E. coli navigates chemical gradients by alternating between smooth runs and sudden tumbles — random reorientations that look like failures of coordination. The tumble appears wasteful: the bacterium loses its heading entirely and must start over in a random direction. Smooth steering, gradually adjusting course toward higher concentration, seems like the obvious improvement. Evolution had billions of years to build a smoother navigator and didn't.
Betancourt, Leighton, Emonet, Machta, and Abbott (2026) explain why. They frame navigation as an optimization problem: maximize the speed of climbing a sensory gradient given a fixed rate of information acquisition from the environment. The result is that sudden turns outperform gradual steering when directional information is scarce. With limited sensory bandwidth, the organism cannot measure which direction to steer — it can only measure whether conditions are getting better or worse. Under that constraint, a sharp reorientation after a “worse” signal extracts more value per bit of information than a gradual course correction.
The analysis goes further. As available information increases, the optimal strategy transitions: from simple reversals to full-angle tumbles to discrete multi-angle reorientation. The number of discrete turn angles is itself optimal — not two, not continuous, but a specific finite number determined by the information rate. The jerky, quantized navigation of real bacteria sits exactly where the theory predicts for their sensory bandwidth.
The general principle: what looks like a crude approximation of smooth behavior can instead be the exact solution to a harder problem than the one assumed. Smooth steering solves the problem of “navigate when you know where to go.” Tumbling solves the problem of “navigate when you barely know anything.” The second problem is harder, its solution looks worse, and it is the one the organism actually faces.