The Sample

In 1873, Camillo Golgi was working by candlelight in a hospital kitchen in Abbiategrasso when he discovered that silver chromate, precipitated inside nervous tissue, would stain individual neurons black against a clear background. The method was capricious. It labeled somewhere between one and five percent of the cells in any given preparation, chosen by a mechanism that remains poorly understood to this day. The rest stayed transparent.

This incompleteness was not a defect. It was the discovery.

If every neuron in a slice of cortex absorbed the silver chromate, the result would be an opaque black rectangle — technically complete, informationally useless. What made the reazione nera revolutionary was precisely what it failed to do. The unstained ninety-five percent created the space in which the stained five percent could be seen. Santiago Ramón y Cajal, beginning in 1887, used Golgi's method to establish that the nervous system was composed of individual cells separated by gaps — the neuron doctrine — contradicting the prevailing view that neural tissue formed a continuous reticulum. Golgi himself held the reticular theory. When they shared the Nobel Prize in 1906, Golgi used his acceptance speech to argue against the conclusion that his own technique had made visible.

The instrument settled a question its inventor got wrong, and it did so because of a property he could not control.

The same principle appears in a different physics. In the years after the Second World War, Martin Ryle at Cambridge began developing aperture synthesis for radio astronomy. Two small radio dishes separated by a distance d can achieve the angular resolution of a single dish with diameter d — not despite the gap between them, but because of it. Each pair of antennas at a given separation measures one spatial frequency component of the sky's brightness distribution. By moving the dishes to different separations over time, Ryle could fill in the Fourier plane and reconstruct an image with resolution determined by the longest baseline, not the largest dish.

The gap between the antennas is the aperture.

Very Long Baseline Interferometry extended this to telescopes on different continents, achieving resolution equivalent to a dish the diameter of the Earth. The Event Horizon Telescope, which produced the first image of a black hole's shadow in 2019, used eight stations spanning from Hawaii to the South Pole. The instrument was mostly gap. The image emerged from what the instrument was not.

Harold Edgerton arrived at the same structure through time rather than space. His electronic stroboscope, developed at MIT in the 1930s, froze motion by illuminating it in microsecond bursts separated by darkness. A bullet passing through an apple. A milk drop forming a coronet. A hummingbird's wings caught mid-stroke. Continuous illumination shows blur — the integration of all positions across the exposure. The stroboscope shows structure by refusing to look most of the time. The darkness between flashes is not dead time. It is what separates one position from the next and makes each position distinguishable.

The principle is the same as Golgi's, rotated into time. Continuous illumination integrates everything into a single blur. Brief illumination selects a moment. The selection is the information.

In 2006, Emmanuel Candès, Justin Romberg, and Terence Tao published the mathematical foundation for compressed sensing: if a signal is sparse in some basis, it can be reconstructed exactly from far fewer measurements than the Nyquist-Shannon theorem requires. The key condition is incoherence — the sampling basis must be maximally unlike the sparsity basis. Random sampling satisfies this condition almost optimally.

The result inverted the engineering intuition. Undersampling was not a compromise forced by limited bandwidth. Under the right conditions, it was the correct strategy. MRI acquisition times dropped by factors of four to eight when compressed sensing was applied to k-space sampling, not because the scanner improved but because the sampling pattern became deliberately incomplete.

The visual cortex may have arrived at the same solution independently. Olshausen and Field showed in 1996 that natural image statistics predict sparse coding: at any moment, roughly one percent of primary visual cortical neurons are active. The other ninety-nine percent are silent, and the silence is part of the code. Dense activation would be metabolically expensive and informationally degenerate — too many neurons saying similar things. Sparse activation creates a high-dimensional space in which each pattern of activity is far from every other, making distinctions easy and errors rare.

The pattern across these cases is not "less is more," which suggests a trade-off. It is that the absence is the mechanism. The unstained neurons are not missing data — they are the contrast that makes structure visible. The gap between antennas is not empty space — it is the baseline that determines resolution. The darkness between flashes is not lost time — it is what separates positions. The silent neurons are not idle — they are the dimensions that make the code work.

In each case, someone could have tried to fill the gap. Stain every neuron. Build a single dish the size of a continent. Illuminate continuously. Sample at Nyquist rate. The result would be technically more complete and informationally worse.

The counter-cases exist. Whole-genome sequencing reads every base pair because any position might carry the variant that matters — the signal is not sparse. Exhaustive search in cryptanalysis checks every key because the correct key has no structural signature that would distinguish it from the rest. These work when what you are looking for has no pattern that sampling could exploit. The principle has a boundary: incompleteness is the instrument only when the space is mostly empty and the emptiness is what you need to see through.

Golgi, working by candlelight in a kitchen he had converted to a laboratory, could not have known this. He thought the incomplete staining was a problem to be solved. He spent years trying to make the method more reliable, more consistent, more complete. The neuroscience that his technique made possible came from someone else, who used the incomplete staining without trying to fix it and saw what the gaps revealed.