Why Turbulence Is Hard To Predict

A $1 million prize from the Clay Mathematics Institute has sat unclaimed since 2000 for anyone who can rigorously settle a basic mathematical question about the equations that govern fluid motion, whether their solutions always stay smooth and well-behaved, or whether they can blow up. Turbulence is visible everywhere: in a garden hose turned up too far, in the wake behind a boat, in the shudder that rattles a plane at cruising altitude. Yet the physics governing it, while fully known, resists exact prediction in a way few other everyday phenomena do. Understanding why gets at one of the deepest distinctions in physics, the gap between knowing the rules of a system and being able to forecast its detailed future.

What turbulence actually is

Turbulence is a state of fluid flow characterized by chaotic, irregular changes in pressure and velocity, in contrast to smooth, orderly laminar flow, where fluid moves in parallel layers with minimal mixing between them. Laminar flow looks like a stack of playing cards sliding past each other; turbulent flow looks like the same cards thrown into a wind tunnel, tumbling, colliding, and mixing unpredictably. Both are described by the exact same underlying physical laws, the difference is a matter of degree, not kind.

The transition between the two regimes is not random. It depends on the geometry of the flow, the speed of the fluid, its viscosity (a measure of a fluid's internal resistance to flow, roughly how "thick" it feels), and the scale of the obstacles or boundaries the fluid moves past. Smoke rising from an extinguished candle demonstrates the transition directly: near the wick, the smoke rises in a smooth, narrow column, laminar flow. A few centimeters up, that column suddenly breaks into curling, chaotic wisps, turbulence. Nothing about the smoke's composition changed in that transition zone; only the balance of forces acting on it did.

Fluid dynamicists estimate that roughly 90 percent of the fluid flows encountered in engineering and nature, from blood in arteries to air over a wing to water in rivers, are turbulent rather than laminar at the scales that matter practically, which is part of why the difficulty of predicting turbulence in detail is not an academic curiosity confined to physics departments. It shapes how airplanes are designed, how weather models are built, and how engineers size pipelines. A phenomenon this common should, in principle, be well understood by now. In an important sense it is: engineers model turbulent flow successfully every day. What remains genuinely unresolved is something narrower and more mathematical, discussed below.

Osborne Reynolds injected a thin stream of dye into flowing water inside a glass pipe in 1883, and by simply adjusting the flow speed, he could switch the dye's behavior from a crisp, straight line to a chaotic, fully mixed cloud, in the same pipe, with the same fluid, using nothing but velocity as the variable. That single, visually dramatic demonstration produced the number engineers still use today to predict the switch.

The Navier-Stokes equations: known laws, unproven guarantees

The motion of essentially all ordinary fluids, whether flowing smoothly or chaotically, is governed by the Navier-Stokes equations, a set of partial differential equations derived from Newton's second law of motion applied to a continuous fluid rather than a collection of discrete particles. They were formulated in the first half of the 19th century, building on work by French engineer Claude-Louis Navier in the 1820s and completed in their modern form by Irish physicist George Gabriel Stokes in the 1840s. Physicists and engineers consider these equations physically accurate for describing fluid behavior across an enormous range of everyday and industrial conditions.

The equations are also nonlinear, meaning they contain terms in which the fluid's velocity is multiplied by itself (a term called the convective acceleration term). Nonlinearity matters because it breaks a shortcut that makes simpler physics problems tractable: in a linear system, you can predict the combined effect of two influences just by adding their individual effects together. In a nonlinear system like turbulent fluid flow, that shortcut fails. Small disturbances can amplify, interact, and compound in ways that make the system's exact evolution extraordinarily sensitive to details that would be irrelevant in a linear world.

This is precisely the mathematical property at stake in the Clay Mathematics Institute's Millennium Prize Problem on Navier-Stokes existence and smoothness, one of seven problems the institute named in 2000 and offered $1 million each to solve. The problem, as formalized by mathematician Charles Fefferman in the institute's official problem statement, asks whether, starting from any smooth, physically reasonable initial velocity field in three-dimensional space, the Navier-Stokes equations always produce a solution that stays smooth and well-defined for all future time, or whether some starting conditions can cause the mathematical solution to develop a singularity, a point where a quantity like velocity becomes infinite, in finite time. As of 2026, neither outcome has been proven for the three-dimensional case. Nobody has found a smooth starting flow that provably blows up, and nobody has proven that no such flow exists.

It is worth being precise about what this unsolved problem does and does not mean, because it gets exaggerated constantly. It does not mean scientists don't understand how fluids behave, and it does not mean the equations themselves are wrong or in doubt. Engineers run Navier-Stokes-based simulations successfully every day to design aircraft, ships, and turbines. What remains open is a narrower, purely mathematical guarantee: a rigorous proof, for the full three-dimensional case with no simplifying assumptions, that solutions never spontaneously misbehave. That gap between practical modeling success and mathematical rigor is unusual, and it is part of why turbulence occupies a strange dual status, a routine engineering problem and a foundational open question in mathematics at the same time.

The Reynolds number: when flow tips from smooth to chaotic

British engineer Osborne Reynolds ran a now-famous series of pipe-flow experiments in Manchester in 1883, injecting a thin filament of dye into water flowing through a glass tube at controllable speeds. At low flow velocities, the dye traveled as a crisp, undisturbed line, the water moving in smooth, parallel layers. As Reynolds increased the velocity, the dye line would, at some point, break apart abruptly into a diffuse, swirling cloud that mixed rapidly through the whole cross-section of the pipe, the signature of turbulence.

Reynolds found that this transition happened at a predictable point defined not by velocity alone, but by a dimensionless combination of the flow's velocity, a characteristic length scale of the flow (like pipe diameter), and the fluid's viscosity. That ratio, now called the Reynolds number, compares the fluid's inertial forces (its tendency to keep moving and carry disturbances forward) against its viscous forces (its internal resistance, which damps disturbances out). At low Reynolds numbers, viscosity wins, and small disturbances get smoothed away before they can grow, keeping flow laminar. At high Reynolds numbers, inertia wins, and small disturbances amplify instead of dying out, tipping the flow into turbulence.

This single number explains a lot of everyday intuition about fluids. Honey poured slowly from a jar flows in smooth, glossy sheets because its high viscosity keeps the Reynolds number low even at modest speeds. Water gushing from a fully open faucet, or air rushing over a car at highway speed, sits at a much higher Reynolds number, and both flows visibly churn into eddies. The specific numerical threshold where the transition happens is not fixed for all situations, it depends heavily on the geometry involved, a pipe transitions at a different value than flow around a sphere, for example, but within any given geometry, the Reynolds number remains the standard engineering tool for predicting roughly when smooth flow will break down.

How is the Reynolds number actually calculated?

The Reynolds number is calculated as Re = (velocity × characteristic length) ÷ kinematic viscosity, where the characteristic length depends on the geometry, such as a pipe's diameter or a wing's chord length. Because the units cancel out, the result is a pure, dimensionless number, which is what makes it comparable across wildly different fluids and situations, from blood in a capillary to air over an aircraft wing.

Sensitivity to initial conditions and the butterfly effect

Turbulent flow belongs to a broader category of systems physicists call chaotic dynamical systems, systems whose long-term behavior is extremely sensitive to the precise details of their starting state, even though the systems themselves follow entirely deterministic, non-random physical laws. This sensitivity is the second, and arguably more important, reason turbulence resists exact prediction, distinct from the raw mathematical difficulty of the Navier-Stokes equations discussed above.

The classic illustration of this sensitivity comes not from fluid dynamics directly but from meteorology, which shares its governing mathematics with turbulent flow. Edward Lorenz, a meteorologist at MIT, was running a simplified computer model of atmospheric convection in the early 1960s when he discovered the effect by accident. Wanting to rerun part of a simulation, he re-entered a variable rounded to three decimal places instead of the six decimal places the computer had originally used internally, expecting the tiny rounding difference to be irrelevant. Instead, the rerun weather pattern diverged completely from the original within a simulated few weeks. Lorenz later described the underlying phenomenon using a memorable image, the idea that the flap of a butterfly's wings in one location could, in principle, set off a chain of atmospheric events leading to a tornado weeks later and thousands of kilometers away, in a 1972 talk titled "Does the Flap of a Butterfly's Wings in Brazil Set Off a Tornado in Texas?" The phrase "butterfly effect" stuck.

Turbulent fluid flow exhibits precisely this same exponential sensitivity. Two turbulent flows that start out nearly identical, differing only in some minuscule, unmeasurably small detail of their initial velocity or pressure, will diverge from one another at an accelerating rate, until their detailed motion looks essentially uncorrelated even though both obeyed the exact same physical equations from the same starting point. Because no real-world measurement of a fluid's starting state can ever be infinitely precise, this sensitivity places a hard practical ceiling on how far into the future the exact, detailed motion of a specific turbulent flow can be predicted, regardless of how much computing power is thrown at the problem. Statistical properties of turbulence, averages, energy distributions, likely ranges, remain predictable even when the exact instant-by-instant motion does not.

Is turbulence actually random, or just unpredictable?

Turbulence is not random in the physical sense. It is fully deterministic, meaning the same starting conditions, followed exactly, will always produce the same outcome, governed by the same Navier-Stokes equations as smooth flow. The apparent randomness comes entirely from extreme sensitivity to starting conditions that can never be measured with perfect precision, not from any inherent chance at the level of individual fluid particles.

Cascades: how energy moves from big eddies to small ones

British mathematician and meteorologist Lewis Fry Richardson described a key structural feature of turbulence in 1922, an idea later put on rigorous mathematical footing by Soviet mathematician Andrey Kolmogorov in 1941. Turbulent kinetic energy, Richardson observed, does not simply dissipate directly at the scale where it enters a flow. Instead, it cascades downward: large eddies (whirling structures roughly the size of the overall flow) break apart into somewhat smaller eddies, which break apart into smaller eddies still, and so on through a whole hierarchy of scales, until the eddies become small enough that viscosity finally converts their kinetic energy into heat.

Richardson captured the idea in a short, often-quoted rhyme paraphrasing Jonathan Swift: "Big whirls have little whirls that feed on their velocity, and little whirls have lesser whirls, and so on to viscosity." The image is not just poetic, it describes a real, measurable structure in turbulent flows, one that shows up in everything from the wake behind a ship to the plumes rising from a wildfire.

This cascade has direct practical consequences for anyone trying to compute turbulent flow rather than just describe it qualitatively. Fully resolving a turbulent flow's motion in a computer simulation, an approach called direct numerical simulation, requires modeling the flow's behavior across the entire range of scales simultaneously, from the largest eddies down to the smallest scale where viscosity dominates, a scale known as the Kolmogorov microscale. The computational cost of resolving that full range of scales grows extremely rapidly as the Reynolds number increases, which is why direct numerical simulation remains computationally prohibitive for most real-world, high-Reynolds-number engineering flows, like airflow over a full aircraft. In practice, engineers instead rely on statistical turbulence models, such as Reynolds-averaged Navier-Stokes (RANS) or large-eddy simulation (LES) approaches, which approximate the effect of the smallest, unresolved scales rather than computing them directly.

Flow regime	Typical Reynolds number range	Visual behavior	Predictability
Laminar flow	Below ~2,000 (pipe flow)	Smooth, parallel layers, minimal mixing	Highly predictable; governed by simple, well-behaved solutions
Transitional flow	Roughly 2,000–4,000 (pipe flow)	Intermittent bursts of disorder within otherwise smooth flow	Partially predictable; onset location and timing are uncertain
Turbulent flow	Above ~4,000 (pipe flow)	Chaotic eddies across a wide range of scales, rapid mixing	Statistically predictable (averages); exact detail is not

Reynolds number thresholds shown are typical values for flow through a smooth cylindrical pipe; exact transition points shift with geometry, surface roughness, and disturbance level.

Why weather and aviation feel this problem directly

Weather forecasting and turbulent fluid dynamics are, mathematically, close cousins, both are governed by fluid equations of the same nonlinear character, applied to the atmosphere instead of a pipe or a wing. That shared mathematical structure is exactly why Lorenz's chaos discovery, made while modeling atmospheric convection, translates so directly to turbulence generally. Weather forecasts remain reasonably reliable a few days out but degrade steadily beyond about a week, not primarily because meteorologists lack good models, but because the atmosphere's chaotic sensitivity to its starting state amplifies small measurement uncertainties into large forecast errors over time.

Clear-air turbulence, the sudden, invisible turbulence that can jolt an aircraft cruising at altitude with no clouds in sight, illustrates the practical stakes especially well. It typically forms near the boundaries of jet streams, where sharp differences in wind speed between adjacent air masses (wind shear) generate the kind of instability that tips smooth airflow into turbulence, but because it produces no visible cloud formation, pilots often cannot see it coming. Forecasters combine numerical weather model output, turbulence diagnostic algorithms, and real-time reports from aircraft instruments and pilots to estimate where clear-air turbulence is likely, but reliable detailed forecasts of moderate-or-greater turbulence in a specific patch of airspace remain difficult to produce with high confidence and low false-alarm rates, according to research summarized by the European Centre for Medium-Range Weather Forecasts (ECMWF). That difficulty traces back to the same root cause discussed throughout this article: turbulence occurs at small spatial scales that are hard to resolve directly within larger-scale weather models, and the flow's chaotic sensitivity limits how far in advance its exact location and intensity can be pinned down.

This is also why aviation turbulence forecasting focuses on probability and risk zones rather than exact predictions of a specific gust at a specific altitude and moment. Forecasters identify regions where atmospheric conditions, wind shear, jet stream proximity, mountain wave activity, make turbulence statistically likely, then update those estimates continuously as new data arrives, rather than attempting to predict the precise timing of individual turbulent eddies days in advance, an approach that mirrors how engineers handle turbulent flow in other contexts: modeling statistical likelihood rather than chasing an impossible level of exact detail.

An everyday analogy: a shuffled deck versus a sorted one

A freshly sorted deck of playing cards, ace through king in each suit, is turbulence's laminar opposite: know the position of one card, and you know a great deal about its neighbors, because the structure is orderly and predictable. Shuffle that deck thoroughly, and even though every single card is still a specific, physical object obeying entirely deterministic rules throughout the shuffling process, no randomness at the level of the cards themselves, the final arrangement becomes practically unpredictable from any reasonable starting knowledge. Tiny differences in exactly how the shuffle was performed, a slightly different riffle, one extra cut, compound rapidly into a completely different final order.

Turbulence works the same way at the level of fluid particles. Nothing about the underlying physics is random, no dice are being rolled inside a turbulent eddy, but the extreme sensitivity to starting conditions, combined with the energy cascade spreading disturbances across a huge range of interacting scales, produces outcomes that are for practical purposes just as unpredictable in exact detail as a shuffled deck's final order, even though a supercomputer with a perfect description of the governing equations is doing the "shuffling."

A second useful comparison is a pinball machine. Drop a ball from a slightly different starting angle or speed, and it might strike the same first bumper, an outcome close enough to predictable in the short run. But that first, tiny difference determines a different angle off the bumper, which determines a different angle off the next one, and within a handful of bounces the ball's path bears no resemblance to the first run at all. Turbulent flow behaves like a pinball machine with an enormous number of bumpers, arranged across every scale from the size of the whole flow down to the Kolmogorov microscale, all interacting simultaneously.

What people consistently get wrong about turbulence

The most common misconception is that turbulence involves genuine physical randomness, as if fluid molecules were making unpredictable choices. They are not. Turbulent flow is fully deterministic and governed by exactly the same Navier-Stokes equations as smooth, laminar flow, with no randomness anywhere in the underlying physics. What looks like randomness is the downstream consequence of extreme sensitivity to starting conditions that can never be measured perfectly, an entirely different phenomenon from actual chance.

A second common misconception, and arguably the more consequential one for public understanding of science, is that the unsolved Navier-Stokes Millennium Prize Problem means turbulence itself is scientifically mysterious or poorly understood. It does not. Engineers and physicists model turbulent flows successfully every day using the same equations, combined with approximation techniques like the Reynolds number framework and statistical turbulence models, to design aircraft wings, predict storm behavior, and model ocean currents with results reliable enough to build real infrastructure on. The unsolved question is a specific, narrow, and highly abstract mathematical one, whether a rigorous proof exists guaranteeing that solutions to the equations never develop singularities in the general three-dimensional case, a question that is genuinely separate from whether the equations produce useful, accurate predictions in practice, which they routinely do.

A third, subtler misconception is that turbulence is simply "more complicated" laminar flow, as though it differs only in degree of detail needed. In fact, turbulence introduces a qualitatively different structure, the multi-scale energy cascade described by Richardson and Kolmogorov, that has no equivalent in laminar flow at all. Laminar flow can often be described adequately by tracking motion at a single scale; turbulent flow inherently spans a continuous range of interacting scales simultaneously, which is part of why turbulence modeling remains a distinct sub-discipline within fluid dynamics rather than simply a more detailed version of laminar-flow analysis.

The takeaway

Turbulence resists exact prediction for two layered reasons that compound each other rather than one single cause. First, the nonlinear Navier-Stokes equations that govern all fluid motion make exact long-term solutions mathematically resistant to prediction, a difficulty so fundamental that a rigorous proof of basic solution behavior remains an unsolved Clay Millennium Prize Problem. Second, turbulent flow is chaotic in the technical sense Edward Lorenz first documented rigorously in weather models, meaning tiny, unmeasurable differences in starting conditions amplify into wildly divergent outcomes over time, even though the underlying physics involves no randomness at all. Layered on top of both is the energy cascade Richardson and Kolmogorov described, which spreads that sensitivity across an enormous range of interacting scales at once, from the largest eddies down to the point where viscosity finally converts motion into heat. None of this means turbulence is poorly understood; it means detailed, exact prediction of a specific turbulent flow's future is fundamentally different from, and much harder than, understanding the physics that produces it, which is precisely why engineers build reliable aircraft and forecasters issue useful storm warnings despite the underlying mathematics remaining formally unsolved.

---

Read next:

• Why Glass Is Not a Normal Solid
• Why Vacuum Chambers Are Useful in Science
• Why Centrifuges Separate Mixtures

FAQ

Q: What is the Reynolds number and why does it matter for turbulence?

The Reynolds number is a dimensionless quantity, developed from Osborne Reynolds' 1883 pipe-flow experiments, that compares inertial forces to viscous forces in a fluid flow. Low Reynolds numbers favor smooth, laminar flow, while high Reynolds numbers favor chaotic, turbulent flow, making it the standard engineering tool for predicting roughly when a given flow will transition between the two.

Q: What are the Navier-Stokes equations?

The Navier-Stokes equations are the fundamental partial differential equations describing fluid motion, derived from Newton's laws applied to a continuous fluid and completed in their modern form by George Gabriel Stokes in the 1840s. Their nonlinearity, terms where the fluid's velocity multiplies itself, is the mathematical root of why exact long-term prediction of turbulent behavior is so difficult, even though the equations are considered physically accurate.

Q: What exactly is unsolved about the Navier-Stokes equations?

The unsolved question, formalized by Charles Fefferman for the Clay Mathematics Institute's Millennium Prize, is whether the three-dimensional Navier-Stokes equations always produce smooth, well-behaved solutions from smooth starting conditions, or whether some starting conditions can cause a solution to develop a singularity in finite time. As of 2026 neither outcome has been proven, but this is a narrow mathematical rigor question, separate from the equations' proven practical usefulness in engineering.

Q: What is the butterfly effect and how does it relate to turbulence?

The butterfly effect, a term popularized by meteorologist Edward Lorenz after he discovered that tiny rounding differences in a weather simulation produced wildly different forecasts, describes extreme sensitivity to initial conditions in a chaotic system. Turbulent fluid flow shows this same sensitivity, meaning minuscule, unmeasurable differences in starting conditions cause exact long-term prediction of a specific flow's motion to become practically impossible.

Q: What is the energy cascade in turbulence?

The energy cascade, described qualitatively by Lewis Fry Richardson in 1922 and formalized mathematically by Andrey Kolmogorov in 1941, describes how turbulent kinetic energy transfers from large eddies down through progressively smaller eddies until it reaches a small enough scale, the Kolmogorov microscale, for viscosity to dissipate it as heat.

Q: Why is clear-air turbulence hard for pilots and forecasters to predict?

Clear-air turbulence forms from wind shear near jet streams but produces no visible cloud formation, so it cannot be seen directly by pilots or ordinary weather radar. Forecasters rely on numerical weather models, turbulence diagnostic algorithms, and real-time aircraft reports to estimate risk zones, but the small spatial scale of turbulence combined with its chaotic sensitivity limits how precisely its exact location and timing can be forecast in advance.

Q: Is turbulence random, or is it deterministic but unpredictable?

Turbulence is fully deterministic, governed by the same Navier-Stokes equations as smooth flow, with no physical randomness at the level of fluid particles. Its apparent unpredictability comes entirely from extreme sensitivity to starting conditions that can never be measured with perfect precision, not from any inherent chance in the underlying physics.