Put a wing model in a wind tunnel and stretch a thin oiled wire across the stream just upstream of it. Heat the wire and a row of smoke threads peels off and wraps over the wing, squeezing together above the upper surface. Beside the tunnel, the textbook shows a neat family of streamlines around the same airfoil. It is tempting to read the photograph and the plot as the same curves. In steady flow they are. In a gust, a pitching wing or the wake of a stall they are not, and reading one as the other gives the wrong picture of where the air goes.
There is an even more basic question first. Air is molecules, of them in a cube one micrometre on a side, each flying at an average m/s in a random direction. What does “the velocity of the air at a point above the wing” mean when the wing moves at only m/s? This lesson answers that and builds the vocabulary of every fluid-dynamics paper and simulation code. The tools are ones you have: the chain rule and Jacobian of 2.2 Linearization, the Chain Rule & the Jacobian, the paddle wheel of 2.8 Curl & Vorticity, the eigenvectors of 1.5 Eigenvectors & Eigenvalues and RK4 from 4.3 Runge–Kutta Methods & Butcher Tableaux. A pathline is an ODE you already know how to integrate.
Transcript
- Stretch a heated wire across a wind tunnel just ahead of a wing, and smoke threads wrap over it, squeezing together above the surface. The textbook draws streamlines around the same wing, and in steady flow the two agree. But what is the velocity of the air at a point? A micrometre cube of air holds twenty-five million molecules, each flying at four hundred fifty-nine metres per second in a random direction, while the wing moves at only fifty.
- The answer is averaging. Air's mean free path is about sixty-eight nanometres, and the Knudsen number is lambda over L. For a one-metre chord it is six point eight times ten to the minus eight, so even a tiny cube averages millions of molecules and the velocity is a smooth field. A channel one micrometre wide has a Knudsen number of zero point zero six eight. That is slip flow: the equations still work, but the gas slides along the walls.
- Now there are two ways to describe one flow. Eulerian: stand still and record. Probes on a grid measure a velocity field, u of x and t. Lagrangian: ride along. Label a parcel and follow its position, which obeys d X by d t equals u at X and t. That is an initial value problem, so each parcel path is integrated with RK4, and we watch where it goes.
- Newton's law applies to the parcel, so we need its acceleration, not the change at a fixed point. The chain rule gives the material derivative: a local change plus a convective one. In a contraction that speeds the air up by thirty percent, every probe reads a constant, yet the parcel accelerates at u times u x, peaking at seven hundred sixty-four point seven metres per second squared, seventy-eight g, at x equals minus zero point six five six. Differentiating an RK4 parcel's speed agrees to within zero point zero two seven.
- Zoom in on one parcel. A neighbour moves with the velocity gradient L times its offset, and L splits into a symmetric strain rate S and an antisymmetric rotation W. Try a shear flow, gamma y comma zero. A horizontal fibre does not move at all. A vertical fibre tips clockwise at gamma. At forty-five degrees S stretches a fibre at gamma over two, and at a hundred thirty-five it squeezes at the same rate. Those are the eigenvectors of S, with eigenvalues plus and minus gamma over two.
- Rates are instantaneous, so let the shear act for a finite time. At gamma t equals one, the square becomes a parallelogram: the forty-five degree diagonal grows by one point five eight one one, the other shrinks to zero point seven zero seven one, and the area stays one. At gamma t of a tenth, the factors give rates near plus and minus one half. A circle becomes an ellipse of axis ratio two point six one eight zero, its long axis at thirty-one point seven two degrees, not forty-five, because rotation keeps turning the material.
- Three different curves come from one velocity field and one source. A streamline is tangent to the field frozen at a single instant. A pathline is one parcel's track. A streakline joins every parcel that has passed the source, which is exactly what a smoke wire shows. Take a wind whose direction swings, u equals one comma a cosine omega t, and watch all three grow from the origin.
- At t equals four and x equals two the three heights differ. The streamline is at minus zero point six five four, the pathline at zero point four five five, and the streakline at minus zero point eight three three. The reason is memory: the streamline knows only the wind's direction now, the pathline one parcel's history, the streakline many parcels released into different phases of the swing. In steady flow all three coincide, which is why a tunnel photograph looks like the textbook; in an unsteady flow, smoke is not streamlines.
- Back to the contraction. The air speeds up thirty percent over the bump, yet a small round parcel is stretched and squeezed and keeps its area, one point zero zero zero. So where does the air go when it speeds up? The vertical velocity, v equals minus y f prime, squeezes the streamlines together. The next lesson turns that into a law: the divergence of u is zero, conservation of mass.
The continuum hypothesis
The distance a molecule travels between collisions is its mean free path ; for air at sea level nm (approximate, from memory; it grows as the density falls). Average the molecular velocities over a volume that holds very many molecules yet is tiny compared with the length over which the flow changes. If such a volume exists, the average is a smooth field and calculus applies. That is the continuum hypothesis, and its yardstick is the Knudsen number
For a wing chord of m, : a micrometre cube, a millionth of the chord, already holds molecules, so the averages are smooth. For a channel one micrometre wide (a MEMS valve), . The usual regimes, from memory and approximate: continuum below ; slip flow from to , where the equations still work but the gas slides along walls instead of sticking; transitional up to about ; free-molecular beyond. The m/s is the mean molecular speed at C. The flow velocity is the small average drift on top of it; the random part appears in the continuum as pressure and temperature.
Two ways to describe a flow
Eulerian: stand still and record. The velocity is a field , what a probe fixed in the tunnel measures and what a grid solver stores.
Lagrangian: ride along. Label a parcel by its starting point and follow its position:
This is an initial value problem. In a steady flow it is autonomous and the parcel paths are the trajectories of a phase portrait (7.3 Phase Portraits & the Trace–Determinant Plane); in an unsteady flow the right-hand side depends on directly.
The material derivative
Newton’s law applies to a parcel, so we need the parcel’s acceleration, not the rate of change at a fixed point. The parcel’s velocity is . The chain rule gives, for the first component,
because and ; the second component is the same with on top. Together,
The first term is the local change, the field changing in time. The second is the convective change, the parcel moving to where the field is different. , the Jacobian of the velocity field, is the velocity gradient. The chain-rule lesson carried a temperature this way; here the velocity carries itself, and that is where the nonlinearity of fluid mechanics comes from.
Worked example: a steady flow that accelerates
Take a made-up, wing-like speed bump on the centreline of a contraction: , m/s, m. The air speeds up by over the crest. The flow is steady, so every fixed probe reads a constant. Yet
The product peaks ahead of the crest, at m. By hand: , so m/s and s, giving m/s, about g. Behind the crest the parcel brakes just as hard ( is even, odd). The force that does this is a pressure gradient, the subject of the Euler-equations lesson (11.3).
Is really the parcel’s acceleration? The lab cell follows one parcel with RK4 ( s), differentiates its speed in time and compares: the largest gap is m/s, finite-difference error. The sim’s contraction is the same bump with (multiply its accelerations by m/s), plus the vertical velocity a two-dimensional flow needs. Its readout shows , and the acceleration measured along the parcel’s own RK4 path; the last two agree to the digits shown.
Strain and rotation
Zoom in on a parcel at . A neighbour at moves with (the linearization of the chain-rule lesson). The first term carries the parcel along unchanged; everything that changes its shape or orientation is in . Split into symmetric and antisymmetric parts:
is the strain-rate tensor, the rotation tensor. The two parts do physically different things.
Worked example: the shear flow, fibre by fibre
The shear flow is the Couette profile of 8.1 Two-Point Boundary Value Problems, and the simplest model of the boundary layer on a wing, where the air goes from rest at the skin to flight speed across a few millimetres. Here and the other derivatives vanish:
Don’t trust the matrices yet. Put short fibres through the centre, with s, and see what does:
- Horizontal, : . Both ends move together; no stretch, no turn.
- Vertical, : . The top runs ahead, so the fibre tips clockwise at rad/s.
- At , : , pushed straight outward: it stretches at rate .
- At , : : it shrinks at rate .
So and are eigenvectors of , eigenvalues : directions only stretches or squeezes, never turns. These are the principal strain directions, and because is symmetric they are always perpendicular (1.6 Change of Basis & Diagonalization). Locally, the strain part of any flow is a stretch along one axis and a squeeze along the other.
The rotation part, , turns every fibre rigidly at : the average of the horizontal fibre’s and the vertical fibre’s . With the vorticity (the 2D curl), the rate is , the paddle-wheel rate of the curl lesson. A flow with perfectly straight streamlines is a stretch plus a clockwise spin. On a wing, almost all the vorticity lives in this thin boundary layer and the wake; outside, the flow is nearly irrotational (from memory), which the potential-flow and airfoil lessons will exploit.
For a general symmetric , the characteristic equation gives
for the angle of the stretching axis. For the shear, , : , . The trace is the divergence (2.7 Divergence: Sources & Sinks), the rate of area growth; here it is zero.
Following the parcel for a finite time
Rates are instantaneous; to see the shape, integrate the pathlines of the parcel’s boundary. In the shear flow , , which the sim’s RK4 reproduces to round-off. At a square of side has corners , , , relative to its centre. The diagonal has grown by , the one shrunk to , and the area factor is . At the factors are and , rates near and : the eigenvalues , exactly so only as .
A circle becomes an ellipse with axis ratio at , long axis at , not . The principal axes of sit at at every instant, but keeps turning the material, sweeping the stretched direction down toward the flow. Problem 2 derives both numbers.
Now pick the vortex and set ; the parcel starts at , far outside the core of radius . The readout says with eigenvalues . The radial fibre turns clockwise at rad/s, the tangential one counterclockwise at rad/s, and their average is zero: the parcel is strained into an arc as it orbits, but it does not spin.
Streamlines, pathlines and streaklines
Three ODE problems, one velocity field, one source point :
- A streamline at time is tangent to the field frozen at : , where is not time. A snapshot.
- A pathline is one parcel’s track: , . A long exposure of one tracer.
- A streakline at time joins all parcels that passed at earlier times : the curve . What a smoke wire shows.
In steady flow the three coincide. The streamline and pathline equations are then the same autonomous ODE, and every parcel released from , at any , starts from the same point, so by uniqueness (3.7 Existence & Uniqueness) it runs along the same curve, only later. When the field changes in time the first step of that argument fails.
Worked example: a swinging wind
Take a uniform wind whose direction swings, , , , source at the origin, releases from .
- Streamline at time : the frozen field has slope , so .
- Pathline: , , that is .
- Streakline: the parcel released at sits at , ; with , .
At and : (streamline), (pathline), (streakline). Three heights, from one source. The mechanism is memory: the streamline knows only the wind’s direction now, the pathline one parcel’s history, and the streak many parcels released into different phases of the swing. Set and the three agree (the sim reads for each).
The sim puts the source at and the probe two units downstream, and its RK4 curves read , and . The travelling wave varies in space too. With the source at and , at (steady) all three cross the probe at ; at they split to , and . At the wave rides with the stream, each parcel keeps one phase, is constant, and the pathlines are straight while the streamlines stay wavy.
Try it
(1) Shear, : read , , , the corners and the diagonal factors; switch to a circle; drag to . (2) Contraction: play, and watch stay zero while swings positive then negative. (3) Vortex from . (4) Swinging wind at , then . (5) Travelling wave at , , . In all five flows the area factor stays up to the polygon’s error.
Computing it
The tolerance is the default of close, relative: with the central difference errs by about on the Taylor-Green field and only by round-off on the quadratic fields. Angles are compared modulo , since an axis has no preferred end.
Concept checks
Problems
Where the air meets the nose of a wing, a model flow near the stagnation point is , s. (a) Find , , , . (b) Find at m. (c) Check (b) from the pathline.
Show worked solution
(a) , , : pure strain, stretching along and squeezing along at s; divergence .
(b) Steady, so m/s.
(c) , , so , : m/s again. The pathlines are the hyperbolas . (These numbers are not in the sim.)
In the shear flow at , points move by with . A circle of radius becomes an ellipse with semi-axes times the square roots of the eigenvalues of , along its eigenvectors. Find the axis ratio and the angle of the long axis.
Show worked solution
: trace , determinant , so and , . The ratio is (the eigenvalues multiply to ). The eigenvector for solves , so and the angle is , as the sim’s circle readout shows. Determinant means the area is unchanged.
For the swinging wind, derive the streakline formula and evaluate it at , . Then find the times at which the streakline lies on the pathline along its whole length.
Show worked solution
The parcel released at has and ; put . Then , as the lab and the sim print. Equality for every needs ; differentiate in at : , so , after whole swings, and there the identity holds. The sim’s slider brackets : the streakline reads at and at , passing the pathline’s .
Key Takeaways
- The continuum works when is small: for a 1 m chord, (slip flow) for a 1 micrometre channel. Flow velocity is an average drift on molecules moving at about m/s.
- Eulerian: a field at fixed points. Lagrangian: parcels with , integrated by RK4 with carried through the stages.
- . Steady flows still accelerate parcels: g over a speed bump at m/s.
- : stretches along its perpendicular eigenvectors, spins at . Shear : at and , .
- Straight streamlines can carry vorticity (the boundary layer); circular ones need not (outside a vortex core).
- Streamlines, pathlines and streaklines coincide in steady flow and differ in unsteady flow: , , in the swinging wind.
- Next, conservation of mass (11.2): over the bump the centreline air speeds up by . Where does the air go when it speeds up? The term squeezes the streamlines together, and every parcel in the sim kept its area. The next lesson makes that a law, .