Can Math Unmix Soil That Worms Have Blended? A Testable Idea, Rated 50 Percent

Earthworms are tireless editors of the past. Over centuries they carry buried objects up and down, until two short visits to a site, a few centimetres apart, look like one long smear. Our research desk asks whether that smear still hides the original two events, recoverable with the right mathematics.

This is an archaeology question, but the lesson is general. It is about when blurred data still contain the answer, and when they do not. That matters for anyone who reads a medical scan, a lab trend or a study chart.

Excavators remove soil in thin slices and count the finds in each. When burrowing animals mix two thin occupation layers, the count profile becomes a single broad hump. Archaeologists call such blended deposits palimpsests and usually interpret them at a coarser time scale rather than trying to reverse the mixing.

The desk points to a family of results in compressed sensing, a branch of applied mathematics. These concern signals made of a few distinct point sources, each with a positive size, blurred by a known bell-shaped spread. The desk cites work by Schiebinger, Robeva and Recht (2018) and by Morgenshtern and Candès (2016) for the claim that, when all sources are positive and the blur is Gaussian, close sources can in principle be separated however near they sit. Noise, not geometry, becomes the limit.

Carried across, the prediction is concrete. Calibrate the blur width from a marker layer, such as a volcanic ash fall. Then, with a few hundred finds per event, a recovery method constrained to positive values should return two events, each within one slice of its true depth, even when they sit half a mixing width apart. Simple peak-finding would return one.

The desk assigns 50 percent to that outcome in a controlled test.

We could not independently open the mathematical papers the desk cites during this session, so we treat the theorem's exact conditions as reported, not checked. The general idea, that positivity makes some blurred signals more recoverable, is familiar from microscopy and spectroscopy.

The archaeological side rests on a real assumption: that worm mixing behaves like diffusion, spreading objects in a smooth bell shape. That is a modelling choice with a long history in sediment science, but the desk itself notes the weak points. Larger objects can sink faster than small ones. Real mixing is often lumpy. And a few hundred finds per layer produces count noise that may swamp the signal below half the mixing width.

We found no published sediment profile where this recovery has been tried. That makes the idea genuinely open, not established.

The proposed test is cheap and clean. Fill containers with sieved soil. Lay two thin layers of glass beads, same size, two colours, at a chosen distance apart. Add earthworms and wait six to eighteen months. Slice each container thinly and count each colour per slice; that is the ground truth.

Then add the colours together and hand the totals to the recovery method blind. A control container with one layer calibrates the blur. Varying the separation and the number of beads maps exactly where the method works and where it fails. Each container gives a yes or a no.

A quicker check uses existing data. Some North Atlantic sediment cores hold pairs of volcanic ash layers from different eruptions, mixed together but distinguishable by glass chemistry. Running the method on total shard counts and comparing with the chemistry would be a natural first trial.

Medicine is full of blurred signals. A blood marker drawn monthly smooths over day-to-day swings. An imaging study averages tissue within each voxel. The core question here, whether a known blur plus a positivity rule lets you recover detail you cannot see by eye, sits under many diagnostic and research tools.

The honest version of the lesson is twofold. Good prior knowledge about how data were blurred can recover real structure. But the method only works if that knowledge is right, and noise sets a hard floor. A clever algorithm cannot invent information the counts do not carry.

• Does real earthworm mixing follow a smooth bell shape, or is it patchy? • Do objects of different sizes and densities move differently enough to break a single blur model? • How many finds per layer are needed before noise stops dominating?

• Any bead-and-earthworm experiment that reports blind recovery against known colours. • Reanalysis of paired ash layers in marine cores using positivity-constrained recovery. • Whether the method's failure point tracks noise, as the desk predicts, or mixing shape.