The physics behind
the tunnel.

Everything AEROLAB shows on screen comes from a handful of physical ideas. Each one below has the equation, the plain-words version, where to point at it in the app, and the one sentence worth saying out loud.

  1. 01 Drag & Cd
  2. 02 Pressure & the wake
  3. 03 Lift
  4. 04 Reynolds number
  5. 05 Vorticity
  6. 06 The solver
  7. 07 Drag → fuel → CO₂
01
Forces

Drag and the drag coefficient

FD = ½ ρ v2 Cd A
ρ
air density, 1.2 kg/m³ at sea level
v
speed of the air relative to the body
A
frontal area the air has to get around
Cd
the shape's number: how badly it disturbs the air

Drag is the force air pushes back with. Speed, size and density are not the shape's fault, so the coefficient divides them out. What remains is a pure measure of shape, and two objects with the same Cd fly equally cleanly no matter how big they are.

In AEROLAB The large lime number is Cd. The Drag force tile is this formula evaluated with the tunnel's speed and the body's height as the reference length. Mission 01 asks you to cut it by 35%.

“Cd is the only number that is purely about the shape. That is why we rank designs by it, not by raw force.”
Flat plate · Cd ≈ 1.3 – 2.0 Cylinder · ≈ 1.1 Modern car · ≈ 0.30 Streamlined body · ≈ 0.05
Textbook ranges at road and flight speeds. The tunnel runs at low Reynolds number, so its values sit higher, but the order is the same.
02
Where drag comes from

Pressure drag and the wake

F = ∮ −p n ds

Air slams into the front of a body and piles up at high pressure. Behind a blunt body it cannot follow the sharp turn around the back edge, so the flow separates and leaves a churning low-pressure region: the wake. High pressure in front, low pressure behind, net force backward. For cars, cylinders and plates this pressure imbalance is most of the drag.

A tapered tail changes the story. The air can stay attached longer, pressure climbs back up as the body narrows, and that recovered pressure pushes the body forward, cancelling some of the front load. That is why a teardrop has a tiny fraction of a cylinder's drag with the same frontal area.

In AEROLAB The dark, slow region behind a shape is the wake. The force readout literally integrates pressure around the body's outline every 50 ms, so when you draw a tail you watch the number drop.

“Cars have tails for pressure recovery. The wake behind a blunt body is a low-pressure hole that the body is dragging around.”
wide wake · low pressure pressure recovers along the tail
Same frontal height, very different wake. Mission 01 is this picture, drawn by hand.
03
Forces

Lift, and how to explain it without the myth

FL = ½ ρ v2 Cl A

Lift is the part of the air's push that is perpendicular to the flow. A tilted or curved surface turns the oncoming air downward. By Newton's third law the air pushes the body upward. Equivalently, the turning shows up as lower pressure on top and higher pressure underneath, and integrating that pressure difference gives the same force. Both descriptions are the same physics.

The ratio L/D is what designers chase: how much lift you get for each unit of drag. A good wing exceeds 15. A flat plate at a steep angle gets lift, but pays for it with a huge wake.

In AEROLAB Cl is the second big number, and the Lift / drag tile is L/D. Mission 02 starts from a flat plate and asks for Cl above 0.3. Tilting the plate shows the downward deflection in the tracers immediately.

“Lift is deflected air. The wing pushes air down, so the air pushes the wing up. The ‘longer path on top so air must go faster’ story is a known misconception, and our tracers show it is wrong.”
lift drag air arrives level… …and leaves angled down
The whole wake is bent downward. That bend is the lift.
04
Regimes

Reynolds number and the vortex street

Re = v Lν
L
a size of the body, here its height
ν
kinematic viscosity: air 1.5 × 10⁻⁵ m²/s, honey about 10⁻²

Re compares momentum, which wants to keep the flow going, with viscosity, which wants to smooth it out. Low Re is thick, orderly, laminar flow: honey around a spoon. High Re is thin, twitchy, turbulent flow: wind around a building. It is the single number that says whether two flows will look alike, which is why a small model in a tunnel can stand in for a full-size car.

Behind a cylinder, somewhere above Re ≈ 50, the wake stops being symmetric and starts peeling off vortices alternately from each side: the Kármán vortex street. Each shed vortex tugs on the body, so drag and lift oscillate. The shedding frequency is set by the Strouhal number, f ≈ 0.2 v/D.

In AEROLAB The regime chip reads LAMINAR, TRANSITIONAL or TURBULENT from the tunnel's Re. Mission 04 raises viscosity until the street vanishes. The cylinder's Cd wobbling between about 0.8 and 1.1 is shedding, which is why missions wait for a settled baseline before scoring.

“That wobble in the drag number isn't noise. It's vortex shedding, the same thing that makes power lines hum.”
alternating sign · f ≈ 0.2 v / D Re ≈ 100 – 1000
Lime vortices spin one way, mint the other. Toggle Vorticity in the app to see exactly this.
05
Fields

Vorticity

ω = ∇ × u = ∂v∂x − ∂u∂y

Vorticity is the local spin of the fluid: drop a tiny paddle wheel in and vorticity is how fast it turns. In two dimensions it is a single signed number per point. Vorticity is born in the thin boundary layer against a wall, where the no-slip condition forces velocity to zero, and it is carried downstream into the wake. Shed vortices alternate in sign because the top and bottom boundary layers spin opposite ways.

In AEROLAB The Vorticity toggle colours the field by sign. A smooth laminar wake shows two thin ribbons; a shedding wake shows a chain of alternating blobs. The 2D solver computes it from the velocity texture in one fragment shader pass.

“Vorticity view is the diagnostic. If the wake is two quiet ribbons, the body is clean. If it's a chain of blobs, it's shedding.”
top boundary layer · ω > 0 bottom boundary layer · ω < 0
No-slip walls make vorticity. The wake is where the two layers meet.
06
Numerics

What the solver does every frame

∂u∂t + (u·∇)u = −∇pρ + ν∇²u
∇·u = 0

These are the incompressible Navier–Stokes equations: momentum is carried along by the flow, pushed by pressure gradients and smeared by viscosity, and the fluid never compresses. There is no general closed-form solution, so the tunnel marches them forward in small time steps using a fractional-step (projection) method on the GPU:

  1. Advect. For each cell, trace backwards along the velocity and pull the value from there (semi-Lagrangian). Stable at any speed.
  2. Diffuse. Apply viscosity, the ν∇²u term. Higher ν, calmer flow.
  3. Pressure. The result is not divergence-free yet. Solve the Poisson equation ∇²p = ∇·u* with Jacobi iterations in ping-pong framebuffers.
  4. Project. Subtract ∇p. Now ∇·u = 0 and the field is physical again.

Solid cells get a no-slip wall. Pressure is read back asynchronously and integrated around the body outline to give force, then Cd and Cl. The 3D mode swaps in a lattice-Boltzmann method (D3Q19): instead of solving the equations directly it streams particle populations along 19 directions and relaxes them toward equilibrium, with bounce-back at walls. Forces come from the momentum the walls exchange with those populations.

In AEROLAB The Numerics panel lists each pass, the grid size and the iteration count. The tracers ride the projected velocity, so what you see is the same field the force integral feels.

“It's a real Navier–Stokes solver, not an animation. Advect, pressure solve, project, every frame, on the GPU. The tracers and the force come from the same field.”
1 · advect 2 · diffuse 3 · pressure 4 · project ∮ −p n ds → Cd, Cl 60 FPS solve · 20 Hz force readback
One frame of the 2D solver. Ping-pong float framebuffers hold u and p.
07
Sustainability

Why drag costs fuel, and why speed costs more

P = FD v = ½ ρ v3 Cd A

Power is force times speed, and drag force already grows with v², so the power spent pushing air grows with v³. Drive 10% faster and you burn about 33% more energy on drag. At highway speed roughly half of a car's engine output goes into the air. That is why aerodynamics is a climate lever and not just a racing one.

Energy over a trip is power times time. Petrol holds about 34.2 MJ per litre, and an engine turns roughly a quarter of it into motion, so every litre buys about 8.5 MJ at the wheels and releases 2.31 kg of CO₂.

Worked example · 100 km at 100 km/h
Cd 0.30, A 2.2 m², v 27.8 m/sP ≈ 8.5 kW
One hour of that≈ 30.6 MJ
÷ 8.5 MJ useful per litre≈ 3.6 L on drag alone
× 2.31 kg CO₂ per litre≈ 8.3 kg CO₂

In AEROLAB The road panel runs exactly this chain for the shape on screen, using its Cd relative to the saved baseline. Mission 03 turns a 20% drag cut on the car into litres and kilograms.

“Drag power goes as speed cubed. Our road panel is drag share only, for comparing shapes, with every assumption listed in the Numerics panel.”
50 80 100 130 km/h → drag power ∝ v³
From 100 to 130 km/h the drag power more than doubles.

Numbers to know

1.2 kg/m³density of air at sea level
1.5 × 10⁻⁵ m²/skinematic viscosity of air
≈ 1.1Cd of a cylinder
0.25 – 0.30Cd of a modern car
≈ 0.05Cd of a streamlined body
≈ 0.2Strouhal number for a cylinder
v³how drag power scales with speed
~50 %engine power spent on drag at highway speed
34.2 MJ/Lenergy in petrol · ~25 % reaches the wheels
2.31 kgCO₂ per litre of petrol burned
> 15lift-to-drag of a decent wing
180 / 1200the tunnel's laminar and turbulent Re thresholds

Questions you will get asked

Is this a real simulation or an animation?

A real one. It integrates the incompressible Navier–Stokes equations with a fractional-step method on the GPU every frame, and the force is the pressure integral around the body. The tracers ride the same velocity field the force is measured from. The Numerics panel lists every pass.

How accurate are the drag numbers?

Good enough to rank shapes, not to certify them. It is 2D, at a lower Reynolds number than a real car, on a grid coarse enough to run at 60 FPS. The ordering and the relative changes match textbooks: a tail cuts drag, a tilted plate makes lift, viscosity kills the vortex street. That is why missions use relative targets.

Why does the number keep moving on the cylinder?

Vortex shedding. Above a modest Reynolds number the wake sheds alternating vortices, and each one tugs on the body. Real cylinders do this too. The missions wait until the last dozen readings agree within 8% before they score a run.

Why 2D? Isn't air 3D?

2D keeps the solve under 16 ms on a Chromebook, which is what makes it usable in a classroom. For long bodies, like a wing section or a car silhouette, 2D captures the mechanism students need to see. The 3D lattice-Boltzmann mode is there for when the shape genuinely needs it, at a lower resolution.

Where do the fuel and CO₂ numbers come from?

Drag power is ½ρv³CdA. We assume 100 km/h, a 2.2 m² frontal area, 34.2 MJ per litre of petrol, 25% engine efficiency and 2.31 kg CO₂ per litre. It is the drag share only, meant for comparing two shapes, and every constant is shown in the app.

How is lift explained to students?

As deflected air. The body turns the flow downward and the air pushes back upward. We deliberately avoid the “equal transit time” story because it is wrong, and the tracers make that visible: air over the top does not meet up with air underneath.

What does the camera actually do?

Colour-keys or hand-detects the object, rasterises its silhouette into the solver mask, and tests it like any drawn shape. Frames stay on the device. Nothing is uploaded.

Why does this matter beyond a classroom?

Half the energy a car spends at highway speed goes into air. Letting a fourteen-year-old feel that in litres, on a device the school already owns, is how you get people who think about efficiency before they think about horsepower.

Say this, not that

  • Air travels farther over the top so it speeds up and lifts the wing.The wing deflects air downward, so the air pushes the wing up.
  • Drag is the air's friction.For blunt shapes drag is mostly pressure: high in front, low in the wake. Skin friction is the small part.
  • The simulation shows exactly what a car would do.The simulation shows the mechanism and ranks shapes. It is 2D at low Reynolds number, and we say so.

What the model leaves out

  • Two-dimensional. Infinite-span bodies; no tip vortices, no 3D relief.
  • Low Reynolds number. Hundreds to low thousands, versus millions for a real car. Coefficients sit higher than road values.
  • Coarse grid. Chosen so a Chromebook holds 60 FPS. Thin boundary layers are under-resolved.
  • Estimates, clearly labelled. Fuel and CO₂ are the drag share under stated assumptions, meant for comparison.

How accurate is it

Measured in the tunnel at the default wind speed on 14 Nov 2026, averaged over six seconds once the wake had settled. Reference values are textbook ranges for the same Reynolds number.

ShapeReAEROLAB CdReferenceResult
Cylinder6921.050.9 – 1.3In range
Cylinder1800.601.2 – 1.5Reads 56% low
Cylinder900.481.3 – 1.6Reads 67% low
Flat plate6921.291.8 – 2.1Reads 34% low
Airfoil, 4.5°6920.060.01 – 0.1In range

Trust it for comparing shapes, not for absolute numbers. The low readings at low Reynolds number come from the coarse grid under-resolving the thick boundary layer, and the plate reads low because its wake is wider than the channel allows. Both move every shape the same way, so rankings hold.

A sixty-second walkthrough

  1. 0:00“A real wind tunnel costs more than a classroom. This one runs in a browser tab on the Chromebooks a school already has.”
  2. 0:10Draw a cylinder. “That number is the drag coefficient. It's wobbling because the wake is shedding vortices, the same physics that makes power lines hum.”
  3. 0:25Draw a tail. “Watch it drop. The air stays attached and pressure recovers along the tail. That is why cars have tails.”
  4. 0:35Open the road panel. “Drag power scales with speed cubed. This shape just saved 3 litres per 100 km and 8 kilograms of CO₂, in drag alone.”
  5. 0:45Hold a hand up to the camera. “Anything you can hold up, you can test.”
  6. 0:52“It's a real Navier–Stokes solver on the GPU, every assumption is on screen, and a class can race designs on a shared board. That's AEROLAB.”