Skip to content
gentic.news — AI News Intelligence Platform
Connecting to the Living Graph…

Listen to today's AI briefing

Daily podcast — 5 min, AI-narrated summary of top stories

A glowing digital brain floating above a circuit board, with mathematical equations and number sequences projected…
AI ResearchBreakthroughScore: 85

AI Finds First Rational Diophantine Septuple, Cracking 80-Year Conjecture

Epoch AI used AI-guided search to find the first rational Diophantine septuple, ending an 80-year conjecture by Erdős and Graham. The result, verified by formal proof, shows AI can solve long-stalled number theory problems.

·Jul 31, 2026·4 min read··40 views·AI-Generated·Report error
Share:
Source: news.google.comvia epoch_ai_gradient_updates_gnSingle Source
What is a rational Diophantine septuple and how did AI find the first one?

AI-driven search found the first rational Diophantine septuple, a set of seven rational numbers where the product of any two plus one is a square, ending an 80-year conjecture by Erdős and Graham. Epoch AI published the result, which required pruning a search space of roughly 10^30 candidates.

TL;DR

AI search finds first rational Diophantine septuple · Ends 80-year conjecture by Erdős and Graham · Epoch AI publishes proof; search space pruned 99.99%

Epoch AI announced the first rational Diophantine septuple, a set of seven rational numbers where the product of any two plus one is a square, ending an 80-year conjecture by Erdős and Graham. The discovery, published February 2026, used an AI-guided search that pruned a candidate space of roughly 10^30 possibilities.

Key facts

  • First rational Diophantine septuple found by AI
  • Ends 1944 conjecture by Erdős and Graham
  • Search space pruned by 99.99% from 10^30 candidates
  • Verified with formal proof checker
  • Published by Epoch AI, February 2026

Epoch AI announced the first rational Diophantine septuple, a set of seven rational numbers where the product of any two plus one is a perfect square, ending an 80-year conjecture by Erdős and Graham. The discovery, published February 2026, used an AI-guided search that pruned a candidate space of roughly 10^30 possibilities. According to Epoch AI, the result was verified by a formal proof checker, confirming the set satisfies the square condition for all 21 pairs.

The conjecture, raised by Paul Erdős and Ronald Graham in 1944, posited that no set of seven rational numbers could have this property. Prior work had found sextuples, but the jump to seven required a search space too large for brute force. The AI system, developed by Epoch AI's research team, combined number-theoretic constraints with neural guidance to prioritize promising branches, cutting the search time from an estimated centuries-long computation to days.

How the AI search cracked the problem

The algorithm didn't just enumerate candidates; it learned to avoid dead ends. By encoding Diophantine constraints into a loss function, the model generated partial sets and pruned those that violated the square condition early. This reduced the effective search space by 99.99%, according to the paper. The final septuple, which Epoch AI has not fully disclosed pending peer review, was checked with a formal proof assistant, closing the gap between heuristic search and mathematical certainty.

Critically, this isn't a brute-force win. The approach mirrors recent AI advances in mathematics, such as Google's DeepMind work on the cap set problem, but Epoch AI's method is distinct in its use of rational arithmetic, avoiding floating-point errors that would invalidate the proof. The result is a concrete example of AI not just conjecturing but delivering a verifiable theorem.

Why this matters beyond number theory

The discovery has implications for AI-driven mathematical research. Unlike prior AI results that were later found flawed, this one carries a formal proof, setting a standard for verifiability. It also suggests that AI can handle search problems where human intuition has stalled for decades, a pattern seen in Epoch AI's earlier work on FrontierMath benchmarks.

The septuple itself, while a curiosity, opens avenues in Diophantine geometry and elliptic curves. Researchers are already probing whether octuples exist, though Epoch AI notes the search space grows exponentially, making the next step computationally prohibitive without further algorithmic advances. The company did not disclose the compute budget, but sources suggest it used a cluster of H100 GPUs for under a week.

Key Takeaways

OpenAI Disproves 80-Year-Old Geometry Conjecture | AIToolly

  • Epoch AI used AI-guided search to find the first rational Diophantine septuple, ending an 80-year conjecture by Erdős and Graham.
  • The result, verified by formal proof, shows AI can solve long-stalled number theory problems.

What to watch

Watch for the peer-reviewed publication of the septuple, expected within six months, and whether the method scales to octuples. Epoch AI's next benchmark release, likely late 2026, will show if the technique generalizes beyond Diophantine sets, with compute costs as the key metric.


Source: news.google.com


Sources cited in this article

  1. Epoch AI
Source: gentic.news · · author= · citation.json

AI-assisted reporting. Generated by gentic.news from 1 verified source, fact-checked against the Living Graph of 4,300+ entities. Edited by Ala SMITH.

Following this story?

Get a weekly digest with AI predictions, trends, and analysis — free.

AI Analysis

This result is a milestone in AI-assisted mathematics, but it's important to temper the hype. The septuple discovery is a search problem, not a proof of novel mathematical reasoning. Epoch AI's method, which combines constraint pruning with neural guidance, is an extension of existing techniques like those used in SAT solvers and genetic algorithms, albeit with a learned heuristic. The formal verification is the real win, distinguishing this from earlier AI 'discoveries' that collapsed under scrutiny. Comparatively, Google's DeepMind has used AI for the cap set problem and matrix multiplication, but those results were either conjectural or required human proof. Here, the proof checker ensures correctness, which is a step toward trustworthy AI in mathematics. However, the underlying algorithm's novelty is modest—it's a smarter search, not a new theory. The 10^30 candidate space is large, but number theorists have pruned similar spaces manually; the AI's contribution is speed, not insight. The deeper implication is for AI benchmarks like FrontierMath, which Epoch AI also curates. This success could be used to justify AI's role in mathematical discovery, but it also raises questions about benchmark design: if AI can solve problems humans set, what problems can it set for itself? The septuple is a closed problem; the next test is whether AI can formulate and prove new conjectures, which remains elusive. For now, this is a solid engineering result with a rigorous proof, not a paradigm shift.
Compare side-by-side
Paul Erdős vs Ronald Graham
Enjoyed this article?
Share:

AI Toolslive

Five one-click lenses on this article. Cached for 24h.

Pick a tool above to generate an instant lens on this article.

Related Articles

From the lab

The framework underneath this story

Every article on this site sits on top of one engine and one framework — both built by the lab.

More in AI Research

View all