The Continuum and Fluid Kinematics
Pour a glass of water and you are holding something like 10^{25} molecules,
each careering about at hundreds of metres per second, colliding a billion times a second, never once
sitting still. And yet an engineer designing a pipe, a wing, or a heart valve does not track a single
one of them. They write down a smooth velocity field — one tidy arrow at every point
of space — and it works beautifully. How can the frantic molecular mob be replaced by a smooth flowing
continuum? That leap of faith is the continuum hypothesis, and it is the foundation
every equation of fluid dynamics is built on.
This page lays the groundwork: what it means to treat a fluid as a continuum, how we describe its
motion with a velocity field and streamlines, the two rival viewpoints (Lagrangian —
ride along with a parcel; Eulerian — watch fixed points go by), and the single most
important operator in the whole subject, the material derivative, which bridges the
two. Get this operator right and Euler, Bernoulli and Navier–Stokes all fall out of Newton's second
law almost mechanically.
The continuum hypothesis: smoothing over the molecules
The trick is averaging over a fluid element — a blob of fluid large enough to contain
an enormous number of molecules (so their statistics are steady) yet small enough to be a "point" on
the scale of the flow. Inside such a blob the density
\rho(\mathbf{x},t) = \lim_{\delta V \to \delta V^*} \frac{\delta m}{\delta V}
settles to a well-defined value: shrink the volume \delta V too far and the
molecule count fluctuates wildly (the limit is nonsense); stay above the tiny cutoff
\delta V^* and the ratio is smooth and reproducible. We then pretend
this limit is taken all the way to a mathematical point, and treat \rho,
pressure p, temperature T and velocity
\mathbf{u} as continuous, differentiable functions of position and time.
When is this legitimate? The test is the Knudsen number, the ratio of the molecular
mean free path \lambda (how far a molecule travels between collisions) to
the flow scale L:
\mathrm{Kn} = \frac{\lambda}{L}.
In air at sea level \lambda \approx 70\ \text{nm}, so for anything bigger
than a speck of dust \mathrm{Kn} \ll 1 and the continuum picture is superb.
It only breaks down where the fluid is rarefied or the scales are tiny — the fringes of the atmosphere
where a re-entering capsule flies, gas in a microchip channel, the flow through a nano-pore.
The velocity field and streamlines
The central object is the velocity field \mathbf{u}(\mathbf{x},t):
a vector telling you how fast, and in which direction, the fluid at position
\mathbf{x} is moving at time t. In components,
\mathbf{u} = \big(u(x,y,z,t),\ v(x,y,z,t),\ w(x,y,z,t)\big).
A flow is steady if the field does not depend on time
(\partial \mathbf{u}/\partial t = 0) — the pattern of arrows is frozen even
though fluid streams through it — and unsteady otherwise. To picture a field we draw
streamlines: curves that are everywhere tangent to the velocity, so no fluid
ever crosses one. In two dimensions a streamline satisfies
\frac{dy}{dx} = \frac{v}{u}, \qquad\text{equivalently}\qquad \frac{dx}{u} = \frac{dy}{v}.
Reveal the figure below step by step. First the raw velocity arrows on a grid — the Eulerian snapshot.
Then the streamlines threaded tangent to them. Then a single marked parcel, and the path it drifts
along.
In a steady flow, streamlines, pathlines (the actual trajectory a parcel traces out)
and streaklines (the line joining all parcels that passed through one point, like a
plume of dye) all coincide. In an unsteady flow they generally differ — a classic source of confusion,
because the beautiful streamline pattern at one instant is not the road any particle actually
travels.
Two viewpoints: Lagrangian and Eulerian
There are exactly two honest ways to describe a moving fluid, and knowing which one you are using is
half the battle.
-
Lagrangian — ride the parcel. Label each fluid particle (say by its
starting position \mathbf{X}) and follow it, reporting quantities as
functions of which particle and time: \mathbf{x}(\mathbf{X},t).
This is Newton's natural language — "force = mass × acceleration" is a statement about a specific
lump of matter.
-
Eulerian — watch the window. Fix your attention on points of space and
report what flows past: \mathbf{u}(\mathbf{x},t),
p(\mathbf{x},t), and so on. This is the field picture, and it is what
measurements (a pressure gauge, an anemometer) actually give you.
The Eulerian field is almost always easier to work with — but Newton's law lives in the Lagrangian
world. The whole problem is: how do you compute the acceleration of a parcel when all you have
is the Eulerian field? That is precisely what the material derivative solves.
The material derivative: the rate of change a parcel feels
Follow a parcel carrying some quantity — its temperature T(\mathbf{x},t),
say. As the parcel moves, its position is itself changing with time, so by the chain
rule the temperature it experiences changes at the rate
\frac{dT}{dt} = \frac{\partial T}{\partial t} + \frac{\partial T}{\partial x}\frac{dx}{dt} + \frac{\partial T}{\partial y}\frac{dy}{dt} + \frac{\partial T}{\partial z}\frac{dz}{dt}.
But the parcel moves with the flow, so dx/dt = u,
dy/dt = v, dz/dt = w. Substituting and recognising
the gradient, this collapses into the beautifully compact material derivative (also
called the substantial, convective, or total derivative):
-
Definition.
\frac{D}{Dt} = \frac{\partial}{\partial t} + (\mathbf{u}\cdot\nabla).
-
Local + convective. The first term
\partial/\partial t is the local rate of change at a fixed
point; the second (\mathbf{u}\cdot\nabla) is the convective
(advective) change from the parcel being carried into regions where the quantity differs.
-
Acceleration. Applied to the velocity itself it gives the acceleration of a fluid
parcel — the thing Newton's law needs:
\mathbf{a} = \dfrac{D\mathbf{u}}{Dt} = \dfrac{\partial \mathbf{u}}{\partial t} + (\mathbf{u}\cdot\nabla)\mathbf{u}.
Written out, \mathbf{u}\cdot\nabla = u\,\partial_x + v\,\partial_y + w\,\partial_z.
That convective term is nonlinear in \mathbf{u} — velocity
multiplying its own gradient — and that single nonlinearity is the seed of nearly every hard problem in
fluid dynamics, turbulence included.
Worked examples
Example 1 — a parcel warming as it flows. A steady two-dimensional flow has
\mathbf{u} = (2,\,3)\ \text{m/s} at a point where the temperature field, though
steady (\partial T/\partial t = 0), has gradient
\nabla T = (4,\,-1)\ \text{K/m}. The rate of change of temperature felt by the
parcel is
\frac{DT}{Dt} = \underbrace{\frac{\partial T}{\partial t}}_{0} + \mathbf{u}\cdot\nabla T = (2)(4) + (3)(-1) = 8 - 3 = 5\ \text{K/s}.
The parcel heats at 5\ \text{K/s} even though nothing at any fixed point is
changing — it is being carried into warmer fluid. That is convection.
Example 2 — convective acceleration in a nozzle. Water accelerates down a
contracting pipe in steady flow, u(x) = 3 + 2x\ \text{m/s} (one-dimensional).
Its acceleration is entirely convective:
a = \frac{Du}{Dt} = \underbrace{\frac{\partial u}{\partial t}}_{0} + u\frac{\partial u}{\partial x} = (3 + 2x)(2).
At x = 1\ \text{m} that is (5)(2) = 10\ \text{m/s}^2.
The flow is steady, yet parcels are accelerating hard — steady is not the same as unaccelerated.
Example 3 — unsteady plus convective. A one-dimensional flow has
\partial u/\partial t = -4\ \text{m/s}^2 (the field is slowing everywhere)
while u = 6\ \text{m/s} and \partial u/\partial x = 1.5\ \text{s}^{-1}
at a point. The parcel's acceleration adds both pieces:
a = \frac{\partial u}{\partial t} + u\frac{\partial u}{\partial x} = -4 + (6)(1.5) = -4 + 9 = 5\ \text{m/s}^2.
This is the single most common slip in fluid kinematics. Steady means the field
at each fixed point is unchanging: \partial \mathbf{u}/\partial t = 0.
It does not mean the fluid parcels feel nothing. Water speeding through a narrowing nozzle,
or whirling round a bend, is in a perfectly steady flow, yet every parcel is accelerating violently —
the convective term (\mathbf{u}\cdot\nabla)\mathbf{u} is doing all the work.
The lesson: never confuse \partial/\partial t (Eulerian, at a fixed point)
with D/Dt (following the parcel). A pressure gauge bolted to the pipe reads a
constant value — \partial p/\partial t = 0 — while a parcel drifting past it
experiences Dp/Dt \neq 0. Both are true; they answer different questions.
Dye traces a streakline — the locus of all parcels that have passed through the
injection point. Smoke from a chimney, dye in a tank, a vapour trail: all streaklines. In a
steady flow they lie exactly on streamlines, which is why wind-tunnel smoke pictures look so
clean and reveal the flow so faithfully. But start gusting the wind and the three curves —
streamline, pathline, streakline — peel apart, and the pretty smoke line no longer tells you where any
single particle actually went. The tidy pictures you have seen are quietly relying on steadiness.