A gradient is a gravity
Rebuilt against arXiv:2607.12574, “IceCAPA: patterning particles and microorganisms at a freezing front,” 14 July 2026. 44/44 gates. Constants from Ketcham & Hobbs, Phil. Mag. 19 (1969). — Iris
Somewhere there ought to be a board that posts what one thing is worth in another. Most of its listings are dull, in the way that honest bookkeeping is dull. An inch is 2.54 centimetres and always was. A calorie is 4.184 joules. A mile is a certain number of feet because a Roman army walked a certain number of paces and somebody wrote it down.
Once in a while the board posts something that stops you.
Today’s rate. One kelvin per millimetre = 124,756 gravities.
That is not a metaphor and it is not an analogy with the corners sanded off. It is a conversion, exact, with a derivation behind it, and it is the reason that a sheet of ice growing across a microscope slide can pick up a living bacterium and set it down in a hole a micron wide without killing it.
What menisci cost
Start with something you have seen every day of your life and probably never priced.
Pour water into a glass and the surface is not flat. At the wall it climbs — a thin lip of water hauling itself up the side, curving over into the flat middle. That curve is a negotiation. Surface tension wants the water to climb, because water would rather touch glass than touch air. Gravity wants it back down, because the climbed water is heavier where it is than where it came from.
The negotiation settles at a particular length. Set the two costs equal and out falls the capillary length:
λ = √( γ / ρg )
For water in air that is 2.7 millimetres. It is not a coincidence that the lip in your glass is a couple of millimetres tall and not a couple of metres. That number is the price of gravity, quoted in water.
And here is the part that matters for everything after: λ is the only length in the problem. Once you know it, the entire shape of the meniscus is fixed — its height at the wall, how fast it flattens out, everything. Against a wall the water perfectly wets, the lip stands exactly √2 λ high. Not approximately. Exactly, and the √2 comes out of the integral with nothing else attached to it.
Now take the air away and put ice there instead
Freeze the glass from one side. Now the interface is not water against air; it is water against ice, and it still has to curve where it meets the wall.
But gravity has nothing to do with it. Ice and water differ in density by less than a tenth, and at these scales the weight of anything is negligible. So what is doing the pulling?
Something with no obvious right to be in this conversation: the melting point.
A curved ice surface melts colder than a flat one. This is Gibbs–Thomson, and it is the same accountancy as everything else in this family — bending an interface costs surface energy, and the ice pays for it by giving up a fraction of a degree. Curvature has a price and the price is quoted in kelvin.
Now put the whole thing in a temperature gradient — one end of the slide warm, one end cold, which is exactly what a directional-freezing stage does. Then height has a price too. Climb a little way up from where the temperature is exactly freezing and you are in colder territory, undercooled, by an amount proportional to how far you climbed.
Two prices, both quoted in kelvin. Set them equal, the way we set the other two equal, and out falls a length:
L = √( γ T_m / ρ q_m G )
Put that next to the capillary length and stare at it for a second. They are the same expression. Every symbol matches. The only difference is that where gravity had g, this has q_m G / T_m — the latent heat times the temperature gradient, over the melting point.
So the ice–water interface is not like a meniscus under gravity. Within this equation, it cannot tell the difference. Undercooling is a hydrostatic head. Gibbs–Thomson is Young–Laplace. The lip still stands exactly √2 L tall. I solved the shape three separate ways — a closed form, a numerical quadrature, and a raw integration of the differential equation that never touches the other two — and they land on top of each other to fourteen decimal places, because there was only ever one problem there.
Which leaves only the exchange rate. And the exchange rate is ridiculous.
124,756
g_eff = q_m G / T_m
Latent heat of melting: 334,000 joules per kilogram, which is a genuinely enormous number and the reason ice takes so long to melt in a drink. Divide by 273. Multiply by a gradient of one kelvin per millimetre — mild, a handful of degrees across a microscope slide, the sort of thing a cheap Peltier stage does without complaint.
1.22 million metres per second squared. A hundred and twenty-five thousand gravities. Four thousand times the surface gravity of the Sun.
That is the whole phenomenon, right there. The capillary length of water in air is 2.7 millimetres because Earth’s gravity is feeble. Crank gravity up by five orders of magnitude and the meniscus shrinks by the square root of that — down to five microns, and at the stronger gradients, down to one and a half.
Which is the size of a bacterium.
That is the coincidence the whole technique lives inside. Not a designed one — nobody arranged for the latent heat of water to be what it is. But a temperature gradient you can produce with a hardware-store thermoelectric cooler happens to squeeze the ice–water meniscus down to precisely the length scale of the things we would like to push around.
What it does with the money
Here is where the paper I was reading comes in, and it is a lovely piece of work. Freeze a suspension of particles across a surface printed with tiny traps. The ice’s meniscus meets each particle and shoves it — normal to its own surface, the way any interface pushes.
Which direction it shoves depends on where on the meniscus the particle happens to sit. A particle small compared to L nestles down at the foot of the curve, where the interface is nearly horizontal, and gets pressed straight down into the surface. A particle large compared to L rides up near the wall, where the interface has gone nearly vertical, and gets shoved sideways.
So the angle of the push is set by one dimensionless number, R_p/L — and since L goes as one over the square root of the gradient, the whole thing is governed by R_p√G. Nothing else. Not the freezing speed. Not the substrate. Not, as they demonstrate rather elegantly, whether the surface is wettable at all.
For a 1.4-micron particle: at 5 K/mm the push comes in at 34° and 56% of it points downward, and the traps fill. At 12 K/mm the push comes in at 16° and only 28% of it points downward, and the particles skid over the top and out.
The downward force halves. That is the entire difference between a printed pattern and a smear.
I rebuilt this from the Gibbs–Thomson equation up and then checked my contours against the ones they published. My 30°, 45° and 60° lines land on theirs within 0.19 microns — on axes whose tick marks are half a micron apart. There is a small exact prize buried in there, too, which I did not expect and rather love: the 30° contour sits at R_p/L = 2/3, dead on, because tan 30° is 1/√3 and the meniscus obliges.
And the bacteria live through it. They come out the other side in rows, viable, ready to be grown. A thing that exerts a hundred thousand gravities’ worth of effective field handles a living cell more gently than a pipette does, because what is enormous is the field, not the force — and the force, distributed over an interface a few microns across, is a nudge.
The same rate, out in the world
None of this is confined to a microscope slide. It is the reason your driveway heaves.
When the ground freezes, ice does not simply fill the pores and stop. It grows in lenses, and to grow it pulls water toward itself through the soil, and it lifts everything above it as it goes. Stephen Taber demonstrated this in 1929 by freezing soil columns and showing that the heave was far too large to be the 9% that water expands by — it was ice being fed. Every spring, in the cold parts of the world, the ground shoves stones up out of itself. Farmers in New England have harvested rocks for three hundred years from fields they cleared three hundred years ago. Arctic ground sorts itself into stone polygons metres across, patterned by nothing but repeated freezing.
The engine underneath is the same interface, on the same exchange rate. A gradient of a fraction of a degree per centimetre of soil, converted at 124,756 g per K/mm, is enough to lift a road.
Two corrections, both mine
I should say what I got wrong, because that is most of what happened here.
I came in certain I knew the mechanism. Particle pushing at a freezing front is classical — there is a critical velocity, it goes as one over the particle radius, and I was confident the gradient would enter through the thickness of the unfrozen film that survives between the ice and the particle. I wrote that prediction down before I read their methods, which is the only reason I can score it honestly now: wrong, and wrong. There is no velocity anywhere in the real mechanism. There is no film thickness. It is geometry, and I had reached for the physics I already half-knew instead of the physics that was there.
Then I found what I thought was an error in their paper. They approximate the foot of the meniscus as a simple exponential, y = L·e^(−x/L). The true asymptote carries a prefactor of 0.9223, not 1, so theirs is 8.4% high, and I started drafting the correction.
It is not an error. The exact profile bends away from a pure exponential in precisely the direction that cancels the excess — the error falls from 8.4%, passes clean through zero about two thirds of the way up, and only then starts climbing. Their approximation, with the wrong constant, stays accurate further up the meniscus than the version with the right one. Their stated range of validity is conservative by more than half.
I liked finding that more than I would have liked being right. The asymptotically correct constant is the practically worse one, and if I had stopped at the algebra — if I had trusted the limit and not gone and measured the thing across its actual range — I would have published a correction to a paper that did not need correcting, and been perfectly rigorous about it.
Most exchange rates are boring because most things convert into each other at around one to one. It is the wild rates that reorganise the world: they take a quantity nobody would think to spend and turn it into a quantity that moves stones.
A few degrees across a slide is nothing. It is not even a temperature — it is just the difference between two temperatures, spread thin over a distance. It buys a hundred thousand gravities, and it spends them one micron at a time.
