The Price of Mistaking Necessity for Sufficiency

What is the shortest route between two cities on the surface of the earth? Between Houston and Tokyo a truly straight line does exist, but you would have to dig it: it runs through the body of the planet, some thirteen hundred miles beneath the Pacific at its deepest, down in the mantle.¹ And nothing is shorter: about fifty-nine hundred miles, more than seven hundred fifty fewer than any route along the surface. Stay on the surface and every route you can actually travel is a curve, so the question becomes which curve is shortest. The answer is an arc of a great circle, one of the giant circles that cut the globe in half, the equator being the most famous one. Trace the shortest route across the surface from Houston to Tokyo and it arcs far north, across the Gulf of Alaska and out along the Aleutian Islands, climbing more than eighteen degrees of latitude above either city. Real flights track close to that arc, bending with the winds and the day’s weather. On the globe that arc is as straight as the surface permits; the flat map is what bends it. And the route that looks straight on the flat map, running out across the middle of the Pacific, is the genuinely bent one, and nearly six hundred miles longer.

An Honest Mistake in Paris

In 1744 Leonhard Euler, whose collected works run to more than eighty volumes, published the general method for finding the optimal curve, whatever optimal means for the problem at hand: the shortest route, the fastest ramp, the flight path that burns the least fuel.² At its center is one equation that the optimal curve has to satisfy. But there was a caveat. The method implied that the optimal curve had to satisfy the equation, and not that any curve satisfying the equation was optimal. In mathematical terms, this means that the equation was a necessary condition for the optimal curve, but it was not a sufficient condition. And in practice, it means that Euler’s equation was not a final solution, and for four decades nobody was able to come up with one.

The problem was still open in 1786, when Adrien-Marie Legendre, one of the leading mathematicians in Paris, presented a memoir to the Academy of Sciences claiming to have solved it.³ Legendre had found a second, sharper test, and he believed the two conditions together were sufficient: a curve that passed both was, he argued, guaranteed to be optimal. But the proof had a hole. Legendre assumed that one step of it, a construction that had to hold along the entire length of the curve, would always work, and he never proved that it would. Joseph-Louis Lagrange, then the greatest living authority on the mathematics of motion and a man whose masterwork Legendre had edited, set out objections in 1797, and for a generation the idea was shelved as a promising instrument that had failed.

Finding the hole took fifty-one years. In 1837 Carl Jacobi, a German mathematician then at the height of his powers, located it.⁴ Start again in Houston. The place on the planet farthest from Houston is the single point directly opposite it, on the other side of the earth, out in the southern Indian Ocean. Fly out of Houston along any great circle and, all the way to that farthest point, flying straight and flying the shortest way are the same thing. One mile past it, they separate. The route is still perfectly straight, but every destination beyond the farthest point sits closer to Houston in the opposite direction, so each additional straight mile now takes you the long way around. Straight has stopped implying shortest, and it stopped at one exact, computable point. This was Legendre’s hole. The step he assumed would always work does work on every arc that ends before the point opposite its start, and fails on every arc that continues past it. His proof was true for arcs shorter than half the circle and false for arcs longer than that, and none of his tests could tell the difference. Mathematicians later named these breakdown places conjugate points, and Jacobi’s check joined the stack as one more necessary condition: a curve carried past its conjugate point cannot be optimal, and a curve short of it still might be.

Even with the hole located, closing it took another generation. A full guarantee, one that certifies a curve against every rival rather than only its nearest neighbors, arrived with Karl Weierstrass, the Berlin professor who spent his career rebuilding calculus on rigorous foundations, in his lectures of 1879.⁵ Legendre had compared the route only against slightly deformed versions of itself, so what his tests truly certified was that the route was the best of the versions he compared it against. Weierstrass compared it against everything the problem allows. He covered the entire region with a family of candidate routes, one passing through every point and none of them crossing, and showed that any rival route, wherever it wanders, can be measured against that family stretch by stretch, with every stretch of deviation costing extra. A route that pays extra everywhere it deviates cannot arrive cheaper. In plain words: best is a relative verdict, the winner of whatever comparison you actually ran, while optimal is absolute, better than every rival there is. Legendre certified a best and believed he had certified the optimum. Weierstrass’s covering comparison is what certifying the optimum costs, and a certificate that reaches every rival is what sufficient means. Euler wrote the equation in 1744; Legendre claimed sufficiency in 1786 and was wrong; Jacobi found the flaw in 1837; Weierstrass delivered the guarantee in 1879. From the necessary condition to the sufficient one, on a problem where nothing was hidden and nothing changed, the crossing took one hundred thirty-five years.

The Reserves Were Real

In March 2023 the second-largest stablecoin in the world was a token called USDC, issued by a company called Circle. A stablecoin is a digital token built to be worth exactly one dollar at every moment, and USDC’s version of the promise had two working parts. The first was redemption: hand Circle a token and receive a dollar, a transfer that settles through ordinary banks. The second was the open market, where tokens change hands around the clock at whatever price buyers will pay. As long as redemption works, the two prices are welded together, because nobody sells for ninety cents what the issuer redeems for a dollar. Behind the redemption promise sat roughly US$40 billion of reserves, most of it in short-term Treasury bills, the remainder held as cash at banks.⁶

How did anyone know the reserves were real? Through an attestation: once a month, an accounting firm examined Circle’s holdings and signed a report confirming that at given instants the assets existed and covered the tokens outstanding.⁷ It is a test with the same shape as Euler’s equation, a condition that any healthy stablecoin must satisfy. In mathematical terms, it is necessary, but nothing about it is sufficient. The market had built a taller structure on that monthly signature anyway: verified at an instant was read as fully backed, fully backed was read as redeemable at a dollar, and redeemable at a dollar was read as always, which is the exact word Circle’s own page used to end the promise. An attestation speaks as of a chosen midnight. The promise covers every instant there is.

On Friday, March 10, 2023, regulators closed Silicon Valley Bank, the sixteenth-largest bank in the country, after depositors had pulled roughly US$42 billion the day before.⁸ Just after ten that night Circle disclosed that US$3.3 billion of its reserve cash was inside, about eight percent of everything standing behind the token. And the closure jammed the promise’s first working part directly: a redemption settles through banks, the banks were shut until Monday morning, and so the machine that turns a token back into a dollar had stopped.

With redemption dark, the weld broke. The market price stopped being an echo of a working exchange window and became the market’s live vote, updated by the minute, on whether the promise would still be standing when the banks reopened. By early Saturday the vote read eighty-eight cents.⁹ Walk through what had failed by that point, because the list is short. The attestations were honest. The reserves were real, and the report describing them was accurate. Even a total loss at Silicon Valley Bank, which nobody expected, would have cost the token eight cents; the market had marked it down twelve. The guarantee had expired at one specific place, exactly as it did on the globe past the point opposite Houston, and it was a place no attestation had ever been pointed at: the rails a dollar travels on.

The failure has a precise mathematical shape. Live for a moment inside the world of fractions, the whole numbers and their ratios, and try to reach the number whose square is exactly two. You can get close, and then closer: 1.4, fourteen tenths, squares to 1.96; 1.41 squares to 1.9881; 1.414 squares to 1.999396, each step an ordinary fraction, the sequence crowding in on its target from below. It can crowd forever, because the target is missing from that world: no fraction multiplied by itself gives exactly two, a fact proved about twenty-five centuries ago. Mathematicians call a world with no missing destinations complete. Their word for such a world is a space, and completeness belongs to the space; the sequence, however disciplined, cannot supply it. The space that matters here is the set of outcomes a token can actually reach through the system, everything a holder can turn a token into, and it needs naming precisely because it is not the set of dollars. The dollars existed all weekend, sitting in accounts across the banking system, which is the honest disanalogy with the missing number: nothing had vanished. What the weekend removed was the dollars’ membership in the reachable world, and a dollar that cannot be reached is, for every purpose inside that world, a point that is not there. Over that weekend the stablecoin system was that world of fractions. Every check inside it could still pass, every verification could land closer to the dollar, and the destination, a dollar actually delivered, was missing from the space the system settled in.

Mathematics made this exact mistake once, inside its own house, and the man who caught it was Weierstrass again. Bernhard Riemann, one of the greatest mathematicians who ever lived, rested a famous argument on the assumption that an optimal function existed, because the quantity he was minimizing could be pushed arbitrarily close to its lower bound.¹⁰ Weierstrass objected that pressing close proves nothing about arrival, since the target can be missing from the space being searched, and the repair, in its mature form, consisted of changing the space the search ran in, completing it, and proving that the minimum was actually attained there. The two cases share their structure exactly. In both, every check available on the inside can pass while the destination sits outside the world being searched, and no discipline in the searcher can put it there. In both, the repair belongs to whoever can change the world rather than the search: mathematics rebuilt its space and proved the arrival, and the world the token settles in could be rebuilt, that Saturday, only by the institutions that own its rails. Which is the limit both cases demonstrate, and it is a limit on verification itself: a check examines the traveler, and whether the world contains the destination was never the traveler’s fact to establish.

The Hypothesis Arrives After the Counterexample

The point was added back on Sunday evening. At 6:15 p.m. Washington time, the Treasury, the Federal Reserve, and the FDIC issued a joint statement: every depositor of Silicon Valley Bank would have access to all of their money on Monday morning, the losses covered under an emergency authority the law reserves for systemic risk.¹¹ As banking policy, it was a depositor rescue. As geometry, it was an edit to the space: from that sentence forward, every route that ended at the failed bank ended at dollars again. Nobody recapitalized Circle. Nobody bought the token. The signers changed the world the token settles in, and the destination missing since Friday night existed again. By Monday the banks were open, redemptions were flowing, and the token traded back at a dollar.

The weekend’s pricing left behind one number worth an autopsy. Take Saturday’s worst case seriously: suppose the US$3.3 billion at the failed bank had been a total loss, every dollar gone. Even then, the assets behind each token would have been worth about ninety-two cents.¹² Credit risk, pushed to its theoretical maximum, could justify a price of ninety-two. The market printed eighty-eight. Those four cents below the worst case are the part the reserve balance sheet alone cannot explain. What they priced sits on no balance sheet at all: the possibility that perfectly good assets cannot become dollars in hand when the rails are gone. For one weekend, the market quoted incompleteness itself, and it quoted it at four cents on the dollar. Across the tokens outstanding, those four cents came to about US$1.7 billion, the market value assigned, for a few hours that Saturday, to a point that was not there.

There had been a private attempt at completion. Circle promised that weekend to stand behind the token and cover any shortfall from its own resources.¹³ The promise was sincere, and it was smaller than the hole: Circle’s later filings put shareholders’ equity at about US$339 million at the end of that year, a tenth of the amount that had been frozen, and equity is the most generous measure available of what the company could have mobilized in a weekend. A promise to complete a space is worth the balance sheet of whoever makes it. Completeness for the dollar’s space was only ever available from the institutions that define the space, which is why the sentence that repaired it carries three government signatures and no price.

Set the two closures side by side. In mathematics the completing hypothesis is named before anyone relies on the result: Jacobi published the condition to check, Weierstrass proved the guarantee, and only then did the discipline treat a best as an optimum. In the dollar’s space the order ran backwards. The reliance came first, years of it, priced at a dollar and stamped Always; the hypothesis arrived at 6:15 on a Sunday evening, forty-four hours after the counterexample started printing. And the two closures do not age alike. Weierstrass’s cannot be taken back, because a proof that reaches every rival is permanent. Sunday’s lasts exactly as long as the signers’ next decision, renewed, or not, one crisis at a time. Mathematics spent one hundred thirty-five years on its gap and closed it once, forever. The dollar’s gap was closed in forty-four hours, by a promise carrying the force of law, and that kind of sufficiency has to be believed before it can be tested.