Flow Lab

Syllabus / Module 11: Fluid Dynamics: Navier–Stokes and Beyond

Lesson 11.8

Boundary Layers

Prandtl's insight, boundary-layer equations, Blasius solution, displacement thickness, separation and stall.

Written byClaude Opus

🌊
Why this matters for simulation: Separation is why wings stall and why golf balls have dimples.

By the end of this lesson you can…

  • Explain Prandtl's split of a high-Re flow into an inviscid outer flow and a thin viscous layer, and derive the boundary-layer equations by scaling, with $\delta/L \sim \mathrm{Re}^{-1/2}$.
  • Reduce the flat-plate layer to the Blasius ODE $f''' + \tfrac12 ff'' = 0$ with a similarity variable, and solve it by shooting on $f''(0)$, checking how the truncated domain moves the answer.
  • Compute $\delta_{99} = 4.910\,x/\sqrt{\mathrm{Re}_x}$, $C_f = 0.664/\sqrt{\mathrm{Re}_x}$ and a laminar plate's friction drag, for a 1.5 m chord at 50 m/s.
  • Explain why an adverse pressure gradient separates a layer, find the Falkner–Skan separation value $m = -0.0904$ ($\beta = -0.1988$) by shooting, and connect separation to stall and golf-ball dimples.
On this page
·
  1. Prandtl’s idea
  2. The boundary-layer equations by scaling
  3. The flat plate: a similarity solution
  4. Solving Blasius by shooting
  5. Skin friction and the drag of a plate
  6. Worked example: the light wing’s laminar layer
  7. Pressure gradients and separation
  8. Falkner–Skan: a family with a dial
  9. Stall, and why golf balls have dimples
  10. Computing it
  11. Concept checks
  12. Problems
  13. Key Takeaways

Last lesson ended with a paradox. Potential flow gave the light-aircraft analog a lift of 1148.41148.4 N/m and a drag of exactly zero, for every shape, because drag’s work had nowhere to go. The culprit is the condition potential flow dropped: it lets air slide along the skin, while real air sticks (no-slip, 11.4 Viscosity & the Navier–Stokes Equations). Ludwig Prandtl’s answer (1904, from memory): at high Reynolds number viscosity matters only in a boundary layer, a sheet so thin that potential flow is right outside it, while inside it no-slip drags the speed down to zero at the wall. Friction drag lives in that sheet; when it tears off the surface, so do pressure drag and stall. Its profile is a boundary value problem on an infinite domain, solved by the shooting of 8.1 Two-Point Boundary Value Problems.

Tools: 11.5 Non-dimensionalization & the Reynolds Number, 11.4 Viscosity & the Navier–Stokes Equations, 11.7 Potential Flow & Stream Functions and 8.1 Two-Point Boundary Value Problems.

▶ Video A thin sheet where no-slip slows the air, the Blasius profile found by shooting, the layer thickening along a wing, and an adverse pressure gradient that stops the slow air and tears the layer offcaptions only
Transcript
  1. Last lesson ended on a paradox. Potential flow gave the wing lift and exactly zero drag, d'Alembert's paradox, because it lets air slide along the skin. Real air sticks. Ludwig Prandtl's answer, in 1904: at high Reynolds number, viscosity matters only in a thin boundary layer hugging the surface. Outside it, potential flow is right. Inside it, the speed falls to zero at the wall.
  2. No-slip pins the air at the wall, and the speed climbs to the outer stream across a thickness delta, far smaller than the length L. Balance inertia, U squared over L, against viscosity, nu U over delta squared, and delta over L scales like Reynolds number to the minus one half. And the pressure does not change across the layer: the outer flow impresses it.
  3. On a flat plate there is no length but x, so the profile is one shape, stretched by the square root of nu x over U. With eta the scaled height and a stream function, the partial differential equation collapses to the Blasius equation: f triple prime plus a half f f double prime equals zero. At the wall, f and f prime are zero. Far away, f prime equals one.
  4. That is a boundary value problem, so we shoot. Guess the wall slope f double prime of zero, integrate with RK4, and measure the miss in f prime at the far end. Starting from 0.2, the profile undershoots. From 0.5, it overshoots. The secant method steers between them, and in six steps lands on 0.332057, with f prime reaching one.
  5. But infinity is not on the grid. End the domain at eta equals 6 and you get 0.332566. At 8, 0.332059. At 10 and 12, seven digits agree, because f double prime dies like a Gaussian. And one shot with slope 1 ends at f prime equals 2.085409. Rescale it: 2.085409 to the minus three halves is 0.332057336. No iteration needed.
  6. Read the layer off the profile. F prime reaches 0.99 at eta equals 4.910: delta ninety-nine is 4.910 x over root Reynolds x. Displacement thickness is 1.7208, momentum thickness 0.6641, in the same units. The wall slope gives skin friction: C f equals 0.664 over the square root of Reynolds x. It is largest at the front, where the layer is thinnest.
  7. Now the wing: a 1.5 metre chord at 50 metres per second. Reynolds number 5.135 times ten to the sixth. If the layer stayed laminar, it would be 3.250 millimetres thick at the trailing edge, and rub 2.693 newtons per metre off both faces: where d'Alembert's drag went, a momentum deficit in a thin sheet. But transition starts near Reynolds x of 5 times ten to the fifth, at only 0.1461 metres. So 2.693 is a lower bound.
  8. Now the pressure gradient. Behind the suction peak the pressure rises: an adverse gradient. At the wall, mu u yy equals p e prime, so the profile bends the wrong way and grows an inflection. Falkner-Skan makes it exact: edge speed K x to the m, with m a dial. Set the wall shear to zero and shoot on m: separation at m equals minus 0.090429, beta minus 0.19884. The edge speed may drop only 6.08 percent per doubling of distance.
  9. Past that point near-wall air flows backwards and the layer lifts off. On a wing that is stall: circulation and lift fall, and drag jumps. A laminar layer tolerates almost no slowing; a turbulent one holds on far longer, which is why dimples help a golf ball. Next, 11.9: past Reynolds x around 5 times ten to the fifth, the layer turns turbulent.

Prandtl’s idea

11.5 showed the trap. The viscous term carries 1/Re1/\mathrm{Re}, and the light wing has Re=5.135×106\mathrm{Re} = 5.135\times10^6, so dropping it looks safe. But dropping the highest derivative loses a boundary condition: Euler can enforce “no flow through the wall”, not “no slip along it”. The escape: near the wall the velocity changes over a distance δ≪L\delta \ll L, so second derivatives there are of size U/δ2U/\delta^2, and viscosity rejoins the balance.

Prandtl’s split:

  • Outside: 11.7’s potential flow. It sets the edge speed Ue(x)U_e(x) and, through Bernoulli, the pressure pe(x)p_e(x).
  • Inside: a thin layer where inertia and viscosity balance, taking pe(x)p_e(x) as given and bringing the speed to zero at the wall.

The boundary-layer equations by scaling

Take steady 2D flow along a wall, xx along it and yy normal to it, with Navier–Stokes and continuity:

uux+vuy=−pxρ+ν(uxx+uyy),uvx+vvy=−pyρ+ν(vxx+vyy),ux+vy=0.uu_x + vu_y = -\frac{p_x}{\rho} + \nu(u_{xx} + u_{yy}), \qquad uv_x + vv_y = -\frac{p_y}{\rho} + \nu(v_{xx} + v_{yy}), \qquad u_x + v_y = 0 .

Inside the layer, xx changes over LL and yy over δ\delta, and uu is of size UU. Continuity sets the size of vv: ux∼U/Lu_x \sim U/L must cancel vy∼v/δv_y \sim v/\delta, so v∼Uδ/Lv \sim U\delta/L, small. Now size each term of the xx equation:

termuuxuu_xvuyvu_yνuxx\nu u_{xx}νuyy\nu u_{yy}
sizeU2/LU^2/L(Uδ/L)(U/δ)=U2/L(U\delta/L)(U/\delta) = U^2/LνU/L2\nu U/L^2νU/δ2\nu U/\delta^2

Both inertia terms are U2/LU^2/L. νuxx\nu u_{xx} is smaller than νuyy\nu u_{yy} by (δ/L)2(\delta/L)^2: drop it. The layer exists because the surviving viscous term balances inertia:

U2L∼νUδ2⟹δL∼νUL=Re−1/2,\frac{U^2}{L} \sim \frac{\nu U}{\delta^2} \quad\Longrightarrow\quad \frac{\delta}{L} \sim \sqrt{\frac{\nu}{UL}} = \mathrm{Re}^{-1/2} ,

11.5’s estimate, now with a mechanism. In the yy equation every term is smaller by δ/L\delta/L, so py/ρ∼U2δ/L2p_y/\rho \sim U^2\delta/L^2, and the pressure changes across the layer by about ρU2(δ/L)2\rho U^2(\delta/L)^2: a part in 5×1065\times10^6 for the wing. So the pressure does not change across the layer: p(x,y)=pe(x)p(x, y) = p_e(x), impressed by the outer flow, where Bernoulli gives −pe′/ρ=UeUe′-p_e'/\rho = U_eU_e'. What remains is

ux+vy=0,uux+vuy=UeUe′+νuyy,u_x + v_y = 0, \qquad uu_x + vu_y = U_eU_e' + \nu u_{yy} ,

with u=v=0u = v = 0 on the wall and u→Ue(x)u \to U_e(x) at the edge.

The flat plate: a similarity solution

On a flat plate in a uniform stream, Ue=UU_e = U and the pressure term vanishes. There is no length in the problem except xx itself, and 11.4’s Stokes problem had the same feature: its profile was one shape stretched by νt\sqrt{\nu t}. Here the analog of tt is x/Ux/U, so try one shape stretched by νx/U\sqrt{\nu x/U}:

η=yUνx,ψ=νUx f(η),u=ψy=Uf′(η).\eta = y\sqrt{\frac{U}{\nu x}}, \qquad \psi = \sqrt{\nu Ux}\,f(\eta), \qquad u = \psi_y = Uf'(\eta) .

The stream function makes continuity automatic. The chain rule, with ηx=−η/(2x)\eta_x = -\eta/(2x) and ηy=U/(νx)\eta_y = \sqrt{U/(\nu x)}, gives

v=−ψx=12νUx (ηf′−f),ux=−Uηf′′2x,uy=Uf′′Uνx,νuyy=U2f′′′x.v = -\psi_x = \tfrac12\sqrt{\frac{\nu U}{x}}\,(\eta f' - f), \qquad u_x = -\frac{U\eta f''}{2x}, \qquad u_y = Uf''\sqrt{\frac{U}{\nu x}}, \qquad \nu u_{yy} = \frac{U^2 f'''}{x} .

Substituting, the ηf′f′′\eta f'f'' pieces of uuxuu_x and vuyvu_y cancel, leaving uux+vuy=−U2ff′′/(2x)uu_x + vu_y = -U^2ff''/(2x). Every term carries U2/xU^2/x, which divides out:

f′′′+12ff′′=0,f(0)=f′(0)=0,f′(∞)=1.f''' + \tfrac12 ff'' = 0, \qquad f(0) = f'(0) = 0, \qquad f'(\infty) = 1 .

This is the Blasius equation (1908, from memory). f′(0)=0f'(0) = 0 is no-slip, f(0)=0f(0) = 0 is no flow through the wall, f′(∞)=1f'(\infty) = 1 joins the outer stream. A PDE became one ODE: every station’s profile is the same curve f′(η)f'(\eta), stretched by νx/U\sqrt{\nu x/U}.

Solving Blasius by shooting

Two conditions at η=0\eta = 0, one at infinity: a BVP. Shooting guesses the missing initial value s=f′′(0)s = f''(0), integrates with RK4, and measures the miss

F(s)=f′(η∞;s)−1.F(s) = f'(\eta_\infty; s) - 1 .

The lab (block A) does this with RK4, h=0.01h = 0.01, and the secant method from s=0.2s = 0.2 and 0.50.5; it converges in six steps to

f′′(0)=0.332057.f''(0) = 0.332057 .

The infinite domain, honestly. We cannot integrate to infinity, so we impose f′=1f' = 1 at η∞\eta_\infty. That is a modelling error; to size it, move η∞\eta_\infty:

η∞\eta_\infty681012
f′′(0)f''(0)0.3325660.3325660.3320590.3320590.33205730.33205730.33205730.3320573

From η∞=10\eta_\infty = 10 on, seven digits stop moving. Why so fast, and why is shooting safe here when it failed on y′′=100yy'' = 100y in 8.1? Write the equation as (f′′)′=−12f f′′(f'')' = -\tfrac12 f\,f'', which integrates to

f′′(η)=f′′(0)exp⁡(−12∫0ηf dη′).f''(\eta) = f''(0)\exp\Big(-\tfrac12\int_0^\eta f\,d\eta'\Big) .

Far out f≈η−1.7208f \approx \eta - 1.7208, so f′′f'' dies like a Gaussian, roughly e−η2/4e^{-\eta^2/4}. No growing mode can swamp the answer, and by η=10\eta = 10 the neglected tail is invisible in the 7th digit.

◆ Interactive Shooting for f''(0): drag the guess, watch f' overshoot or undershoot 1 at the domain end, run the secant method, and move the end

Loading simulation…

The profile’s numbers (lab block A; the plate panel prints the first two):

  • 99% thickness: f′=0.99f' = 0.99 at η99=4.910\eta_{99} = 4.910, so δ99=4.910 x/Rex\delta_{99} = 4.910\,x/\sqrt{\mathrm{Re}_x}, with Rex=Ux/ν\mathrm{Re}_x = Ux/\nu. The textbook "δ≈5x/Rex\delta \approx 5x/\sqrt{\mathrm{Re}_x}".
  • Displacement thickness: δ∗=1.7208 x/Rex\delta^* = 1.7208\,x/\sqrt{\mathrm{Re}_x} (it is η∞−f(η∞)\eta_\infty - f(\eta_\infty)): the slowed air pushes the outer flow away as if the body were thicker by δ∗\delta^*.
  • Momentum thickness: θ=0.6641 x/Rex\theta = 0.6641\,x/\sqrt{\mathrm{Re}_x}, the lost momentum as a thickness. Its coefficient equals 2f′′(0)2f''(0): next section.

Skin friction and the drag of a plate

The wall shear is τw=μuy∣y=0=μUf′′(0)U/(νx)\tau_w = \mu u_y|_{y=0} = \mu Uf''(0)\sqrt{U/(\nu x)}. Divide by 12ρU2\tfrac12\rho U^2, using μ=ρν\mu = \rho\nu:

Cf=τw12ρU2=2f′′(0)Rex=0.664Rex.C_f = \frac{\tau_w}{\tfrac12\rho U^2} = \frac{2f''(0)}{\sqrt{\mathrm{Re}_x}} = \frac{0.664}{\sqrt{\mathrm{Re}_x}} .

Friction is largest at the leading edge, where the layer is thinnest. Per metre of span, over one face of length LL:

Done face′=∫0Lτw dx=ρU2f′′(0)νU⋅2L=2f′′(0) ρU2LReL.D'_{\text{one face}} = \int_0^L \tau_w\,dx = \rho U^2 f''(0)\sqrt{\frac{\nu}{U}}\cdot2\sqrt L = \frac{2f''(0)\,\rho U^2L}{\sqrt{\mathrm{Re}_L}} .

Where d’Alembert’s drag went. A box around the plate loses momentum flux ρU2θ\rho U^2\theta per face through the trailing-edge station (von Kármán’s momentum balance, stated), and with the lab’s θ=2f′′(0) L/ReL\theta = 2f''(0)\,L/\sqrt{\mathrm{Re}_L} that is exactly Done face′D'_{\text{one face}}. The missing drag is a momentum deficit in a millimetre-thin sheet, carried off in the wake.

Worked example: the light wing’s laminar layer

A 1.51.5 m chord at 5050 m/s in sea-level air (ν=1.4607×10−5\nu = 1.4607\times10^{-5} m2^2/s, ρ=1.225\rho = 1.225 kg/m3^3, as in 11.5). Model each face as a flat plate (a thin wing at small incidence nearly is).

  1. ReL=50⋅1.5/1.4607×10−5=5.135×106\mathrm{Re}_L = 50\cdot1.5/1.4607\times10^{-5} = 5.135\times10^6, so ReL=2266\sqrt{\mathrm{Re}_L} = 2266.
  2. If the layer stayed laminar to the trailing edge: δ99=4.910⋅1.5/2266=3.250\delta_{99} = 4.910\cdot1.5/2266 = 3.250 mm and δ∗=1.139\delta^* = 1.139 mm.
  3. Wall shear at the trailing edge: τw=1.225⋅502⋅0.33206/2266=0.4488\tau_w = 1.225\cdot50^2\cdot0.33206/2266 = 0.4488 Pa, Cf=2.931×10−4C_f = 2.931\times10^{-4}.
  4. Friction drag, two faces: 2⋅2⋅0.33206⋅1.225⋅2500⋅1.5/2266=2.6932\cdot2\cdot0.33206\cdot1.225\cdot2500\cdot1.5/2266 = 2.693 N/m. Against 11.7’s lift of 1148.41148.4 N/m, that is 0.23%0.23\%.

Does it stay laminar? No. A smooth plate in a quiet stream starts transition near Rex=5×105\mathrm{Re}_x = 5\times10^5 (from memory; roughness, noise and pressure gradients move it), reached at x=5×105ν/U=0.1461x = 5\times10^5\nu/U = 0.1461 m, under a tenth of the chord, where δ99=1.014\delta_{99} = 1.014 mm. Behind it the layer turns turbulent, so 2.6932.693 N/m is a lower bound, not the wing’s friction drag (11.9).

◆ Interactive A Blasius layer along a plate: 99% and displacement thickness, profiles at five stations, transition marked at Re_x = 5e5; slide the speed and the chord

Loading simulation…

Slide UU down to 55 m/s: ReL\mathrm{Re}_L drops tenfold, the layer is 10\sqrt{10} times thicker, and the transition line moves back to 1.4611.461 m, nearly the trailing edge. Below 4.94.9 m/s it leaves the plate.

Pressure gradients and separation

On a wing’s upper surface the air speeds up over the nose (Ue′>0U_e' \gt 0, pressure falling: favourable), then slows behind the suction peak toward the trailing edge (Ue′<0U_e' \lt 0, pressure rising: adverse). 11.7’s cylinder is the extreme: CpC_p from −3-3 on the shoulder back to +1+1.

The mechanism. Outside the layer, air climbs the pressure rise by trading kinetic energy for pressure. Inside, the same rise acts (pressure does not change across the layer), but near-wall air has already lost most of its kinetic energy to friction: it slows first, stops, and reverses. At the wall, where u=v=0u = v = 0, the boundary-layer equation reads

0=−pe′ρ+νuyy∣y=0⟹μ uyy∣y=0=pe′.0 = -\frac{p_e'}{\rho} + \nu u_{yy}\big|_{y=0} \quad\Longrightarrow\quad \mu\,u_{yy}\big|_{y=0} = p_e' .

An adverse gradient makes the profile curve the “wrong” way at the wall; since it must curve the usual way far out, it has an inflection point. As the gradient persists the wall slope is pushed to zero: zero wall shear, the separation point. Behind it near-wall air flows backwards, the layer lifts off, and the outer flow no longer follows the body.

Falkner–Skan: a family with a dial

To put numbers on this, take edge speeds Ue=KxmU_e = Kx^m. 11.7’s complex potential produces them: w=Kzm+1/(m+1)w = Kz^{m+1}/(m+1) has dw/dz=Kzmdw/dz = Kz^m, so the speed along the real axis is KxmKx^m, the flow filling a corner of angle 180∘/(m+1)180^\circ/(m+1) (ψ=0\psi = 0 on both walls). With η=yUe/(νx)\eta = y\sqrt{U_e/(\nu x)}, the same substitution as Blasius (Problem 2) gives

f′′′+m+12 ff′′+m(1−f′2)=0,f(0)=f′(0)=0,f′(∞)=1.f''' + \frac{m+1}{2}\,ff'' + m\big(1 - f'^2\big) = 0, \qquad f(0) = f'(0) = 0, \quad f'(\infty) = 1 .

m=0m = 0 is Blasius, m=1m = 1 a stagnation point, m<0m \lt 0 adverse. Tables quote Hartree’s β=2m/(m+1)\beta = 2m/(m+1) (label from memory). At the wall, f′′′(0)=−mf'''(0) = -m: the curvature result above.

For each mm the sim bisects on s=f′′(0)s = f''(0) in [0,2][0, 2] (F(s)F(s) increases with ss). It prints f′′(0)=1.2326f''(0) = 1.2326 at m=1m = 1, 0.332060.33206 at 00 and 0.213480.21348 at m=−0.05m = -0.05, where the profile has an inflection at η=1.65\eta = 1.65. Separation is where the wall shear reaches zero, so set s=0s = 0 and shoot on mm instead, solving f′(η∞;m,s=0)=1f'(\eta_\infty; m, s = 0) = 1. Lab block C (secant) and the sim (bisection) find

msep=−0.090429,βsep=−0.19884,m_{\text{sep}} = -0.090429, \qquad \beta_{\text{sep}} = -0.19884 ,

with β\beta the same to five digits at η∞=10\eta_\infty = 10 and 1212 (mm moves in its sixth decimal). That matches the tabulated β=−0.1988\beta = -0.1988 (from memory). Below msepm_{\text{sep}} even s=0s = 0 overshoots: no attached similar solution exists.

How gentle is that? The edge speed drops only 6.08%6.08\% each time the distance doubles (2−0.0904=0.93922^{-0.0904} = 0.9392): a corner of 197.9∘197.9^\circ, a wall bending away by under 18∘18^\circ. A laminar layer tolerates almost no deceleration.

◆ Interactive Falkner–Skan: slide the pressure-gradient exponent m from accelerating to adverse flow; the profile grows an inflection and the wall shear falls to zero at m = -0.0904

Loading simulation…

Honest limits. Falkner–Skan layers are self-similar. On a real wing the gradient changes along the chord and separation depends on the layer’s history, found by marching (or integral methods such as Thwaites’, from memory). Falkner–Skan gives the picture and the order of magnitude, not a wing’s separation point.

Stall, and why golf balls have dimples

Stall. Raise a wing’s angle of attack and the suction peak near the nose deepens, so the climb back to trailing-edge pressure steepens. Past some angle the upper-surface layer separates. The Kutta condition that fixes the circulation (11.7, 12.1) assumes the flow leaves smoothly at the trailing edge; once it leaves earlier, circulation and lift fall and drag jumps. 12.1 sets ideal lift lines beside real ones, which bend over at stall. Given Falkner–Skan’s tiny tolerance, a laminar layer would separate soon behind the suction peak; the turbulent layer after transition holds on much longer (11.9), which is why wings work at useful angles.

Golf balls (qualitative, from memory). A smooth sphere at the golf ball’s Re=2.05×105\mathrm{Re} = 2.05\times10^5 (11.5) keeps a laminar layer that separates near its shoulder, leaving a wide low-pressure wake. Dimples trip the layer turbulent. Turbulent mixing brings fast outer air down to the wall, so near-wall air carries more momentum and climbs further up the adverse gradient. Separation moves aft, the wake narrows, and pressure drag falls, by about half (from memory), though skin friction rises: 11.5’s drag crisis, pulled down to golf-ball speeds.

Computing it

🐍 PythonLab: Blasius by secant shooting and the domain end (A), the light wing's laminar layer (B), Falkner–Skan separation (C)
⌘/Ctrl + Enter

Block A also checks the rescaling trick of Problem 3: one shot at s=1s = 1 gives f′(10)=2.085409f'(10) = 2.085409, and 2.085409−3/2=0.3320573362.085409^{-3/2} = 0.332057336, the secant answer, with no iteration at all.

🧪 Coding exerciseExercise: shoot the Blasius equation for f''(0), then price a laminar wing's friction
⌘/Ctrl + Enter

Graded: the residual’s sign on both sides of the root, f′′(0)f''(0) to 10−610^{-6}, the truncated η∞=6\eta_\infty = 6 answer, and the wing’s 2.6932.693 N/m.

Concept checks

✅ Concept check

A 1.5 m chord flies at 50 m/s in sea-level air ($\mathrm{Re}_L = 5.135\times10^6$). If its boundary layer stayed laminar, how thick would it be at the trailing edge, in mm ($\delta_{99} = 4.910\,x/\sqrt{\mathrm{Re}_x}$)?

✅ Concept check

At a fixed station $x$ on a plate, the free-stream speed is quadrupled. The laminar layer's thickness there:

✅ Concept check

Why does an adverse pressure gradient separate the boundary layer, while the outer flow climbs the same pressure rise without trouble?

✅ Concept check

Shooting for Blasius with the far condition imposed at $\eta_\infty = 6$ gives $f''(0) = 0.332566$; at $\eta_\infty = 10$ it gives $0.332057$. What is the right reading?

Problems

✏️ Pen & paperFriction drag from the wall shear●○○

Integrate τw=ρU2f′′(0)/Rex\tau_w = \rho U^2 f''(0)/\sqrt{\mathrm{Re}_x} from 00 to LL to get the drag per unit span on one face. Show that CD=D′/(12ρU2L)=1.328/ReLC_D = D'/(\tfrac12\rho U^2L) = 1.328/\sqrt{\mathrm{Re}_L}, and evaluate the two-face drag of a 1.51.5 m plate at 5050 m/s in sea-level air.

Show worked solution

τw=ρU2f′′(0)ν/U x−1/2\tau_w = \rho U^2 f''(0)\sqrt{\nu/U}\,x^{-1/2}, and ∫0Lx−1/2dx=2L\int_0^L x^{-1/2}dx = 2\sqrt L, so D′=2f′′(0)ρU2νL/U=2f′′(0)ρU2L/ReLD' = 2f''(0)\rho U^2\sqrt{\nu L/U} = 2f''(0)\rho U^2L/\sqrt{\mathrm{Re}_L}. Dividing by 12ρU2L\tfrac12\rho U^2L: CD=4f′′(0)/ReL=1.328/ReLC_D = 4f''(0)/\sqrt{\mathrm{Re}_L} = 1.328/\sqrt{\mathrm{Re}_L}, twice the local CfC_f at the trailing edge (the front, with its thin layer, rubs harder). With ReL=2266\sqrt{\mathrm{Re}_L} = 2266: one face 1.3461.346 N/m, two faces 2.6932.693 N/m, as in the lab. The singular τw\tau_w at x=0x = 0 is integrable, and real leading edges are rounded anyway.

✏️ Pen & paperFalkner–Skan from the boundary-layer equation●●●

With Ue=KxmU_e = Kx^m, η=yUe/(νx)\eta = y\sqrt{U_e/(\nu x)} and ψ=νUex f(η)\psi = \sqrt{\nu U_ex}\,f(\eta), show that u=Uef′u = U_ef' and that the boundary-layer equation becomes f′′′+m+12ff′′+m(1−f′2)=0f''' + \frac{m+1}{2}ff'' + m(1 - f'^2) = 0. Then show f′′′(0)=−mf'''(0) = -m, and explain what its sign means for m<0m \lt 0.

Show worked solution

Ue/(νx)∝x(m−1)/2\sqrt{U_e/(\nu x)} \propto x^{(m-1)/2} and νUex∝x(m+1)/2\sqrt{\nu U_ex} \propto x^{(m+1)/2}, so u=ψy=νUexUe/(νx) f′=Uef′u = \psi_y = \sqrt{\nu U_ex}\sqrt{U_e/(\nu x)}\,f' = U_ef'. With ηx=m−12ηx\eta_x = \frac{m-1}{2}\frac{\eta}{x}: ux=Uex(mf′+m−12ηf′′)u_x = \frac{U_e}{x}\big(mf' + \frac{m-1}{2}\eta f''\big), and v=−ψx=−νUe/x(m+12f+m−12ηf′)v = -\psi_x = -\sqrt{\nu U_e/x}\big(\frac{m+1}{2}f + \frac{m-1}{2}\eta f'\big), uy=Uef′′Ue/(νx)u_y = U_ef''\sqrt{U_e/(\nu x)}, νuyy=Ue2xf′′′\nu u_{yy} = \frac{U_e^2}{x}f''', UeUe′=mUe2xU_eU_e' = \frac{mU_e^2}{x}. In uux+vuyuu_x + vu_y the ηf′f′′\eta f'f'' terms cancel, leaving Ue2x(mf′2−m+12ff′′)\frac{U_e^2}{x}\big(mf'^2 - \frac{m+1}{2}ff''\big). Setting it equal to Ue2x(m+f′′′)\frac{U_e^2}{x}(m + f''') and dividing gives the equation. At η=0\eta = 0, f=f′=0f = f' = 0, so f′′′(0)+m=0f'''(0) + m = 0. For m<0m \lt 0, f′′′(0)>0f'''(0) \gt 0: the profile curves upward at the wall but must curve downward as it approaches 11, so it has an inflection point (the sim marks it, at η=1.65\eta = 1.65 for m=−0.05m = -0.05).

✏️ Pen & paperShooting without iterating●●●

Show that if ff solves f′′′+12ff′′=0f''' + \tfrac12 ff'' = 0, so does F(η)=af(aη)F(\eta) = a f(a\eta) for any a>0a \gt 0. Use this to get f′′(0)f''(0) from a single shot with f′′(0)=1f''(0) = 1 that ends at f′(∞)=λf'(\infty) = \lambda.

Show worked solution

F′=a2f′(aη)F' = a^2f'(a\eta), F′′=a3f′′(aη)F'' = a^3f''(a\eta), F′′′=a4f′′′(aη)F''' = a^4f'''(a\eta), and FF′′=a4ff′′FF'' = a^4ff''. So F′′′+12FF′′=a4(f′′′+12ff′′)=0F''' + \tfrac12FF'' = a^4(f''' + \tfrac12ff'') = 0, and F(0)=F′(0)=0F(0) = F'(0) = 0 still. If gg has g′′(0)=1g''(0) = 1 and g′(∞)=λg'(\infty) = \lambda, then G=ag(aη)G = ag(a\eta) has G′(∞)=a2λG'(\infty) = a^2\lambda; choose a=λ−1/2a = \lambda^{-1/2} to make it 11. Then G′′(0)=a3=λ−3/2G''(0) = a^3 = \lambda^{-3/2}. Lab block A: λ=2.085409\lambda = 2.085409, so f′′(0)=0.332057336f''(0) = 0.332057336. (The trick is usually credited to Töpfer, from memory; it works because Blasius has no length scale, and it fails for Falkner–Skan with m≠0m \ne 0, whose m(1−f′2)m(1 - f'^2) term does not scale the same way.)

Key Takeaways

  • Prandtl: at high Re, viscosity acts only in a thin wall layer. Potential flow outside sets Ue(x)U_e(x) and pe(x)p_e(x); inside, inertia balances νuyy\nu u_{yy}, so δ/L∼Re−1/2\delta/L \sim \mathrm{Re}^{-1/2}, and pressure is constant across the layer.
  • xx acts like time: a code can march the layer downstream until separation.
  • A flat plate’s layer is one shape stretched by νx/U\sqrt{\nu x/U}: Blasius, f′′′+12ff′′=0f''' + \tfrac12ff'' = 0. Shooting on f′′(0)f''(0) gives 0.3320570.332057; test the cut-off at η∞\eta_\infty (66 gives 0.3325660.332566, 1010 and 1212 agree to seven digits).
  • δ99=4.910 x/Rex\delta_{99} = 4.910\,x/\sqrt{\mathrm{Re}_x}, δ∗=1.7208 x/Rex\delta^* = 1.7208\,x/\sqrt{\mathrm{Re}_x}, Cf=0.664/RexC_f = 0.664/\sqrt{\mathrm{Re}_x}; the drag is the layer’s momentum deficit.
  • The light wing at 5050 m/s: a laminar layer would be 3.2503.250 mm thick at the trailing edge and rub 2.6932.693 N/m off both faces, but Rex\mathrm{Re}_x reaches 5×1055\times10^5 at 0.1460.146 m.
  • An adverse gradient stops the slow near-wall air first: μuyy=pe′\mu u_{yy} = p_e' at the wall makes an inflection, then zero wall shear. Falkner–Skan separates at m=−0.0904m = -0.0904 (β=−0.1988\beta = -0.1988): a 6%6\% slowdown per doubling of distance.
  • Separation causes bluff-body pressure drag and wing stall; dimples trip a golf ball’s layer turbulent so it separates later.
  • Next, 11.9: past Rex≈5×105\mathrm{Re}_x \approx 5\times10^5 (from memory) the laminar layer we just solved breaks down and the boundary layer becomes turbulent, a chaotic layer that mixes fast air down to the wall, rubs harder and holds on longer. Describing it takes averages, a closure problem and an energy cascade: turbulence.
Finished this lesson?

Mark it complete to track your progress (saved in this browser). Reading every section does this for you.