DeepMind researcher asks what comes after Riemann — number theorists have no answer
Amid rumors that the Riemann Hypothesis may have been proven, DeepMind researcher Geoffrey Irving posted a structural observation: for P≠NP, mathematics has a rich theory of "natural stronger conjectures" (such as the various variants of Circuit-SETH), whereas the Riemann Hypothesis has no corresponding body of follow-up conjectures. He asked several number theorists "what is the stronger conjecture after Riemann," and the answers were essentially "we don't know."
Confirmed
- Irving noted that the Riemann Hypothesis only yields a bound of O(√n log n) for the largest prime gaps, whereas under the random model of primes in Cramér's conjecture, the gaps should be O(log² n)—one of the most直观 examples of "what we know being far weaker than what actually holds." (one of the most直观 examples of what we know being far weaker than what actually holds)
- He mentioned the existence of Montgomery pair correlation conjectures (the spacing distribution of zeta zeros correlated with eigenvalues of random Hermitian matrices), but this does not constitute a systematic "post-RH" theory.
- In cryptography, hardness conjectures are stated in terms of exact cycle counts, such as "no SHA-256 collision can be found below 2^128 cycles," belonging to the same tradition of precise, computation-cycle-centered formulations as the prime gap conjectures.
- He analyzed two reasons for the gap: first, the Riemann Hypothesis is more "canonical" than P≠NP, making it harder to look past it toward stronger statements; second, number theorists long viewed RH as unreachably far off—"envisioning what comes after a proof belonged to the future"—so the community never seriously planned a research program for a post-RH world.
- He remarked that "it turns out the future was November 21, 2026," and wondered whether a proof of RH would spur the field to develop such a theory.
Why it matters
- This discussion steers public attention from "has RH been proven" toward a deeper issue: mathematics lacks a ladder of conjectures beyond RH, meaning theoretical construction after a proof would start nearly from scratch.
- Contrasted with the mature "stronger conjecture" traditions in computational complexity and cryptography, this highlights a planning vacuum in this direction of number theory, and also suggests that if RH is proven, it could open an entirely new era of conjecture-building.
2026-11-21 ~ 2026-11-21 · 8 related posts
Primary sources
- P≠NP has rich stronger-conjecture theories; Riemann has none — will one emerge? — geoffreyirving · 2026-11-21
- Prime gap comparison shows Riemann is far weaker than Cramér's O(log²n) prediction — geoffreyirving · 2026-11-21
- Crypto hardness conjectures are phrased in sharp cycle counts — geoffreyirving · 2026-11-21
- [source] DeepMind's Irving: number theorists can't name the conjecture beyond Riemann — geoffreyirving · 2026-11-21
- Irving: Riemann felt so out of reach no one planned what comes after — geoffreyirving · 2026-11-21
- "The future is 21 November 2026": reimagining math after Riemann falls — geoffreyirving · 2026-11-21
- [source] AI researcher proposes harder test: no O(n²) randomized algorithm for n-bit primes conjecture — geoffreyirving · 2026-11-21
- With Riemann Done, DeepMind's Irving Proposes Harder Prime-Generation Conjecture for AI — geoffreyirving · 2026-11-21