50 years of studies couldn't explain why widening a pipe past a small angle made it work worse.
Between 1900 and about 1950, engineers measured it again and again. A diffuser is just a passage that spreads out, slowing the flow and raising the pressure. Open the angle a little, and the pressure rise grows. Open it more, and it drops.
Everyone had the numbers. Nobody had the picture.
So a team stopped calculating and started looking.
The tool was a wire 2 mils thick, stretched across a water channel. Run a DC current through it and electrolysis strips hydrogen out of the water. Tiny bubbles peel off and ride the flow. Small enough to follow it almost exactly.
Insulate parts of the wire, and the bubbles come off in bands. Pulse it, and they come off as squares.
Those squares turned out to be the key.
In a flat, 2-dimensional flow of water, each square keeps the same area forever. So when a square stretches long, the water there is fast. When it squeezes short and wide, the water is slow.
The flow draws its own speed map.
Put an airfoil in the channel at 0 lift, and a line of bubbles splits around it and meets again at the back. Same speed above and below.
Tilt it for lift, and the top half of the line pulls ahead of the bottom half.
Then the film shows why this work is harder than it looks.
There are 4 different lines you can draw in a flow. The path of 1 drop. The trail of every drop that passed 1 point. A line of drops marked at the same instant. And the streamline, the line the math uses, which no dye or bubble can ever show directly.
In steady flow, 3 of them are the same line. You see 1, you know the other.
Swap the steady flow for a plate swinging back and forth, and they split apart.
Two drops enter at the exact same point, a moment apart. Their paths cross almost at right angles. One goes around the top of the plate. The other goes around the bottom.
None of the bubble trails match either path. Neither matches the streamline.
The water looks like it's telling you something. It's telling you something else.
For 50 years, the diffuser problem lived in exactly that gap.
The math came later. First, someone had to see the water.