- Gómez-Serrano, J. et al. (2025). "Unstable singularities in the Euler equations" (arXiv 2509.14185) — DeepMind-assisted computer-assisted progress on blow-up.
- Kolmogorov, A.N. (1941). The three foundational papers on local structure and the −5/3 law; translated in Proc. Royal Society A 434 (1991).
- Williams, P. & Storer, L.N. (2024). University of Reading CAT trend analysis, Geophysical Research Letters.
- Deco, G. et al. (2024). "Whole-brain turbulent dynamics predict antidepressant response," Molecular Psychiatry.
- Tao, T. (2016). "Finite-time blowup for an averaged three-dimensional Navier-Stokes equation," J. AMS — the modified-equations blow-up result.
- to verify: FAMU-FSU 2025 quantum-vortex reconnection paper (universal law of separation faster than approach).
Further reading
- Turbulence by Uriel Frisch (1995) — still the cleanest exposition of the K41 theory and its corrections, written by a working theorist.
- Clay Mathematics Institute official Navier-Stokes problem statement at claymath.org — the actual prize criteria, surprisingly readable.
- Steven Strogatz, Nonlinear Dynamics and Chaos (2nd ed., 2014) — the chaos foundations that turbulence sits on top of.
- The DeepMind blog post accompanying the Gómez-Serrano collaboration — useful for seeing how ML now enters rigorous mathematics, not just engineering.
- The Feynman Lectures, Volume I, Chapter 41 — Feynman's own take on why turbulence beat him, written long before the Millennium Prize existed.
See Also
- concept arrow of time — time-reversible equations producing irreversible flow, the same asymmetry as thermodynamic entropy in a different substrate
- concept emergence — simple rules generating irreducible complexity; turbulence is the canonical fluid case
- concept fermi paradox — plasma turbulence has gated fusion for 70+ years, a candidate energetic bottleneck for any technological civilisation
- concept hard problem consciousness — if brain turbulence amplitude tracks integration capacity, the cascade may be more than analogy
- event bronze age collapse — cascade failure in over-coupled networks; the mathematics of system collapse and fluid collapse share more than vocabulary
- dest trappist 1 — atmospheric turbulence sets the habitability question for any exoplanet we can resolve