08 · Gallery II: The Modern Solves

THE SHORTEST PATH — A History of the Art and Science of Problem-Solving

The five solves in this gallery all happened within living or near-living memory, three of them broadcast or published in enough detail that we can watch the reframe happen almost in real time. The lesson changes shape from the previous gallery. Where chapter 07's solvers each found a cheap measurement to stand in for an expensive world, this gallery's solvers each found something narrower and often more decisive: the one constraint buried inside a sprawling problem that, once identified, made everything else fall into place — or fall apart.

The Theorem the Germans Never Saw Coming — Warsaw, 1932

By 1932, the German military's Enigma machine had been in service for years, and the Polish Cipher Bureau's small codebreaking staff had made almost no progress against it. Enigma's security did not rest on secrecy of design — commercial versions were sold openly — but on the machine's internal wiring, a configuration of three rotors and their electrical connections that the Poles had never seen and could not obtain. Every established approach to cryptanalysis assumed you needed either the machine or a great deal of matching plaintext and ciphertext. Neither was available in useful quantity.

Marian Rejewski, a 27-year-old mathematician recently recruited from Poznań University, took a different route entirely. German operators, following procedure, encrypted a three-letter message key twice in a row at the start of every transmission — a doubled indicator meant to guard against transmission errors. Rejewski realized that this doubling, combined with the fact that an Enigma rotor's wiring is a mathematical permutation, meant the relationships between the first and fourth, second and fifth, third and sixth letters of thousands of intercepted message openings encoded, in aggregate, the internal wiring of the machine itself. Working with pure permutation-group theory — and with two French-supplied operating manuals that pinned down a few of the unknowns — Rejewski wrote out a system of equations and, by December 1932, reconstructed the wiring of a machine he had never touched (Engineering and Technology History Wiki, "First Breaking of Enigma Code by the Team of Polish Cipher Bureau, 1932–1939"; AMS, "Marian Rejewski and the First Break into Enigma"). Later popular accounts, including a widely used mathematics-education article, would call this "the theorem that won the war" — a title that belongs to a 2015 teaching piece rather than to Rejewski himself, who never claimed anything so grand, and the credit properly belongs jointly to him and his Poznań colleagues Jerzy Różycki and Henryk Zygalski, who built the mechanical and tabular tools that turned the theorem into daily practice (Mathematical Association of America, "The Theorem that Won the War").

The load-bearing constraint Rejewski found was procedural, not mechanical: a repeated indicator that operators were required to send. When Germany dropped that procedure in 1938 and tightened Enigma's security further, the Polish team's mathematical technique stopped working, and on 25–26 July 1939 — with war weeks away — Poland handed Britain and France everything it had: its reconstructed Enigma replicas, its methods, and the design for its bomba kryptologiczna, an electromechanical device built to automate the search (GCHQ, "The Pyry Forest Meeting"). At Bletchley Park, Alan Turing and Gordon Welchman then performed a second, equally sharp reframe on top of the first. Rather than trying to break individual messages, they built a machine — the Turing–Welchman bombe — to break the daily key that unlocked every message sent that day, working from "cribs," fragments of probable plaintext (a routine weather report was a favorite) whose position in a ciphertext could be guessed from operator habit (The National Museum of Computing, "The Turing-Welchman Bombe"). The load-bearing constraint here was Enigma's one mechanical guarantee: it could never encipher a letter as itself, a fact that let cribs be positioned and logical contradictions be chased electrically through loops of wiring until only one candidate rotor setting survived. Two generations of solvers, a few years apart, had each found the single condition — a doubled indicator, then a letter that could never map to itself — that made an otherwise astronomical search collapse to a morning's work.

The move: the machine was secret, but its procedure leaked a theorem.

The Mailbox Built From What Was Already Aboard — Apollo 13, 1970

On April 13, 1970, an oxygen tank exploded aboard Apollo 13, roughly 200,000 miles from Earth, crippling the command module and forcing all three astronauts into the lunar module — a craft built as a two-man, two-day lifeboat, now asked to keep three men alive for four. Within days a slower, quieter emergency was building alongside the obvious ones: carbon dioxide. The lunar module's lithium hydroxide canisters, which scrubbed CO2 from the cabin air, were designed for two occupants and were running out; the command module carried plenty of spare canisters, but they were square, built for a different air-scrubbing system, and the lunar module's receptacles were round (Space Center Houston, "Apollo 13 Infographic: How did they make that CO2 scrubber?").

The constraint that mattered was not chemistry — lithium hydroxide would happily absorb CO2 in a square or round housing alike — it was geometry, and one further condition sat underneath it, tighter than the shape of a filter: the solution had to be built entirely from objects already on board a spacecraft in flight. Nothing could be sent up. A NASA engineer named Ed Smylie, who led the crew systems team, gathered a small group and imposed exactly that rule, then had them lay out, on a table, only what the astronauts had within reach: plastic bags meant for spacesuit liners, cardboard ripped from the covers of a flight plan, a roll of gray duct tape, and the hose from a spacesuit (The Washington Post, "Ed Smylie, engineer who helped save Apollo 13 crew with duct tape, dies"). Duct tape, notably, was the one material that could seal a joint reliably in zero gravity by unskilled hands under stress — which made it, on this flight, a load-bearing engineering material rather than a household convenience.

The team built and tested a working adapter — soon nicknamed "the mailbox" — on the ground in under a day, then radioed step-by-step assembly instructions up to the crew, who built their own copy from identical spare parts already stowed aboard (NASA, "Apollo 13 Lunar Module 'Mail Box'"). Cabin CO2 partial pressure, which mission rules treated as safe below about 8 mm Hg and dangerous approaching 15, had been climbing toward that danger zone with no working scrubber; once the crew wired the mailbox into the lunar module's cabin fan, the reading fell and held below 2 mm Hg for the rest of the flight home (Universe Today, "13 MORE Things That Saved Apollo 13, part 5: The CO2 Partial Pressure Sensor"). What makes this a reframe rather than a repair is the discipline mission control imposed on itself before any design work started: every other spaceflight problem admits, in principle, an answer involving more material, more mass, more redundancy sent up in advance. This one did not. By declaring the parts list closed before the design began, Houston turned an open-ended engineering problem — build a working scrubber — into a narrow, solvable one: build a working scrubber from this exact list of thirteen items, a constraint tight enough that a handful of engineers could exhaust the possibilities in an afternoon.

The move: nothing could go up, so nothing except what was already there was allowed to count.

A Clamp, a Glass of Ice Water, and a Committee — Washington, 1986

The Space Shuttle Challenger broke apart 73 seconds after launch on January 28, 1986, killing all seven crew members, and the presidential commission convened to investigate — chaired by former Secretary of State William Rogers — spent weeks accumulating testimony, telemetry, and competing engineering theories about what had failed. The physicist Richard Feynman, appointed to the commission somewhat reluctantly, grew frustrated with a process he felt was drowning a simple mechanical question in procedural caution.

The question Feynman wanted answered was narrow: did the rubber O-rings sealing the joints of the solid rocket boosters lose their resiliency — their ability to spring back into shape after being compressed — at the unusually cold temperature on the morning of the launch, around 36°F (2°C)? He arrived at a public hearing on February 11, 1986, with a piece of O-ring material, a small C-clamp bought that morning at a hardware store, and a glass of ice water requested from the hearing staff. During testimony from a NASA engineer, Feynman clamped the sample, submerged it in the ice water for a few minutes, then released the clamp on camera: the rubber stayed compressed instead of springing back, demonstrating in under a minute, in front of live television cameras, that cold rubber loses exactly the elastic property the seal depended on (Caltech, Feynman's personal appendix to the Rogers Commission report, "F.: An Additional Comment"; NASA, "Report of the Presidential Commission on the Space Shuttle Challenger Accident," Volume IV).

The demonstration is often told as a lone genius's improvisation, and Feynman's showmanship was certainly his own. But the tip that let him aim so precisely came from elsewhere, and the commission's own members kept that quiet for over two decades. Astronaut Sally Ride, also a commission member, had been shown internal NASA data linking O-ring damage to low launch temperatures by a source she wanted to protect; rather than raise it herself, she passed the chart to Air Force general Donald Kutyna, another commissioner. Kutyna, in turn, invited Feynman to his garage, showed him a leaking carburetor on a car he was restoring, and asked, apparently offhand, whether cold might affect an O-ring's resilience the same way — a question that pointed Feynman directly at the joint's temperature sensitivity without ever mentioning Ride's data (Britannica, "Sally Ride"). Only after Ride's death in 2012 did Kutyna disclose her role; Feynman, who died in 1988, may never have known exactly whose evidence had reached him. The load-bearing constraint of the entire disaster — a rubber seal's resiliency threshold, crossed by a specific, knowable number of degrees — existed in NASA's own data well before the launch. What the ice-water demonstration did was compress a multi-volume technical argument that risked disappearing into bureaucratic hedging into thirty seconds that a television audience, a Congress, and a grieving public could see and understand without a single equation.

The move: the failure had one number attached to it, and a glass of ice water made the number visible.

Seven Years in an Attic — Cambridge and Princeton, 1986–1995

Fermat's Last Theorem — that no three positive integers a, b, c can satisfy aⁿ + bⁿ = cⁿ for any integer n greater than 2 — had resisted proof for over 350 years since Pierre de Fermat scrawled his famous, almost certainly mistaken claim to have found "a truly marvelous proof" in the margin of a book in 1637. By the twentieth century it had become less a live research problem than a graveyard of failed attempts, the kind of statement mathematicians warned graduate students away from.

In 1986, mathematicians Gerhard Frey, Jean-Pierre Serre, and Ken Ribet established a bridge that changed the problem's shape entirely. They showed that if the Taniyama–Shimura conjecture — a deep statement claiming that every elliptic curve over the rational numbers is secretly connected to a highly structured object called a modular form — were true for a specific restricted class of curves (semistable curves), then Fermat's Last Theorem would follow as a corollary. In other words: stop trying to prove Fermat's statement about three numbers and an exponent. Prove instead that a hypothetical counterexample to Fermat would generate an elliptic curve so strange it could not possibly be modular — and prove separately that no such non-modular curve can exist. Andrew Wiles, a number theorist at Princeton who had been fascinated by Fermat's Last Theorem since childhood, recognized that this reframe had, for the first time, connected the old problem to the living, well-resourced center of modern number theory. He set out in near-total secrecy, working essentially alone in the attic study of his house for seven years, to prove the modularity conjecture for semistable curves — enough to make Fermat fall out for free (Wikipedia, "Wiles's proof of Fermat's Last Theorem"). He told almost no one what he was working on — not department colleagues, not most collaborators, only his wife Nada, whom he later said "has only known me while I was working on Fermat" — periodically surfacing old, unrelated results to keep up the appearance of ordinary output while the real work stayed hidden (Nautilus, "How Math's Most Famous Proof Nearly Broke").

Wiles announced a complete proof across three lectures at Cambridge's Isaac Newton Institute in June 1993 — reportedly finishing the final line of the argument in the closing minutes of the final lecture, to a stunned silence and then a standing ovation from an audience that had gradually realized, over three days, what it was witnessing — and the announcement drew global press coverage, with the New York Times putting it on the front page. Then, during the routine but rigorous process of peer review, a referee named Nick Katz found a gap: a step in the argument's construction of an "Euler system," a technical device load-bearing enough that without it the whole edifice risked being merely a very good sketch rather than a proof (American Mathematical Society, Faltings, "The Proof of Fermat's Last Theorem by R. Taylor and A. Wiles," Notices of the AMS, 1995). For over a year, Wiles worked to repair the gap, at first alone and then, from summer 1994, with his former student Richard Taylor. On September 19, 1994, examining precisely the approach he had abandoned years earlier for a different reason, Wiles saw why it had failed before and why, combined with the Euler-system machinery, it now succeeded — closing the gap with a method distinct enough from the original that Taylor and Wiles published it as a companion paper. The corrected proof appeared in 1995 (Wikipedia).

The gap is not an embarrassing footnote to this story; it is the story's clearest illustration of the book's larger claim. Wiles did not find the short path by avoiding error — he found it by refusing to let one uncorrected error force him back to the old problem space (attacking Fermat's equation directly), when the error was local to one technical tool inside the new one (the Euler system within the modularity bridge). The load-bearing constraint of the whole seven-year project was never Fermat's equation. It was always the bridge — and a bridge with one weak plank is repaired, not abandoned.

The move: don't lift the boulder; prove the bridge that makes the boulder irrelevant.

A Fifty-Year Problem Closed by Ignoring Physics — CASP14, 2020

Predicting the three-dimensional shape a protein folds into, given only the one-dimensional sequence of amino acids that builds it, had stood as one of biology's grand open problems since the molecular biologist Cyrus Levinthal pointed out in 1969 that a protein could not possibly be finding its shape by randomly trying configurations — the number of possible foldings for even a modest protein chain is so large that exhaustive search would take vastly longer than the age of the universe, yet real proteins fold correctly in milliseconds (Wikipedia, "Levinthal's paradox"). Since 1994, the biennial Critical Assessment of protein Structure Prediction (CASP) competition had measured the field's slow, incremental progress against experimentally solved structures kept secret from competitors until scoring (CASP history, cited via DeepMind, "AlphaFold: a solution to a 50-year-old grand challenge in biology").

Most serious attempts at the problem tried, in one form or another, to simulate the underlying physics: model the forces between atoms and let a folding trajectory play out computationally, an approach that is faithful to how nature actually does it but computationally so expensive that it remained impractical for anything but small proteins over short timescales. DeepMind's AlphaFold team, entering CASP13 in 2018 and then rebuilding the system almost entirely for CASP14 in 2020, made a different bet: rather than simulate the physical process of folding, treat structure prediction as a problem of completing a pattern buried in evolutionary history. A protein family that has been conserved across millions of years of evolution carries statistical fingerprints in its sequence variations — when one amino acid position mutates, a second, often physically distant position frequently mutates to compensate, because the two must remain close together in the folded, functional structure for the protein to keep working. AlphaFold's neural network learned to read those co-evolutionary correlations out of large alignments of related sequences and infer three-dimensional distances between amino acids directly, without ever simulating a single physical folding step (accounts of the co-evolutionary approach summarized in reviews of AlphaFold's method, e.g. PMC, "Chasing long-range evolutionary couplings in the AlphaFold era").

The result, announced at CASP14 on November 30, 2020, startled a field accustomed to grinding, decade-scale progress: AlphaFold achieved a median score of about 92.4 on the Global Distance Test across the competition's target proteins, a score competitive with the resolution of experimental methods like X-ray crystallography, and it did so consistently across proteins the model had never seen (C&EN, "Baker, Hassabis, and Jumper win 2024 Nobel Prize in Chemistry"). John Moult, the computational biologist who co-founded CASP in 1994 specifically to measure progress on this problem, told Nature at the time: "This is a big deal... In some sense the problem is solved" — a verdict CASP's own founder is entitled to render, and one he still qualified with "in some sense" (Nature, Callaway, "'It will change everything': DeepMind's AI makes gigantic leap in solving protein structures," 30 November 2020). In October 2024, the Nobel Prize in Chemistry recognized DeepMind's Demis Hassabis and John Jumper for AlphaFold, sharing the prize with University of Washington biochemist David Baker for related work in computational protein design (Nature, "Chemistry Nobel goes to developers of AlphaFold AI that predicts protein structures").

The honest boundary of the achievement matters as much as its size. AlphaFold predicts a static structure — the shape a protein settles into — not the pathway or timing by which it gets there, and it struggles with proteins that adopt multiple functional shapes, with the effects of mutations and binding partners on structure, and with the roughly one-third of the human proteome made up of intrinsically disordered regions that never settle into one fixed shape at all (PubMed, "AlphaFold and Protein Folding: Not Dead Yet! The Frontier Is Conformational Ensembles"). Levinthal's paradox — how folding happens so fast — remains genuinely unsolved; AlphaFold answered a narrower, extremely valuable question (what shape does this sequence produce) by declining to answer the dynamical one at all.

The move: stop simulating how the protein gets there; read the shape out of a hundred million years of relatives.

The Move

Move 08 — Find the load-bearing constraint: the one condition that, honored or broken, decides everything.

Every problem this large contains hundreds of true facts and dozens of plausible-looking obstacles. Almost none of them matter equally. Each solver in this gallery found the single condition that the rest of the problem actually hinged on, and treated everything else as background.

• Ask which fact, if it changed, would collapse the whole difficulty. Enigma's daily-key search is astronomically large — until a doubled indicator, or a letter that can never encipher to itself, turns it into an afternoon's logical deduction. Look for the one rule the system cannot break.
• Separate the dramatic obstacle from the deciding one. Apollo 13's crisis produced a dozen simultaneous emergencies; the CO2 problem was solved by fixing the parts list, not the chemistry. Challenger's failure had a hundred contributing organizational causes; the physical cause was one seal's behavior below one temperature.
• When an old attack keeps failing, ask whether you are fighting the real constraint or a decoy. Three and a half centuries of failed attempts hit Fermat's equation directly. Frey, Serre, and Ribet found the actual load-bearing structure one layer beneath it — modularity — and Wiles spent his effort there instead.
• A found constraint survives its own repair. Wiles's 1993 gap did not mean the reframe was wrong; it meant one plank of the bridge needed replacing. Distinguish a flaw in your central insight from a flaw in a technical tool serving it — the first sends you back to the old problem, the second just costs a year.

Next — Chapter 09 · The Aesthetics of Solution: why the short path feels the way it does — beauty as the signal of compression, argued through Erdős's Book, Kolmogorov complexity, and the neuroscience that can now watch an insight arrive.