Nowhere for it to accumulate
Shrink a swimmer far enough and water stops behaving like water. Purcell put a number on it in 1977. Take something a micron across, moving at the thirty microns a second that bacteria actually manage, give it a shove and then let go: it coasts about a tenth of an angstrom, which is a tenth of the width of a hydrogen atom, and it takes six tenths of a microsecond to do it. Nothing about that is coasting. It is stopping.
So there is no gliding, and there is also no memory. What the animal is doing at any instant depends on the forces on it at that instant and on nothing at all that came before. Purcell’s consequence: flap something open and then shut it again, and you retrace your own path exactly and end up where you began. It does not help to close fast and open slowly. Speed cannot enter, because nothing at that scale keeps a record of speed.
His example was the scallop, which opens its shell slowly and slams it shut, squirting water. At our scale that works. At a micron it is worthless, and his explanation of why is the sentence this whole piece is about:
It can’t swim because it only has one hinge, and if you have only one degree of freedom in configuration space, you are bound to make a reciprocal motion. There is nothing else you can do.
Read that twice. The reason is not that the scallop chose a bad stroke. The reason is that its space of possible shapes is a line, and on a line the only way back is the way you came.
People have been looking for holes in this for fifty years, mostly by cheating on the assumptions: make the fluid non-Newtonian, put a wall nearby, let one of the oars bend. A paper posted to arXiv four days ago tries a different door. At a micron the hinge is Brownian, so its angle jitters, and the animal is a cloud of possible shapes instead of one shape. Drive that cloud with two knobs instead of one, a rest angle and a stiffness, take the knobs around a closed loop, and you should pick up a geometric phase the way a Foucault pendulum picks one up from the turning Earth. The claim is a small steady drift, powered by nothing but noise and a good protocol.
I spent this evening checking it, and it is zero. Not small. Zero.
Here is the part I think is prettier than the claim it kills. For a swimmer like this, where you are is not a history. It is an address.
Write down the swimmer’s speed at any moment and you find it is fixed by how the cloud of hinge angles is drifting at that moment, weighted by how much thrust each angle is worth. Integrate that over time and the whole thing collapses: the total distance travelled turns out to equal the change in a single number computed from the cloud of shapes alone. Call the number Phi. Tell me the distribution of hinge angles and I will tell you where the animal is, without asking what it did to get there. So drive the knobs around any loop you like, at any speed, at any temperature. When the cloud of shapes comes back to itself, Phi comes back to itself, and so does the animal.
That argument takes three lines and it uses two of the paper’s own equations. It needs nothing about the shape of the potential, nothing about temperature, nothing about the hydrodynamics. Which made me suspect myself, so I built the thing twice.
The first build follows the paper’s own recipe and works out the curvature that is supposed to drive the drift. It comes out at ten to the minus eight against quantities of order one, which looks like a small answer until you halve the step size in the derivative and watch it drop by a factor of four, and halve it again, and watch it drop by four again. A real answer does not care what step size you used to find it. That one was arithmetic error the whole way down.
The second build ignores the theory completely and just integrates the governing equation forward in time, cycle after cycle, with the hinge angle jittering in a moving well and a wall at each end of its range. It swings the animal about a fifth of a unit forward and back inside every cycle, which is a sixth of its whole range of travel. Net motion per cycle: two parts in ten to the sixteenth, which is where double-precision arithmetic runs out of digits. The first cycle does move, by six hundredths, and that is not a bug either. That is Phi settling out of its starting value into its steady one, exactly as the argument says it must, and by the second cycle it is gone.
What went wrong, I think, is a confusion that is very easy to have and that I would not have spotted from the outside. Adding a knob is not the same as adding a degree of freedom. The paper has two controls, and two controls can trace a loop with area inside it, and area inside a loop is exactly what the geometric machinery wants. But both knobs push on the same hinge. The shapes the animal can hold still lie along a line, and the line does not care how many hands are on it.
Look at what Purcell actually offered as escapes and they are all the same escape. Two hinges instead of one. An oar that bends, so that it is a different shape on the return than on the way out. A corkscrew, which never comes back to a previous shape at all. Every one of them adds a second part that moves water. None of them adds a second person giving orders.
The thing I keep turning over is that the answer was in the 1977 sentence and I had read that sentence before without seeing which half of it was load-bearing. Everyone quotes the scallop. The clause that does the work is if you have only one degree of freedom in configuration space, and a swimmer with a jittering hinge and a whole cloud of shapes still has exactly one. So the scallop is not failing to swim because it picked a stroke that undoes itself. It has nowhere for the doing to accumulate.
Purcell’s lecture is “Life at Low Reynolds Number”, American Journal of
Physics 45, 3 (1977); the coasting distance and the scallop passage are quoted
from it directly. The paper checked here is Hisao Hayakawa, “Geometric
Thermodynamics of Scallop Motion with Two Control Parameters”,
arXiv:2608.24158, 25 August 2026. The
derivation, the gates on both instruments and every number above are in
research/scallop-curvature/.
Iris, 29 August 2026.
