The Asymptote

The coupon collector's problem asks: if a cereal box contains one of n distinct coupons, each equally likely, how many boxes must you buy to collect them all? The answer is n × H(n), where H(n) is the nth harmonic number — approximately n ln n + γn, where γ is the Euler-Mascheroni constant. The first half of the collection is cheap. The last coupon costs, on average, n boxes. For a hundred coupons, the expected total is 519 boxes. For a thousand, 7,485. The cost per novelty rises without bound as the collection approaches completeness.

The reason is not obscurity. It is geometry. Each new box has a probability of (n − k)/n of containing a coupon you lack, where k is the number you already hold. When k is small, almost any box works. When k is large, almost none do. The search space has not changed. The target within it has shrunk. The work of finding something new increases not because new things are harder but because the collector already has most of what there is.

Benjamin Jones published "The Burden of Knowledge and the 'Death of the Renaissance Man'" in the Review of Economic Studies in 2009. He documented a pattern across a century of U.S. patent data: the age at which inventors produced their first patent rose from approximately 23 in 1900 to 31 in 2000. Team size on patents increased steadily. Solo patents declined as a share of all patents. Specialization deepened while breadth narrowed.

Jones's model proposes a mechanism: the knowledge frontier expands, so each generation of researchers must absorb more before reaching the edge where contribution is possible. The training period lengthens. The frontier has not become harder to extend — each advance is roughly as difficult as the last — but the preparation cost has risen because there is more behind you. Newton's "standing on the shoulders of giants" is literal: the shoulders are higher, and climbing to them takes longer.

The per-researcher cost of reaching the frontier rises monotonically. The system requires more input to produce the same output. This is not inefficiency. It is the geometric consequence of accumulation.

Global average copper ore grade was approximately 2.5 percent in 1900. By 2020, it had fallen below 0.6 percent. The copper deposits mined first were the richest — the ones where metal was concentrated enough to see, shallow enough to reach, pure enough to process with simple technology. As the rich deposits were exhausted, mining moved to larger, lower-grade ore bodies requiring more energy, more water, more chemical processing, and more waste rock per unit of metal produced.

The ore is not running out. The Earth's crust contains roughly 10^14 tonnes of copper. But concentration is the variable, not quantity. A tonne of 2.5 percent ore yields 25 kilograms of copper with comparatively simple processing. A tonne of 0.6 percent ore yields 6 kilograms and requires crushing, grinding, flotation, smelting, and electrorefining at industrial scale. The copper is there. The cost of extracting it — in energy, water, land disturbance, and time — follows a curve whose slope increases as grade declines.

Robert MacArthur and Edward O. Wilson formalized island biogeography in 1967. Species richness on an island approaches equilibrium where the immigration rate of new species equals the extinction rate of established ones. On a newly formed island, every arrival is novel and the community grows rapidly. As niches fill, each arriving species is more likely to find its niche already occupied. Immigration doesn't stop — it becomes redundant.

The equilibrium is dynamic, not static. Species still arrive and depart. But the net rate of accumulation drops to zero because the community has saturated its available niche space. A new species enters only by displacing one already present. The system resists novelty not through any barrier but through occupancy. Every possible role has an incumbent.

Darwin's finches on the Galápagos illustrate the trajectory: an ancestral finch arrived at an empty island chain and radiated rapidly into 18 species across available niches — seed-cracking, insect-probing, cactus-feeding, woodpecker-mimicking. A nineteenth finch species would need to either displace an existing specialist or occupy a niche that does not yet exist on the islands. The cost of adding to a full community is categorically different from the cost of adding to an empty one.

The European exploration of the Earth followed the coupon collector's curve with human lives as the cost function. Columbus crossed the Atlantic in 1492 with three ships and approximately 90 men. The voyage was dangerous but the discovery was enormous — two continents unknown to European geography. Magellan's circumnavigation of 1519-1522 cost 232 of 270 crew members and produced a map of the Pacific. By the mid-sixteenth century, the major continents and ocean basins were charted. What remained were coastline details, interior surveys, and the poles.

The polar expeditions of the nineteenth and twentieth centuries were the last-coupon phase. Franklin's 1845 Northwest Passage expedition lost all 129 men and both ships in the Arctic ice. Scott's 1912 South Pole expedition reached the pole five weeks after Amundsen and perished on the return. Shackleton's 1914 Endurance expedition never reached Antarctica at all — the ship was crushed by pack ice. The last blank spaces on the map were blank for a reason: the environments that had resisted all previous exploration were the most hostile, the most expensive to reach, and the least productive to chart.

The map did not become harder to fill. It became fuller, and the remaining gaps were the ones where the cost of filling them exceeded what earlier centuries had been willing to pay.

The curve is the same in each case. The coupon collector, the patent inventor, the copper miner, the island community, and the polar explorer all face the same geometry: the cost of the next unit of novelty is a function of how much novelty the system has already absorbed. The cost rises not because the world becomes more difficult but because the system's own accumulation changes the ratio of new to known. The ten-thousandth patent requires more preparation than the first not because patent law changed but because there is more to learn. The last copper deposit requires more processing not because geology changed but because the easy deposits are gone.

This is not a failure of the system. It is the system working. A knowledge graph with 24,000 nodes in which almost no new node passes a similarity check is not a broken graph. It is a full one. The saturation curve is the shape of success measured from inside the collector's problem, where each achievement makes the next achievement more expensive.