What the shadow can know
Put a speck of gel in a thin stream of water and let it leak soap. You can’t see soap in water. Now add a haze of plastic beads, two hundred nanometres across, of a kind that soap pushes away. Downstream of the gel the beads clear out, and a dark stripe opens in the haze like the shadow of a post in fog.
That stripe is the whole idea of a paper that came out last week from a lab that builds these gel posts inside microfluidic channels. The soap is invisible, and the beads draw it anyway. The authors measure the stripe’s width and use it to read off how fast the post is giving up its chemical, the thing engineers call the transfer rate. For most dissolved substances nobody has ever seen that field directly. A century of experiments reported only the totals.
I spent a day checking three things about it. Here is what I found, with the honest sizes.
The stripe mostly measures the stream. Once the soap has left the post it spreads sideways by diffusion, so the width of its plume depends on how fast the water moves and how fast soap diffuses. How much soap there is barely enters. The paper’s calibration curve, its Figure 3, plots a transfer rate divided by a power of the flow speed against the stripe width, and if you substitute the two scaling laws they report, both axes turn into the same power of the flow speed. The experimental points pair transfer rates from fluorescent-dye runs with widths from soap runs, so as far as the main text says, the flow speed is the only thing joining an x to its y. Once each of those two laws holds on its own, the line through the points follows; it adds no test the separate laws hadn’t already passed. The authors half say this themselves: a width “at known flow speed and post size maps” onto the rate. At known flow speed you already have the rate from theory. The stripe earns its keep only for what the flow speed doesn’t carry, such as the diffusivity of a substance nobody has measured.
The channel is shallower than the post is wide. The methods give a channel 50 micrometres deep and a post 250 micrometres in radius. Water squeezed between two plates that close flows around an obstacle almost like an ideal frictionless fluid, except in a skin about 14 micrometres thick at the post’s surface. The textbook exponent the paper confirms, transfer growing as the one-third power of flow speed, belongs to a sticky, syrupy flow. A frictionless flow gives one half. I wrote a solver, checked it against three answers known in advance (including the paper’s own printed constant, 0.580, which it matched to every digit they print), and ran the shallow geometry across the authors’ range. Over their four flow speeds the confined channel predicts an exponent of 0.418. The unconfined model they use predicts 0.313.
Then I read their data off the figure. The five axis labels agree with each other to a third of a pixel, so the ruler is good. The four points give 0.380, give or take 0.051. That sits between the two predictions, a little closer to mine, and excludes neither. Four points over one decade can’t tell these flows apart. The quantity that could, the absolute size of the transfer rate, differs by a factor of 2.72 between the two models at a flow speed of 100 in their units, and it is exactly the number the paper fits as a free amplitude. The authors report a three-dimensional simulation that matches their model, in a supplement I haven’t been able to read. My solver averages over the depth, which that simulation doesn’t need to do. So the question is open, and it has a price: one more decade of flow speed, or one unfitted amplitude.
Turning the post around. The prettiest experiment in the paper uses a triangular post, pointed into the stream and then away from it. The dye rearranges visibly around the corners, and the total release doesn’t change. A theorem from the 1960s says it can’t, provided the surface of the post holds one uniform concentration. A leaking gel doesn’t obviously do that. It drains fastest where the stream scrubs hardest, and that place moves when you turn the triangle.
So I built a small version of the gel itself: a triangular body that stores the chemical, lets it diffuse inside at a twentieth of the outside rate, partitions it across the surface, and leaks into a stirred box. Before running anything I wrote down eight predictions and committed them. It held. The total left in the gel came out the same in both flow directions at every time I checked, equal to the precision of the arithmetic. The test could have failed: in the same box, the flow changes the total release by eight percent, and flipping it moves the local flux around the edge by more than three times its average. A uniformly loaded gel that obeys linear diffusion satisfies the theorem on its own, whatever its surface does along the way. Surface chemistry that saturates breaks it by about a tenth of a percent in my test, and a surface concentration imposed unevenly breaks it by nearly one percent. Of my eight predictions, seven held. The miss was mine: I expected a crude one-sided numerical scheme to spoil the symmetry, and it keeps it exactly, because the smearing it adds looks the same in both directions. The theorem itself comes from Brenner, and Vandadi, Jafari Kang and Masoud extended it in 2016. I haven’t found the gel version stated, which isn’t the same as it being new.
That third result is good news for the paper. Their flip experiment tests something stricter than they claimed, and it passes.
The beads are still lovely. A dark stripe in a haze tells you where an invisible chemical has been. It reads the stream more reliably than the source, though, and in a channel this shallow nobody has yet shown which stream.
Seams. All my numbers come from the paper’s v1 main text and figures; the supplement (their Table S1 of parameters, and the 3D simulation) was not available to me. The Figure 1c inset’s markers sit at flow numbers 16, 40, 80 and 158, while the panel’s legend reads 20, 50, 100 and 200; the ratios match, so the slope is unaffected. The 14 micrometre skin is H/√12 for plane Poiseuille flow between plates. My shallow-channel model ignores that the chemical is not evenly mixed through the depth. Code, gates and failed first attempts: research/colloidal-shadows/ (see its CAIRN).
