Fluid Dynamics

Navier-Stokes Equations

ρ(∂u/∂t + u·∇u) = −∇p + μ∇²u + f — Newton's second law for fluids

The Navier-Stokes equations are the master equations of viscous fluid flow: ρ(∂u/∂t + (u·∇)u) = −∇p + μ∇²u + f, where u is velocity, ρ density, p pressure, μ dynamic viscosity, and f body forces. Coupled with the continuity equation ∇·u = 0 (incompressible) or ∂ρ/∂t + ∇·(ρu) = 0 (compressible). Derived by Claude-Louis Navier (1822) and George Stokes (1845). Despite being "just" Newton's law for a fluid, proving smooth solutions exist for all time in 3D is one of the seven Clay Millennium Problems ($1M prize) — open since 2000. Solving them numerically requires CFD methods (FVM, FEM, spectral); industrial-scale simulations involve grids of 10⁹+ cells (DNS of turbulence).

  • Equationρ(∂u/∂t + u·∇u) = −∇p + μ∇²u + f
  • AuthorsNavier 1822, Stokes 1845
  • Clay Millennium Prize$1M (open)
  • Continuity∇·u = 0
  • ReynoldsRe = ρUL/μ
  • DNS turbulence10⁹+ grid cells

Interactive visualization

Press play, or step through manually. The visualization is yours to drive — try it before reading on.

Open visualization fullscreen ↗

Watch the 60-second explainer

A condensed visual walkthrough — narrated, captioned, under a minute.

Why Navier-Stokes matters

  • Aerodynamics. Every aircraft wing, propeller, fan blade, and rocket nozzle is shaped to control Navier-Stokes solutions. Drag, lift, separation, and stall are all features of the equations.
  • Weather and climate. Atmospheric general circulation models solve Navier-Stokes on a rotating sphere with thermodynamics. Ocean models do the same for seawater. Climate forecasts are NS plus radiation, chemistry, and ice.
  • Biological flow. Blood through arteries, air through bronchi, and cerebrospinal fluid all obey NS — at low to moderate Reynolds numbers, often with non-Newtonian rheology.
  • Engineering. Pipe networks, heat exchangers, turbines, pumps, mixing vessels, lubrication films, and microfluidic chips are designed against NS solutions.
  • Geophysics. Mantle convection (huge viscosity, tiny Re), magma chambers, lake circulation, sediment transport.
  • Astrophysics. Stellar convection, accretion disks, galactic gas — NS plus magnetic terms (MHD) and gravity.
  • Combustion. Flame propagation, jet-engine combustors, internal-combustion cylinders — reactive NS with species transport and chemistry.

Reading the equation term by term

  • ρ(∂u/∂t). Local acceleration — how velocity changes at a fixed Eulerian point.
  • ρ(u·∇)u. Convective acceleration — the velocity change a parcel experiences as it moves through a non-uniform field. This term is nonlinear and is what makes NS hard.
  • −∇p. Pressure-gradient force per unit volume — fluid is pushed from high to low pressure.
  • μ∇²u. Viscous diffusion of momentum — friction smooths velocity differences across layers.
  • f. Body force per unit volume — gravity ρg, Lorentz J×B in MHD, Coriolis −2ρΩ×u in rotating frames.

The companion: continuity

NS is one momentum equation per spatial direction, but velocity has three components and pressure adds a fourth unknown. The system closes only when paired with mass conservation:

  • Incompressible: ∇·u = 0. Velocity field is divergence-free; pressure becomes a Lagrange multiplier enforcing the constraint.
  • Compressible: ∂ρ/∂t + ∇·(ρu) = 0, plus an equation of state p(ρ, T) and an energy equation. Now sound waves and shocks appear.

Reynolds number — one knob to rule the regime

Non-dimensionalize NS with U, L, and ρ; viscous term acquires a 1/Re prefactor. Re = ρUL/μ is the only dimensionless control parameter for incompressible isothermal flow.

  • Re ≪ 1. Stokes flow — viscous-dominated, time-reversible, swimming bacteria, lubrication films, mantle creep.
  • Re ~ 1–1000. Laminar with growing inertial structure — vortex shedding, periodic wakes, slow boundary layers.
  • Re > ~2300 (pipes), ~5×10⁵ (flat plate). Transition to turbulence — flow becomes chaotic, dissipative, and three-dimensional.
  • Re ~ 10⁶–10⁸. Fully turbulent — aircraft, ships, pipelines, atmospheric boundary layer.

How NS is actually solved

  • Analytical. A handful of exact solutions: Couette flow, Poiseuille flow, Stokes' first/second problem, Hagen-Poiseuille pipe, Burgers vortex. Useful as benchmarks.
  • Finite Volume Method (FVM). The dominant industrial CFD approach. Conservation laws applied to control volumes; ANSYS Fluent, OpenFOAM, STAR-CCM+ all use FVM.
  • Finite Element Method (FEM). Variational formulation, well suited to complex geometries and unstructured meshes; common in biomedical and structural-fluid coupling.
  • Spectral and pseudo-spectral. Expand u in Fourier or Chebyshev modes; ultra-high accuracy in simple geometries; backbone of DNS turbulence research.
  • Lattice Boltzmann. Solves a kinetic equation that recovers NS at the macroscopic scale; trivially parallel.
  • Turbulence closures. RANS averages out fluctuations and adds models (k-ε, k-ω SST, Reynolds-stress); LES resolves large eddies and models the small ones.

Famous problems and progress

  • Clay Millennium Problem. Posed 2000. Prove (or disprove) that smooth, finite-energy initial data in ℝ³ produces a smooth solution for all time. Open.
  • Leray's weak solutions. Jean Leray (1934) showed weak solutions always exist globally — but they may lose smoothness. The gap between weak existence and strong regularity is the heart of the Millennium Problem.
  • Partial regularity. Caffarelli-Kohn-Nirenberg (1982) showed the singular set of any Leray-Hopf weak solution has 1D Hausdorff measure zero — singularities, if any, are very sparse.
  • Tao's averaged NS. Terence Tao (2014) showed a related averaged equation can blow up, suggesting that pure-energy methods will not suffice.

Common misconceptions

  • "It's trivially derived from F = ma." Writing it down is straightforward. The convective term (u·∇)u is nonlinear and is the source of every analytical and numerical difficulty.
  • "Always laminar at low speed." Reynolds number, not absolute speed, sets the regime. Honey at high speed can stay laminar; air at low speed in a wide channel can turn turbulent.
  • "DNS solves real-world flow." DNS is restricted to small domains and modest Re. Aircraft and weather use RANS or LES with empirical closures.
  • "Existence is settled." Existence of weak solutions is settled (Leray 1934). Existence of smooth solutions in 3D is open.
  • "Pressure is a thermodynamic variable." In incompressible NS, pressure is a Lagrange multiplier enforcing ∇·u = 0 — it is determined non-locally by velocity, not by an equation of state.
  • "Viscosity always dissipates." Viscosity diffuses momentum and dissipates kinetic energy on average — but it can also generate vorticity at walls (no-slip) and stabilize otherwise unstable flows.

Frequently asked questions

What does each term in Navier-Stokes mean physically?

ρ(∂u/∂t) is local acceleration — the rate of change of velocity at a fixed point. ρ(u·∇)u is convective acceleration — the change a fluid parcel sees as it moves through a non-uniform velocity field; this term is nonlinear and is the heart of turbulence. −∇p is the pressure-gradient force (high to low pressure). μ∇²u is the viscous diffusion of momentum — friction between layers. f represents body forces like gravity or electromagnetic forces. Together: mass × acceleration = sum of forces, applied to a fluid element.

Why is the existence/smoothness problem so hard?

In 3D, the convective term (u·∇)u can transfer energy from large scales to ever-smaller scales (the energy cascade). Mathematically, this raises the question of whether velocity gradients can blow up to infinity in finite time — a singularity. Energy and L²-norm bounds are not strong enough to prevent this in 3D. The Clay Millennium Problem asks for a proof that smooth, finite-energy initial data always yields globally smooth solutions, or a counterexample. In 2D the problem is solved (smoothness holds); in 3D it has been open since 2000.

What's the difference between Euler and Navier-Stokes?

The Euler equations drop the viscous term μ∇²u, treating the fluid as inviscid. They describe an idealized fluid with zero internal friction — useful for high-Reynolds flows away from walls and for shock-wave problems. Navier-Stokes adds viscosity, allowing no-slip boundary conditions at walls (fluid sticks) and predicting boundary layers, dissipation, and the transition to turbulence. Euler equations have been around since 1757; Navier added viscosity in 1822 by introducing molecular friction.

What's a Reynolds number and how does it select regimes?

Re = ρUL/μ is the ratio of inertial to viscous forces, where U is a characteristic velocity and L a characteristic length. Low Re (< ~2,000 in pipes): viscosity dominates, flow is laminar, layers slide smoothly. High Re (> ~4,000 in pipes): inertia dominates, flow is turbulent with chaotic eddies. Transition Re is geometry-dependent. Re also sets which terms in Navier-Stokes are leading order — viscous term scales as 1/Re after non-dimensionalization.

What is direct numerical simulation (DNS)?

DNS solves Navier-Stokes directly, resolving every scale from the largest eddies down to the Kolmogorov dissipation scale η ≈ L·Re^(−3/4). To capture all scales in 3D requires roughly Re^(9/4) grid points and Re^(11/4) time steps. At Re = 10⁵, that is ~10¹¹ grid points — feasible only on the largest supercomputers and only in modest domains (a turbulent channel, a small wing section). For aircraft and weather, DNS is impossible; engineers use RANS (Reynolds-averaged) or LES (large-eddy simulation) instead.

Why are turbulent flows so hard to compute?

Turbulence is multi-scale, three-dimensional, and chaotic. The Kolmogorov cascade transfers energy from large eddies (set by geometry) down to viscous scales over many decades; capturing the full range demands enormous grids. The flow is sensitive to initial and boundary conditions — small perturbations grow. Closure for averaged equations (RANS) requires turbulence models (k-ε, k-ω, Reynolds-stress) which are empirical and limited to flow classes they were tuned for. No closed-form solution exists for general turbulent flow.