The last lesson ended with a cylinder that feels no drag. Euler’s air slides along a surface as if oiled, forbidden only to pass through it. Real air sticks: at the skin of a wing it moves with the wing, and within a few millimetres its speed climbs to flight speed. The shearing in that thin layer is where skin-friction drag comes from, and where, much later, stall begins.
The missing physics is viscosity; adding it to Euler gives the Navier–Stokes equations, the equations a wind-tunnel simulator solves. One idea is new: momentum diffuses, exactly as heat does. The tools you own: the strain-rate tensor from 11.1 Continuum Hypothesis & Flow Kinematics, the momentum balance from 11.3 Momentum: Pressure & the Euler Equations, BVPs from 8.1 Two-Point Boundary Value Problems, and stability of explicit steps from 4.5 Stability Regions & Stiffness.
Transcript
- The last lesson ended with a cylinder that feels no drag: Euler's air slides along a surface as if oiled, which is d'Alembert's paradox. Real air sticks. At the skin of a wing it moves with the wing, and within a few millimetres its speed climbs to flight speed. That shearing is where skin friction comes from. The missing physics is viscosity.
- Picture two layers of air, the upper one faster. Molecules hop across the boundary carrying momentum: the slow layer is dragged forward, the fast one held back. Momentum per second per area is a shear stress, and Newton's law says tau equals mu du dy. Divide by density and you get nu, a diffusion coefficient: one point four seven times ten to the minus five for air, ten to the minus six for water. Water is fifty-five point six times stickier by mu, yet air's nu is fourteen point seven times larger.
- Slide a plate at fifty metres per second over a one millimetre film of air. The shear rate is five times ten to the fourth, so the stress is zero point nine pascals. To handle any face we need a tensor, sigma, that turns a face normal into the force on it. Shears on the top and bottom, balanced by the sides, plus pressure, give minus p on the diagonal and mu gamma off it: minus p I plus two mu S. Tilt the face to forty-five degrees and the force is purely normal: plus or minus zero point nine pascals.
- Take the divergence of that stress. Continuity kills the extra term and leaves mu times the Laplacian. Divide by rho and you have the Navier-Stokes equations: Euler plus nu del squared u. The new term is diffusion, so the equation is second order in space, and a wall needs a second condition. That is no-slip: the fluid's velocity equals the wall's.
- Plate flows turn Navier-Stokes into a two-point boundary value problem. With no pressure gradient, Couette is a straight line. Between fixed plates, Poiseuille is a parabola. For air in a one centimetre slot with G equal to one pascal per metre, the peak is zero point six nine four four metres per second, the mean zero point four six three zero, and the flow rate four point six three times ten to the minus three. Start it from rest: every layer first gains zero point eight one six, then the walls' brake diffuses inward, and the slowest mode decays at one point four five zero two per second.
- Now start air at rest and snap the plate to speed U. Momentum diffuses upward, and a similarity argument turns the heat equation into an ODE whose solution is U times erfc of y over two root nu t. The edge of the layer is four root nu t: four point eight five millimetres after a tenth of a second, fifteen point three after one second, growing like root t. The dots are a finite-difference run, and they sit on the curve.
- Why must r be small? Feed forward Euler the sharpest pattern a grid can hold, a sawtooth. Each step multiplies it by one minus four r. Up to r equal to one half that factor stays between minus one and one. At point four the error is eight point seven times ten to the minus four. At one half it is two point one times ten to the minus three and never decays. At point five one it has grown to zero point zero nine eight, and at point six, three times ten to the eleven.
- Where does the energy go? Only strain heats the fluid: the dissipation is two mu S double dot S. In the Couette film that is forty-five thousand watts per cubic metre, forty-five per square metre, exactly the plate's power: zero point nine pascals times fifty. Next lesson: the equations hold one viscosity, one speed and one length, but only one combination matters, the Reynolds number. For this film it is three thousand four hundred three.
Viscosity: momentum leaking between layers
Picture two layers of air, the upper faster. Molecules hop across the boundary at random (kinetic theory, from memory). One crossing downward carries the upper layer’s larger -momentum into the slow layer; one crossing upward brings a deficit. No net mass crosses, but momentum does: the slow layer is dragged forward, the fast one held back. Momentum per second per unit area is a force per unit area, a shear stress .
For most everyday fluids it is proportional to the shear rate, Newton’s law of viscosity:
is the dynamic viscosity (Pa s). Divided by density it gives the kinematic viscosity , in m/s, the units of a diffusion coefficient. Sea-level values, approximate and from memory: air Pa s, so m/s; water , . Water is times stickier by , yet air’s is times larger: air is so light that a little friction moves it a lot.
Worked example: a plate over a thin film of air
A plate slides at m/s over a mm gap of air above a fixed floor, the air sheared linearly. Then s and
That is of the dynamic pressure, Pa (11.3’s pitot reading). A tiny stress, but it acts along the flight direction over the whole wetted surface, while pressure forces largely cancel in that direction: hence skin friction is a major part of a clean airplane’s drag (from memory).
The stress tensor, built from a shear flow
In 11.3 a face felt only pressure, along . Now it also feels a sideways pull. We want a machine taking a face’s normal to the force per area on it, the traction ; it is a matrix, the stress tensor: .
Build it for the shear flow of 11.1, at pressure :
- Top face, : the faster air above drags it forward with ; pressure pushes down with . So , the second column of .
- Right face, : pressure pushes left with . A shear too? On a tiny square parcel the top and bottom shears form a couple; it scales like the side squared while the moment of inertia scales like its fourth power, so the parcel would spin up infinitely fast unless the side faces carry an opposite couple. So the right face feels along : , the first column.
The second matrix is exactly this flow’s strain-rate tensor from 11.1. Now tilt the face to , , the eigenvector of along which the fluid stretches. Then : a pure normal force, pressure reduced by , a viscous tension. On the face, the squeezing direction, it is : extra compression. In the mm film that is Pa. Shear and normal stress are one tensor seen from different faces.
The Newtonian law
For a general flow, a Newtonian incompressible fluid has
Stated, not derived, it assumes: viscous stress is linear in the velocity gradient (air and water, not paint or blood); it ignores the rotation , since rigid spinning rubs nothing; it is isotropic, no preferred direction, leaving one constant ; and it is symmetric, by the torque argument. Since , is still minus the average normal stress. (A compressible gas adds a term in ; 11.12.)
From stress to Navier–Stokes
The box argument of 11.3 carries over: the -force on a box is the stress on its right face minus its left, plus top minus bottom, and by Taylor, per unit volume,
the divergence of the first row of . With and :
Continuity kills the last bracket and leaves the Laplacian (2.11 The Laplacian & Vector Identities). The -row works the same way. Newton’s law per unit volume, divided by , gives the incompressible Navier–Stokes equations:
One new term, . The Laplacian measures how far a value sits below its neighbours’ average, so viscosity pulls each parcel’s velocity toward its neighbours’: diffusion. The count still closes and pressure is still the multiplier enforcing , but the equation is now second order in space, which changes the boundary conditions.
No-slip
At a solid wall the fluid velocity equals the wall’s, both components: . Physically (from memory): molecules striking a microscopically rough wall bounce off in random directions, forgetting their tangential momentum, so the gas touching the wall shares its velocity on average. It fails only in rarefied gas, where the mean free path rivals the body size.
Euler, first order in space, takes one wall condition, . Navier–Stokes, second order, needs the tangential one too. As the highest derivative fades but the extra condition stays, so the solution must change abruptly in a thin layer to meet it. That layer is the boundary layer (11.8).
Plate flows: Navier–Stokes as a two-point BVP
Seek a flow between plates at (fixed) and (sliding at ). Continuity holds automatically, and the nonlinear term vanishes identically: that is why these flows have exact solutions. The -equation gives (ignoring gravity), so depends on only, and the -equation is
The left side depends only on , the right only on , so both equal a constant, with . The BVP , , , integrates twice:
Couette (): a straight line, the mm film. Plane Poiseuille (): a parabola, which central differences reproduce exactly at the nodes.
Worked example: Poiseuille in a 1 cm slot of air
Pa/m, cm. The peak, at mid-gap, is m/s, the mean is two thirds of that, m/s, and the flow rate per unit width is m/s. The wall shear is Pa, for a reason: a slab of length is pushed by of pressure and held by of friction, and steady means they balance.
Momentum diffuses like heat
Start the channel from rest. With the nonlinear term still vanishes:
Without the forcing this is the heat equation, , with velocity as temperature. Read it as a stencil, : a layer slower than its neighbours’ average speeds up, a faster one slows, at a rate set by . Units fix the time scale: is m/s, so momentum crosses a gap in a time of order . For cm of air that is s, a scale, not a rate.
Separating variables gives the rate. Write , steady; the error obeys with at both plates. Try : . Every mode decays, sharp ones fastest; the last is , at rate s, time constant s: the scale divided by . Expanding the initial error, itself, in these sines gives the exact transient.
At first every layer accelerates at m/s; the brake at the walls diffuses inward. The centre passes half its final speed at about s (the exercise’s Fourier series gives m/s then). On the log plot the error turns straight: one mode survives. The readout measures its decay rate over the last second as s, against for the PDE; the gap is mostly forward Euler’s factor per step. The error drops below at s.
Stokes’ first problem: the impulsively started plate
Air at rest fills , and at the plate at jumps to speed .
Similarity. The problem contains no length except , so the profile must be one fixed shape stretched by it: with . By the chain rule, and , so
The PDE has become an ODE, separable in : . With and , using ,
Thickness. At , : call that the edge, . In air is mm after s, and it grows like .
Wall shear. Differentiating, , so : infinite at the jump, then decaying as the layer thickens.
Checking erfc numerically, and the stability limit
Discretise with forward Euler in time and central differences in space (FTCS):
Why must be small? Feed it the sharpest pattern a grid can hold, the sawtooth . Then , and each step multiplies the sawtooth by . For that factor lies in ; above, the sawtooth flips sign and grows every step. The stiffness lesson said the same: the discrete Laplacian has eigenvalues down to , and forward Euler needs . Halve and the step must shrink fourfold.
With mm, run to s. At ( ms, steps) the largest error is . At it is : the jump at the wall seeded a sawtooth with factor exactly , which never decays. At the factor is and the error has grown to ; at , factor , it reaches in steps.
Energy dissipation
The power of the viscous stress per unit volume, , splits in two: changes the kinetic energy, and the rest, , heats the fluid. Since is symmetric and antisymmetric, sums to zero (pairs cancel), so the heating rate per unit volume is
Only strain dissipates. For Couette, , so and . In the mm film, W/m: W per square metre of plate. Check: the plate pushes against Pa at m/s, W/m. If the heat stayed, the film would warm at K/s ( J/(kg K), from memory).
Computing it
The solution prints m/s at s against the series’ (graded to m/s) and at s, when the slowest mode is down to about (graded to of ).
Concept checks
Problems
Plates cm apart, air, top plate sliding at m/s. (a) Find the wall shear for the Couette–Poiseuille profile. (b) For what is it zero? (c) What happens below that?
Show worked solution
(a) , so .
(b) Pa/m.
(c) : backflow beside the fixed plate. With the channel panel’s steady wall shear is Pa at and Pa at .
For plane Poiseuille flow, show that , where . Evaluate for Pa/m, cm, air.
Show worked solution
, . Then , so . And , so the dissipation equals , the pressure drop’s pumping power. Numerically W per square metre of plate (lab block C).
(a) Verify that satisfies and the boundary conditions. (b) Find the wall shear stress. (c) For air and a plate at m/s, evaluate and at s.
Show worked solution
(a) For , , so , which is what reduced to. At , ; as or with , and .
(b) , which at is : , holding the plate back.
(c) m, so mm and Pa (lab block D).
Key Takeaways
- Viscosity is momentum diffusion: ; is m/s for air, times water’s.
- turns a face normal into a force: shear on flat faces, normal stresses at .
- Its divergence is : Navier–Stokes is Euler plus , second order, so walls need no-slip.
- Plate flows are BVPs: Couette ( Pa at m/s over mm of air), Poiseuille .
- Start-up is the heat equation; the slowest mode decays at s, at s.
- Impulsive plate: , ; FTCS reproduces it only for .
- Dissipation is : W/m in the Couette film, exactly the plate’s power .
- Next, 11.5: the equations hold a viscosity , a speed and a length , yet only one combination matters, the Reynolds number . For the mm film it is , and whether that film stays smooth depends on it alone.