The Navier–Stokes Equations
Honey oozing off a spoon, air peeling over a wing, blood pulsing through an artery, the smoke
curling off a candle, the weather over an entire hemisphere — all of them, every last one, obey the
same equations. Write them once and you have written down the law of motion for
practically every fluid on Earth. They are the Navier–Stokes equations, and they
are simply Euler's equation
— Newton's second law for a fluid — with one extra ingredient bolted on: viscosity,
the internal friction that makes honey thick and water thin.
That one extra term changes everything. It is what makes a fluid stick to a wall, what turns smooth
flow into roaring turbulence, and what makes the equations so ferociously hard that nobody on the
planet knows whether their solutions in three dimensions always stay smooth or can blow up into a
singularity. The Clay Mathematics Institute will pay one million
dollars to whoever settles the question. This page builds the equations term by term,
reads each one out loud, meets the no-slip condition that ties a fluid to a solid
surface, and solves them exactly for flow down a channel — before pointing at the million-dollar
hole in our understanding.
Starting point: Euler, then add friction
For an ideal (frictionless) fluid, Newton's law \mathbf{F} = m\mathbf{a}
applied to a fluid parcel gives Euler's equation. Using the
material derivative
D/Dt = \partial/\partial t + (\mathbf{u}\cdot\nabla) for the acceleration
a parcel actually feels, it reads
\rho\,\frac{D\mathbf{u}}{Dt} = -\nabla p + \rho\mathbf{g}.
The left side is mass-per-volume times acceleration; the right side is the two forces per unit
volume acting on the parcel — the pressure pushing it from high pressure toward
low, and gravity (or any body force). This is exact and elegant, but it tells a
lie: it says a fluid feels no friction. Real fluids drag on themselves. A fast-moving layer
pulls its slower neighbour along and gets held back in return. That internal drag is
viscosity, and for a Newtonian fluid — one whose internal stress
is simply proportional to the rate at which layers shear past each other — it adds exactly one term:
\rho\,\frac{D\mathbf{u}}{Dt} = -\nabla p + \mu\,\nabla^2\mathbf{u} + \rho\mathbf{g}.
Here \mu is the dynamic viscosity (units
\text{Pa·s}) — big for honey, tiny for air. The new term
\mu\nabla^2\mathbf{u} is a diffusion of momentum: the
Laplacian \nabla^2 compares the velocity at a point with the average of
its neighbours, so wherever a parcel is moving differently from the fluid around it, this term nudges
it back toward the local average — smearing sharp velocity differences out, exactly as heat
conduction smears out temperature.
The equations, read out loud
A fluid also has to conserve mass. For an incompressible fluid (constant density —
an excellent approximation for liquids and for gases well below the speed of sound) that conservation
law takes the beautifully simple form \nabla\cdot\mathbf{u} = 0: whatever
flows into any little region must flow back out, so the velocity field has no sources or sinks. Put
the momentum balance and the mass balance together and you have the whole system.
-
Momentum (Newton's second law for the fluid):
\underbrace{\rho\frac{\partial\mathbf{u}}{\partial t}}_{\text{local inertia}} + \underbrace{\rho(\mathbf{u}\cdot\nabla)\mathbf{u}}_{\text{convective inertia}} = \underbrace{-\nabla p}_{\text{pressure}} + \underbrace{\mu\nabla^2\mathbf{u}}_{\text{viscous diffusion}} + \underbrace{\rho\mathbf{g}}_{\text{body force}}.
-
Incompressibility (mass conservation):
\nabla\cdot\mathbf{u} = 0.
-
No-slip boundary condition: at any solid surface the fluid velocity equals the
surface's velocity — right at the wall the fluid does not slide, \mathbf{u} = \mathbf{u}_{\text{wall}}
(so \mathbf{u} = \mathbf{0} at a stationary wall).
Term by term, the momentum equation says: the mass of a parcel times its acceleration
(\rho\,D\mathbf{u}/Dt, split into the local change
\rho\,\partial\mathbf{u}/\partial t and the convective
change \rho(\mathbf{u}\cdot\nabla)\mathbf{u}) equals the
pressure force -\nabla p, plus the viscous
force \mu\nabla^2\mathbf{u}, plus the body force
\rho\mathbf{g}. Delete the viscous term and Navier–Stokes collapses back
into Euler. Everything hard and everything real lives in that one Laplacian.
Two viscosities: dynamic μ and kinematic ν
Divide the whole momentum equation by the density \rho and the viscous
term becomes (\mu/\rho)\nabla^2\mathbf{u}. That combination is so common it
earns its own name and symbol, the kinematic viscosity:
\nu = \frac{\mu}{\rho} \qquad (\text{units } \text{m}^2/\text{s}).
The two are genuinely different animals. \mu (dynamic)
measures the force a fluid's shearing generates; \nu
(kinematic) measures how quickly momentum diffuses through it, and it is
\nu that governs whether a flow stays orderly or goes turbulent. A famous
surprise: air is far runnier than water to the touch, yet air's kinematic viscosity is about
fifteen times larger than water's, because air is so much less dense. Low
\mu does not guarantee low \nu — always check
which one a formula wants.
No-slip and viscous shear stress
The no-slip condition is one of the most counter-intuitive facts in all of physics:
at a solid surface, the fluid touching it moves with the surface — velocity zero at a stationary wall,
no matter how fast the flow rushes by just a hair's breadth away. It is why a fan blade or a car
keeps a stubborn film of dust: the air right at the surface is not moving to blow it off. Between the
stuck-still fluid at the wall and the fast free stream lies a thin boundary layer
where the speed ramps up, and inside it the fluid layers slide across one another.
That sliding is where viscosity earns its keep. In a simple shear where the speed
u varies in the cross-stream direction y, a
Newtonian fluid resists with a shear stress (force per unit area) proportional to
the velocity gradient:
\tau = \mu\,\frac{du}{dy}.
The steeper the velocity gradient — the harder neighbouring layers are forced to slide past each
other — the bigger the drag. This single relation is the definition of a Newtonian fluid,
and integrating it over a surface is how you turn a velocity profile into the actual force felt by a
wall (skin-friction drag). Drag the top plate below and watch the no-slip condition pin the fluid to
both walls while a straight-line velocity profile builds across the gap.
An exact solution: flow down a channel
Navier–Stokes is usually unsolvable by hand, but in a few clean geometries the nonlinear convective
term (\mathbf{u}\cdot\nabla)\mathbf{u} vanishes and the equations become
gentle. The classic case is steady flow between two parallel plates a distance
2h apart, with the walls at y = \pm h. The flow
is purely horizontal, \mathbf{u} = (u(y), 0), so
(\mathbf{u}\cdot\nabla)\mathbf{u} = u\,\partial u/\partial x = 0 and the
horizontal momentum equation reduces to a simple balance between the pressure push and the viscous
term:
0 = -\frac{dp}{dx} + \mu\frac{d^2u}{dy^2}\quad\Longrightarrow\quad \frac{d^2u}{dy^2} = \frac{1}{\mu}\frac{dp}{dx}.
Integrate twice and impose no-slip at both walls (u = 0 at
y = \pm h). With a constant pressure gradient driving the flow you get
plane Poiseuille flow — a parabola:
u(y) = u_{\max}\left(1 - \frac{y^2}{h^2}\right), \qquad u_{\max} = -\frac{h^2}{2\mu}\frac{dp}{dx}.
Zero at both walls (no-slip), fastest right down the middle. If instead there is no pressure
gradient and the top wall is simply dragged at speed U, the parabola
straightens into a line — plane Couette flow,
u(y) = U(y+h)/(2h). Toggle the two profiles below and drag the sliders to
feel how the peak speed and wall speed reshape them; notice both are pinned to zero at any stationary
wall.
Worked examples
Example 1 — kinematic from dynamic viscosity. Water has dynamic viscosity
\mu \approx 1.0\times 10^{-3}\ \text{Pa·s} and density
\rho \approx 1000\ \text{kg/m}^3. Its kinematic viscosity is
\nu = \frac{\mu}{\rho} = \frac{1.0\times 10^{-3}}{1000} = 1.0\times 10^{-6}\ \text{m}^2/\text{s}.
Air, by contrast, has \mu \approx 1.8\times 10^{-5}\ \text{Pa·s} but
\rho \approx 1.2\ \text{kg/m}^3, giving
\nu \approx 1.5\times 10^{-5}\ \text{m}^2/\text{s} — about fifteen times
larger than water's, even though air's \mu is fifty times smaller.
Example 2 — shear stress in Couette flow. Oil of viscosity
\mu = 0.2\ \text{Pa·s} fills a gap
h = 2\ \text{mm} = 0.002\ \text{m} between plates; the top plate slides at
U = 0.5\ \text{m/s}. The profile is linear, so
du/dy = U/h and the shear stress on the plate is
\tau = \mu\frac{du}{dy} = \mu\frac{U}{h} = 0.2\times\frac{0.5}{0.002} = 50\ \text{Pa}.
Example 3 — reading the Poiseuille parabola. A channel flow has centreline speed
u_{\max} = 3\ \text{m/s} and half-gap h = 1\ \text{cm}.
Halfway to the wall, at y = h/2, the speed is
u = u_{\max}\left(1 - \frac{(h/2)^2}{h^2}\right) = 3\left(1 - \tfrac{1}{4}\right) = 3\times 0.75 = 2.25\ \text{m/s}.
Right at the wall y = h it gives 3(1-1) = 0 —
no-slip, exactly as it must.
A tempting picture is that viscosity is "friction", so \mu\nabla^2\mathbf{u}
must be a brake that everywhere reduces the speed. Not so. The term is a
diffusion of momentum, not a drag force: it drives velocity toward the local
average of its surroundings. Where a parcel is moving slower than its neighbours, the
Laplacian is positive and the term speeds it up — it is how the fast free stream
drags the near-wall fluid along and how the moving wall in Couette flow accelerates the layers above
it. Viscosity smears velocity differences out in both directions; only the fluid near a
faster region gets a push, and the fluid near a slower region gets held back. Whether the net effect
on a given parcel is to speed it up or slow it down depends entirely on the shape of the profile
around it.
Because being able to write an equation is a world away from being able to solve
it — or even to prove a solution exists. The trouble is the innocent-looking convective term
(\mathbf{u}\cdot\nabla)\mathbf{u}: it is nonlinear,
velocity multiplying its own gradient, and nonlinearity is what breeds turbulence and mathematical
chaos. The Clay Millennium Prize Problem asks a very basic question about the
3-D incompressible equations: given smooth initial data, do the solutions stay smooth and finite for
all time, or can they spontaneously develop a singularity — a point where velocity
becomes infinite — in finite time? Nobody knows. We use these equations to design aircraft and
forecast weather with spectacular success, yet we cannot prove they never break. The viscous term
\mu\nabla^2\mathbf{u} is thought to be the peacemaker that keeps things
smooth, but proving it always wins against the nonlinearity is the open problem. Solve it, either
way, and the million dollars is yours.